|
|
|
 |
Stammering
Research
A
Journal Published by the
British Stammering Association
Editor
Peter Howell
Email p.howell@ucl.ac.uk
Associate Editors
|
|
Dr Robin Lickley Speech and Language Sciences Faculty of Health and Social
Sciences 0131 317 3686 Email Rlickley@QMUC.ac.uk |
|
Stammering
Research
A
Journal Published by the
British Stammering Association
Volume
1, Issue 4, January
2005
Contents
|
|
|
Pages |
|
|
Contents of volume one, issue four |
330 |
|
|
|
EDITORIAL |
|
|
|
|
PETER HOWELL Editorial for Stammering Research |
331-332 |
|
|
|
|
TARGET ARTICLE ADRIAN DAVIS & PETER HOWELL Elements of statistical treatment
of speech and hearing science data |
333-343 |
|
|
|
TARGET
ARTICLE
S. Z. LEVINE, K. V. PERTRIDES,
|
344-363 |
|
|
|
RESEARCH DATA, SOFTWARE AND
ANALYSIS SECTION REPORTS OF ANALYSES OF UCLASS DATA PETER HOWELL The effect of using time intervals
of different length on judgements about stuttering |
364-374 |
|
|
|
CHLOE
The impact of word-end phonology and morphology on
stuttering
|
375-391 |
|
|
|
Indexes to
Volume 1, 2004-2005 |
392-403 |
|
Notice
The British
Stammering Association is a UK-based charity which seeks to promote
understanding into the causes, treatment and understanding of
stammering. Its
activities include research into stammering which it supports through
its
vacation studentship scheme (http://www.stammering.org/research_schol.html) and the publication of
Stammering Research
(provided free of charge to all-comers).
Stammering
Research is intended to promote public understanding of high quality
scientific
research into stammering and allied areas
If individuals
wish to make a donation to support either of these initiatives, they
should
forward a cheque (payable to the British Stammering Association) to The
British
Stammering Association, 15 Old Ford Road, London E2 9PJ, or call the
BSA on 020
8983 1003 (+44 20 8983 1003 from abroad) with their credit card
details. If
they wish this to be used specifically for either the vacation
studentship
scheme or Stammering Research, they should mark it accordingly on the
back of
the cheque. For information on tax-effective ways to support the
charity’s
research activities, please go to http://www.stammering.org/donations.html.
Donors will be
listed in the last issue of the appropriate volume of the journal
unless they
indicate otherwise. Companies wishing to make a donation or who wish to
make
enquiries about advertising in Stammering Research should address
correspondence to Norbert Lieckfeldt at nl@stammering.org.
‘Stammering
Research’.
An on-line
journal published by the British Stammering Association
ISSN 1742-5867
Description
Stammering Research is an
international journal published
in electronic format. Currently it appears as four quarterly issues per
volume
(officially published March 31st, June 30th,
September 30th and
December 31st). The first issue of volume one
appeared on
Editor
Peter Howell
Department of Psychology
SUMMARY OF
STEP BY STEP PROCEDURE FOR AUTHORING AN ARTICLE
TO STAMMERING RESEARCH
1. Contact the editor with a
brief outline of the proposed
article. The editor and other board members make initial decisions only
as to
the suitability of the general area proposed. The primary function in
this step
is to ensure the topic is of sufficiently broad interest for, and
within the
remit of, the readership of Stammering Research. The intent behind this
initial
contact is to ensure authors do not spend time preparing articles on
unsuitable
topics. Review, empirical and theoretical work are all appropriate.
Authors
will be informed whether the judgement is that the proposed topic has a
suitable, or too narrow, a focus.
Indication that the scope is too narrow does not
imply anything about
the scientific standard of the proposed work. Neither does notification
that a
topic is suitable indicate that the submitted work will necessarily be
accepted
for publication (all submitted material has to go through the normal
processes
of peer review). Articles should be submitted by email to p.howell@ucl.ac.uk.
2.
Submitted
articles are peer reviewed in the normal way and an indication as to
suitability of publication or not (possibly after revision) is notified
to the
author by the editor.
3. After an article has been
accepted, the author cannot
change the article. It is then made available for open peer commentary.
Details
how the accepted article can be accessed are posted on the British
Stammering
Association’s website (www.stammering.org). Indications that the
article is
available for access are posted on http://www.mnsu.edu/dept/comdis/kuster/Internet/Listserv.html for ASHA members, the
British
Stammering Association’s website (http://www.stammering.org, the stut-l list
(stutt-l@listmail.
temple.edu), the stutt-x mailing list (stutt-x@asu.edu), and on the
stuttering
home page (www.stutteringhomepage.com). The primary function in posting
details
about access available to an accepted article, is to alert potential
commentators. A list of commentators is being drawn up and individuals
are
encouraged to submit their nominations (for themselves or others).
4. See the next page for
precise details how to prepare a
commentary and the timetable allowed for this. When preparing a
commentary,
authors might find it helpful to consult a recent issue of Stammering
Research
to see the range of comments that are appropriate, the style and format
of
commentary
5. All accepted commentaries
are available to the author
of a target article from receipt until two weeks after invitations for
commentaries has closed. In this time, the author can prepare a
response to
commentaries. The response will be peer-reviewed by the editorial
board.
Further details are given on the next page and authors should again
consult a
recent issue of Stammering Research to see the sorts of comments that
are
appropriate, style and formatting of a submission.
6. On completion of this
process, the target article,
commentaries and response to commentaries will be published together in
Stammering Research. Authors are responsible for preparing their
articles
according to the stipulated format. The current and previous issues of
the
journal are available as PDF files at
http://www.speech.psychol.ucl.ac.uk/.
Notes about
commentaries for Stammering Research
ISSN 1742-5867
Once a manuscript has been accepted as a target article, the authors cannot change it. The manuscript needs to be available for commentary before it is officially published so that commentaries and the author’s responses can appear simultaneously.
Manuscripts are posted for commentary on http://www.psychol.ucl.ac.uk/ under Stammering Research. Commentators are alerted as indicated on the previous page.
Manuscripts will be available for peer commentary for six weeks. Commentaries have to reach the editor, or associate editor, responsible for the article within that time (late submissions will not be accepted). Commentaries should ordinarily not exceed a total (including references and other material) of 1,000 words. The commentaries have to conform to APA style conventions.
In order to appear in the same issue as the target article, commentaries should be sent by email as soon as possible within the period the article is open for peer commentary. The commentary should appear within the body of the email text (not as an attachment) and be sent to p.howell@ucl.ac.uk. Authors of target articles will receive commentaries as they are accepted and have two weeks from close of submission of commentaries to complete their responses. Commentaries that appear outside this timetable may appear as continuing commentaries in subsequent issues (these are considered in the same way as commentaries that appear at the same time as the target article.
Commentaries will be peer-reviewed and edited for style as well as content. Authors of commentaries need to establish the relevance of their submission to the target article at the outset, and preferably also show an awareness of the wider work of the target article’s author.
If there are several commentaries which raise the same point, the editorial board reserves the right to group them together and prepare them as a single coauthored commentary. In this (probably rare) eventuality, the authors will have the opportunity to see the manuscript and decide whether they wish to be included on the list of authors.
Editing and revision of commentaries will be completed within two weeks of close of submission. Revisions that are not satisfactorily completed in this period, or that are received late, may be published as continuing commentaries.
Formatting
Accepted Publications in Stammering Research
Peter
Howell1, John Smith2,
and John Doe3
1Department
of Psychology, University College London Gower St. , London WC1E 6BT England
P.Howell@ucl.ac.uk
2Stuttering
Treatment Clinic, Somewhere, Some Country
Smith@some.email.address
3For
private citizens, house number and
street, city and postcode/ZIPcode, Country
Doe@email.address.if.you.have.one
Abstract.
A
short abstract summarizing the significant content and contribution
of the paper should be included here. This page illustrates and
describes the
format for paper submissions. Authors are requested to adhere as
closely as
possible to this format once an article is accepted. The abstract
should be in
Times New Roman 9-point font, justified with left and right margins
indented 1
cm in from the margins of the main text. Keywords.
A few key terms that will be used by abstracting services to make sure
your
article reaches those who will be interested in reading what you say.
1.
Introduction
Articles and commentaries should be submitted for review in APA format. After an article or commentary is accepted, it needs to be prepared according to the journal format as indicated next. Articles and commentaries must be in Word format. An article will typically be up to 15,000 words. A commentary should preferably be up to 1,000 words. Authors may submit longer articles or commentaries for consideration but these may be reduced in length by the editor. Articles with fewer than 15,000 words and commentaries with fewer than 1,000 words are acceptable if the author can demonstrate sufficient content and contribution. Typically commentaries will have an abstract, usually only a single section in the text headed so as to identify the target article. If an author needs to use more than one section heading and diagrams or figures, then they should follow the same instructions as for preparation of a target article. Each page of an article should consist of single column, of single-spaced text in a 16cm x 24cm column using A4 or US Letter settings on your word processor as illustrated in Figures 1 and 2. Figures should be numbered consecutively and appear close to the text where they are mentioned.
|
|
|
|
Figure
1: First page format |
Figure
2: Subsequent page format |
2.
Detail of styles
The article or commentary title should be bold and centred using 14 point Times New Roman font. Authors’ names, affiliations and email details should be centred using 10 point Times New Roman font. The author's name and affiliation should be italicized. The main text and the bibliographical references must be left and right justified and single line spaced. The main text should be in 10 point Times New Roman font with numbered section headings in 11 point bold font.
All references should be cited using APA referencing styles. For example a publication which is referred to as support for a statement would be cited in the text this way (Howell & Sackin, 2002) whatever the number of authors. When an article is referred to directly in the text as in "… in the work of Howell and Sackin (2002) the …" only the year is placed in brackets. If there is more than one reference from the same authors in the same year then they are distinguished by using different letter designations after the year as in 1996a, 1996b etc. In the references below, examples are given of how a conference paper, a journal paper and a book would be listed. All references should be listed at the end of the paper using 9 point Times New Roman font.
All figures, and diagrams must be good quality black and white images suitable for readers to display and print. Colour illustrations or text can be used, but bear in mind readers who want to print articles may not have access to a colour printer. When an article is accepted, figures and pictures must be inserted in the word file in the exact position they will appear in the publication. Any format for figures, pictures and diagrams may be used provided they allow good quality reproduction for readers who wish to print off a copy.
References
Howell,
P. (2002). The EXPLAN theory
of fluency control applied to the treatment of stuttering by altered
feedback
and operant procedures. In E. Fava (Ed.), Pathology
and therapy of speech disorders (pp. 95 -118).
Howell,
P., & Sackin, S. (2002).
Timing interference to speech in altered listening conditions. Journal of the Acoustical Society of
Rosen,
S., & Howell, P.
(1991). Signals and Systems for Speech and Hearing.
Stammering
Research
A
Journal Published by the
British Stammering Association
Volume
1, Issue 4, January 2005
Contents
|
|
|
Pages |
|
|
Contents of volume one, issue four |
330 |
|
|
|
EDITORIAL |
|
|
|
|
PETER HOWELL Editorial for Stammering Research |
331-332 |
|
|
|
|
TARGET ARTICLE ADRIAN DAVIS & PETER HOWELL Elements of statistical treatment
of speech and hearing science data |
333-343 |
|
|
|
TARGET
ARTICLE
S. Z. LEVINE, K. V. PERTRIDES,
|
344-363 |
|
|
|
RESEARCH DATA, SOFTWARE AND
ANALYSIS SECTION REPORTS OF ANALYSES OF UCLASS DATA PETER HOWELL The effect of using time intervals
of different length on judgements about stuttering |
364-374 |
|
|
|
CHLOE
The impact of word-end phonology and morphology on
stuttering
|
375-391 |
|
|
|
Indexes to
Volume 1, 2004-2005 |
392-403 |
|
Editorial for Stammering
Research
This is the fourth and final issue of volume 1 of Stammering Research. It includes two target articles and a report of an analysis of some of the Howell and Huckvale (2004) data that were published previously in the journal. The two target articles both concern statistical issues relevant to research into stammering. The first article, by Adrian Davis and Peter Howell, is an introduction to some basic statistical issues in the area and sets the scene for future statistical articles. The second article gives background details about one statistical approach (Structural Equation Modeling, SEM) that can be used to assess alternative multivariate models of various aspects of stammering. Statements to the effect that stammering is a disorder that is determined by a multiplicity of factors abound. There are few, if any, models that specify the structural relations between the various factors in a way that they can be evaluated against alternative models. Simply acknowledging that many variables affect stammering behavior (as in the Demands and Capacities model) does not indicate whether the variables have a significant effect, how each variable interacts with others that the author would wish to include and so on. SEM is one technique that allows such models, once constructed to be evaluated.
Both these articles are published as target articles and, as such, commentaries on them will be reviewed on receipt. This allows me to air, once more, some changes the editorial board has made to submission procedures for articles and commentaries. Submissions should now be sent by email to p.howell@ucl.ac.uk (this applies to prospective target articles as well as commentaries) and commentaries on any article will be received, reviewed and, if accepted, published in issues other than the one in which the target article appeared. Formerly commentaries appeared in the same issue as the target article but from feedback I received, it appeared that some commentators would have appreciated more time. Personally, I think it is important for authors to see immediately the impact a target article has had. For this reason, I would urge individuals to submit commentaries promptly. A second issue concerns what type of commentary would be appropriate for a statistical tutorial? The introductory article by Davis and Howell probably will not attract commentaries because of the nature of the material. The SEM paper probably will raise comments that need addressing. To pick two important ones, first, SEM presupposes a step of theory construction. Hopefully, we will begin to see precisely specified theories (maybe with statistical evaluation) as submissions to Stammering Research. It would be particularly nice to see models of treatment outcome. Second, the approach put forward in the SEM article is only one approach to these issues. Again, following the philosophy behind Stammering Research, we would like to see authors presenting alternative approaches (as potential target articles if they are substantive enough) so that a comprehensive box of tools can be built up for investigations in this area. A more general point a propos publication of these articles, is that we hope to see Stammering Research as the forum for introducing, tutoring and discussing a wide range of statistical issues that are pertinent to research in this area.
In volume 1, issue 2 of Stammering Research, a data archive was released of spontaneous monologue speech from speakers who stutter covering a wide age-range in alternative formats appropriate for use with different free software packages (Howell & Huckvale, 2004). It is intended that this database (the UCL archive of stuttered speech, or UCLASS for short) be used for a range of different analyses. The Howell (2005) article that appears in this issue of Stammering Research, is provided as an example how to report one type of analysis that can be done with audio data. It is hoped that the international community will take up the prospects and new challenges that have been opened up by release of the UCLASS and other data (Howell, Davis, Bartrip & Wormald, 2004) through the pages of Stammering Research. A workshop conference has been arranged for June 27th 2005 in London which will include tutorials describing ways to deal with the UCLASS material described in Howell and Huckvale (2004) with talks by Rose and MacWhinney (on CHILDES) and Huckvale (on SFS). This has been timed to precede the Oxford Dysfluency Conference so people can attend both. Details of the conference can be found at http://www.speech.psychol.ucl.ac.uk/.
At the end of this first year involved in editing and publishing Stammering Research, I feel there have been significant advances in research opportunities available to the whole spectrum of the journal’s readers. The open peer commentary format is unique in this area of research and this journal provides the only forum I know of which releases software and data specifically intended to enhance and enable people’s ability to research into stammering. I thank all of those who have made this venture possible.
Peter
Howell, December 2004
References
Howell, P. (2005). The effect of using time intervals of different length for judgements about stuttering. Stammering Research, 1, 364-374.
Howell, P., Davis, S., Bartrip, J. & Wormald, L. (2004). Effectiveness of frequency shifted feedback at reducing disfluency for linguistically easy, and difficult, sections of speech (original audio recordings included). Stammering Research, 1, 309-315.
Howell, P. & Huckvale, M. (2004) Facilities to assist people to research into stammered speech. Stammering Research, 1, 130-242.
TARGET
ARTICLE
Elements of
statistical treatment of speech and hearing science data
1Director
NHS Newborn Hearing Screening Programme
MRC
Hearing and Communication Group
adrian@mrchear.man.ac.uk
2 Department of Psychology, Centre for Human
Communications, Institute
of Cognitive Neuroscience, and Institute of Movement Neuroscience,
University
College London, Gower St., London WC1E 6BT, England
p.howell@ucl.ac.uk
Abstract: Many of the statistical issues
involved in speech and
hearing research are shared with other areas of medicine. This article
is the
first in a series intended to stimulate examination of research data in
speech
and hearing areas using a wide variety of techniques. This article
specifically
deals with two essential, but elementary, issues. The first is
concerned with
experimental design and choice of test data. The second, defines and
explains
statistical terms, concentrating particularly on
the inference to the population mean from the
sample mean.
Keywords: Statistics, experimental design,
choice of data,
inferential statistics.
1. Background
This article is the first in
a series that will discuss statistical treatment of data from studies
into
communication and its disorders. Application areas include where data
from
people with speaking or hearing disorders are available and researchers
want to
visualize or quantify how performance of these individuals relate to
fluent
populations. The range of topics that can be covered is massive
starting with simple
ones like how to display and make summary statements about data,
through to
highly detailed technical ones like implementation and assessment of
causal
models of the problems. This article is at the elementary end (the
appropriate
place to start) and introduces a) issues concerned with experimental
design and
choice of test data, b) definition of statistical terms involved with
inference
of the sample mean from the population mean. We start by outlining our
motivation in initiating this series of articles and then emphasize how
this
article is different from a textbook on the subject.
Our experience in teaching
and supervising projects with some trainee professionals who will
deliver
speech and hearing therapy to clients has shown that difficulties are
experienced with statistics. It is our belief that part of the reason
for this
is that the texts and approaches often appear to be remote from the
concerns of
the students. In particular, texts for teaching students are often from
other
application areas about the study of behaviour (mainly psychology)
that, though
closely allied to speech and hearing science, are not directly
applicable. When
we have used examples from the speech and hearing sciences in our
teaching, we
have found that students seem able to access the material more readily.
Even
though we have limited ourselves to elementary statistics, the
treatment is not
comprehensive. There are two major omissions from these ‘elements’. The
first
concerns depiction of data. The second is statistical treatment when
only a
single case is available. It is hoped that both these omissions will be
the
subject of a future article.
The
article is not a précis of a statistics
course. There are many good such texts and other resources around (see
the
bibliography). These books mostly contain detailed information about
how to
describe data (only touched on here), hypothesis testing and cover the
logic
and mathematics behind different tests. The article should be used as a
gateway
to use one of the statistical techniques that give details of specific
tests
and procedures (though a couple of tests are considered to convey some
important details). We do not mean to imply that all available
textbooks lack
such an overview. However, as textbooks are more extensive than this
brief
article, there is a tendency for this information to be distributed
widely
making it hard for students to relate it together. We envisage students
reading
this article and, once they have understood the material, being able to
use it
to access the appropriate information in other texts.
2.
Experimental design and
choice of data
When talking about procedural
considerations in statistical analysis, it may help to make things
concrete from
the start by using an example drawn from hearing science. This is
intended to
introduce some concepts which are needed both for a researcher to
evaluate
studies and to allow an experimenter to do their own statistics. Let us
assume
that a company has developed a hearing aid (Aid A). It is to be
employed in a
country where all the inhabitants who have a hearing impairment might
want to
use it. The company wants some idea about how its performance compares
with
another aid on the market (Aid X). The company calls on a hearing
clinic to
evaluate the device and the clinic decides to assess it using read
texts that
can be carefully controlled, even though the aid will eventually have
to
operate with spontaneous speech to be useful for clients. Some of the
questions
the assessment team commissioned to do the work may decide to address
are:
These points highlight some of the
statistical analysis issues that feature in everyday clinical
decisions.
Similar questions would arise for a speech therapist when considering
whether
some new treatment they have learned about produces improvement in
speech
control so they can decide whether it is worth changing from the
current
procedure they deliver. In each case, the speech or hearing
professional may or
may not conduct the research themselves. When they rely on someone
else’s
published report about the speech or hearing procedure, they need to
know
whether the analysis was conducted properly and how to interpret the
results.
Moreover, the specific questions raised, though pertaining to a
particular
issue of concern, are
illustrative of many similar problems that clinicians
encounter. Now we will set about attempting to provide answers to these
(and
other) questions.
Statistical
and
experimental procedures for analyzing data
In
the first part of this section, some fundamental ideas in statistics
will be
illustrated through selected examples drawn from the speech and hearing
sciences.
Statistics is the acquisition, handling,
presentation and interpretation of numerical data. Speech and hearing
scientists have considerable experience acquiring, handling, presenting
and
interpreting communication data.
Populations, samples
and other terminology
A population
is usually defined as the collection of units to which inference (from
the
sample) is desired (the units may be people, phonemes etc.). In the
earlier
example, all hearing impaired individuals in the country are the
population.
Here everyday use of the term 'population' corresponds with its use in
a
statistical sense. Though population in a statistical sense can have
the same
meaning as the geographical sense, it need not be the case. Thus, for
instance
the population of users of a hearing aid specifically developed for
presbyacusis would only comprise individuals with this disorder.
Population
does not only refer to humans - for example, the population of /p/
phonemes of
a speaker would be all of the instances of that phoneme a speaker ever
produces.
A variable
ranges over numerical values associated with each unit of the
population.
Variables are classed as either independent
or dependent variables. An
independent (or, as some statisticians prefer, explanatory) variable is
one
that is controlled or manipulated by the researcher. So, for example,
when
setting up a test for a hearing aid or some treatment for a speech
problem, the
experimenter might consider it necessary to ensure that as many females
are
recorded in the test data as males. Sex would then be an independent
variable
(independent variables are also referred to as factors, particularly in
connection with the statistical technique Analysis
of Variance, ANOVA
discussed in a
later section). A dependent variable is a variable that the
investigator
measures to determine the effect of the independent variable.
When a variable is measured on all units
of a population, a full census has been taken. If it were always
possible to
obtain census data, there would be no need for inferential statistics.
However,
since most speech and hearing applications (and, indeed, in many other
aspects
that require measurement), involve very large or infinite populations
(such as
those illustrated earlier of speakers or phonemes), it is not possible
to
measure variables on all units: In these circumstances, a finite sample
is
taken. This sample is used to study the variable of concern in the
population.
So, if you wanted an idea of the average voice fundamental frequency of
men,
you might make measurements on a sample of 100 men. This sample is then
studied
as if it is representative of the population. The statistician is able
to
provide information about the relationship between variables measured
on the
sample (here its mean) and, what the investigator is really interested
in, the
mean voice fundamental frequency of the population.
Sampling
The main problem in treating data
statistically is how to ensure the reliability of information about the
population obtained from a sample. The main requirement to achieve this
is to
take a simple random sample. A sample is simple
random if every member of the population has the same chance
of being
selected as every other member. Thus, if some voice recognition
equipment is
needed for a research application, and a check is made on its
performance using
employees from the clinic, the sample would not be simple random: It is
unlikely that the employees from the clinic are from all social strata,
there
may be gender imbalances, and they would only include people of
working-age.
Biases
Selection of a sample that is not a simple
random sample is one of the main sources of bias in assessments. Bias
can be
defined as a systematic tendency to misrepresent the population. So, if
the
speech recognizer mentioned in the previous section is intended to be
used by
all members of the population, you cannot select an unbiased sample of
speakers
from a sample of people recorded just between
If you take a sample, how sure can you be
that if you measure variables such as the mean of the sample is close
to the
mean of the population? This sort of problem is termed estimation
and is considered in the following.
Estimating sample
means, proportions and
variances
Estimation is used for making decisions
about populations based on simple random samples. A truly random sample
is
likely to be representative of the population; this does not mean that
a
variable measured on a second sample taken will be the same as the
first. The
skill involved in estimating the value of a variable is to impose
conditions
which allow an acceptable degree of error in the estimate without being
so
conservative as to be useless in practice (an extreme case of the
latter would
be recommending a sample of the same order of magnitude as the
population). The
necessary background skill is to understand how quantities like sample
means,
proportions and variances are related to means, proportions and
variances in
the population. The following notation is used in the discussion: M is
the
sample mean, S is the sample standard deviation, and S2
is the
sample variance; μ, is the population mean, σ is the population
standard deviation, and σ2 is the population
variance. The
abbreviations sd and S.D. are
sometimes used for standard deviation; S.E. is used for standard error,
Z is
used for z-scores, >> pest
<< stands for estimated probability, and p
stands for proportion.
Estimating means
A fundamental step towards this goal is to
relate the sample statistic to a probability distribution: What this
means is:
if we repeatedly take samples from a population, how do the variables
measured
on the sample relate to those of the population? To translate this to
an empirical
example: How sure can you be about how close your sample mean lies to
the
population mean? Even more concretely, if we obtain the mean of a set
of
samples, how does the mean of a particular sample relate to the mean of
the
population? As has already been said, the value of the mean of the
first of two
samples is unlikely to be exactly the same as the second. However, if
repeated
samples are taken, the mean value of all the samples will cluster
around the
population mean; this is usually regarded as an unbiased estimator of
the
population mean.
The usual way this is shown in textbooks
is to take a known distribution (i.e., where the population mean is
known) and
then consider what the distribution would be like when samples of a
given size
are taken. So, if a population of events has equally likely outcomes
and the
variable values are 1, 2, 3 and 4, the mean would be 2.5. If all
possible
combinations are taken (1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4, 3
and 4),
the mean of the mean values for all pairs is also 2.5 (taking all pairs
is a
way of ensuring that the sample is simple random). An additional
important
finding is that if the distribution of sample means (the sampling
distribution)
are plotted as a histogram, the distribution is no longer rectangular
(rectangular because each option was assumed to occur with the same
frequency)
but has a peak at 2.5 (1 and 4, and 2 and 3 both have a mean of 2.5 and
none of
the means of the other pairs have the same value, thus, the peak at
2.5).
Moreover the distribution is symmetrical about the mean and
approximates more
to a normal (Gaussian) distribution even though the original
distribution was
not. As sample size gets larger the approximation to the normal
distribution
gets better. Moreover, this tendency applies to all distributions, not
just the
rectangular distribution considered. The tendency of large samples to
approximate the normal distribution is, in fact, a case of the Central Limit Theorem.
This particular result has far-reaching
implications when testing between alternative hypotheses (see below).
As a rule
of thumb sample sizes of 30 or greater are adequate to approximate the
mean of
a normal distribution (though this depends on the nature of the parent
distribution).
The statistical quantity standard
deviation (S) is a measure of
how a set of observations x1
– xn where n is the number of observations scatter
about the mean (x with the bar above
it). It is defined numerically as:
Later the related quantity of the variance
will be needed. This is simply the sd
squared:
An important aspect of the situation
described is that the sample means themselves (rather than the
observations)
have a standard deviation (sd). The sd of the sample means (here the sd of all samples of size two for the
rectangular distribution) is related to the sd
of the samples in the original distribution by the formula:
This quantity is given a particular name
to distinguish it from the sd - it
is
called the standard error (S.E.). In practice, the standard
deviation of the population is often not known. In these circumstances,
provided the sample is sufficiently large, the standard deviation of
the sample
can be used to approximate that of the population and the above formula
used to
calculate the S.E. The S.E. is used in the computation of another
quantity, the
z score of the sample mean:
The importance of this quantity is that
the measure can be translated into a probabilistic statement relating
the
sample and population means. Put another way, from the z
score, the probability of a sample mean being so far
from the
population mean can be computed.
To show how this is used in practice: if a
sample of size 200 is taken, what is the probability that the mean is
within
1.5 S.E.s of the population mean? Normal distribution tables give the
desired
area. Here is a section of a table giving the proportion of the area of
a
normal distribution associated with given values of z
(the stippled section in the figure indicates what area is tabulated)
where the mean is the line at the peak and the line to its right is 1.5
S.E.
away:
The sketch of the normal distribution is
symmetrical and the symmetry is about the mean value (i.e., the peak of
the
distribution). The z values above the mean are tabulated, and the row
with a z
value of 1.5 indicates that 0.4332 of the area on the right half of the
distribution lie within 1.5 S.E.s above the mean. Since it has already
been
noted that the distribution is symmetrical, 0.4332 of the area will lie
within
1.5 S.E.s below the mean. Thus, the area within 1.5 S.E.s above or
below the
mean is 0.4332 + 0.4332, or 0.8664. Thus, converted to percentages,
approximately 86.6 of all samples of size 200 will have means within
1.5 S.E.s
of the population mean. If, as in any real experiment, one sample is
taken, we
can assign a statement about how likely that sample is being within the
specified distance of the mean.
Another, related, use of S.E.s is in
stipulating confidence intervals. If you look at the areas associated
with
particular z values in the way just described, you should be able to
ascertain
that the area of a normal distribution enclosed within z
values ±1.96 S.E.s of the mean M is 95. Thus, if the S.E. and M of
a sample are known, you can specify a measurement interval that
indicates the
degree of confidence (here 95%) that the population mean will be within
these
bounds. This is between the value 1.96 times the S.E below the sample
mean and
1.96 times the S.E. above the sample mean. This particular interval is
called
the 95% confidence interval. Other levels of confidence can be adopted
by
obtaining the corresponding z
values.
Since this topic is so important, an
example is given: Say a random sample of mean voice fundamental of 64
male
university students has a mean of 98 Hz and a standard deviation of 32
Hz. What
is the 95% confidence interval for mean voice fundamental of the male
students
at this university? The maximum error of the estimate is approximated
(using
sample standard deviation S rather than that of the population σ as an
approximation, see above) as:
Thus, the 95% confidence interval is from
98 - 7.84 = 90.16 to 98 4- 7.84 = 105.84. Often, the confidence
intervals are
presented graphically along with the means: the mean of the dependent
variable
is indicated on the y axis with
some
chosen symbol; a line representing the confidence interval extends from
(in
this case) 90.16 to 105.84 and it is drawn vertically and passes
through the
mean.
Before leaving this section, it is
necessary to consider what to do when wanting to make corresponding
statements
about small-sized samples which cannot be approximated with the normal
distribution. Here computation of the mean and standard error proceeds
as
before. Since the quantity z is
used
in conjunction with the normal distribution tables, it cannot be used.
Instead
the analogous quantity t is
calculated:
The distribution of t
is dependent on sample size n
and so (in essence) the t value has
to be referred to different tables for each size of sample. The tables
corresponding to the t distribution
are usually collapsed into one table and the section of the table used
is
accessed by a parameter related to the sample size n
(the quantity used for accessing the table is n
-
1 and is called the degrees of freedom). Clearly, since
several different
distributions are being tabulated, some condensation of the information
relative to the z tables is
desirable. For this reason, t
values
corresponding to particular probabilities are given. Consideration of t tables emphasizes one of the
advantages of the Central Limit Theorem insofar as one table can be
used to
address a wide variety of issues rather than is the case for t.
Estimating proportions
Here the problem faced is similar to that
with means: A sample has been taken and the proportion
of people meeting some criterion and those not meeting
that criterion are
observed. The question is with what degree of confidence
can you assert that the proportions observed reflect
those in the population? Once again the solution is directly related to
that
discussed when estimating how close a sample mean lies to the
population mean
using z scores. Essentially the z
score for means measures:
The only difference here is that binomial
events are being considered (meet/not meet the criterion). The z score
associated with a particular sample based on the estimated probability
and the
population proportion is (where q = 1-p):
>>phat<<
Normal distribution tables can again be
used to assign a probability associated with this particular outcome.
To illustrate with an example: Suppose
that it is expected that as many men will use the speech recognizer as
will
women (p (man) = p (woman) 0.5). What size of sample is needed to be
95%
certain that the proportion of men and women in the sample differs from
that in
the population by at most 4%?
Solving for n
gives 600.25. Therefore, a sample of size at least 601 should be
used.
Now what are
the effects if we want to be more than 4 confident, say if the
difference is
reduced to 2%. The required sample size jumps to 2401.
Estimating variance
The relationship between the variance of a
sample and that of the population is distributed as χ2
(chi
squared) with n - 1 degrees of
freedom.
Thus, if we have a sample of size 10 drawn
from a normal population with population variance 12, the probability
of its
variance exceeding 18 is:
This has associated with it 9 degrees of
freedom. Because χ2 values are only tabulated
for particular
probabilities (as with t), the
probability can only be estimated for limited probabilities. In this
case
χ2 lies between 0.2 and 0.1.
Ratio of sample
variances
If two independent samples are taken from
two normal populations with variance σ12
and σ22,
the ratio of the two variances (S12
and S22)
has the F distribution:
If the two samples (which can differ in
size) from the same normal population are taken, then the ratio of the
variances will be approximately 1. Conversely, if the samples are not
from the
same normal population, the ratio of their variances will not be 1 (the
ratio
of the variances is termed the F ratio).
The F tables can be used to assign
probabilities that the sample variances were or were not from the same
normal
distribution. The importance of this in the Analysis of Variance
(ANOVA) will
be seen later.
3. Statistical
terms involved in inference to the population mean from the sample mean
Simple hypothesis
testing
Many practical and research applications
in speech and hearing science require testing of hypotheses. An example
from
the scenario given at the outset was testing whether there were
differences
between read and spontaneous speech with respect to selected
statistics. If the
statistic was mean vowel duration in the two conditions where speech
was
recorded, we have a situation calling for simple hypothesis testing.
This
situation is called simple hypothesis testing since it involves a
parameter of
a single population.
Following the approach adopted so far, the
concepts involved in such testing are illustrated for this selected
example.
The first step is to make alternative assertions about what the likely
outcome of
an analysis might be. One assertion is that the analysis might provide
no
evidence of a difference between the two conditions. This case is
referred to
as the null hypothesis (conventionally abbreviated as Ho) and might
assert here
that the mean syllable duration in the read speech is the same as that
in the
spontaneous speech. Other assertions might be made about this
situation. These
are referred to as alternative hypotheses.
One alternative hypothesis would be that the vowel duration in the read
speech
will be less than that of the spontaneous speech. A second would be the
converse, i.e. the vowel duration in the spontaneous speech will be
less than
that of the read speech. The decision about which of these alternate
hypotheses
to propose will depend on factors that lead the speech or hearing
student or
investigator to expect differences in one direction or the other. These
instances are referred to as one-tailed
(one-directional) hypotheses
as
each predicts a specific way in which there will be a difference
between read
and spontaneous speech. If the investigator wants to test for a
difference but
has no theoretical or empirical reasons for predicting the direction of
the
difference, then the hypothesis is said to be two-tailed. Here, large
differences between the means of the read and spontaneous speech, no
matter
which direction they go in, might constitute evidence in favor of the
alternative hypothesis.
The distinction between one and two-tailed
tests is an important one as it affects what difference between means
is needed
to assert a significant difference (i.e., support the null hypothesis).
In the
case of a one-tailed test, smaller differences between means are needed
than in
the case of two-tailed tests. Basically, this comes down to how the
tables are
used in the final step of assessing significance (see below). There are
no
fixed conventions for the format of tables for the different tests, so
there is
no point in illustrating how to use them. The tables usually contain
guidance
as to how they should be used to assess significance.
Hypothesis testing involves asserting what
level of support can be given in favor of, on the one hand, the null,
and, on
the other, the alternate hypotheses. Clearly no difference between the
means of
the read and spontaneous speech would indicate that the null hypothesis
is
supported for this sample. A big difference between the means would
seem to
indicate that there is a statistical difference between these samples
if the
direction in which the means differs is in the same direction as
hypothesized
for a one-tailed hypothesis or if a two-tailed test has been
formulated. The
way in which a decision whether a particular level of support (a
probability)
is provided is described next.
In the read-spontaneous example that we
have been working through, we are interested in testing for a
difference
between means for two samples where, it is assumed, the samples are
from the
same speaker. The latter point requires that a related groups test as
opposed
to an independent groups test is used. In this case, the t
statistic is computed from:
Thus if the read speech for 15 speakers
had a mean vowel duration of 40.2 milliseconds and the spontaneous
speech 36.4
milliseconds and the standard deviation of the difference between the
means is
2.67, the t value is 1.42. The t value is then used for establishing
whether two sample means lying this far apart might have come from the
same
(null hypothesis) or different (alternate hypothesis) distributions.
This is
done by consulting tables of the t
statistic using n-l degrees of
freedom (here n refers to the
number
of pairs of observations).
In assessing a level of support for the
alternate hypothesis, decision rules are formulated. Basically this
involves
stipulating that if the probability of the means lying this far apart
is so low
then a more likely alternative is that the samples are drawn from
different
populations, assuming that the samples are from the same distribution.
The
"stipulation" is done in terms of discrete probability levels and,
conventionally, if there is a less than 5% chance that the samples were
from
the same distribution, then the hypothesis that the samples were drawn
from
different distributions is supported (the alternative hypothesis at
that level
of significance). Conversely, if there is a greater than 5 in a hundred
chance
that the samples are from the same distribution, the null hypothesis is
supported. In the worked example, with 14 degrees of freedom, a t value of 1.42 does not support the
hypothesis that the samples are drawn from different populations, thus
the null
hypothesis is accepted. It should be noted that support or rejection of
these
alternative hypotheses is statistical rather than absolute. In 1/20
(5%) cases
where no difference is asserted, a difference does occur (referred to
as a Type
II error, accepting the null hypothesis when it is false) and in cases
where a
5% significance level is adopted and differences found, 1 occasion out
of 20
will also lead to an error (referred to as a Type I error, rejecting
the null
hypothesis when it is in fact true).
Analysis
of Variance
As was said earlier, this is not supposed
to be a substitute for your statistics textbook as, in particular, it
does not
cover all statistical tests that might be encountered. It only offers
an
overview and a means of accessing relevant material in a textbook.
However, some comments on Analysis of
Variance (ANOVA) are called for as it is a technique that has a
widespread use
in speech and hearing assessment. ANOVA is a statistical method for
assessing
the importance of factors that produce variability in responses or
observations. The approach is to control for a factor by specifying
different
values (or, treatment levels) for it in order to see if there is an
effect. It
can be thought of as having sampled a potentially different population
(different in the sense of having different means). Factors that have
an effect
change the variation in sample means, where "factor" refers to a
controlled
independent variable. When the experimenter controls the levels of the
factors,
this is referred to as treatment level.
For example, in the ANOVA approach, two
estimates of the variances are obtained: the variance between the
sample means,
between groups variance, and the variance of each of the scores about
their
group mean, within groups variance. If the treatment factor has had no
effect,
then variability between and within groups should both be estimates of
the
population variance. So, as discussed earlier when the ratio of two
sample
variances from the same population was considered, if the ratio of
between
groups to within groups is taken, the value should be about 1 (in which
case,
the null hypothesis is supported). The ratio of two variances is called
the F ratio. Statistical tables of
the F distribution can be consulted
to
ascertain whether the F ratio is
large enough to support the hypothesis that the treatment factor has
had an
effect resulting in larger variance of the between group to the within
group
means (the alternative hypothesis is supported). Another way of looking
at this
is that the between groups variance is affected by individual variation
of the
units tested plus the treatment effect whereas the within groups
estimate is
only affected by individual variation of the units tested.
ANOVA is a powerful tool which has been
developed to examine treatment effects involving several factors. Some
examples
of its scope are that it can be used with two or more factors. Factors
that are
associated with independent and related groups can be tested in the
same
analysis, and so on. When more than one factor is involved in an
analysis, the
dependence between factors (interactions) comes into play and has major
implications for the interpretation of results.
Non-parametric tests
Parametric tests cannot be used when
discrete, rather than continuous, measures are obtained since the
Central Limit
Theorem does not approximate the normal distribution in these
instances. The
distinction between discrete and continuous measures is the principal
factor
governing whether a parametric or non-parametric test can be employed.
Continuous and discrete measures relate to another taxonomy of scales -
interval, nominal and ordinal: interval scales are continuous and the
others
are discrete. Statisticians consider this taxonomy misleading, but
since it is
frequently encountered in the behavioral sciences in general, the
nature of
data from the different scales is described. Interval data are obtained
when the
distance between any two numbers on the scale are of known size and is
characterized by a constant unit of measurement. This applies to
physical
measures like duration and frequency measured in Hertz (Hz) which have
featured
in the examples discussed to now. Nominal scales are obtained when the
measures
are obtained from symbols to characterize objects (such as sex of the
speakers). Ordinal scales give some idea of the relative magnitude of
units
that are measured but the difference between two numbers does not give
any idea
of the relative size. Examples would be responses to questionnaires
where there
is no guarantee of equal distance between the response choices offered
(e.g.
strongly agree, agree, disagree, strongly disagree). In cases where
parametric
tests cannot be used, non-parametric (also known as distribution-free)
tests
have to be employed. The computations involved in these tests are
straightforward and covered in any elementary text book. A reader who
has
followed the material presented thus far should find it easy to apply
the
previous ideas to these tests.
A number of representative questions a
speech and hearing investigator might want to answer were considered at
the
start of this section. Let us just go back over these and consider
which ones
we are now equipped to answer. First there was how to check whether
there are
differences between spontaneous and read speech.
If the measures are parametric (such as
would be the case for many acoustic variables), then either an
independent or
related t test would be appropriate
to test for differences. An independent t
test is needed when samples of spontaneous speech and read speech are
drawn
from different speaker sets; a related t
test is used when the spontaneous and read samples are both obtained
from the
same group of speakers.
If the measures are non-parametric (e.g.
ratings of clarity for the spontaneous and read speech) then a Wilcoxon test would be used when the
read and spontaneous versions of the speech are drawn from the same
speaker and
a Mann-Whitney U test otherwise.
If you find differences between read and
spontaneous speech (see application described), how can you check
whether
language statistics on your sample of recordings is representative of
the
language as a whole - or, what might or might not be the same thing,
how can
you be sure that you have sampled sufficient speech? For this, the
background
information provided to estimate how close sample estimates are to
population
estimates is appropriate.
4.
Conclusions
This has been a whirlwind tour of some
elementary concepts in treatment of data from the speech and hearing
areas
starting with issues concerned with choice of data and inferential
statistics.
A final aim is to draw readers attention to other simple, clearly
written
resources for examining data in these (and medical) areas in general.
For this
purpose, a selected bibliography follows.
Acknowledgement. The second author is supported by the Wellcome Trust grant 072639.
Bibliography
Armitage,
P.,
Bland,
M. (2000). An
introduction to Medical Statistics, 3rd edition.
Hand,
D. J. (2004). Measurement - theory and practice.
Huff,
D (1993). How to lie with statistics.
Moser,
C & Kalton, G. (1979), Survey
methods in social investigation. Heinemann Educational:
British Medical Journal: Statistics Notes
Perhaps the finest series of short articles on the use of statistics is the occasional series of Statistics Notes started in 1994 by the British Medical Journal. It should be required reading in any introductory statistics course. The full text of the articles is available is available on the World Wide Web. The articles are listed here chronologically.
TARGET ARTICLE
The
Use of Structural Equation Modeling in Stuttering Research: Concepts
and
Directions
Stephen Z. Levine
Adjunct Lecturer,
Department of Behavioral Sciences
K. V. Petrides
Stephen Davis
Department of
Psychology, Centre for Human Communications, Institute of
Cognitive Neuroscience, and Institute of Movement Neuroscience,
University
College London, Gower St., London WC1E 6BT, England
Chris J. Jackson
Peter Howell
Department of
Psychology, Centre for Human Communications, Institute of
Cognitive Neuroscience, and Institute of Movement Neuroscience,
University
College London, Gower St., London WC1E 6BT, England
Abstract.
This
article provides a brief introduction to the
history and applications of the class of data analytic techniques
collectively
known as Structural Equation Modeling (SEM). Using an example based on
psychological factors thought to affect the likelihood of stuttering,
we
discuss the issues of specification, identification, and model fit and
modification in SEM. We also address points relating to model
specification
strategies, item parceling, advanced modeling, and suggestions for
reporting
SEM analyses. It is
noted that SEM
techniques can contribute to the elucidation of the developmental
pathways that
lead to stuttering.
Keywords: Structural equation
modeling, LISREL, stuttering models.
1.
Without
going into the nuances of the many
definitions of stuttering that are available, the disorder involves
difficulty
in handling language and is usually manifest as problems in overt
speech
control. This does not necessarily mean that language problems are
paramount
and the sole cause of stuttering. The anxiety a speaker is
experiencing, the
situation in which he or she is interacting and so on will determine
whether or
not the speaker stutters at that particular time.
The
aim of this paper is to establish how the
multiple and various factors associated with stuttering behavior, work
together
in influencing stuttering. Before introducing the Structural Equation
Modeling
(SEM) technique, it is necessary to look at some of the issues and
debates that
have been raised in connection with the process of theory construction
in the
field of stuttering. This shows that more detailed specification of
theories is
needed before SEM can be employed as a way of implementing, assessing
and
evaluating alternative multifactor models of stuttering (the intention
is that
this article will stimulate work on both theory construction and
evaluation).
The
first thing to note is that there are several
models of stuttering that are specified in detail that have been tested
experimentally. In the main, these have been developed to try to
explain how
language influences speech-motor performance and how this then results
in
stuttering. Kolk and Postma’s (1997) covert repair hypothesis (CRH)
offers one
example. This explains stuttering in terms of a linguistic system that
can
malfunction and produce errors which then affect speech-motor
performance. Stuttering,
according to CRH, can be characterized as the response an individual
makes to
linguistic errors, which produces different types of stuttering events
(such as
repetitions and prolongations) in overt speech (Smith, 1999). Put
another way,
stuttering is viewed by CRH as a linear causation process in which
errors in
the linguistic processes lead directly to observed problems in speech
output
(Smith, 1999). CRH does not concern itself with other factors that can
influence stuttering. It is simply an account of the particular link
between
language processes and speech output behaviour. In tests that have been
made of
CRH, other factors that affect stuttering would have minimal influence
provided
good experimental practice is followed. An example of such practices
can be
seen for gender, which is known to affect stuttering behaviour, and
language
more generally. Any influences of gender per se can be controlled by
selecting speakers
of the same sex for both stuttering and control groups. It is not that
CRH maintains
that gender or other factors are not relevant to stuttering, simply
that with
suitable precautions, authors approaching their research from CRH’s
perspective
are able to focus on what interests them – the link between language
processes
and speech performance.
Second,
also as a consequence of using
experimental control principles, the problem of the relationship
between
language and speech output is simplified into one of linear causation.
This
does not imply that if other factors are added to the model, they too
have to
be placed at some point in a linear chain. If, for instance, CRH was
extended
to account for the effects of anxiety on stuttering, anxiety might
influence
the linguistic system directly by, for instance, increasing the
likelihood that
a speaker who stutters makes more errors in that system. Alternatively,
anxiety
might be one factor that increases overall arousal level, making
speakers
attempt utterances faster than they otherwise might. This could then,
in turn,
put pressure on both the linguistic and motor systems (if the
linguistic system
operates more rapidly, it may make more errors, Kolk and Postma, 1997;
if the
motor system executes speech plans more rapidly, it may show more
variable
performance, Smith and Kleinow, 2000). The arousal system might then be
modeled
as an independent parallel process that has indirect influences on both
language and speech processes. If anxiety level is controlled, this
nonlinear
causal chain is reduced to a linear one for the purpose of study.
The
third issue is whether there is only
one appropriate measure of output behavior (the dependent variable).
CRH
focuses on events that are regarded as being associated with
stuttering, such
as the repetitions and prolongations mentioned earlier. These are
suited to
their authors’ aims where they wish to examine specifically the
influence of
the linguistic system on failures in speech output. Smith uses an
output that
reflects the (possibly nonlinear) influences of a variety of additional
factors. Others who
are interested in the effects of a
clinical procedure on speech outcomes might want to use a standardized
instrument as their output measure (such as SSI-3, Riley, 1994).
Each
of these output measures is
appropriate for the particular research needs and no one measure meets
all the requirements.
This is not a point of view that is shared by all workers. Smith
(1999), for
instance, sees serious limitations in using speech event measures. One
argument
she puts forward is that studying stuttering by examining the overt
manifestations of the problems (the stuttering events like part-word
repetitions or prolongations) is analogous to studying volcanoes by
examining
the shape and type of effused eruptive material. In both these
examples, the surface
form (of language, and of the geological forms, respectively) is the
output
object. Smith points out that examination of surface forms of volcanoes
did not
advance ability to forecast eruptions. Plate tectonics provided the key
to the
underlying forces that allowed eruptions to be forecasted. The
underlying
dynamic movements of the plates lead to volcanoes at the points where
they
join. The analogy suggests that an underlying dynamic representation of
speech
(rather than events) might provide a unifying framework for
understanding
stuttering. This is
not an argument we
would buy (instead we prefer to retain our view that different output
measures
are suitable for different purposes). Moreover, we think the same may
apply to
the study of volcanoes: An important (and tried, tested and confirmed)
prediction of plate tectonics is that volcanoes only occur at points
where two
plates abut. Clearly, some output measure that looked at the
geographical
distribution of volcanoes was needed to test this prediction.
This
article is not concerned with issues
in multifactor theory construction per se.
It does seek to establish, once a multifactor theory has been
specified, how a
researcher can establish how well the theory (or model) fits the data
and how
that model can be compared with others. Modeling necessarily relies on
having a
theory specified in detail. Current efforts at theory construction in
the
stuttering area do not meet this requirement. For instance, Smith
(1999, p.34)
is a multifactor models that predict ways in which various component
factors
interact, but offer little in the way of specification of the exact
form of
that interaction (whether by direct or indirect influences as discussed
in
connection with anxiety). One major goal we hope to achieve is to
stimulate
authors to offer their theories in more detail (which would be welcomed
as
submissions to Stammering Research).
The second major goal is to introduce SEM as an approach to
discriminate
between alternative theories. As there are no detailed and explicit
multifactor
theories at present, we use specified hypothetical models as the basis
of our
discussion. Before model specification, some general remarks about SEM
are
made.
2.
Introduction II: Structural Equation Modeling and stuttering
Recent years have witnessed
an unprecedented
rate of growth in advanced quantitative techniques that are potentially
applicable across a wide range of disciplines.
With respect to non-experimental research in the
social sciences, the
development of new methods for data analysis has largely revolved
around
Structural Equation Modeling (SEM).
Indeed, although SEM is mostly used for analysis of
data collected by non-experimental
methods, it is possible to conduct analyses of data obtained using
experimental
designs. Actually,
SEM is a broad class
of techniques covering confirmatory factor analysis, time growth
analysis,
multi-level latent modeling, and simultaneous equation modeling. The
early
origins of SEM can be traced back to the technique of path analysis,
introduced
in the field of biology by Wright (1934). However, the field of SEM as
we now
know it was created by the seminal contributions of Karl J. Jöreskog
and his
colleagues (notably Dag Sörbom), who together managed to integrate the
data
analytic traditions of multiple regression and factor analysis (e.g.,
Jöreskog,
1969, 1971). The
breakthroughs in
statistical theory were implemented in the computer program LISREL (LInear
Structural RELations;
e.g., Jöreskog & Sörbom, 1996), which
has had a major impact across the social sciences and beyond.
The
SEM literature has been growing
rapidly over the last decade as reflected in the ever-increasing number
of
introductory chapters (e.g., Fife-Schaw, 2000), textbooks (e.g. Kline,
2004;
Maruyama, 1998), and journal articles (e.g., Muthén, 2002). The results of a recent
review of the
relevant literature (Hershberger, 2003) showed that: (a) SEM has
acquired dominance
among multivariate techniques; (b) the number of journals publishing
SEM
articles continues to grow; and (c) ‘Structural
Equation Modeling: An Interdisciplinary Journal’ has become
the primary
outlet for the publication of technical developments in SEM. The
journal
contains tutorials by leaders in the field (e.g., Raykov &
Marcoulides,
1999), and is a particularly useful resource for applied researchers,
as is the
online discussion group SEMNET (http://www.gsu.edu/~mkteer/semnet.html).
There
are five main reasons for the recent
surge in the popularity of SEM. First, it allows for statistical
analyses that
account for measurement error in the dependent and independent
variables
through the use of multiple observed indicators per latent variable.
Second, it
provides a rigorous approach to model testing, which requires careful
theoretical
development as discussed in the previous section. Third, it is
extremely
flexible as it can accommodate experimental designs, group differences
designs,
longitudinal designs, and multi-level designs, all of which are used in
stuttering research. Fourth, it can accommodate tests of mediation and
moderation (Baron & Kenny, 1986). Fifth, it allows for the
statistical
testing of multiple competing theoretical models (such as the two
models
proposed earlier about how anxiety could be added to CRH). While the
typical
approach to data analysis involves testing hypotheses based on a
particular
theory, a more powerful approach involves comparing alternative
theories and
establishing, based on a priori
criteria and statistical indices, which model best accounts for the
data. This
elegant approach to developing theories and testing hypotheses has been
applied
in many different domains, including occupational psychology (e.g.,
Earley
& Lituchy, 1991), personality (Petrides, Furnham, Jackson
& Levine,
2003), criminological psychology (Levine & Jackson, 2004), and
counselling
psychology (Quintana & Maxell, 1999).
The
main objective of this article is to
introduce the new possibilities that SE modeling offers for the
analysis of
stuttering data. For the purposes of this exposition, we will employ a
hypothetical example, illustrated in Figure 1 and based on
socio-psychological
variables that are implicated in stuttering (see Furnham &
Davies, 2004).
First some elementary steps that construct two theories of this
process: Model
1 suggests that cognitive ability is influential in addressing whether
a child
is bullied which, in turn, influences the severity of stuttering (like
the
linear causation discussed in connection with anxiety and CRH). Model 2
differs
in suggesting that cognitive ability directly influences the severity
of
stuttering, in addition to its indirect effect mediated via being
bullied (in a
similar way to the second alternative way of adding anxiety to CRH). In
everyday terms, the first model maintains that a child with low IQ
might be
more prone to bullying, which then causes anxiety and leads to
stuttering
(linear causation model). The second model assumes IQ affects severity.
IQ can
also affect the degree to which a child is bullied that will then
affect
stuttering severity. SEM provides techniques that allow the best
alternative
model of the process involving these factors to be determined.
Path
diagrams, like the one in Figure 1,
are pictorial representations of an underlying system of mathematical
equations. Observed indicators (items or scales) are designated by a
box. In
Figure 1, each latent variable (where a latent variable reflects the
relationship between variables) is represented by three observed
indicators in
square boxes. Directional relationships are represented by straight
lines with
an arrowhead pointing toward the endogenous variable. Non–directional
relationships (i.e., correlations) are represented by curved lines with
arrowheads at both ends. Latent
variables are represented by circles or ellipses in SEM and can be
either
exogenous or endogenous. Exogenous latent variables are
‘upstream’ variables that are not influenced by other variables in the
model. Endogenous latent variables are
‘downstream’ variables that are
influenced by other variables (exogenous, endogenous, or both) in the
model. For
instance, in Figure 1, IQ is
an exogenous variable, whereas bullying and severity of stuttering are
endogenous variables.
Figure
1.
Two possible Competing Structural Models of Stuttering that propose a
relationship between IQ, bullying and stuttering severity.
Model
1
Model
2
Note.
IQ =
General intelligence factor, vo = vocabulary test, si = similarities
test, co =
comprehension (example scales taken from WAIS – III tests). Bu =
bullying, srb
= self-reports bullying, prb = peer-reports of bullying, trb =
teacher-reports
of bullying. SS = Severity of stuttering, Srs = Self-reported severity
of
stuttering, prs = peer-reported severity of stuttering, trs =
teacher-reports
severity of stuttering.
Specification
The
first stage in SEM involves specifying
a model. A model is a statistical statement about the relationships
between
variables. Models take different forms, depending on the analytical
approach
that is adopted. In SEM, relationships are specified on the basis of
the
hypotheses outlined by theory. Specification refers to the translation
of a
theory into a structural model stating specific relationships between
the
variables. These relationships entail parameters that have magnitude
and
direction (+ / -) and can be either fixed or free. Fixed
parameters are pre-specified by the researcher, rather than
estimated from the data, and are usually, but not necessarily, set at
zero. Free parameters are
estimated from the
data and are usually, again albeit not necessarily, significantly
different
from zero. In the example shown in Figure 1, the direct relationship
between IQ
and severity of stuttering is a fixed parameter, as it is set to zero
in model
1 (i.e., no relationship is specified). In model 2, the relationship
between IQ
and severity of stuttering is a free parameter (i.e., estimated from
the data).
The
various parameters in an SEM define
its two components, namely the measurement and the latent parts of the
model.
‘The measurement model is that
component of the general model in which latent variables are
prescribed. Latent
variables are unobserved variables implied by the covariances among . .
.
[three] or more indicators’ (Hoyle, 1995, p. 3). Latent variables are
often
termed factors and are free of the random error and uniqueness
associated with
their observed indicators. Scores on the observed indicators comprise
common
factor variance (i.e., variance shared with the other indicators of the
construct), specific factor variance (i.e., reliable variance specific
to the
observed indicator), and measurement error. The specific variance and
measurement error parts are collectively described as a variable’s
uniqueness.
In
Figure 1 there are three observed
indicators (represented by boxes) defining each of the three latent
variables
(represented by circles). It is noted that there are three observed
indicators
for each latent variable in Figure 1. The exact ratio of observed to
latent
variables remains a topic of debate, and appears to be somewhat
dependent on
the data at hand. In Figure 1, for example, the latent variable
‘severity of
stuttering’ influences the observed indicators of self, peer, and
teacher
ratings of stuttering. The underlying rationale is that each of these
ratings
partially reflects the target’s standing on the general latent variable
of
severity of stuttering.
One
of the main strengths of the SEM
approach is that it allows the decomposition of the relationships
between
variables. There are five different types of effects. Direct
effects refer to direct relationships between variables,
whereas indirect effects refer to
relationships that are mediated via intervening variables. In both models in Figure
1, IQ has a direct
effect on bullying, which is a mediating variable that transmits the
effect of
IQ on severity of stuttering. In addition, it is possible for the
relationship
between two variables to be caused by a third variable, which is
referred to as
a non-causal relationship due to shared
antecedents. For example, it could be that the relationship
between being
bullied and stuttering is caused by a third factor.
It is also possible to specify nondirectional relationships
between variables, which are sometimes referred to as unanalyzed
prior associations. In the example in Figure 1, it could
be that a person is bullied because they stutter, or that being bullied
results
in more stuttering. Lastly, reciprocal
effects may also be accommodated by allowing variables
simultaneously to
influence each other.
Identification
A
model is said to be identified when
there is a single best value or unique solution for
each of its unknown (free) parameters. If a model is not identified it
is
possible to find an infinite number of values for the parameters that
would
produce a good fit of the data to the model. Just-identified
models possess the same number of equations as
unknown free parameters. A consequence of this is zero degrees of
freedom. Under-identified models
occur when there
are too many parameters to estimate the number of observed measures.
The
consequences of under-identification may include, impossible parameter
estimates, fit tests that are not valid and large standard errors. Over identified models are those where
there are fewer possible parameter estimates than possible equations.
Perhaps
the best example of such a model is the well-known multiple regression
model. Identified
models provide the best evidence in favor of the proposition that the
theoretical model represents the data, as there is a unique solution to
the
data.
Identification
is an important concern in
SEM, as the methodology provides the end user with the freedom to
specify
models that are not identified. An SEM is said to be "identified" if
the model's restrictions and a population covariance matrix imply
unique values
for the model's parameters. Assessing identification is complex, given
the
complexities of matrix algebra. Fortunately, most computer programs
provide
information in their output if there is a problem regarding
identification.
Unfortunately, the locus of the problem is not specified.
Estimation
Once
a model is specified, the next step
is to obtain estimates of the free parameters from the data. To
estimate the
model based on the data, SEM programs use iterative algorithms. The
matrix
based on the data, usually the covariance matrix, in SEM is known as S. Taking this matrix SEM software
calculates a 
One
of a number algorithms can be used to
minimize the similarity between S and
The
decision about which algorithm to use
can be complex. It depends on the nature of the data and the research
aims of
the study. For example, WLS, GLS, and ML all assume multivariate
normality (i.e.,
the assumption that each variable that is considered is normally
distributed,
but with respect to each other variable).
ULS is unduly affected by the metric of the variables
and usually requires
prior standardization. This standardization is problematic and should
be
avoided because standardized covariance (i.e., correlation) matrices
result in
poorly estimated models (Cudeck, 1989). WLS needs an asymptotic
covariance
matrix (an estimate of the large sample covariance matrix used to
generate
weights) and a matrix of correlations to work properly. Current
research
indicates that the most appropriate estimation method in most cases,
including
those where the multivariate normality assumption has been violated, is
ML
(Olsson, Foss, Troye & Howell, 2000).
The
estimation of parameter values is
achieved by means of an iterative process. This process starts with
initial
estimates that are sequentially adjusted until model fit cannot be
improved.
The parameter estimates after iteration constitute the final solution.
It is
worth noting that, in some cases, the algorithm may fail to explore the entire range of potential values
under the fit function and
become trapped into a local minimum point. Under such circumstances,
the
resulting estimates will be suboptimal. In complicated models, it is
always
sensible to provide the software with alternative sets of starting
values to
establish whether the resultant solutions converge (this option is not
provided
by all software). It is also possible for a final solution to contain
illogical
parameter estimates (e.g., negative variances or correlations over
1.00). These
are known as Haywood cases and are indicative of ill-fitting or poorly
specified models.
Model fit
Once
the estimation process has been
completed, it is important to assess how well the data fit the proposed
model.
A large number of indices have been developed to evaluate the goodness of fit of a model and some
attention should be given as to which fit indices should be reported.
The most
widely reported fit index is the
The
reason for this is that ideally S
and
Two
points should be kept in mind in the
discussion of fit indices. First, these indices apply to some types of
SEM,
like the examples in Figure 1, but not to others (e.g., tests of
factorial
invariance; Cheung & Rensvold, 2002). Second, SEM values
parsimony, which
can be operationalized as ‘the ratio of the degrees of freedom in the
model
being tested to the degrees of freedom in the null model’ (Raykov
&
Marcoulides, 1999, p. 293). Thus, some fit indices attempt to take
parsimony
into account by penalizing complex models with many free parameters.
Absolute
indices
Absolute
fit indices attempt to assess how
well the theoretical model reproduces the sample data.
The Goodness of Fit Index (GFI) compares the
specified model to no model at all.
It
ranges from 0 to 1, with higher values indicating better fit. The Adjusted GFI (AGFI)
introduces an
adjustment based on degrees of freedom that penalizes model complexity. It is interpreted in the
same way as the GFI.
The Standardized Root Mean Square Residual (SRMR) expresses the average
discrepancy between observed and expected correlations across all
parameter
estimates in a model. It
ranges from 0
to 1, with lower values indicating better model fit.
A
widely used index of absolute fit is the
Root Mean Square Error of Approximation (RMSEA), which essentially asks
“How
well would the model, with unknown but optimally chosen parameter
values, fit
the population covariance matrix if it were available?” (Browne
& Cudeck,
1993, p. 137-138). The
RMSEA expresses
fit discrepancy per degree of freedom, thus addressing the parsimony of
the
model. It is
generally insensitive to
sample size. In
contrast to most other
indices, it is possible to estimate confidence intervals (usually, 90%)
around
the point estimate of the RMSEA.
Incremental
fit indices
Incremental
fit indices measure the
proportionate improvement in fit by comparing a target model with a
more
restricted, nested baseline model (Hu & Bentler, 1999, p.2).
Further note,
that if two models are equivalent except for a subset of free
parameters in
model one that are fixed in model two, then model one is said to be a
nested
model. The Non-Normed Fit Index (NNFI) is an example of such a fit
index.
Essentially the NNFI compares the specified model with a baseline model
(usually an independence model,
i.e.,
a model that stipulates that the variables are unrelated. NNFI
penalizes model complexity, such that complex models with many free
parameters
have lower NNFI values.
Bentler
(1990) proposed the Comparative
Fit Index (CFI), which is based on the non-central
Modification
Once
model fit has been assessed,
researchers may wish to adjust their model to account for aspects of
the data
that do not accord to the theory. This is known as the model
modification stage
of SEM. It is contentious because it involves freeing up previously
fixed
parameters on a post-hoc data driven basis. For example, a researcher
may fit
the model in Figure 1 and subsequently discover that an additional
direct path
from IQ into severity of stuttering is necessary to improve fit. This
modification would result in an improvement of model fit.
Modification
indices quantify the expected
drop in
As noted, many are critical of the modification process because it can result in high levels of capitalization on chance. Two strategies are available to avail of the advantages of post-hoc modification strategies without shouldering the attendant pitfalls. First, assuming a sufficient sample size, it is possible to split the data randomly into two sets of approximately equal size. The first data set can be used for post-hoc modifications and model exploration, while the second data set can be used for cross-validation purposes. The second strategy is to collect an independent new dataset and use it to fit the revised model including all the post-hoc modifications.
Modeling
strategies
A number of different strategies to conduct SEM have been proposed in
the
literature. Although all strategies involve the essential steps of
model
specification, identification, estimation, and modification, the exact
manner
in which they are carried out can vary very considerably. Perhaps the
best
known strategy, giving excellent results in psychological studies, is
known as
the Two Step. Step 1 involves the development of an individual
congeneric model
for each latent variable. Observed variables which produce goodness of
fit in
predicting a latent variable are identified. Factor score regression
weights
(transformed to sum to 1) are used to calculate a composite observed
variable
from these observed variables. A lower bound estimate of the
reliability of the
composite observed variable is Cronbach's alpha, but a better measure
of
reliability can be computed by hand. This is relatively easy to do if
there are
no error covariances (see Gerbing & Anderson, 1988), but can be
difficult
if there are (see Werts, Rock, Linn & Joreskog, 1978). Once the
reliability
is known, Munck (1979) showed that it is possible to fix both the
regression
coefficients (which reflect the regression of each composite variable
on its
latent variable) and the measurement error variance. At step 2, the
overall
model is considered. This involves putting each latent variable and its
associated composite observed variable into the whole model. As with
most
issues there are those who support (Mulaik & Millsap, 2000) and
those who
refute the use of this method (Hayduk & Glaser, 2000).
Item
parceling
In
some cases, researchers may be unsure
as to how to represent their latent variables.
An important choice is between using item parcels
versus single items. The
sometimes controversial technique of
parceling involves summing up a number of individual items in order to
construct item parcels and use them as indicators of the latent
variables. Critics
note that, when variables are not
truly unidimensional, parceling may result in model mis-specification
and in
the acceptance of models that in fact provide a poor fit to the data.
On the
other hand, MacCallum, Widaman, Zhang and Hong (1999) provide several
reasons
to use item parcels as indicators of the latent factors. They point out
that
parceled data are more parsimonious (i.e., there are fewer parameters
to be
estimated both locally in defining a construct and globally in
representing the
entire model). Parceled data are also less likely to produce correlated
residuals or multiple cross-loadings and may lead to reductions in
various
sources of sampling error.
In
the delinquency literature it has been
noted that the use of item parcels overcomes the extreme skewness that
is often
found in individual items. Thus,
a
number of authorities in criminology recommend the use of parcels to
help
achieve multivariate normality (e.g., Farrell & Sullivan, 2000). It is worth noting that
the same may apply to
stuttering data, particularly in cases of severe skewness. Little,
Cunningham,
Shahar and Widaman (2002) reviewed the use of parceling in the
literature and
noted that ‘in the end two clear conclusions may be drawn from our
review of
the issues. On the one hand, the use of parceling techniques cannot be
dismissed out of hand. On the other, the unconsidered use of parceling
techniques is never warranted’ (p. 171).
Interactions
SEM
is not restricted to linear structural
relationships, but can also accommodate interactions and polynomial
effects.
Indeed, testing interactions through SEM has advantages over the
conventional
multiple regression approach because SEM overcomes many of the problems
that
are associated with interaction terms in regression models (e.g., low
internal
consistency, which reduces the power of the corresponding statistical
test).
There are several approaches to testing interactions through SEM, their
main
differentiating characteristic being how the interaction term is
represented
and estimated (Schumacker, 2002; see also Jaccard & Wan, 1995,
Moulder
& Algina, 2002).
Criticisms
Critics of SEM methodology (notably Cliff, 1983)
have drawn attention to
a number of contentious issues and limitations, foremost among which is
the
erroneous assumption of some practitioners that modeling correlational
data can
somehow help to establish causal relationships between variables. This criticism perhaps
originates from the
early use of the term “causal models” to describe SEM analyses.
Another
criticism is that researchers
often fail to consider alternative models, other than their preferred
one,
which could provide an equally good or even superior fit to their data. The problem is especially
difficult to
resolve in cases where a researcher has to grapple with a number of
theoretically conflicting, but mathematically equivalent, models (see
MacCallum, Wegener, Uchino, & Fabrigar, 1993). Other oft-quoted
limitations
of SEM per se or of the manner in
which it is applied include the routine violation of the assumptions on
which
the analyses are based, the excessive or uncritical reliance on
modification
procedures, and an unwarranted preoccupation with model fit at the
expense of
substantive considerations.
Reporting
SEMs
A
number of texts provide directions on
how to report an SEM (e.g., Boomsma, 2000; Hoyle & Panter,
1995). In general
the texts suggest the following. A diagram showing the theoretical
relations between
the elements in the model should be presented in the introduction. This
should
be much like the one in Figure 1, although it should only include the
latent
factors. In the results section, prior to conducting the SEM, the
strategy
adopted (e.g., the Two Step), matrix used (e.g., covariance matrix),
and
algorithm employed (e.g., MLE) should be reported. The fit indices that
should
be reported remain an issue of debate. The fit criteria of Hu and
Bentler
(1999) have become widely adopted in the cases of path analysis and
confirmatory factor analysis. It is recommended that a balance be
struck
between model fit and direct effects. With regard to direct effects,
the results
should contain all the standardized coefficients and their associated p values. Correlations between error
terms should also be reported with an explanation of why they were set
to
correlate. These values, together with the standardized estimates of
the
parameters, are reported in a figure, much like the one in Figure 1.
Finally,
the covariance matrix should be included in the appendix, so that
researchers
can reproduce the original solutions of authors.
Conclusion
SEM
provides a powerful approach to
hypothesis testing. It is becoming increasingly popular and, in many
cases
(even in cognitive psychology where experimentation prevails), it is
replacing
conventional data analytic techniques, such as exploratory factor
analysis and
multiple regression. Because SEM encourages researchers to explicitly
state
their theories and hypotheses in an a
priori manner, it often leads to more comprehensive and
precise theoretical
statements. The aim
of this short
exposition was to bring this flexible and powerful class of modeling
methods to
the attention of substantive researchers, in the hope that they will
prove a
useful data analytic tool in the study of the developmental pathways of
stuttering. Appendix A takes the reader through a LISREL analysis of
manufactured
data (these data are presented in Appendix B).
Acknowledgement. The third and the final authors are supported by the Wellcome Trust grant 072639.
References
Baron,
R 90. M., & Kenny, D. A. (1986). The
moderator-mediator variable distinction in social psychological
research:
Conceptual, strategic and statistical considerations. Journal
of Personality and Social Psychology, 51,
1173-1182.
Bentler,
P. M. (1990). Comparative fit indices in structural
models. Psychological Bulletin, 107, 238 – 246.
Boomsma,
A. (2000). Reporting
analyses of covariance structures. Structural
Equation Modeling: A Multidisciplinary Journal, 7,
461-483.
Browne,
M. W., & Cudeck, R. (1993). Alternative ways of assessing
model fit. In K. A. Bollen & J. S. Long (Eds.), Testing
Structural Equation Models.
Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9, 233-255.
Cliff,
N. (1983). Some cautions concerning the application of causal
modelling methods. Multivariate
Behavioral Research, 18, 115-126.
Cudeck, R. (1989). Analysis of correlation matrices using covariance structure models. Psychological Bulletin, 105, 317-327.
Earley,
P. C., & Lituchy, T. R. (1991) Delineating Goal and
Efficacy Effects: A Test of Three Models. Journal
of Applied Psychology. 76
81-98.
Farrell, A.
D., & Sullivan, T. N. (2000). Structure of the Weinberger
Adjustment
Inventory Self-Restraint Scale and its relation to problem behaviors in
adolescence.
Psychological Assessment, 12,
394 -401.
Furnham, A., & Davies, S. (2004). Involvement of social factors in stuttering: A review and assessment of current methodology. Stammering Research, 1, 112-122.
Fife-Schaw.
C. (2000). Introduction to Structural Equation Modeling.
In G. M. Breakwell,
Gerbing, D. W., &
Anderson, J. C. (1988). An updated paradigm for scale development
incorporating
unidimensionality and its assessment. Journal
of Marketing Research, 25,
186-192.
Hayduk, L. A., & Glaser, D. N. (2000). Jiving the four step, waltzing around factor analysis and other serious fun. Structural Equation Modeling, 7, 1-35.
Hershberger, S. L. (2003). The Growth of Structural Equation Modeling: 1994-2001. Structural Equation Modeling, 10, 35 -47.
Hoyle, R. H. (1995). The
Structural
Equation Modeling Approach: Basic concepts and fundamental issues.
In R.
H. Hoyle (Ed.), Structural equation
modeling: Concepts Issues and Applications (pp. 1-18).
Hoyle, R. H., &
Panter, (1995).
Writing about structural equation models. In R. H. Hoyle
(Ed.), Structural equation modeling: Concepts
Issues and Applications (pp. 158-176).
Hu, L., & Bentler,
P. M.
(1995). Evaluating model fit. In R. H. Hoyle (Ed.), Structural Equation Modeling: Concepts Issues and
Applications (pp.
76-99).
Hu, L., & Bentler, P. M. (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424-453.
Hu, L., & Bentler,
P. M.
(1999). Cutoff criteria for fit indexes in covariance structure
analysis: Conventional
criteria vs. new alternatives. Structural
Equation Modeling, 6, 1-55.
Jaccard,
J., & Wan, C. K. (1995). Measurement error in the
analysis of interaction effects between continuous predictors using
multiple
regression: Multiple indicator and structural equation approaches. Psychological Bulletin, 116,
348-357.
Jöreskog,
K.G. (1969). A general approach to confirmatory maximum
likelihood factor analysis. Psychometrika,
34, 183-202.
Jöreskog,
K.G. (1971). Simultaneous factor analysis in several
populations. Psychometrika, 36, 409-426.
Jöreskog,
K., & Sörbom, D. (1996). LISREL 8:
User’s reference guide. SSI Scientific software
International.
Kline,
R. B. (2004). Principles
and Practice of Structural Equation Modeling (Second Edition).
NY:
Kolk,
H., & Postma, A. (1997). Stuttering as a covert repair
phenomenon. In R. F. Curlee & G. Siegel, (Eds.). Nature
and treatments of stuttering: New directions.
Levine,
S. Z., & Jackson, C. J. (2004). Eysenck’s theory of crime
revisited: Factors or primary scales? Legal
and Criminological Psychology, 9,
135-152.
Little,
T. D., Cunningham, Shahar, G., Widaman, K. F. (2002). To
parcel or not to parcel: Exploring the question, weighing the merits. Structural Equation Modeling, 9,
151-173.
MacCallum,
R. C., Wegener, D. T., Uchino, B. N., & Fabrigar, L. R.
(1993). The problem of equivalent models in applications of covariance
structure analysis. Psychological
Bulletin, 114, 185-199
MacCallum,
R. C., Widaman, K. F., Zhang, S., & Hong, S. (1999).
Sample size in factor analysis. Psychological
Methods, 4, 84-99.
Maruyama,
G. M. (1998). .Basics
of Structural Equation Modeling.
Moulder,
B. C., & Algina, J. (2002). Comparison of methods for
estimating and testing latent variable interactions. Structural
Equation Modeling,
9, 1-19.
Mulaik, S. A., & Millsap, R. E. (2000). Doing the four step right. Structural Equation Modeling, 7, 36-73.
Munck,
Muthén,
B. (2002). Beyond SEM: General latent variable modeling. Behaviormetrika,
29, 81-117.
Olsson,
U. H., Foss, T., Troye, S. V., & Howell, R. D. (2000). The
performance of ML, GLS, and WLS estimation in structural equation
modeling
under conditions of misspecification and nonnormality. Structural
Equation Modeling, 7,
557-595.
Quintana,
S. M., & Maxell, S. E. (1999). Implications of recent
developments in structural equation modeling for counseling psychology.
The Counseling Psychologist, 27,
485-587.
Petrides,
K. V., Furnham, A., Jackson, C. J., & Levine, S. Z.
(2003). Exploring issues of personality measurement and structure
through the
development of a revised short form of the Eysenck Personality
Profiler. Journal of Personality Assessment,
81, 272-281.
Raykov,
T., & Marcoulides, G.A. (1999). On desirability of
parsimony in structural equation model selection. Structural
Equation Modeling, 6,
292-300.
Raykov,
T., & Marcoulides, G. A. (2001). Can there be infinitely many
models equivalent to a given covariance structure model? Structural
Equation Modeling, 8, 142-149.
Riley,
G. D. (1994). The
Stuttering Severity Instrument for Adults and Children (third
edition).
Schumacker,
R. E. (2002). Latent variable interaction modeling. Structural
Equation Modeling, 9,
40-54.
Smith,
A. (1999). Stuttering: A unified approach to a multifactorial
dynamic disorder. In N. Bernstein Ratner & E. C. Healey (Eds.),
Stuttering research and practice:
Bridging
the gap. Mahwah:
Smith, A., & Kleinow, J. (2000). Kinematic correlates of speaking rate changes in stuttering and normally fluent adults. Journal of Speech, Language, Hearing Research, 43, 521-536.
Werts, C. E., Rock, D. R.,
Linn, R. L., & Jöreskog,
K.
G. (1978). A general method of estimating the reliability of a
composite. Educational and Psychological
Measurement,
38, 933-938.
Wright,
S. (1934). The method of
path coefficients. Annals of Mathematical
Statistics, 5, 161-215.
Appendix
A
This Appendix takes readers through the steps involved in a LISREL 8.3 analysis (software published by Scientific Software International, Inc.). The problem for analysis was based on the IQ/bullying examples used in the body of the text using model one as a basis (it was not possible to test the other model in the paper as it did not converge.) As shown in Figure 1, model one has three latent variables (IQ, stuttering severity, SS and bullying BU) which correspond to ach (achievement), ss (stuttering severity) and vic (victimization) in the current analysis. According to model one, IQ is reflected in vo (vocabulary) si (similarities) and co (comprehension) test scores and the equivalent here is that achievement is reflected in IQ and GPA scores. In model one, overall stuttering severity (ss) is reflected in self (Srs), peer (Prs) and teacher’s (Trs) reports of severity of stuttering, and here SS is reflected in parent, child and clinical (PARENT, CHILD SSITOTAL, respectively) scores. The final latent variable of bullying (Bu) in model one is reflected in self (srb) peer (prb) and teacher (trb) reports of bullying and the measure of bullying here (vic) is reflected in peer (VICTIM), parent (VICTIM2) and teacher’s assessments of bullying (SBS).
The data are made up for this exercise and, therefore, the conclusions from the analysis have no meaning. The data are given in Appendix B so that readers can reproduce the analysis if they wish. The data contain the following observed variables (labels are those used in the LISREL analysis that follows).
SSITOTAL - clinical assessment of stuttering severity - high scores
= more severe
PARENT - parent report of stuttering severity - high scores =
more severe
CHILD - child report of stuttering severity - high scores = more
severe
IQ
-
score on the Otis-Lennon Mental Abilities Test
GPA
- grade point average
SBS - teacher report of bullying - low scores = bullied a lot
VICTIM - peer reports of bullying - high scores = more
nominations as
bully victim
VICTIM2
- parent report of bullying - high scores = bullied
a lot
The following is the SIMPLIS syntax used to produce the SEM in LISREL.
Title Testing a Path Model of
Stuttering
Observed
Variables: SSITOTAL
PARENT CHILD IQ GPA
SBS VICTIM VICTIM2
Covariance
Matrix from File Made.cov
Sample
Size: 112
Latent
Variables:
ach
ss vic
Relationships:
ach
-> IQ GPA
ss
-> PARENT CHILD SSITOTAL
vic
-> VICTIM VICTIM2 SBS
vic
= ach
ss
= vic
ss
= ach
Path
Diagram
End
of Problem
A description of each of the steps in the SIMPLIS code now follows. The code itself is given in Courier font and the comments are in Times New Roman:
- Title
Testing a Path Model of Stuttering – the keyword ‘TITLE’
gives a name to the analysis for later reference.
- Observed
Variables: SSITOTAL
PARENT CHILD IQ GPA SBS VICTIM VICTIM2 – The
command ‘Observed variables’ tells the program that the labels that
follow are the observed variables. For instance, the IQ measures are
scores on the Otis-Lennon Mental Abilities Test
- Covariance
Matrix from File Made.cov - This is the basic covariance
matrix which may be obtained by a package such as SPSS. The covariance
matrix has been saved in file Made.cov.
- Sample
Size: 112 – This tells the computer the number of
observation in the sample. A sample size of 112 is small for such
analyses. It is difficult to give concrete
advice as to what sample size is appropriate. One way is to take 200 as
an absolute minimum, but it should also be ensured that the ratio of
variables to factors is as high as possible. Naturally, like more
conventional forms of analysis Power Analysis may serve as a useful
guide.
- Latent
Variables: ach ss vic – This line identifies which are the
latent variables. There are three here, achievement (ach), severity of
stuttering (ss) and victimization (vic).
- ss
-> PARENT CHILD SSITOTAL – This line indicates
that the latent variable severity of stuttering ‘ss’, causes PARENT,
CHILD and SSITOTAL (parent, child and clinical assessment of severity
of stuttering, respectively).
- vic
-> VICTIM VICTIM2 SBS This line indicates that the
latent variable ‘vic’ (victimization), causes VICTIM VICTIM2 SBS (peer, parent
and teacher indications about bullying).
- vic
= ach This line indicates that the latent variable ‘vic’
(victim of bullying), causes the latent variable ach.
- ss
= vic - This line indicates that the latent variable ‘ss’
(severity of stuttering) causes the latent variable ‘vic’.
- ss = ach - This
line tells the program that latent severity of stuttering is caused by
achievement
- Path Diagram This
line tells the program to produce a path diagram.
- End of Problem This
line tells the program that it is the end of the problem.
LISREL then produces the following output (comments that have been added for explanation are given in bold Times New Roman font):
DATE:
TIME:
L I S R E
L 8.30
BY
Karl G. Jôreskog &
Dag Sôrbom
This program is published
exclusively by
Scientific Software
International, Inc.
Phone: (800)247-6113,
(847)675-0720, Fax: (847)675-2140
Copyright by Scientific Software
International, Inc., 1981-2000
Use of this program is subject to the
terms specified in the
Universal Copyright
Convention.
Website:
www.ssicentral.com
The following lines were
read from file C:\LISREL83\DATA\STUT.SPL:
Title
Testing a Path Model of Stuttering
Observed Variables: SSITOTAL PARENT
CHILD IQ GPA
SBS VICTIM VICTIM2
Covariance Matrix from
File Made.cov
Sample Size: 112
Latent Variables:
ach
ss vic
Relationships:
ach -> IQ GPA
ss
-> PARENT CHILD SSITOTAL
vic
-> VICTIM VICTIM2 SBS
vic
= ach
ss
= vic
ss
= ach
Path Diagram
End
of Problem
Sample
Size =
112
Testing
a Path Model of Stuttering
Covariance Matrix to be Analyzed
SSITOTAL
PARENT
CHILD
SBS
VICTIM
VICTIM2
--------
--------
--------
--------
--------
--------
SSITOTAL
68.65
PARENT
38.99
40.27
CHILD
29.91
21.19
27.62
SBS
-21.73
-13.11
-8.61
14.75
VICTIM
15.27
8.77
6.26
-5.16
6.46
VICTIM2
6.00
3.94
2.55
-2.80
1.66
1.38
IQ
-83.65
-58.84
-42.09
27.95
-17.62
-7.81
GPA
-4.95
-3.66
-2.47
1.67
-1.02
-0.43
Covariance
Matrix to be Analyzed
IQ
GPA
--------
--------
IQ
170.13
GPA
7.83
0.70
Testing a Path Model of
Stuttering
Number
of Iterations = 35
LISREL
Estimates (Maximum Likelihood)
Measurement Equations
Below
are the measurement
equations
SSITOTAL
= 7.77*ss, Errorvar.= 8.32 , R² =
0.88
(2.19)
3.81
PARENT = 5.15*ss, Errorvar.= 13.72, R² =
0.66
(0.42)
(2.08)
12.38
6.61
CHILD = 3.76*ss, Errorvar.= 13.45, R² =
0.51
(0.39)
(1.92)
9.70
7.00
SBS = 2.90*vic, Errorvar.= 6.34 , R² =
0.57
(1.05)
6.04
VICTIM =
- 1.96*vic, Errorvar.= 2.62 , R² = 0.59
(0.24)
(0.45)
-8.01
5.87
VICTIM2 =
- 0.85*vic, Errorvar.= 0.65 , R² = 0.53
(0.11)
(0.10)
-7.54
6.27
IQ = 11.44*ach, Errorvar.= 39.16, R² =
0.77
(1.02)
(8.69)
11.25
4.51
GPA = 0.68*ach, Errorvar.= 0.23
, R² = 0.67
(0.067)
(0.040)
10.17
5.84
These
are the relationships
between the latent variables.
Structural Equations
ss =
- 0.46*vic - 0.59*ach, Errorvar.= 0.0098, R² = 0.99
(0.12)
(0.12)
(0.037)
-3.78
-5.01
0.26
vic = 0.80*ach, Errorvar.= 0.36 , R² =
0.64
(0.11)
(0.12)
6.97
3.12
Reduced Form Equations
ss =
- 0.96*ach, Errorvar.= 0.088, R² = 0.91
(0.082)
-11.60
vic = 0.80*ach, Errorvar.= 0.36, R² =
0.64
(0.11)
6.97
Correlation Matrix of Independent
Variables
ach
--------
1.00
Covariance Matrix of Latent
Variables
ss
vic
ach
--------
--------
--------
ss
1.00
vic
-0.93
1.00
ach
-0.96
0.80
1.00
These are the ‘Goodness of Fit
Indices’
Goodness of Fit
Statistics
Degrees of Freedom
= 17
Minimum Fit Function Chi-Square
= 25.31 (P = 0.088)
Normal Theory Weighted Least Squares
Chi-Square = 24.55 (P = 0.11)
Estimated Non-centrality
Parameter (NCP) = 7.55
90 Percent Confidence Interval
for NCP = (0.0 ; 24.86)
Minimum Fit Function
Value = 0.23
Population Discrepancy Function
Value (F0) = 0.068
90 Percent Confidence Interval
for F0 = (0.0 ; 0.22)
Root Mean Square Error of
Approximation (RMSEA) = 0.063
90 Percent Confidence Interval for
RMSEA = (0.0 ; 0.11)
P-Value
for Test of Close Fit (RMSEA <
0.05) = 0.32
Expected Cross-Validation
Index (ECVI) = 0.56
90 Percent Confidence Interval for
ECVI = (0.50 ; 0.72)
ECVI for Saturated
Model = 0.65
ECVI
for
Chi-Square for
Model AIC =
62.55
Saturated
AIC = 72.00
Model CAIC =
133.20
Saturated CAIC =
205.87
Normed Fit Index
(NFI) = 0.96
Non-Normed
Fit Index (NNFI) = 0.98
Parsimony Normed Fit Index
(PNFI) = 0.58
Comparative Fit Index
(CFI) = 0.99
Incremental Fit Index
(IFI) = 0.99
Relative Fit Index
(RFI) = 0.93
Critical N (CN) =
147.56
Root Mean Square Residual
(RMR) = 0.78
Standardized RMR =
0.033
Goodness of Fit Index
(GFI) = 0.95
Adjusted Goodness of Fit
Index (AGFI) = 0.89
Parsimony Goodness of Fit
Index (PGFI) = 0.45
These
are the suggested
parameters to add, i.e., the modification indices.
The Modification Indices Suggest to Add
the
Path to
from
Decrease in
Chi-Square New
Estimate
SSITOTAL
vic
7.9
-5.85
The
Modification Indices Suggest to Add an
Error Covariance
Between
and
Decrease in
Chi-Square New
Estimate
VICTIM
SSITOTAL
9.6
2.30
The Problem used
12760 Bytes (=
0.0% of Available Workspace)
Time used:
0.047 Seconds
Path analysis
The path
diagram with standardized estimates is produced in the final step in
the
analysis. Note that all direct effects are significant with the
exception of SS
to SSITOTAL and the latent variable relating vic to SBS. The latent
variable SS
reflects estimates of child and parent indications about severity of
the problem,
but not clinical indications about severity of stuttering. The model
indicates
that both achievement and victimization influence severity of
stuttering, and
that achievement has both a direct effect on severity of stuttering and
an
indirect effect that is mediated by victimization. This model was
estimated
using a covariance matrix with maximum likelihood estimation. It
appeared to
fit the data relatively well (according to the criteria of Hu &
Bentler,
1999) although it is limited by the small sample size (c2
= 24.55,
df=17, P =0.11, RMSEA=0.06; Standardized RMR = 0.03, CFI = 0.99, NNFI =
0.98).
Appendix
B
Fictitious data compiled for modeling purposes only. These are included so that readers may reproduce the results or practice developing different forms of SEM analysis.
SSI = clinical assessment of stuttering severity - high scores = more severe; Parent = parent report of stuttering severity - high scores = more severe; Child = child report of stuttering severity - high scores = more severe; IQ = score on the Otis-Lennon Mental Abilities Test; GPA = grade point average; SBS (social behavior at school checklist) = teacher report of bullying - low scores = more bullying; Victim1 = peer reports of bullying - high scores = more nominations as bully victim; Victim2 = parent report of bullying - high score = more bullying.
|
SSI |
Parent |
Child |
IQ |
GPA |
SBS |
Victim1 |
Victim2 |
|
30 |
27 |
28 |
79 |
1.67 |
29 |
1 |
2 |
|
23 |
21 |
20 |
108 |
4.00 |
38 |
1 |
1 |
|
14 |
10 |
11 |
131 |
3.75 |
42 |
0 |
1 |
|
25 |
10 |
25 |
100 |
2.50 |
40 |
2 |
1 |
|
18 |
20 |
19 |
107 |
3.50 |
41 |
1 |
1 |
|
12 |
13 |
17 |
115 |
4.00 |
42 |
0 |
1 |
|
15 |
14 |
19 |
92 |
2.23 |
42 |
1 |
2 |
|
8 |
10 |
8 |
111 |
3.00 |
41 |
1 |
1 |
|
9 |
17 |
17 |
95 |
3.00 |
40 |
0 |
1 |
|
6 |
14 |
13 |
106 |
3.75 |
41 |
0 |
1 |
|
9 |
9 |
9 |
107 |
3.00 |
42 |
1 |
2 |
|
29 |
25 |
22 |
88 |
2.25 |
38 |
3 |
3 |
|
19 |
18 |
15 |
109 |
2.25 |
42 |
1 |
3 |
|
19 |
14 |
8 |
102 |
2.75 |
40 |
2 |
1 |
|
25 |
20 |
34 |
81 |
2.00 |
39 |
3 |
2 |
|
36 |
23 |
20 |
93 |
2.00 |
35 |
7 |
2 |
|
19 |
17 |
18 |
106 |
2.75 |
38 |
1 |
1 |
|
30 |
20 |
20 |
97 |
2.67 |
25 |
4 |
5 |
|
21 |
10 |
20 |
112 |
3.00 |
42 |
2 |
2 |
|
12 |
15 |
15 |
95 |
2.75 |
40 |
0 |
1 |
|
29 |
39 |
33 |
86 |
1.00 |
33 |
1 |
2 |
|
24 |
34 |
22 |
90 |
2.50 |
40 |
0 |
3 |
|
15 |
18 |
13 |
118 |
3.00 |
42 |
1 |
2 |
|
13 |
20 |
21 |
115 |
3.75 |
42 |
0 |
1 |
|
11 |
19 |
16 |
105 |
2.75 |
40 |
0 |
1 |
|
22 |
26 |
18 |
81 |
1.50 |
41 |
5 |
2 |
|
35 |
35 |
26 |
83 |
.67 |
30 |
6 |
3 |
|
27 |
22 |
29 |
85 |
1.75 |
40 |
1 |
3 |
|
24 |
19 |
25 |
83 |
2.75 |
38 |
0 |
5 |
|
20 |
19 |
16 |
105 |
2.50 |
39 |
1 |
1 |
|
8 |
18 |
20 |
115 |
4.00 |
42 |
1 |
2 |
|
29 |
24 |
30 |
102 |
2.75 |
42 |
3 |
3 |
|
28 |
32 |
17 |
81 |
1.50 |
40 |
3 |
2 |
|
16 |
13 |
15 |
118 |
3.00 |
40 |
1 |
2 |
|
9 |
17 |
15 |
120 |
4.00 |
42 |
0 |
1 |
|
8 |
13 |
11 |
129 |
3.75 |
41 |
1 |
2 |
|
11 |
19 |
9 |
111 |
3.00 |
40 |
0 |
2 |
|
19 |
20 |
10 |
100 |
2.50 |
35 |
0 |
2 |
|
11 |
17 |
22 |
109 |
2.25 |
40 |
1 |
1 |
|
25 |
25 |
29 |
81 |
2.00 |
39 |
1 |
1 |
|
12 |
20 |
17 |
112 |
3.00 |
42 |
1 |
2 |
|
31 |
22 |
18 |
93 |
2.00 |
40 |
2 |
1 |
|
29 |
30 |
22 |
86 |
1.00 |
36 |
5 |
3 |
|
30 |
26 |
24 |
85 |
2.50 |
32 |
5 |
3 |
|
18 |
25 |
12 |
97 |
2.67 |
42 |
1 |
1 |
|
37 |
23 |
22 |
85 |
1.75 |
31 |
7 |
2 |
|
28 |
26 |
22 |
79 |
1.67 |
35 |
3 |
4 |
|
20 |
19 |
15 |
95 |
3.00 |
42 |
2 |
4 |
|
33 |
26 |
24 |
90 |
2.50 |
40 |
7 |
2 |
|
34 |
27 |
24 |
83 |
.67 |
39 |
8 |
2 |
|
34 |
25 |
22 |
92 |
2.23 |
38 |
4 |
3 |
|
16 |
14 |
19 |
115 |
3.75 |
42 |
2 |
2 |
|
28 |
26 |
22 |
88 |
2.25 |
40 |
2 |
2 |
|
18 |
10 |
13 |
106 |
2.75 |
42 |
2 |
1 |
|
10 |
15 |
13 |
108 |
4.00 |
41 |
0 |
1 |
|
11 |
14 |
11 |
115 |
3.75 |
41 |
0 |
2 |
|
17 |
19 |
16 |
106 |
3.75 |
40 |
2 |
3 |
|
25 |
27 |
23 |
124 |
2.00 |
37 |
6 |
3 |
|
24 |
27 |
14 |
95 |
2.75 |
36 |
2 |
3 |
|
25 |
15 |
13 |
85 |
2.75 |
32 |
2 |
4 |
|
22 |
19 |
21 |
99 |
3.50 |
42 |
1 |
2 |
|
28 |
24 |
21 |
99 |
2.67 |
40 |
8 |
3 |
|
28 |
22 |
21 |
95 |
2.75 |
36 |
4 |
3 |
|
25 |
24 |
18 |
92 |
3.50 |
39 |
2 |
2 |
|
36 |
32 |
25 |
83 |
2.75 |
30 |
9 |
5 |
|
22 |
20 |
19 |
95 |
1.50 |
40 |
2 |
2 |
|
25 |
23 |
21 |
92 |
2.23 |
39 |
4 |
3 |
|
23 |
21 |
15 |
105 |
2.75 |
35 |
1 |
2 |
|
30 |
24 |
23 |
85 |
1.50 |
34 |
2 |
3 |
|
8 |
15 |
16 |
115 |
4.00 |
42 |
0 |
1 |
|
14 |
17 |
15 |
111 |
3.00 |
42 |
4 |
1 |
|
18 |
16 |
18 |
112 |
3.00 |
41 |
1 |
2 |
|
11 |
15 |
17 |
118 |
3.00 |
42 |
0 |
2 |
|
23 |
18 |
14 |
100 |
2.50 |
36 |
0 |
3 |
|
25 |
32 |
21 |
95 |
1.50 |
36 |
2 |
3 |
|
27 |
22 |
23 |
88 |
2.25 |
31 |
1 |
2 |
|
24 |
27 |
26 |
86 |
1.00 |
42 |
0 |
2 |
|
29 |
30 |
22 |
85 |
1.75 |
40 |
4 |
3 |
|
10 |
13 |
17 |
115 |
3.75 |
42 |
1 |
1 |
|
8 |
10 |
12 |
129 |
3.75 |
41 |
0 |
2 |
|
15 |
22 |
11 |
106 |
2.75 |
40 |
2 |
1 |
|
31 |
24 |
21 |
90 |
2.50 |
30 |
10 |
4 |
|
15 |
25 |
23 |
99 |
3.00 |
41 |
4 |
2 |
|
34 |
28 |
23 |
81 |
2.00 |
31 |
9 |
5 |
|
32 |
35 |
25 |
79 |
1.67 |
33 |
6 |
4 |
|
30 |
28 |
24 |
93 |
2.00 |
40 |
5 |
3 |
|
18 |
19 |
18 |
109 |
2.25 |
42 |
2 |
2 |
|
26 |
25 |
23 |
83 |
2.75 |
40 |
4 |
3 |
|
28 |
26 |
23 |
85 |
2.50 |
38 |
6 |
2 |
|
26 |
25 |
17 |
85 |
2.75 |
40 |
3 |
3 |
|
33 |
30 |
25 |
83 |
.67 |
35 |
5 |
4 |
|
19 |
20 |
20 |
105 |
2.75 |
42 |
4 |
1 |
|
19 |
16 |
18 |
115 |
3.00 |
40 |
0 |
2 |
|
18 |
20 |
19 |
115 |
3.00 |
39 |
1 |
1 |
|
31 |
29 |
25 |
83 |
.67 |
31 |
6 |
6 |
|
9 |
15 |
18 |
115 |
4.00 |
42 |
1 |
1 |
|
25 |
22 |
22 |
99 |
3.00 |
40 |
2 |
2 |
|
11 |
19 |
15 |
115 |
4.00 |
42 |
0 |
2 |
|
12 |
15 |
16 |
115 |
3.75 |
42 |
0 |
3 |
|
24 |
21 |
21 |
99 |
2.75 |
41 |
4 |
2 |
|
29 |
26 |
22 |
88 |
2.25 |
40 |
5 |
3 |
|
23 |
20 |
19 |
109 |
2.25 |
38 |
2 |
2 |
|
26 |
22 |
22 |
100 |
2.50 |
35 |
0 |
3 |
|
40 |
32 |
29 |
85 |
1.75 |
29 |
10 |
6 |
|
30 |
25 |
22 |
93 |
2.00 |
37 |
6 |
5 |
|
31 |
22 |
22 |
90 |
2.50 |
36 |
5 |
2 |
|
20 |
15 |
18 |
118 |
3.00 |
41 |
0 |
1 |
|
30 |
33 |
23 |
95 |
1.50 |
35 |
0 |
4 |
|
29 |
22 |
23 |
92 |
3.50 |
38 |
5 |
3 |
|
28 |
30 |
21 |
92 |
2.23 |
38 |
4 |
3 |
|
35 |
29 |
30 |
86 |
1.00 |
30 |
8 |
4 |
|
25 |
19 |
22 |
97 |
2.67 |
40 |
2 |
2 |
The effect of using time intervals
of
different length on judgements about stuttering
Peter
Howell
Department of
Psychology, Centre for Human Communications, Institute of Cognitive
Neuroscience, and Institute of Movement Neuroscience, University
College
London, Gower St., London WC1E 6BT
Abstract.
Conventional
clinical procedures for assessment of stuttering are
reported to have poor reliability. Time interval analysis procedures
have been
reported to produce greater reliability than the conventional
procedures. In
time interval procedures, successive intervals of the same duration are
extracted from a sample of speech and judged by participants as
stuttered or
fluent. There is a problem insofar as the amount of speech judged
stuttered
depends on the length of the interval used. This problem is illustrated
in an
experiment in which 1-s and 5-s intervals were drawn from the same
samples of
speech and judged by participants as stuttered or fluent. It is also
shown that
the problem of lack of sensitivity when longer intervals are used is
more acute
for individuals who exhibit severe stuttering. Since ability to detect
changes
in stuttering rate is dependent on the length of interval used (as well
as
stuttering severity), the procedure can highlight or disguise changes
in
stuttering rate depending on parameterization of interval length and
choice of
participants to study. Thus, use of different length intervals across
studies
can distort whether particular treatments have an effect on speech
control.
Therefore, it is concluded that time interval analysis, as it is
currently
used, is an unsatisfactory procedure. If a standard-length interval
could be
agreed, comparison across studies or analyses would be possible. Keywords: Stuttering assessment, time
interval analysis procedure.
1. Introduction
A
conventional
method for assessing stuttering is to make a count of the disfluent
events
directly from recordings (by using, for instance, a manually operated
counter).
Studies have shown that agreement between judges using such methods is
only
around 60% (e.g., Curlee, 1981; Martin & Haroldson, 1981).
Ingham and his
colleagues have proposed that time-interval (TI) procedures produce
higher
agreement. In
TI
procedures applied to stuttering, listeners
hear (and in some cases see) a fixed-length extract of speech and
designate it
as stuttered (STUT) or fluent (FLU).
The
Ingham group claim that such procedures ".. could
certainly lead to
different ways of overcoming the problem of judgement reliability for
stuttering" (Ingham, Cordes & Gow, 1993, p.512). If this claim
is
true, then TI procedures should be used for clinical assessment in
preference
to the standard procedures. The current study examines the claim that
TI
procedures provide a reliable indication about stuttering.
Reliability can be defined as the relationship
between observed and true
scores (Cordes, 1994). Ingham and co-workers used expert judges to
establish
the true responses to each time interval. They employed
audio-visual
recordings of speech from a selected sample of speakers who stutter.
The samples
of speech from each speaker were divided up into adjacent,
non-overlapping 4-s
intervals and these were played to judges in semi-random order. Each
interval
was presented for judgement to the panel of four experienced judges who
independently assessed each sample twice. For each 4-s sample, these
judges
were instructed to identify whether the sample contained stuttering or
not
(Ingham et al., 1993, p.506). These data were then used to locate which
intervals the experienced judges agreed on (a criterion of 7/8
judgements given
the same response was used for this purpose) and to establish what the
agreed
response should be (the response given on the majority of occasions for
the
agreed intervals). 110 of the 143 intervals tested (77%) were agreed by
the
experienced judges and 64 of these were judged to be STUT.
A
major potential source of bias in such
procedures stems from the fact that intervals of different lengths have
been
used across different TI studies. Howell, Staveley, Sackin and Rustin
(1998)
pointed out that a) when long intervals are used, the chance of the
interval
containing signs of stuttering increases, and b) long intervals will
tend to
result in a ceiling effect with all intervals judged as being
stuttered. These
effects limit the use of TI procedures for assessing whether a
treatment
reduces stuttering rate. To make such an assessment, a researcher may
decide to
have participants’ speech judged before and after treatment. If an
interval-length is used which results in a ceiling effect before and
after
treatment, it would not be possible to detect a change due to the
treatment.
Changes due to the treatment may have been evident if a shorter
interval had
been used. A worked example shows this. Say there are two samples of
speech, each
100 s in length. One contains 30 stutterings and the other 20
stutterings and
in both cases these are evenly distributed over the sample1.
Assuming
a speech rate of five syllables per second (Perkins, 2001) this would
correspond to stuttering rates of 6%, and 4%, respectively. If the
samples are
partitioned into 5-s TIs, there will be an average of one stutter for
the 20
stutter sample and 1.5 for the 30 stutter sample. If these are reliably
judged,
100% of intervals will be designated STUT for both samples. This would
suggest
that there is no difference in stuttering rate between the samples.
Different
outcomes would be expected if the same samples were partitioned into
1-s TIs.
The 1-s TIs made from the first sample would include 30 stuttered
intervals and
the second sample 20 stuttered intervals and observers making accurate
judgements would reflect a 10% difference in stuttering rate between
the two
sets of intervals. The ceiling effect when the longer intervals are
used shows
that these intervals may fail to detect any fluency-enhancing effects
of
treatment procedures, which would have been evident if shorter
intervals had
been used. There are indications that this is more than a hypothetical
possibility. For instance, Ingham, Moglia, Frank, Costello-Ingham and
Cordes (1997) assessed the
effects of frequency-shifted feedback on the speech of people who
stutter using
5-s TIs. The long interval appears to have a) resulted in a ceiling
effect in
both the pre- and post-treatment conditions (both conditions showed
around 100%
intervals judged to be STUT), and b) led in turn to failure to find
effects of
the treatment that the majority of participants
reported (i.e., increased fluency under frequency-shifted feedback)
(Howell et
al., 1998).
This
analysis shows that the authors might
very well have found fluency-enhancing effects of frequency-shifted
feedback,
in line with what their participants reported, if they had used a
shorter (more
sensitive) TI. Clearly, a procedure that can result in misleading
conclusions
about treatment outcome is not satisfactory. To provide evidence on
this, the
study compared STUT/FLU judgements of the same material when it was
segmented
into 5-s and 1-s TIs. The earlier analysis predicts that more speech
intervals
will be judged STUT when the longer (5-s) intervals are used than when
the
shorter (1-s) intervals are used. The influence of interval length on
speakers
with different severity of stuttering is also examined. The above
analysis
predicts that speakers with more severe stutters will be less affected
(as a
higher proportion of their intervals will be judged stuttered whatever
the
length of interval used).
2.
Method
Speech
samples. Eight 2-minute samples of
speech from the UCLASS archive of stuttered speech (Howell &
Huckvale,
2004) were selected for use in this study. These represent one sample
each from
eight male speakers ranging in age from 7 years 7 months to 17 years 9
months.
The samples were chosen to cover a broad range of both age and
stuttering rate.
The samples can be down-loaded and examined by visiting
http://speech1.psychol.ucl.ac.uk/index.htm.
The samples employed were 0030_17y9m.1, 0061_14y8m.1, 0078_16y5m.1,
0095_7y7m.1, 0098_10y6m.1, 0138_13y3m.1, 0210_11y3m.1, 0234_9y9m.1 (as
listed
in the first column of Table 1). Further details about these (and
other) speech
samples from the UCLASS archive can be obtained from Howell and
Huckvale
(2004).
Procedure. A MATLAB program was
written to divide a file into either
1-s or 5-s intervals. The MATLAB program played one set of intervals
and
recorded listeners’ responses to each interval as STUT or FLU. The
program had an
option that allowed intervals to be replayed at random with no
replacement, or
in sequence. The responses of the experts were used to obtain intervals
that
were agreed as fluent (FLU) or stuttered (STUT)2.
These provided
criteria against which the responses of naïve judges were assessed.
These are
described in more detail below.
Selection
of intervals for
assessment and experts’ criteria judgements. Four expert
judges, each of
whom had 10 or more years’ intensive experience of assessing stuttered
speech,
were employed. The procedure for assessing the intervals was broadly
similar to
that which Ingham et al. (1993) adopted
(see the start of the introduction). The differences
in procedure were
1) that the experts judged each sample of speech both when it was
segmented
into 1-s, and into 5-s, TIs, 2) that judgements about the 1-s, and 5-s
TIs were
made twice, once when the TIs were sampled at random without
replacement, and
once when the TIs were presented in sequential order (the reasons for
the
latter are given below), and 3) they were instructed as to what events
should
be used to judge an interval as STUT, (this ensured that the task was
less
open-ended than in other studies).
The reason that judgements were made in sequential
and random order was
that in initial work with the experts, they indicated that their
designations
might well have differed if they had had the preceding interval
available as
context. (Howell, Sackin & Glenn, 1997 also found that
judgements were
better if some of the prior context, the preceding word in their
report, was
given before the judged extract.) The segmentation process led to two
artifacts: 1. Sometimes an interval started at a point in speech that
gave the
sound an apparent hard onset which led judges to rate it as STUT. 2.
Truncation
at the end of an interval sometimes made an interval sound as if the
first
sound was part of a repetition sequence. In both cases the sound would
have
been designated STUT while hearing the preceding context would have led
the
judge to revise their view and designate the interval as FLU. There is
a five
times greater chance of these artifacts affecting 1-s than 5-s
intervals. A 5-s
TI was dropped when the judgement about one or more of the five 1-s
intervals
that made up the 5-s interval was judged STUT while the 5-s interval
itself was
judged FLU (indicative of these problems) in more than one of the eight
judgements made by the experts.
It is also possible that there are cases where
stutterings were not
apparent on 1-s intervals but were on 5-s intervals. An example would
be where
a prolonged phone is split between two adjacent 1-s TIs leaving each
section of
the prolonged phone below the duration-threshold of a prolongation that
then
leads to each interval judged FLU. A 5-s TI was dropped when the
judgement
about the 5-s interval was STUT while all the five 1-s intervals it
contained
were judged FLU (indicative of such a problem) for more than one of the
eight
judgements made by the experts.
To ensure consistency in what events were considered
as STUT, the
experts were told:
The experts were also allowed to indicate any
intervals that they
considered to be ambiguous with repect to STUT/FLU status. One
situation where
this arose was when the duration of a sound was prolonged only
marginally as
this could have been done for emphasis or might have been a brief
period of
disfluency.
Ingham et al.’s (1993) procedure depended on the
experts using their own
judgement as to what was, or was not, stuttered. This would inevitably
lead to
difference of opinion and, for this reason, the procedure employed here
where
judges were told what events to consider STUT was considered
preferable.
The experts next indicated when there were
extraneous sounds, as these
might have affected productions and/or judgements in intervals that
contained
them. These usually arose because the participant knocked the
microphone or
some other object in the recording environment. There were also
occasional
prompt questions by the researcher making the recordings (when, for
instance, the
speaker ran out of things to say). These intervals were noted and
intervals
with these events were not included in the results of the naïve judges.
During any one session, judgements were made about
either the 1-s or 5-s
set of intervals (interval length) and the intervals were either
presented in
random, or sequential, order (order type). The speech of each person
who
stutters was assessed in turn. The interval length by order type
judgements
were carried out in different random orders by the expert judges, and
the
judgements were separated by at least a week to prevent carry-over
effects on
judgements. The experts indicated which rejection criteron applied to
an
excluded interval. The numbers of 5-s intervals excluded by the
different
criteria discussed above are summarised in Table 1. 86 5-s intervals
were
excluded in total (corresponding to 430 1-s intervals).
Table
1. Summary of number of 5-s intervals that
were excluded (under the column labeled N) and the percentage of total
speech
available this represents (under the column labeled %) for the
exclusion
criteria indicated at the left of each row.
|
|
N |
% |
|
Artefacts
of segmentation process: |
|
|
|
5-s FLU, one of the five 1s-intervals agreed
STUT |
24 |
12 |
|
5-s
STUT, all of the five 1s-intervals agreed FLU |
10 |
5 |
|
Ambiguous |
28 |
14 |
|
Extraneous
noises |
24 |
12 |
|
Total
|
86 |
43 |
Note
that there were more cases
where the 5-s interval was judged FLU but one of the constituent 1-s
intervals
was judged STUT (24) than vice versa (10). This suggests that the hard
onset
and spurious repetition artifacts were more prevalent than cases where
the more
extensive 5-s context allowed disfluencies to be detected that were
missed when
the shorter 1-s segments were used.
Some of the 1-s intervals that were agreed by the
experts occurred
within a 5-s interval that included other 1-s intervals that the
experts did
not agree on (six cases in total). All constituent 1-s intervals of a
5-s
interval were dropped when one or more of the 1-s intervals were not
agreed.
This permitted direct comparison – i.e. all 1-s intervals and the 5-s
interval
they comprised were agreed.
The exclusion criteria affected speakers
differentially and this
depended on the severity of their disorder. Table 2 gives some details
of how
the exclusion criteria affected individual speakers and how this
relates to
stuttering incidence in the 5-s intervals that remain. The overall
rejection
rate of 5-s samples was 46% (i.e., the 43% in Table 1 and the
additional 3%
where 5-s agreed intervals contained at least one 1-s interval not
agreed on).
The same applies (though to a lesser extent) to Ingham et al. (1993).
As the
main point here is to evaluate the TI procedure for intervals that are
precisely
defined, more strict criteria were applied (leading to the higher
rejection
rate than in Ingham et al., 1993).
Table
2. Speakers are labeled at the top of the
Table. The first row gives the duration (in s) of the original file
(based on
5-s intervals) and results for across speakers (labeled N). The second
row
indicates the duration (again in s) after all exclusion criteria were
applied.
The amount of speech lost (in s) is given in the third row (i.e. the
difference
between what was available initially and after the exclusion criteria
were
applied) and row four gives this as a percentage of the total material
available. Row five gives the TIs that were agreed to be STUT (in s)
for the
data after the exclusion criteria were applied and row six represents
this as a
percentage of all material (row two).
|
|
Speaker |
||||||||
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
N |
|
Initial duration |
125 |
130 |
125 |
125 |
120 |
130 |
125 |
120
|
1000 |
|
After
exclusion |
55 |
105 |
75 |
75 |
55 |
40 |
85 |
50 |
540 |
|
Lost |
70 |
25 |
50 |
50 |
65 |
90 |
40 |
70 |
460 |
|
%age
lost |
56 |
19 |
40 |
40 |
54 |
69 |
32 |
58 |
46 |
|
TIs
where judged STUT |
40 |
105 |
45 |
75 |
5 |
20 |
70 |
45 |
405 |
|
%age
stuttered |
73 |
100 |
60 |
100 |
9 |
50 |
82 |
90 |
75 |
The experts’ judgements served two roles: 1) to select intervals for testing with the naïve judges as described in the next section; 2) to determine whether the selected intervals were responded to correctly by naïve judges (correspond with experts’ responses) or not (did not correspond with experts’ responses).
Assessment
of intervals by naïve
judges.
The
naïve judges were undergraduates aged betwen 20 and 22 from a variety
of
humanity disciplines who reported that they had no experience of
judging
stuttered speech (speech science students were explicitly excluded).
Eight
naïve judges assessed the sets for each interval length (1-s and 5-s
intervals)
in random order for each speaker separately in the same way as the
experts. The
two assessments of the same material (1-s or 5-s intervals) were done
at least
a week apart. TI judgements were made with intervals presented as with
the
experts. All intervals were judged (i.e., material in the row labeled
‘initial
duration’ in Table 2) but the results are only reported for those
intervals
that the experts agreed on (i.e., material in the row labeled ‘after
exclusion’
in Table 2). Using all material ensured that there was at least one FLU
interval agreed by the experts for each of the speakers, so judges
should have
used both available responses and the context in which they made these
judgements was the same as that of the experts so that judgements could
be
compared (Parducci, 1965).
3. Results
Prediction
one
The experts’ response
designations (STUT or FLU) for the agreed intervals, were used as the
criterion
against which to assess the accuracy of the naïve judges. The expert
judgements
for 1-s intervals excluded all five 1-s intervals when the 5-s interval
they
comprise was agreed by the experts to be FLU but one of its constituent
1-s TI
was judged STUT and also excluded all five 1-s intervals when the 5-s
interval
they comprise was agreed by the experts to be STUT but all the
constituent 1-s
TI were judged FLU. Cases were, however, included where 5-s intervals
were
agreed to be STUT by the expert judges which contained one or more 1-s
intervals that were agreed by the experts to be FLU (though at least
one 1-s interval
has to be expert-agreed STUT). This arises when the experts agree that
there is
a stuttering of less than 5-s in length and leads to agreed FLU 1s-TI
from the
expert judges for where there were no agreed FLU 5-s TIs for speakers 2
and 4
(see below where responses of each judge to each speaker are
presented).
Responses from 5-s intervals were converted to responses to 1-s intervals so that the results on different interval lengths could be compared directly. To do this, 1) a 5-s FLU interval was considered to be made up of five 1-s FLU intervals, and 2) a 5-s STUT interval was considered to be made up of five 1-s STUT intervals. The second assumption operationalizes the view that too much material is designated STUT when longer intervals are used (i.e., the main topic addressed in this paper). After this response translation, comparison can be made between intervals of different length. Table 3 gives the mean percentage correct (and SD) for the naïve judges separately for each speaker and separately for both interval lengths using the experts’ responses to the agreed intervals as the criterion (more extensive data from individual judges are given in Table A.1 at the end of this article).
Table
3. Mean percentage correct and standard
deviation across judges for 1-s and 5-s intervals and for each speaker.
|
|
1-s |
5-s |
||
|
Speaker |
Mean |
SD |
Mean |
SD |
|
S1 |
68.41 |
6.87 |
94.32 |
4.70 |
|
S2 |
79.40 |
2.83 |
91.07 |
5.37 |
|
S3 |
76.33 |
3.62 |
75.00 |
4.71 |
|
S4 |
79.83 |
3.93 |
90.83 |
4.95 |
|
S5 |
78.18 |
5.99 |
79.55 |
4.42 |
|
S6 |
83.44 |
5.17 |
98.44 |
4.42 |
|
S7 |
88.82 |
2.52 |
95.59 |
5.21 |
|
S8 |
87.25 |
3.37 |
93.75 |
5.18 |
The mean percentages for each speaker from Table 3 are presented in histogram form in Figure 1. Speakers are indicated on the abscissa, blue bars represent 1-s judgements and red bars represent 5-s judgements. It can be seem that for all but one speaker (S3), there were more 5-s intervals judged STUT than 1-s intervals judged STUT. Thus for seven out of eight of the speakers, the judgments about STUT across 1-s and 5-s intervals were in the expected direction according to prediction one.
Figure 1. Percent correct judgements of naïve judges (relative to experts’ agreed responses). Results are shown for individual speakers (labeled along the abscissa) separately for 1-s and 5-s TIs.
To test the first prediction statistically, STUT intervals alone were examined to see whether more TIs were judged STUT when the longer (5-s) intervals were used than when the shorter (1-s) intervals were used. A mixed model Analysis of Variance was employed with the within-groups factor of interval length (1-s versus 5-s) and the between-groups factor of speaker (the eight speakers whose speech was judged) and the dependent variable was proportion of intervals judged stuttered). There was a significant effect of interval length (F(1,56) = 20.4, p < .001) which arose because more speech was judged STUT when 5-s intervals were used than when 1-s intervals were used. There was also a significant effect of speaker (F(7,56) = 5.410, p < .001) which indicated total stuttering rate (across 1-s and 5-s TI) differed (i.e., showed the speakers differed in severity of stuttering). There was also an interaction between interval length and speaker (F(7,56) = 3.27, p < 0.01). This arose because the effect on stuttering rate of changing interval length (stuttering rate increase going from 1-s to 5-s intervals) depended on speaker. Inspection of individual speaker data revealed that more severe stutters showed less effect than milder ones. This effect is explored in more detail in the next section.
Prediction
two
The
second prediction tested was
that the participants who had a more severe stutter had less chance of
losing
intervals than milder ones. This
prediction was based on the fact that the first two exclusion criteria
in Table
1 would apply less to speakers with a severe stutter, as few of their
intervals
did not include a real stutter before and after these exclusion
criteria were
applied. Essentially this implies that the first two exclusion criteria
were
less applicable to speakers with a severe stutter than speakers with a
milder
stutter. This predicts that there should be a negative correlation
between
amount of speech lost from the expert judges and the percentage of TIs
judged
STUT after the exclusion criteria were applied (i.e. a one-tail
prediction).
The Pearson product moment correlation coefficient was -.512 which was
in the
correct direction but not significant (p
= .10, one tail) which is not surprising given the small N. The related
correlation
coefficient between amount of speech lost and total length of those 5-s
intervals designated STUT, correlated negatively r
= -.818, p = .013 with
an N of 8. This gives qualified support to the view that speakers with
a less
severe stutter lose more intervals due to the exclusion criteria than
speakers
with a more severe stutter. Thus, TI assessments using 5-s intervals
affect
speakers with different stuttering severity differentially.
Examination
of data from individual
judges and speakers
The data for the individual judges for each speaker are given in Table A.1. These data show which length intervals are judged more consistently by the naïve judges, and some additional information concerning variability between judges. Looking at interval length first, the right-most section gives the proportion of the total correct responses the naïve judges made (relative to the experts) separately for 1-s (first column of this section) and 5-s (middle column of this section) intervals and the signed difference between the two (5-s – 1-s). The majority of the latter signed differences are positive, which indicates that naïve judges were more consistent with the experts for long intervals.
Looking at variability across interval length (Table
A.1), there appears
to be higher numbers of false positives (i.e., calling FLU intervals
STUT) in
naïve judges’ ratings of the 1-s than the 5-s intervals. For example,
judge
four assessing speaker three had 29 out of the total 75 1-s intervals
rated as
STUT, which included 17 false positives. So 17/29 (more than 58%) of
this
judge’s STUT responses are wrong. For 5-s intervals for this same
speaker and
judge, 11 out of the total 15 intervals were rated as STUT, including
three false
positives. So 3/11 (27.27%) of this judges STUT responses were false
positives
for the 5-s intervals. 48 speaker by judge sets of data were available
for whom
this calculation was possible (there were no agreed fluent intervals
judged for
speakers two and four). For these data, the percentage of false
positives was
52.6% for the 1-s intervals and 16.9% for the 5-s intervals which was
highly
significant by related t test (t(47)
= 10.5, p < .001) Thus
judgements
about 1-s intervals are more prone to false positive STUT responses
than 5-s
intervals.
4.
Discussion
The main result is that estimated stuttering rate
depends on interval
length with longer intervals more likely to be judged STUT. The second
main finding
was that the effect of interval size depended on the speaker’s
stuttering
severity (more severe stutterers tend to have more intervals
consistently
judged ‘stuttered’ than milder ones). The experiment controlled for
decison
context between the expert judges that provided the criteria responses
and the naïve
judges by having both sets of judges assess all materials. If only
expert-agreed intervals had been assessed, different range and
frequency
effects would have applied to the different materials and this would
have
affected the responses given (Parducci, 1965). The results suggest that
users of
TI procedures should not have free choice over interval length;
otherwise the
results across clinics and with different clients are not comparable.
Moreover
use of long intervals (5-s and over) is not recommended for detecting
changes
in stuttering frequency across conditions as the procedures are
insensitive
even to large changes in
stuttering rate
(this applies to Ingham et al., 1997).
The
study points to major issues of
reliability of the TI procedure such as difficulty in selecting a good
number of
samples, particularly in fluent speech, with the rejection criteria
being as
they are; large proportion of false positives; inability of longer
intervals to
measure differences in more severe stuttering. Although TI procedures
can be
automated and would then provide a relatively efficient method for
assessing
speech, these problems rule out using these procedures in clinics. TI
procedures may have other uses with respect to stuttering, however. The
method
as applied in the current study has provided intervals which are
completely
fluent or contain one type of stuttering. These could be used for
training and
testing material for procedures that automatically count stuttering
events
(Howell & Huckvale, 2004). Also, the intervals can be used to
establish what
acoustic information is salient for detecting stutterings.
Acknowledgement.
This research was supported by
programme grant 072639
awarded by the
Wellcome Trust.
References
Cordes,
A. K. (1994). The reliability of observational data: I. Theories and
methods
for speech-language pathology. Journal of
Speech and Hearing Research, 37,
264-278.
Curlee,
R. F. (1981). Observer agreement on disfluency and stuttering. Journal of Speech and Hearing Research, 24, 595-600.
Howell,
P., & Huckvale, M. (2004). Facilities to assist people to
research into stammered speech. Stammering
Research, 1, 130-242.
Howell, P., Sackin,
S., & Glenn, K. (1997). Development of a two-stage procedure
for the
automatic recognition of dysfluencies in the speech of children who
stutter: I.
Psychometric procedures appropriate for selection of training material
for
lexical dysfluency classifiers. Journal
of Speech, Language and
Hearing Research,
40, 1073-1084.
Howell, P.,
Staveley, A., Sackin, S., & Rustin, L. (1998). Methods of
interval selection,
presence of noise and their effects on detectability of repetitions and
prolongations. Journal of the Acoustical
Society of
Ingham,
R. J., Cordes, A. K., & Gow, M. L. (1993). Time-interval
measurement of
stuttering: Modifying interjudge agreement. Journal
of Speech and Hearing Research, 36,
503-515.
Ingham, R. J.,
Moglia, R. A., Frank, P., Costello-Ingham, J., & Cordes, A,
(1997).
Experimental investigation of the effects of frequency-altered feedback
on the
speech of adults who stutter. Journal
of Speech, Language and Hearing Research, 40,
361-372.
Martin, R. R., &
Haroldson, S. K. (1981). Stuttering identification: Standard definition
and
moment of stuttering. Journal of Speech
and Hearing Research, 24,
59-63.
Parducci, A. (1965). Category judgement: A range-frequency model. Psychological Review, 72, 407-418.
Perkins,
W.H. (2001). Stuttering: A matter of bad timing. Science,
294, 786.
1. The
interval starts and ends (both 1-s
and 5-s) are imposed irrespective of where utterances start. Providing
interval
durations are as long as the utterance duration or longer, there will
be no
effect due to the tendency of stutterings to occur at the beginning of
sentences,
2. The
expert judges also indicated which of
the intervals contained part-word repetitions or prolongations alone.
These are
intended for use in future studies and are not reported here as the
procedures
and treatment of results parallel those that Ingham and co-workers used
in
their studies.
Table
A.1.
The three sections (going from
left to right) give results for 1-s, and for 5-s, TIs and comparisons
between
5-s and 1-s TIs. Speaker (S1-S8) and naïve judge (J1-J8) are labeled at
the
left of each row. The three rows under the two sections headed
‘Correct/total
1-s’ and ‘Correct/total 5-s’, are, form left to right, number of
responses the naïve
judge got correct relative to the expert for a) fluent TI (FLU), b)
stuttered
TI (STUT) and c) all TI (A). The section headed ‘Comparison of 5-s and
1-s’
gives (going left to right), the total proportion of TI that were
judged
correct for 1) 1-s, and 2) 5-s intervals and 3) the signed difference
between
the 5-s and 1-s proportions.
|
Correct/total 1-s |
Correct/total
5-s |
Comparison 5-s and 1-s |
|||||||||||||||
|
|
FLU |
STUT |
A |
FLU |
STUT |
A |
PropA 1-s |
PropA 5-s |
DiffA (5-s
- 1-s) |
||||||||
|
S1J1 |
39/49 |
3/6 |
42/55 |
8/8 |
3/3 |
55/55 |
0.76 |
1.00 |
0.33 |
||||||||
|
J2 |
33/49 |
4/6 |
37/55 |
8/8 |
2/3 |
50/55 |
0.67 |
0.91 |
0.24 |
||||||||
|
J3 |
39/49 |
3/6 |
42/55 |
7/8 |
3/3 |
50/55 |
0.76 |
0.91 |
0.15 |
||||||||
|
J4 |
34/49 |
3/6 |
37/55 |
8/8 |
3/3 |
55/55 |
0.67 |
1.00 |
0.33 |
||||||||
|
J5 |
35/49 |
4/6 |
39/55 |
8/8 |
3/3 |
55/55 |
0.71 |
1.00 |
0.29 |
||||||||
|
J6 |
34/49 |
3/6 |
37/55 |
8/8 |
2/3 |
50/55 |
0.67 |
0.91 |
0.14 |
||||||||
|
J7 |
33/49 |
4/6 |
37/55 |
7/8 |
3/3 |
50/55 |
0.67 |
0.91 |
0.14 |
||||||||
|
J8 |
26/49 |
4/6 |
30/55 |
8/8 |
2/3 |
50/55 |
0.55 |
0.91 |
0.36 |
||||||||
|
|
|
|
|
|
|
|
|
||||||||||
|
S2J1 |
33/42 |
52/63 |
85/105 |
|
20/21 |
100/105 |
0.81 |
0.95 |
0.14 |
||||||||
|
J2 |
35/42 |
53/63 |
88/105 |
|
19/21 |
95/105 |
0.84 |
0.90 |
0.06 |
||||||||
|
J3 |
31/42 |
47/63 |
78/105 |
|
17/21 |
85/105 |
0.74 |
0.80 |
0.06 |
||||||||
|
J4 |
33/42 |
50/63 |
83/105 |
|
20/21 |
100/105 |
0.79 |
0.95 |
0.16 |
||||||||
|
J5 |
34/42 |
50/63 |
84/105 |
|
19/21 |
95/105 |
0.80 |
0.90 |
0.10 |
||||||||
|
J6 |
33/42 |
52/63 |
85/105 |
|
20/21 |
100/105 |
0.81 |
0.95 |
0.14 |
||||||||
|
J7 |
32/42 |
49/63 |
81/105 |
|
20/21 |
100/105 |
0.77 |
0.95 |
0.18 |
||||||||
|
J8 |
33/42 |
50/63 |
83/105 |
|
18/21 |
90/105 |
0.79 |
0.86 |
0.07 |
||||||||
|
|
|
|
|
|
|
|
|
|
|||||||||
|
S3J1 |
46/60 |
13/15 |
59/75 |
4/6 |
7/9 |
55/75 |
0.79 |
0.73 |
0.06 |
||||||||
|
J2 |
50/60 |
12/15 |
62/75 |
5/6 |
5/9 |
50/75 |
0.83 |
0.67 |
0.16 |
||||||||
|
J3 |
43/60 |
11/15 |
54/75 |
4/6 |
7/9 |
55/75 |
0.72 |
0.73 |
0.01 |
||||||||
|
J4 |
43/60 |
12/15 |
55/75 |
3/6 |
8/9 |
55/75 |
0.73 |
0.73 |
0.00 |
||||||||
|
J5 |
44/60 |
12/15 |
56/75 |
5/6 |
7/9 |
60/75 |
0.75 |
0.80 |
0.05 |
||||||||
|
J6 |
46/60 |
13/15 |
59/75 |
5/6 |
6/9 |
55/75 |
0.79 |
0.73 |
0.06 |
||||||||
|
J7 |
45/60 |
13/15 |
58/75 |
6/6 |
6/9 |
60/75 |
0.77 |
0.80 |
0.03 |
||||||||
|
J8 |
43/60 |
12/15 |
55/75 |
5/6 |
7/9 |
60/75 |
0.73 |
0.80 |
0.07 |
||||||||
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
S4J1 |
32/39 |
30/36 |
62/75 |
|
13/15 |
65/75 |
0.83 |
0.87 |
0.04 |
||||||||
|
J2 |
30/39 |
26/36 |
56/75 |
|
15/15 |
75/75 |
0.75 |
1.00 |
0.25 |
||||||||
|
J3 |
31/39 |
30/36 |
61/75 |
|
13/15 |
65/75 |
0.81 |
0.87 |
0.06 |
||||||||
|
J4 |
33/39 |
30/36 |
63/75 |
|
13/15 |
65/75 |
0.84 |
0.87 |
0.03 |
||||||||
|
J5 |
29/39 |
26/36 |
55/75 |
|
14/15 |
70/75 |
0.73 |
0.93 |
0.20 |
||||||||
|
J6 |
31/39 |
28/36 |
59/75 |
|
14/15 |
70/75 |
0.79 |
0.93 |
0.14 |
||||||||
|
J7 |
32/39 |
30/36 |
62/75 |
|
13/15 |
65/75 |
0.83 |
0.87 |
0.04 |
||||||||
|
J8 |
31/39 |
30/36 |
61/75 |
|
14/15 |
70/75 |
0.81 |
0.93 |
0.12 |
||||||||
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
S5J1 |
39/53 |
0/2 |
39/55 |
7/10 |
1/1 |
40/55 |
0.71 |
0.73 |
0.02 |
||||||||
|
J2 |
47/53 |
2/2 |
49/55 |
9/10 |
1/1 |
50/55 |
0.89 |
0.91 |
0.02 |
||||||||
|
J3 |
43/53 |
1/2 |
44/55 |
8/10 |
1/1 |
45/55 |
0.80 |
0.82 |
0.02 |
||||||||
|
J4 |
42/53 |
1/2 |
43/55 |
7/10 |
1/1 |
40/55 |
0.78 |
0.73 |
-0.05 |
||||||||
|
J5 |
40/53 |
2/2 |
42/55 |
8/10 |
1/1 |
45/55 |
0.76 |
0.82 |
0.06 |
||||||||
|
J6 |
40/53 |
1/2 |
41/55 |
8/10 |
0/1 |
40/55 |
0.75 |
0.73 |
-0.02 |
||||||||
|
J7 |
39/53 |
1/2 |
40/55 |
7/10 |
1/1 |
40/55 |
0.73 |
0.73 |
0.00 |
||||||||
|
J8 |
44/53 |
2/2 |
46/55 |
9/10 |
1/1 |
50/55 |
0.84 |
0.91 |
0.07 |
||||||||
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
S6J1 |
26/34 |
6/6 |
32/40 |
4/4 |
4/4 |
40/40 |
0.80 |
1.00 |
0.20 |
||||||||
|
J2 |
30/34 |
6/6 |
36/40 |
4/4 |
4/4 |
40/40 |
0.90 |
1.00 |
0.10 |
||||||||
|
J3 |
27/34 |
6/6 |
33/40 |
4/4 |
4/4 |
40/40 |
0.83 |
1.00 |
0.17 |
||||||||
|
J4 |
28/34 |
6/6 |
34/40 |
4/4 |
4/4 |
40/40 |
0.85 |
1.00 |
0.15 |
||||||||
|
J5 |
25/34 |
5/6 |
30/40 |
4/4 |
¾ |
35/40 |
0.75 |
0.88 |
0.13 |
||||||||
|
J6 |
26/34 |
6/6 |
32/40 |
4/4 |
4/4 |
40/40 |
0.80 |
1.00 |
0.20 |
||||||||
|
J7 |
30/34 |
6/6 |
36/40 |
4/4 |
4/4 |
40/40 |
0.90 |
1.00 |
0.10 |
||||||||
|
J8 |
28/34 |
6/6 |
34/40 |
4/4 |
4/4 |
40/40 |
0.85 |
1.00 |
0.15 |
||||||||
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
S7J1 |
46/53 |
29/32 |
75/85 |
2/3 |
14/14 |
80/85 |
0.88 |
0.94 |
0.06 |
||||||||
|
J2 |
49/53 |
27/32 |
76/85 |
3/3 |
14/14 |
85/85 |
0.89 |
1.00 |
0.11 |
||||||||
|
J3 |
48/53 |
29/32 |
77/85 |
2/3 |
13/14 |
75/85 |
0.91 |
0.93 |
0.02 |
||||||||
|
J4 |
46/53 |
29/32 |
75/85 |
3/3 |
14/14 |
85/85 |
0.88 |
1.00 |
0.12 |
||||||||
|
J5 |
49/53 |
29/32 |
78/85 |
3/3 |
14/14 |
85/85 |
0.92 |
1.00 |
0.08 |
||||||||
|
J6 |
44/53 |
27/32 |
71/85 |
3/3 |
12/14 |
75/85 |
0.84 |
0.93 |
0.09 |
||||||||
|
J7 |
46/53 |
29/32 |
75/85 |
2/3 |
14/14 |
80/85 |
0.88 |
0.94 |
0.06 |
||||||||
|
J8 |
48/53 |
29/32 |
77/85 |
3/3 |
14/14 |
85/85 |
0.91 |
1.00 |
0.09 |
||||||||
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
S8J1 |
28/33 |
14/17 |
42/50 |
0/1 |
9/9 |
45/50 |
0.84 |
0.90 |
0.06 |
||||||||
|
J2 |
30/33 |
14/17 |
44/50 |
1/1 |
8/9 |
45/50 |
0.88 |
0.90 |
0.02 |
||||||||
|
J3 |
28/33 |
13/17 |
41/50 |
1/1 |
8/9 |
45/50 |
0.82 |
0.90 |
0.08 |
||||||||
|
J4 |
30/33 |
13/17 |
43/50 |
1/1 |
8/9 |
45/50 |
0.86 |
0.90 |
0.04 |
||||||||
|
J5 |
29/33 |
16/17 |
45/50 |
0/1 |
9/9 |
45/50 |
0.90 |
0.90 |
0.00 |
||||||||
|
J6 |
30/33 |
13/17 |
43/50 |
1/1 |
9/9 |
50/50 |
0.86 |
1.00 |
0.14 |
||||||||
|
J7 |
29/33 |
16/17 |
45/50 |
1/1 |
9/9 |
50/50 |
0.90 |
1.00 |
0.10 |
||||||||
|
J8 |
30/33 |
16/17 |
46/50 |
1/1 |
9/9 |
50/50 |
0.92 |
1.00 |
0.08 |
||||||||
Note:
As indicated in the text, there were cases
where 5-s intervals were agreed to be STUT by the expert judges which
contained
one or more 1-s intervals that were agreed by the experts to be FLU
(though at
least one 1-s interval was expert-agreed STUT). This arose when the
experts
agreed that there was a stuttering of less than 5-s in length. This
results in
agreed FLU 1-s TI from the expert judges for subjects 2 and 4 where
there were
no agreed FLU 5-s TIs.
The
impact
of word-end phonology and morphology on stuttering
Chloe
Centre for Developmental Language Disorders
and Cognitive Neuroscience,
Department of Human Communication Science,
Abstract.
This
paper investigates whether stuttering rates in
English-speaking adults and children are influenced by phonological and
morphological complexity at the ends of words. The phonology of English
inflection is such that morphological and phonological complexity are
confounded, and previous research has indicated that phonological
complexity influences
stuttering. Section 1 of this paper considers how to disentangle
phonological
and morphological complexity so that the impact of each on stuttering
can be
tested. Section 2 presents an analysis of some adult corpus data, and
shows
that phonological and morphological complexity at the word end do not
influence
stuttering rates for English-speaking adults, at least in spontaneous
speech. Section
3 presents results from a non-word repetition task and a past tense
elicitation
task which reveal that while word-end phonological and morphological
complexity
do not affect stuttering rates in most of the adults and children
tested, a
small proportion of adults and children do stutter over morphologically
complex
words in an elicitation task. Taken as a whole, these results suggest that morphology has an impact on
stuttering for some individuals in certain circumstances. Keywords:
Developmental
stuttering, word-end phonology, word end morphology.
1.1
Introduction
Several
recent models of stuttering
hypothesise that the linguistic characteristics of the word being
attempted can
trigger stuttering (e.g. Au-Yeung & Howell, 1998; Packman,
Onslow, Richard
& van Doorn, 1996). While the role of phonological factors in
stuttering
has received some attention in the literature, the impact of word-end
phonology
has not been studied independently from other aspects of phonology. Nor
has the
role of inflectional morphology (plural marking on nouns; tense,
agreement and
aspectual marking on verbs) in stuttering been investigated. Word-end
phonology
and inflectional morphology are considered together in this paper for
the
following reason: In English, inflectional morphology occurs at the
ends of
words, and changes the phonology of the word end. For example, the
plural form
of ‘cat’, cats,
ends in a cluster, whereas the singular form, cat,
does not. Similarly the verb ‘lie’
ends in a vowel in the forms I lie
and you lie, but in a consonant in
the past tense forms I lied and you lied. As a consequence of this close
relationship between
phonology and morphology, the impact of morphology on the production of
words
is best studied alongside that of phonology.
The
research questions addressed in this
paper are as follows: (1) does phonological complexity at the ends of
words
influence stuttering rates, (2) does inflectional morphology influence
stuttering rates, and (3) if morphology does influence stuttering, is
this effect
independent of phonological complexity? The paper attempts to respond
to Demuth’s
plea that the role of phonology and morphology in stuttering be
investigated
(Demuth, 2004).
1.2 The
impact of phonology on stutttering
Researchers
use differing measures of what
constitutes phonological or phonetic complexity, and therefore they
differ in
what phonological and phonetic characteristics they consider are likely
to
cause stuttering. As a rule these measures are based on what typically
developing children find hard to acquire. For example, Throneburg,
Yairi and
Paden (1994) investigated the impact of late developing sounds,
consonant
clusters and multisyllabicity. A more comprehensive metric is the Index
of
Phonetic Complexity (IPC; Jakielski, 1998; Weiss & Jakielski,
2001),
whereby words are scored according to how many ‘difficult’ structures
they
contain, with the most difficult words being those that get a high
score. The
IPC is outlined in Table 1 below.
Table 1. Index of Phonetic Complexity (Jakielski, 1998)
|
Factor |
No score |
One point each |
|
1.
consonant by place |
Labials,
coronals, glottals |
dorsals |
|
2.
consonant by manner |
stops,
nasals, glides |
fricatives,
affricates, liquids |
|
3.
singleton consonants by place |
Reduplicated |
variegated |
|
4.
vowel by class |
monophthongs,
diphthongs |
rhotics |
|
5.
word shape |
ends with a
vowel |
ends with a
consonant |
|
6.
word length (syllables) |
monosyllables,
disyllables |
three or
more syllables |
|
7.
contiguous consonants |
no clusters |
consonant
clusters |
|
8.
cluster by place |
Homorganic |
heterorganic |
Results
from these studies have been
mixed. Throneburg et al. (1994) found that neither late-developing
sounds,
consonant clusters nor multisyllabicity had any effect on stuttering
rates in
children aged between two and a half and five years. Howell and
Au-Yeung (1995)
confirmed this finding for a wider range of ages, up to twelve years
old.
The
IPC has been used in various
investigations (Dworzynski & Howell, 2004; Howell, Au-Yeung,
Yaruss &
Eldridge, submitted; Weiss & Jakielski, 2001). Weiss and
Jakielski (2001)
studied children between the ages of six and eleven and a half. They
found a
trend for prosodic complexity to influence stuttering rates more in
younger
children than in older children, although this difference was not
significant.
In contrast, Howell et al. (submitted) found that phonetic complexity
only
influenced stuttering rates in children older than eleven and in
adults.
Studies using the IPC have revealed different patterns
cross-linguistically.
For example, Dworzynski and Howell (2004) found that for German people
who
stutter (PWS), words ending in consonants are more likely to be
stuttered than
words ending in a vowel, and that this effect was present for both
adults and
children over the age of six. This effect was not found for English
speakers (Howell
et al., submitted), a finding that Dworzynski and Howell (2004) put
down to the
fact that words ending in consonants are more frequent in English than
in
German.
One
of the disadvantages of the IPC is
that, apart from the factor ‘word shape’ (i.e. whether the word ends in
a
consonant or a vowel), it does not specify where in the word phonetic
complexity occurs. It is well-established that word-initial sounds
trigger
disfluencies (e.g. Conture, 1990; Howell, Au-Yeung & Sackin,
2000; Natke,
Sandreiser, van Ark, Pietrowski & Kalveram, 2004; Wingate,
1982, 1988), but
the aim of this paper is to investigate the impact of the word end. Dworzynski and Howell
(2004) suggest that
word-final factors might indeed play a role in the planning and
retrieval time
of words, given that word shape affects German PWS. Otherwise, there is
as yet
little evidence that word-end phonology influences stuttering.
1.3 The
relationship between phonology and inflectional morphology
In
this paper I consider the impact not
only of phonological complexity at the word end, but also that of
inflectional
morphology. English morphology is sparse, with a mere two forms for
nouns (cat, cats)
and a maximum of five for verbs (sew,
sewed, sews,
sewing, sewn).
I am concerned only with suffixes
that add a consonant – past tense t/d
(e.g. hoped, sewed),
plural s/z
(e.g. cats, dogs)
and third person singular s/z
(e.g. hopes, sews).
By adding a consonant, these suffixes increase phonological
complexity at the word end, and may create clusters (compared sewed with hoped).
I leave aside the syllabic suffixes -ed
(e.g. wanted, needed),
-ez
(e.g. horses, reaches)
and -ing.
Any theory of phonological and morphological complexity should mirror what we know happens developmentally. Children’s first words end in a vowel (e.g. mama) and only later, at about the age of one and half or two do they produce words ending in a consonant, e.g. man. Only later still will they produce words ending in a consonant cluster. Therefore word-final clusters are more complex in some way than word-final singleton consonants. In terms of children’s early word productions, their first words are generally uninflected, and only later are inflections produced. -Ing and plural –s are among the first inflections used, while third person singular –s and past tense –ed take longer to become reliably established (see Bernhardt & Stemberger, 1998, for an overview of phonological development).
Howell
et al. (submitted) investigated the
effect that a word-end consonant has on stuttering rates in adults, and
found that
words ending in a consonant were actually stuttered less frequently
than those
ending in a vowel. The first indications would therefore appear to be
that
word-end morphology that adds a consonant might not actually be
relevant to
stuttering, at least in adults. However, the possibility that
morphology does
affect stuttering cannot be ruled out on the basis of Howell et al.’s
(submitted) data. To do so, a systematic analysis of phonological and
morphological factors at the word end is needed.
In
the analyses presented in this paper, word
forms that end in a singleton consonant are termed ‘phonologically
simple’
Such a classification enables the following questions to be raaised: (1) does phonological complexity at the ends of words influence stuttering rates independent of morphology, and (2) does inflectional morphology influence stuttering rates independent of phonology?
The pattern of results that might plausibly be expected to arise are now discussed. If phonological complexity at the ends of words influences stuttering rates independently of morphology, the following pattern of results would be predicted – higher stuttering rates on 3+4 (i.e. phonologically complex words) than on 1+2 (i.e. phonologically simple words), but no difference between 1 and 2 (i.e. between morphologically simple and complex words that are phonologically simple) and no difference between 3 and 4 (i.e. between morphologically simple and complex words that are phonologically complex) (see Figure 1a). If morphological complexity influences stuttering rates then higher stuttering rates would be seen on 2+4 (i.e. morphologically complex words) than on 1+3 (i.e. morphologically simple words). If morphological complexity influences stuttering independently of phonology, then stuttering rates would be expected to be equivalent for 2 (i.e. phonologically simple) and 4 (i.e. phonologically complex) (see Figure 1b). A third possibility is that there may be an interaction between phonological and morphological complexity, with stuttering rates highest for those words that are both phonologically and morphologically complex (Figure 1c).
Figure 1a.
Phonological complexity influences
stuttering independent of
phonology
Figure 1b. Morphological complexity
influences stuttering independent of phonology
Figure 1c. Interaction between phonological and morphological complexity
In order to investigate which of these patterns of performance is found amongst PWS, two types of data are considered. In Study 1, a corpus of teenage and adult speech is analysed, looking at the patterns of relative performance on phonologically and morphologically complex forms (as outlined above). In Studies 2 and 3, some experimental data from children and young adults, from a non-word repetition task and from a task designed to elicit inflected verbs are analysed. The non-word repetition task allows phonologically simple and complex non-words to be compared in a context that is by its very nature morphologically simple. The elicitation task allows comparison between phonologically simple and complex words in the presence of morphological complexity.
2.1
Study 1.
This
study is an analysis of
words selected from the University College London Archive of Stuttered
Speech
(UCLASS) data base (Howell & Huckvale, 2004). The material was
produced
spontaneously in response to prompt questions by an interviewer and
conforms to
“casual” speech in terms of Labov’s (1978) stylistic continuum in
sociolinguistics.
The speech was recorded from 16 male participants, diagnosed as showing
signs
of stuttering by a speech pathologist in a clinic in the
Table 2. Details of speakers, total number of words in the sample, number of words stuttered, percentage stuttering rate in the sample and age in years
|
Participant
number |
Number
of words |
Number
of words stuttered |
%
stuttered |
Age |
|
1 |
1420 |
37 |
2.61 |
47 |
|
2 |
1720 |
46 |
2.70 |
25 |
|
3 |
1016 |
31 |
3.10 |
38 |
|
4 |
1255 |
61 |
4.90 |
26 |
|
5 |
1390 |
76 |
5.47 |
28 |
|
6 |
188 |
11 |
5.90 |
24 |
|
7 |
219 |
14 |
6.40 |
18 |
|
8 |
221 |
15 |
6.80 |
16 |
|
9 |
338 |
23 |
6.80 |
17 |
|
10 |
1315 |
125 |
9.51 |
20 |
|
11 |
272 |
28 |
10.30 |
16 |
|
12 |
763 |
85 |
11.10 |
25 |
|
13 |
1593 |
179 |
11.24 |
21 |
|
14 |
378 |
43 |
11.40 |
16 |
|
15 |
943 |
154 |
16.33 |
35 |
|
16 |
1082 |
220 |
20.33 |
40 |
In this analysis stuttering rates are compared across four different types of words, whose characteristics (and some examples) are set out in Table 3 below. Phonologically, words differ only in whether they contain a word-final cluster. The words selected have either the shape Consonant-Vowel-Consonant (CVC) or Consonant-Vowel-Consonant-Consonant (CVCC), i.e. none contain onset clusters, and none consist of more than one syllable.
Table 3. Characteristics of the words
being
analysed and some examples
|
|
|
Morphology |
|
|
|
|
simple |
complex |
|
phonology |
simple |
bad, house |
died, days |
|
complex |
month, think |
looked, wants |
|
A Perl program (Brown, 2001) identified words ending in the letters s, z ,d and t. These were then hand-coded as being morphemic or otherwise. Irregular verbs were then removed from the data set, again by hand. These included irregular past tenses (e.g. hit, had, got, was), irregular third person singular has, and past participles (e.g. done). They were removed because irregulars are not strictly speaking decomposable into verb stem + suffix in the way that regular verbs are (e.g. lies = lie + s but has ¹ ha + s; therefore lies is clearly morphologically complex, but has is not).
Scoring
Each word was classified as stuttered (1) or not stuttered (0), based on such events marked in the original transcriptions (see Howell & Huckvale, 2004 for details).
2.2
Results
Table
4 shows the total number of words of
each type that were extracted from the database, and then in brackets
the
number that were stuttered.
Table 4. Total number of words in each
type (number of stuttered words in brackets); M = mean number of words
stuttered, expressed as a percentage
|
|
|
morphology |
|
|
|
|
simple (uninflected) |
complex (inflected) |
|
Phonology |
simple (CVC) |
509
(48) M
= 9.43% |
23
(2) M
= 8.70% |
|
complex (CVCC) |
94
(7) M
= 7.45% |
190
(16) M
= 8.42% |
|
Overall, 8.95% of words were stuttered. Differences in the percentage of stuttered items in each cell were very small, and chi-square analysis shows them to be statistically insignificant (for phonologically simple words, Cramer’s V = 0.005, p = 0.906; for phonologically complex words, Cramer’s V = 0,017, p = 0.777; for morphologically simple words, Cramer’s V = 0.025, p = 0.539; for morphologically complex words, Cramer’s V = 0.003, p = 0.964).
Ideally
it would be desirable to analyse individual
performance, in order to determine whether any speakers were
significantly
affected by phonological and/ or morphological complexity. However,
this
analysis was not possible because the speakers produced greatly
different
numbers of words in the different categories of interest, and some
speakers
produced no tokens within a particular category, either fluent or
stuttered.
From
these results it can be concluded
that for English-speaking adults and adolescents the presence of a
cluster at
the end of a word does not influence stuttering rates in natural
speech. Nor is
there an effect of inflectional morphology on stuttering. However, it
is
stressed that the data are not amenable to individual analysis because
often speakers
did not produce any tokens within a particular category. The
possibility
remains that for certain speakers either or both types of complexity
might have
an effect. Experimental studies provide an opportunity to elicit a
larger
number of tokens of the particular word shapes that are of interest.
3.1.
Studies
2 and 3: Experimental data from children and young adults.
In this section two experimental studies are reported. Study 2 uses a repetition task, which is designed to elicit nonsense words with or without clusters at the word end. Study 3 uses an elicitation task, and is designed to elicit third person singular (e.g. weighs, wraps) and past tense forms (e.g. weighed, wrapped) with or without word-final clusters. These tasks permit an investigation of the impact of phonological complexity on stuttering rates, first within morphologically simple, and then within morphologically complex, words.
As was seen in Study 1, samples of spontaneous speech might not contain enough tokens of the sort that are of interest, so that even if group analyses are possible, individual analyses are not. Elicitation tasks have been used previously for investigating syntactic effects on stuttering. For example, Silverman and Bernstein Ratner (1997) asked children to repeat sentences with differing types of complex syntactic structures, such as wh questions and centre embedded sentences. A standard technique used in the atypical phonological development literature involves children playing with toys that have particular names so that the experimenter can elicit the phonological forms of interest (e.g. Chiat, 1989). This technique offers a halfway house between free spontaneous speech and formal elicitation methods, but as far as is known, it has not been used for studying the phonological development of children who stutter (although it has been used to ascertain whether such children have awareness of their speech problem, Yairi & Ambrose, 2004).
Experimental studies allow balanced numbers of the types of words that are of interest to be gathered, controlled for factors such as lexical frequency, in a short period of time. A disadvantage is that if very different results are obtained as compared to analyses of spontaneous speech, then these would need to be interpreted. For example, if higher rates of stuttering were discovered for certain classes of words in experimental studies compared to spontaneous speech, is that because the experimental task is forcing speakers to make errors that they wouldn’t normally make, or is it because in spontaneous speech speakers can avoid using words that they are aware they might stutter over? And yet, even if errors are forced in an experimental situation, that can still give us valuable information about the cognitive processes underlying stuttering. The merits of analyzing spontaneous material versus material obtained experimentally in young children have been discussed at length by Savage and Lieven (2004), who conclude that both are essential research tools.
3.2.1
Study
2 – non-word repetition
In
non-word repetition tasks the
participant hears nonsense words and repeats them. These tasks have
been widely
used with children with specific language impairment (SLI) and
dyslexia.
Children with SLI and dyslexia typically make repetition errors as the
non-words get longer (Bishop, North & Donlan, 1996; Gathercole
&
Baddeley, 1990; Martin & Schwartz, 2003). Controversy still
exists over the
precise locus of the deficit that gives rise to non-word repetition
difficulties. The evidence was initially interpreted as revealing that
children
with SLI have verbal short-term memory limitations, but this
interpretation has
been challenged from many quarters.
Marshall,
Harris and van der Lely (2003) contend
that non-word repetition tasks can be used to probe the status of
phonological
skills in children. Manipulating the phonological structure of
non-words allows
investigation of which structures are easy to repeat (and hence which
are
easily represented/ processed) and which are hard to repeat (less
easily
represented/ processed). Marshall et al. (2003) and Gallon, Harris and
van der
Lely (submitted) used the Test of Phonological Structure (TOPhS, van
der Lely
& Harris, 1999) to reveal that SLI children have difficulty
repeating non-words
that contain clusters and unfooted syllables (a real word example of an
unfooted syllable would be the initial unstressed syllable in gorilla). Note that the children in van
der Lely and colleagues’ studies did not have verbal dyspraxia or
disfluencies
– their articulation was clear and fluent for known words.
The aim of the non-word repetition task included
here is to investigate
whether phonological complexity at the end of nonwords, as indexed by
the
presence of a consonant cluster, influences stuttering rates. Note that
other
studies of non-word repetition (e.g. Hakim & Bernstein Ratner,
2004) have
investigated both whether PWS make more repetition errors than people
who do
not stutter, and whether stuttering rates increase as the length of the
non-word increases. Hence they have investigated both the non-word
repetition
accuracy of PWS and whether the properties of non-words trigger
stuttering. In
this analysis, interest is solely in the latter question.
3.2.2
Method
Participants
19
speakers participated, and their details are presented in Table 5.
Their
average age is 14;0. For the purposes of analysis they were divided
into 2
groups: the participants in Group 1 are aged 14 years and younger, and
those in
Group 2 are 15 and older. The motivation for dividing the group at 14
years of
age was that around this age, stuttering changes form (the type of
stuttering
that occurs and the words on which the stuttering is located differ
from before
to after teenage). Thus the two age groups might operate differently
with
respect to stuttering.
Table 5. Details of speakers who participated in the experimental tasks
|
Participant
number |
Age |
Group |
Gender |
|
9 |
8;0 |
1 |
M |
|
10 |
9;10 |
1 |
M |
|
11 |
9;8 |
1 |
M |
|
12 |
10.5 |
1 |
M |
|
14 |
10;5 |
1 |
F |
|
15 |
10;10 |
1 |
M |
|
13 |
11;4 |
1 |
M |
|
3 |
12;10 |
1 |
M |
|
1 |
13;5 |
1 |
M |
|
16 |
14;2 |
1 |
F |
|
19 |
14;4 |
1 |
F |
|
4 |
14;5 |
1 |
M |
|
18 |
15;5 |
2 |
M |
|
6 |
15;8 |
2 |
M |
|
17 |
17;3 |
2 |
F |
|
2 |
17;4 |
2 |
M |
|
5 |
17;7 |
2 |
F |
|
8 |
19;5 |
2 |
M |
|
7 |
21;10 |
2 |
M |
Items
There were two experimental conditions and one filler condition, with items adapted from the TOPhS (van der Lely & Harris, 1999). For Experimental Condition 1, one-syllable items (N=8) were chosen that ended in a singleton consonant. Half of these items have a simple onset (e.g. kEt) and half have a complex onset (e.g. klEt). For Condition 2, one-syllable items (N=8) were chosen that end in a two-consonant cluster. Again, half have a simple onset (e.g. kEst) and half have a complex onset (e.g. klEst). The TOPhS was designed to allow investigation into the impact of word-final clusters on the repetition of non-words that are otherwise segmentally identical. This match for segmental content is rarely possible when using real word stimuli. A further point is that, by the very nature of the task, these stimuli are morphologically simple. For the Filler Condition (N = 12), three and four syllable (i.e. multisyllabic) items were chosen, none of which contains a cluster (e.g. dEp@ri, s@pIfi; stressed vowel underlined). The filler condition was chosen to contrast with the monosyllabic experimental items, with the aim of adding variety and maintaining participants’ attention to the task.
Procedure
The experimenter tells the participant ‘In this game I’m going to say some funny, made-up
words which I would
like you to repeat after me. We’ll start with some practice ones so you
can see
what you have to do. Can you say ‘zIk’?’ The experimenter gives the
participant time to repeat the non-word, and repeats it if necessary.
The
experimenter carries on down the list of practice items, just saying
the non-word
with no introduction.
Before the experimenter starts the experimental
items, he or she says ‘Those were the practice
words. You did
really well with those. Now we’re going to start for real.’
The
experimenter says each experimental item with no introduction, and this
time
does not repeat any of them if the participant is unsure about them –
he or she
just moves on to the next item.
Scoring
Answers
were scored for the presence (1)
or absence (0) of stuttering. Note that repetition accuracy is not of
concern
in this analysis.
Predictions
If phonological complexity at the word end influences stuttering, we predict higher stuttering rates for non-words ending in a cluster.
3.2.3
Results
The
results are shown in Table 6
below.
Table 6. Total number of words in each type (number of stuttered words in brackets); M = mean number of words stuttered, expressed as a percentage; SDi = standard deviation by items; SDp = standard deviation by participant
|
|
Experimental conditions |
Filler condition |
|
|
Participant group |
no
word-final cluster (e.g.
kEt,
klEt) |
word-final
cluster (e.g.
kEst,
klEst) |
multisyllabic (e.g.
s@pIfi) |
|
1
(14 years and younger) |
96
(1) M = 1.04% SDi = 4.42 SDp
= 2.41 |
96
(2) M = 2.08% SDi = 3.24 SDp
= 3.24 |
144
(21) M = 14.58% SDi = 10.74 SDp
= 20.41 |
|
2
(15 years and older) |
56
(1) M = 1.79% SDi = 4.42 SDp
= 5.40 |
56
(1) M = 1.79% SDi = 4.42 SDp
= 5.40 |
84
(11) M = 13.10% SDi = 2.41 SDp
= 59.39 |
The data in Table 6 indicate that very few of the experimental items are stuttered, and that these are fairly evenly distributed between those that end in a cluster and those that do not. Obviously, with such low numbers of stuttered items, it is not possible to test for a significant difference between non-words with and without final clusters. However, multisyllabic filler items were stuttered much more frequently than the monosyllabic experimental items, and a chi-square analysis shows that across participant group this difference is significant (Cramer’s V = 0.241, p < 0.001). Caution is expressed on interpreting these results, however, because they hide a range of individual patterns, as indicated by the very large figures for the standard deviations by participant (SDp). Only one participant from Group 2 (participant 2) stuttered on multisyllabic words, whereas five from Group 1 (participants 3, 10, 13, 16 and 19) stuttered at least once on these items. These details are shown in Table 7.
Table 7. The number of multisyllabic items stuttered by individual speakers
|
Group |
Number multisyllabic items stuttered |
|
|
9 |
1 |
0 |
|
10 |
1 |
4 |
|
11 |
1 |
0 |
|
12 |
1 |
0 |
|
14 |
1 |
0 |
|
15 |
1 |
0 |
|
13 |
1 |
3 |
|
3 |
1 |
2 |
|
1 |
1 |
0 |
|
16 |
1 |
5 |
|
19 |
1 |
7 |
|
4 |
1 |
0 |
|
18 |
2 |
0 |
|
6 |
2 |
0 |
|
17 |
2 |
0 |
|
2 |
2 |
11 |
|
5 |
2 |
0 |
|
8 |
2 |
0 |
|
7 |
2 |
0 |
As for which multisyllabic items were stuttered most, these data are shown in Table 8. As the standard deviations in Table 6 indicate, there is little variability across items.
Table 8. Multisyllabic items and how often they were stuttered
|
Item |
Number
of times item was stuttered |
|
|
Group 1 |
Group 2 |
|
|
dEp@ri |
2 |
1 |
|
fIp@l@ |
0 |
1 |
|
kEt@l@ |
3 |
1 |
|
pIf@t@ |
2 |
1 |
|
b@dEp@ |
2 |
1 |
|
d@fIpl |
2 |
1 |
|
f@kEte |
1 |
1 |
|
s@pIfi |
0 |
1 |
|
b@dEp@ri |
4 |
1 |
|
d@fIp@l@ |
0 |
1 |
|
f@kEt@l@ |
3 |
0 |
|
s@pIf@t@ |
2 |
1 |
3.2.4
Discussion
The
fact that a non-word repetition task
is capable of inducing stuttering is shown by the high stuttering rates
for
multisyllabic items. Yet rates of stuttering are lower for those same
phonological
forms that were produced in spontaneous speech (see Table 3)
Figure 2. Percent syllables stuttered
in spontaneous speech and imitation tasks for individual participants
who
stutter. Redrawn from Silverman and Bernstein Ratner (1997) with
participants
ordered in increasing percentage of syllables stuttered in spontaneous
speech.
Participants have the same numbers as in Silverman and Bernstein Ratner
(1997).
Given
the high rate of stuttering in this
experiment on multisyllabic words, it would be interesting to compare
this rate
with the rate of stuttering for multisyllabic words in spontaneous
speech
samples, but that particular issue is beyond the scope of this paper.
3.3.1
Study
3
Elicitation tasks have been widely used to probe inflectional abilities, particularly in children with SLI who are known to have difficulties with inflectional morphology (Berko, 1958; Leonard, Eyer, Bedore & Grela, 1997; Rice, Wexler & Cleave, 1995; van der Lely & Ullman, 2001). Many of these tasks use real verbs, but some use nonsense verbs as a way of investigating whether the child’s knowledge of morphology is truly productive (as opposed to the child just supplying a memorised inflected form).
As
the non-word repetition task (Study 2)
gave such low stuttering rates compared to real words (as measured from
spontaneous speech, Study 1), real verbs were used for this task. The
aim is to
determine whether phonological complexity affects stuttering rates for
inflected words by comparing inflected words that end in a cluster
(e.g. wrapped) with those that end
in just a
single consonant (e.g. weighed)
3.3.2
Method
Participants
The
same speakers participated
as in Study 2. See Table 5 for their details.
Items
Two
experimental conditions (no fillers)
were selected. Condition 1 consists of 5 verbs whose 3rd
person
singular and past tense forms both end in a single consonant (e.g. weighs, weighed).
Condition 2 consists of five verbs whose inflected forms
end in a two-consonant cluster (e.g. wraps,
wrapped). The aim is to
compare
stuttering rates on the phonologically simple (no cluster) and
phonologically
complex (cluster) forms. None of the verbs has an onset cluster.
Therefore the
effect of complexity is investigated only at the verb end. Hubbard and
Prins
(1994) report frequency effects on stuttering rates, so we ensured that
both
conditions consisted of high frequency verbs that were matched for
frequency
Procedure
The
items are presented verbally, and the
response is requested verbally (i.e. the participant is not required to
read or
write anything). The experimenter says to the participant ‘In this game we’re going to talk about things that
I like to do, and
that my friend also likes to do. You’re going to help me to say those
things.
For example, if I say ‘Everyday I go to school’, this is something my
friend
does too. So you can tell me ‘Your friend also goes to school’. Can you
say
that? And you can also tell me what we both did yesterday: ‘Yesterday
you both
went to school’. Can you say that?
‘Let’s
do another one to practice. ‘Everyday I eat ice cream.’ So you can say
‘Your
friend also (leave blank for the participant to fill in). And you can also say ‘Yesterday you both (leave
blank for the participant to fill in).
That’s it. Let’s do one more practise one before we start for real.’
Most
participants get the hang of this
procedure very quickly. The experimenter gives the participant as much
prompting as they need with the initial part of the sentence, but
always lets
them fill in the verb. If the participant can’t remember the verb, then
the
experimenter repeats the first sentence (Everyday I …). Verbs are based
on
those used in
Scoring
Responses were scored for the presence (1) or absence (0) of stuttering. Accuracy of inflection itself is not of concern.
Predictions
If
phonological complexity is a factor
influencing stuttering on morphologically complex words, then more
items ending
in a consonant cluster would be predicted to be stuttered.
3.3.3
Results
The
results are set out in
Table 9 below.
Table 9. Total number of words in each
type (number of stuttered words in brackets); M = mean number of words
stuttered,
expressed as a percentage; SDi = standard
deviation by items; SDp
= standard deviation by participant
|
|
Experimental condition |
|
|
Participant group |
No
cluster (e.g.
weighs, weighed) |
Cluster (e.g.
wraps, wrapped) |
|
1 (14 years
and under) |
120
(14) M = 11.67% SDi = 5.16 SDp
= 22.90 |
120
(14) M = 11.67% SDi = 6.99 SDp
= 20.65 |
|
2 (15 years
and older) |
70
(3) M = 4.29% SDi = 4.83 SDp
= 7.87 |
70
(4) M = 5.71% SDi = 5.16 SDp
= 15.12 |
An examination of the data in Table 9 indicates that the presence of a cluster at the inflected verb end has no influence on stuttering rates. The standard deviations by items (SDi) in each cell are low, reflecting that there is little variability across verbs. The only item for which none of the 19 speakers stuttered was hums. The items that were stuttered most were wrapped, hummed and sews, each stuttered by 3 of the 19 speakers.
Variability across speakers, as revealed by SDp, is much greater, and the scores for each speaker are presented in Table 10. Only three of the twelve participants in Group 1 stutter on more than one item (participants 3, 10 and 16), and eight stutter on none of the items. One participant in Group 2 (participant 2) is responsible for all the items stuttered by his group, and he is the only one who stutters more on items containing a final cluster. Interestingly, he was the participant who stuttered on the multisyllabic non-words in the repetition task in Study 2. Although it appears that participants in the younger group stutter more on the items in the elicitation task than the older group, this difference was not tested for significance because of the large individual differences in performance in both groups.
Table 10. Number of morphologically complex items stuttered by individual speakers
|
Participant
number |
Group |
Number
items stuttered – no final cluster |
Number
of items stuttered – final cluster |
|
9 |
1 |
0 |
0 |
|
10 |
1 |
7 |
6 |
|
11 |
1 |
0 |
0 |
|
12 |
1 |
0 |
0 |
|
14 |
1 |
0 |
0 |
|
15 |
1 |
0 |
0 |
|
13 |
1 |
0 |
0 |
|
3 |
1 |
4 |
4 |
|
1 |
1 |
0 |
0 |
|
16 |
1 |
3 |
3 |
|
19 |
1 |
0 |
0 |
|
4 |
1 |
0 |
0 |
|
18 |
2 |
0 |
0 |
|
6 |
2 |
1 |
0 |
|
17 |
2 |
0 |
0 |
|
2 |
2 |
2 |
4 |
|
5 |
2 |
0 |
0 |
|
8 |
2 |
0 |
0 |
|
7 |
2 |
0 |
0 |
3.3.4
Discussion
A
few individuals stutter on
morphologically complex forms, but they are in a minority. For those
individuals, with the possible exception of Participant 2, there was no
effect
of phonological complexity upon stuttering: the presence of a cluster
at the
verb end did not influence stuttering rates.
4.
Conclusions
The
analysis of spontaneous speech (Study
1) revealed no differences in stuttering rates within words that are
either
phonologically complex (as indexed by the presence of a word-final
cluster),
morphologically complex (as indexed by the presence of inflection) or
both. In a non-word
repetition task
(Study 2), where items are by their very nature morphologically simple,
the
presence of a word-end cluster has no influence on stuttering rates. In
a task
designed to elicit inflected words (Study 3), phonological complexity
has no
effect either, with the possible exception of one of the speakers in
the older
group. However, a small proportion of individuals do stutter on
inflected
forms, indicating that for certain PWS morphological complexity can
affect
their stuttering.
These preliminary results reveal that phonological complexity at the word end has no effect on stuttering amongst speakers of English, while morphological complexity affects only a minority of PWS. However, this is by no means the full story. For example, these studies did not consider inflections that add a syllable, e.g. ed, ez and –ing. The higher rates of stuttering on multisyllabic words in the non-word repetition test suggest that syllabic inflections might cause fluency difficulties, particularly for verb stems that are themselves more than one syllable long, e.g. oranges, recorded, balancing. Also beyond the scope of this paper, but perhaps worth investigating in future work, is the impact of derivational morphology, as this adds one or more syllables to either the beginning or the end of words (e.g. heavier, muddiest, unreal, substandard, overcater). Certainly, on the non-word repetition task presented in Study 2, stuttering rates were high in the repetition of multisyllabic non-words with similar metrical shapes to derived words, e.g. s@pIfi, pIf@t@.
It
would also be worth investigating the
effects of cumulative complexity. It is possible that word final
clusters only
have a measurable effect on stuttering rates when they occur in words
that
contain other complex phonological structures, such as unfooted
syllables and
onset clusters. In support of this hypothesis, research using the IPC
metric
suggests that the effects of phonological complexity are indeed
cumulative
(e.g. Howell et al., submitted).
Nor
can we conclude from the results
presented here that effects of word-end phonology and of inflectional
morphology will not be among the factors implicated at the onset of
stuttering,
which is most common at the age of 3. Tasks requiring children to
repeat
non-words and produce morphologically complex words have been carried
out
successfully with children this young (e.g. Roy & Chiat, 2004,
non-word
repetition; Rice & Wexler, 2001, inflection), and experimental
material
similar to that presented in Studies 2 and 3 of this paper would be
suitable
for this purpose.
Finally, one of the main aims of this paper has been to instruct the experimenter on how to tease apart phonology and morphology, so that the effects of each can be studied independently. It is my hope that it will stimulate further work in this field, improving our understanding of how word-end factors affect stuttering both at its onset and in persistent stutterers.
Acknowledgements
I would like
to thank Pete Howell, for encouraging me to think about the issues
raised in
this paper and for his support during every stage of the writing
process. Huge
thanks are also due to Jenny Hayes, for writing the computer program
that
selected words from the corpus of spontaneous speech, and for carrying
out
experimental tasks. I am grateful to Heather van der Lely and John
Harris for
letting me use items from the Test of Phonological Structure. The bulk
of this
work was carried out while I was in receipt of an ESRC studentship,
which is
gratefully acknowledged.
Appendix
1. Non-words
used in the repetition task.
Practice items
zIk,
b@nEli, fIn@ri,
wEf
Experimental items – no word-final
cluster
frIp, kEt
,
pIf, prIf, dEp, drEp, fIp, klEt
Experimental items –word-final
cluster
dEmp, klEst, kEst, frImp, fImp, drEmp,
pIlf, prIlf
Filler items
dEp@ri, s@pIf@t@, kEt@l@, d@fIpl, pIf@t@, b@dEp@ri, b@dEp@, d@fIp@l@, s@pIfi, f@kEt@l@, f@kEt@, fIp@l@
Appendix
2. Stimuli
for the third person singular and past tense elicitation task
Practice
items
Everyday
I go to school. Your
friend also (goes) to school. Yesterday you both (went) to school.
Everyday
I eat ice cream. Your
friend also (eats) icecream. Yesterday you both (ate) ice cream.
Everyday
I swim in the sea. Your
friend also (swims) in the sea. Yesterday you both (swam) in the sea.
Experimental
items – no word-final cluster
Everyday
I pour a drink. Your
friend also (pours) a drink. Yesterday you both (poured) a drink.
Everyday
I pay a bill. Your
friend also (pays) a bill. Yesterday you both (paid) a bill.
Everyday
I sew a shirt. Your
friend also (sews) a shirt. Yesterday you both (sewed) a shirt.
Everyday
I weigh a parcel. Your
friend also (weighs) a parcel. Yesterday you both (weighed) a parcel.
Everyday I lie a little bit. Your friend also (lies) a little bit. Yesterday you both (lied) a little bit.
Experimental
items – word-final cluster
Everyday
I lick a lollipop. Your
friend also (licks) a lollipop. Yesterday you both (licked) a lollipop.
Everyday
I cough a lot. Your
friend also (coughs) a lot. Yesterday you both (coughed) a lot.
Everyday
I wrap a present. Your
friend also (wraps) a present. Yesterday you both (wrapped) a present.
Everyday
I pack a suitcase. Your
friend also (packs) a suitcase. Yesterday you both (packed) a suitcase.
Everyday
I hum a tune. Your
friend also (hums) a tune. Yesterday you both (hummed) a tune.
References
Au-Yeung, J. & Howell, P.
(1998). Lexical and
syntactic content and stuttering. Clinical
Linguistic and Phonetics, 12,
67-78.
Baayen, H., Piepenbrock, R.
& van Rijn, H. (1993). The CELEX
lexical database, CD-ROM.
University of Pennsylvania: Linguistic Data Consortium.
Berko, J. (1958). The child’s
learning of English
morphology. Word, 14,
150-177.
Bernhardt, B. & Stemberger,
J. (1998). Handbook of phonological
development.
Bishop, D., North, T. &
Donlan, C. (1996). Nonword
repetition as a behavioural marker for inherited language impairment:
Evidence
from a twin study. Journal of Child
Psychiatry, 4, 391-403.
Brown, M. C. (2001). Perl:
The complete reference second edition.
Chiat, S. (1989). The relation
between prosodic
structure, syllabification and segmental realization: evidence from a
child
with fricative stopping. Clinical
Linguistics and Phonetics, 3, 223-242.
Conture, E.G. (1990). Stuttering.
Demuth, K. (2004). Production
approaches to stuttering.
Stammering Research, 1, 297-298.
Dworzynski, K. & Howell, P.
(2004). Predicting
stuttering from phonetic complexity in German. Journal
of Fluency Disorders, 29,
149-173.
Gallon,
N., J. Harris & van der Lely, H.K.J. (submitted). Nonword
repetition: An
investigation of phonological complexity in children with G-SLI.
Gathercole, S. &, Baddeley,
A. (1990). Phonological
memory deficits in language disordered children: Is there a connection?
Journal of Memory and Language,
29, 336-360.
Hakim, H.B. & Ratner, N.B.
(2004). Nonword
repetition abilities in children who stutter: An exploratory study. Journal of Fluency Disorders, 29,
179-199.
Howell, P.
Au-Yeung, J. (1995). The association between
stuttering, Brown’s factors
and phonological categories in child stutterers ranging in age between
2 and 12
years. Journal of Fluency Disorders,
20: 331-344.
Howell, P., Au-Yeung, J. &
Sackin, S. (1999).
Exchange of stuttering from function words to content words with age. Journal
of Speech, Language and Hearing Research, 42,
345-354.
Howell, P., Au-Yeung, J. &
Sackin, S. (2000).
Internal structure of content words leading to lifespan differences in
stuttering. Journal of Fluency Disorders,
25, 1-20.
Howell, P., Au-Yeung, J., Yaruss, S.
& Eldridge, K.
(submitted). Phonetic difficulty and stuttering in English. Journal of Fluency Disorders.
Howell, P.,
&
Dworyznski, K. (in press). Planning and
execution processes in speech control by fluent speakers and speakers
who
stutter. Journal of Fluency Disorders.
Howell, P. & Huckvale, M.
(2004). Facilities to
assist people to research into stammered speech. Stammering
Research, 1,
130-242.
Hubbard C.P. & Prins D.
(1994). Word familiarity,
syllabic stress pattern, and stuttering. Journal
of Speech and Hearing Research, 37,
564-571.
Jakielski, K.J. (1998). Motor organization in the acquisition of consonant
clusters.
Dissertation/PhD thesis,
Kadi‑Hanifi,
K., & Howell, P. (1992). Syntactic analysis of the spontaneous
speech of
normally fluent and stuttering children. Journal
of Fluency Disorders, 17,
151-170.
Labov, W. (1978). Sociolinguistic
patterns.
Leonard, L.B., Eyer, J.A., Bedore,
L.M. & Grela,
B.G. (1997). Three accounts of the grammatical morpheme difficulties of
English-speaking children with Specific Language Impairment. Journal of Speech, Language and Hearing
Research, 40, 741-753.
Marshall, C.R. (2004). The morpho-phonlogical interface in children with
Specific Language
Impairment. Unpublished doctoral dissertation,
Marshall, C.R., Harris, J. &
van der Lely, H.K.J.
(2003). The nature of phonological representations in children with
Grammatical-Specific Language Impairment (G-SLI). In D. Hall, T.
Markopoulos,
A. Salamoura & S. Skoufaki (eds) Proceedings of the
University of
Cambridge First Postgraduate Conference in Language Research, 1, 511-517
Marton, K. & Schwartz, R.
(2003). Working memory
capacity and language processes in children with Specific Language
Impairment. Journal of Speech, Language and
Hearing
Research, 46, 1138-1153.
Natke, U., Sandreiser, P., van Ark,
M., Pietrowski, R.
& Kalveram, K.T. (2004). Linguistic stress, within-word
position and
grammatical class in relation to early childhood stuttering. Journal of Fluency Disorders, 29, 109-122.
Packman, A., Onslow, M., Richard, F.
& van Doorn,
J. (1996). Syllabic stress and variability: A model of stuttering. Clinical Linguistics and Phonetics, 10, 235-263.
Rice, M.L. & Wexler, K.
(2001). Test of Early Grammatical Impairment.
Psychological Corporation.
Rice, M.L., Wexler, K. &
Cleave, P.L. (1995).
Specific Language Impairment as a period of extended optional
infinitive. Journal of Speech and Hearing
Research, 38, 850-863.
Roy, P. & Chiat, S. (2004).
A prosodically
controlled word and nonword repetition task for 2- to 4-year-olds:
evidence
from typically developing children. Journal of Speech,
Language, and Hearing
Research, 47, 223-234.
Savage, C. & Lieven, E.
(2004). Can the Usage-based
approach to language development be applied to analysis of
developmental
stuttering? Stammering Research, 1, 83-111.
Silverman, S., & Bernstein
Ratner, N. (1997).
Syntactic complexity, fluency, and accuracy of sentence imitation in
adolescents. Journal of Speech, Language,
and Hearing Research, 40,
95-106.
Throneburg, R.N., Yairi, E.
& Paden, E.P. (1994).
Relation between phonologic difficulty and the occurrence of
disfluencies in
the early stage of stuttering. Journal of
Speech and Hearing Research, 37,
504-509.
van der Lely, H.K.J. &
Harris, J. (1999). The Test of Phonological
Structure.
Unpublished test available from the first author, Centre for
Developmental
Language Disorders and Cognitive Neuroscience, Department of Human
Communication Science, University College London, London UK.
van der Lely, H.K.J. &
Ullman, M. (2001). Past
tense morphology in specifically language impaired and normally
developing
children. Language and Cognitive
Processes, 16: 177-217
Weiss, A. & Jakielski, K.
(2001). Phonetic
complexity measurement and prediction of children’s disfluencies: a
preliminary
study. Proceedings of the 4th
International Speech Motor Control Conference,
Wingate, M. (1982). Early position
and stuttering occurrence.
Journal of Fluency Disorders, 7, 243-258.
Wingate, M. (1988). The
structure of stuttering: A psycholinguistic analysis.
Yairi, E., & Ambrose, N. G.
(2004). Early childhood stuttering.
AustinTX:
Pro-Ed.
Stammering
Research
Index
to Volume 1, 2004-2005
Author index
375-378
Subject
index
379-382
Title index
383-386
Author
index
Acton,
C. (2004). Broadening the stuttering research base: the possible merits
of a
conversation analytic perspective. Stammering
Research, 1, 18-20.
Acton, C. (2004). A conversation analytic perspective
on stammering: some reflections and
observations. Stammering Research, 1, 249-270
Anderson,
J. D., & Musolino, J. (2004). How Useful is the Usage Based
Approach to Stuttering Research?
Stammering
Research, 1, 295-296
Bartles,
S., & Ramig, P. R. (2004). Clinical Research into use of the
SpeechEasyTM
device. Stammering Research, 1, 66.
Blood, G. & Blood,
Dahm,
B. (2004). Commentary on Partnerships between Clinicians,
Researchers, and People
Who Stutter in the Evaluation of Stuttering Treatment Outcomes by J.
Scott
Yaruss and Robert W. Quesal. Stammering
Research, 1, 16-17.
Davis,
A., & Howell, P.
(2005). Elements
of statistical treatment of speech and hearing science data. Stammering
Research, 1, 333-343
Davis,
S., & Furnham, A. (2004). Authors response to commentaries on
‘Involvement
of social factors in stuttering: A review and assessment of current
methodology’. Stammering Research, 1, 128-129.
Davis, S., & Furnham,
A. (2004). Authors response to commentaries on ‘Involvement of social
factors
in stuttering: A review and assessment of current methodology’. Stammering Research, 1,
307-308
Demuth, K.
(2004). Production approaches to stuttering. Stammering
Research, 1, 297-298
Everard,
R. (2004). Commentary on Partnerships between Clinicians, Researchers
and
People who Stutter in the Evaluation of Stuttering Treatment Outcomes. Stammering Research, 1,
24-25.
Furnham, A., & Davis, S. (2004). Involvement of social factors in stuttering: A
review and assessment
of current methodology. Stammering Research, 1,112-122
Gershkoff-Stowe,
L. (2004). Commentary on ‘Can
the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 101-102.
Ghazi, S.,
& Lickley, R. (2004). Bilingual
issues in stammering intervention: an exploratory study. Stammering Research, 1,
327
Gooding,
S., &
Howell,
P. (2004). Effects of delayed auditory feedback and frequency-shifted
feedback
on speech control and some potentials for future development of
prosthetic aids
for Stammering. Stammering
Research, 1,
31-46.
Howell,
P. (2004). Response to commentaries on ‘Effects of delayed auditory
feedback
and frequency-shifted feedback on speech control and some potentials
for future
development of prosthetic aids for stammering’. Stammering
Research, 1, 68-77.
Howell, P. (2005). The effect
of using time intervals of different length on judgements about
stuttering. Stammering Research, 1, 364-374.
Howell,
P., Davis, S., Bartrip, J. & Wormald, L. (2004).Effectiveness
of frequency
shifted feedback at reducing disfluency for linguistically easy, and
difficult,
sections of speech (original audio recordings included). Stammering
Research, 1, 309-315.
Howell, P., & Huckvale, M.
(2004). Facilities to
assist people to research into stammered speech. Stammering
Research, 1, 130-242
Hussain, K., Khan,
S., Howell, P., & Davis,
S. (2004). Do children who stutter have word finding difficulties? Stammering
Research, 1, 326
Joukov, S. (2004). Trial
software for frequency shifted and delayed auditory feedback Stammering Research, 1, 316-325.
Levine,
S. Z., Petrides, K. V., Davis
Lowitt,
A., & Brendel, B. (2004). The response of patients with
Parkinson’s Disease
to DAF and FSF. Stammering Research, 1, 58-61
Manning,
W. H. (2004). The social phenomena of stuttering. Stammering
Research, 1, 123-124.
Marshall,
C. (2005). The impact of word-end phonology and morphology on
stuttering. Stammering Research, 1, 375-391.
Miller,
D. (2004). Effects of using an
Neumann,
K., & Euler, H. (2004). Are alerts sufficient to smooth speech?
Stammering Research, 1, 48-50.
Osborne,
V. (2004). Comments on the implications of Yaruss and Quesal (2004) on
seeking,
and measuring outcomes of therapy from the perspective of an adult
covert
stammerer. Stammering Research, 1, 26-27.
Pfordresher,
P. (2004). Can altered auditory information affect planning? Evidence
from
music performance. Stammering Research, 1, 51-53
Reed,
P. (2004). Altered Recurrent Auditory Information and Stammering:
Contributions
from Learning Theory. Stammering Research,
1, 55-57
Ryan,
B. (2004). Programmed Stuttering Therapy for Children and Adults using
ARAI. Stammering Research, 1, 62
Sage, R. (2004). How
interviews with adults who stammer inform research directions
concerning what
we think about our stammer: Participants views from the Apple House
Fluency
Course,
Savage,
C. (2004). Responses to commentators. Stammering
Research, 1, 107-111.
Savage, C., & Lieven, E.
(2004). Can the Usage-Based
approach to language development be applied to analysis of
developmental stuttering?
Stammering
Research, 1, 83-100
Savage, C., & Lieven,
E. (2004). Can the Usage-Based approach to language development be
applied to analysis
of developmental stuttering? Stammering
Research, 1, 299-300
Schwartz, M. (2004). Commentary
on ‘Can
the
Usage-Based approach to language development be applied to analysis of
developmental
stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 106.
Skljarov, O. P.
(2004). Commentary
on ‘Can
the Usage-Based approach to language development be applied to analysis
of developmental
stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 103-105.
Skljarov, O. P. (2004). Commentary on ‘Involvement of social factors in
stuttering: A review and
assessment of current methodology’ by A. Furnham and
Skotko,
J. (2004). Experience of a speech pathologist providing clients with
the
SpeechEasyTM device. Stammering
Research, 1, 63-65.
Stenner,
P. (2004). On reaching the parts that quantitative researchers cannot:
Commentary on Sage, R. (2004) “How interviews with adults who stammer
inform
research directions”. Stammering
Research, 1, 290-292
Vanryckeghem,
M. (2004). Commentary
on
"Involvement of social factors in stuttering: A review and assessment
of
current methodology" By A. Furnham and
Wade,
J. (2004). Extending the scope of stuttering research and
treatment. Stammering Research, 1, 125-126.
Wormald,
L., Davis, S., & Howell, P. (2004). The Family History of
Stuttering in Recovered and Persistent Groups. Stammering
Research, 1, 326
Yaruss,
J. S.,& Quesal, R. W. (2004). Partnerships between Clinicians,
Researchers,
and People Who Stutter in the Evaluation of Stuttering Treatment
Outcomes. Stammering Research, 1,
1-15.
Yaruss,
J. S., & Quesal, R. W. (2004). The
Subject
index
Target articles
Yaruss,
J. S.,& Quesal, R. W. (2004). Partnerships between Clinicians,
Researchers,
and People Who Stutter in the Evaluation of Stuttering Treatment
Outcomes. Stammering Research, 1,
1-15.
Howell,
P. (2004). Effects of delayed auditory feedback and frequency-shifted
feedback
on speech control and some potentials for future development of
prosthetic aids
for Stammering. Stammering
Research, 1,
31-46.
Savage, C., & Lieven, E.
(2004). Can the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering? Stammering Research, 1, 83-100
Furnham, A., & Davis, S. (2004). Involvement of social factors in stuttering: A
review and assessment
of current methodology. Stammering Research, 1,112-122
Acton, C. (2004). A conversation analytic perspective
on stammering: some reflections and
observations. Stammering Research, 1, 249-270
Sage, R. (2004). How
interviews with adults who stammer inform research directions
concerning what
we think about our stammer: Participants views from the Apple House
Fluency
Course,
Levine,
S. Z., Petrides, K. V., Davis
Davis,
A., & Howell, P. (2005). Elements of statistical treatment
of speech and hearing science data. Stammering
Research, 1, 333-343
Commentaries and response to
commentaries
Yaruss and Quesal
Dahm,
B. (2004). Commentary on Partnerships between Clinicians,
Researchers, and People
Who Stutter in the Evaluation of Stuttering Treatment Outcomes by J.
Scott
Yaruss and Robert W. Quesal. Stammering
Research, 1, 16-17.
Acton,
C. (2004). Broadening the stuttering research base: the possible merits
of a
conversation analytic perspective. Stammering
Research, 1, 18-20.
Gooding,
S., &
Everard,
R. (2004). Commentary on Partnerships between Clinicians, Researchers
and
People who Stutter in the Evaluation of Stuttering Treatment Outcomes. Stammering Research, 1,
24-25.
Osborne,
V. (2004). Comments on the implications of Yaruss and Quesal (2004) on
seeking,
and measuring outcomes of therapy from the perspective of an adult
covert
stammerer. Stammering Research, 1, 26-27.
Yaruss,
J. S., & Quesal, R. W. (2004). The
Howell
Neumann,
K., & Euler, H. (2004). Are alerts sufficient to smooth speech?
Stammering Research, 1, 48-50.
Pfordresher,
P. (2004). Can altered auditory information affect planning? Evidence
from
music performance. Stammering Research, 1, 51-53
Reed,
P. (2004). Altered Recurrent Auditory Information and Stammering:
Contributions
from Learning Theory. Stammering
Research, 1, 55-57
Lowitt,
A., & Brendel, B. (2004). The response of patients with
Parkinson’s Disease
to DAF and FSF. Stammering Research, 1, 58-61
Ryan,
B. (2004). Programmed Stuttering Therapy for Children and Adults using
ARAI. Stammering Research, 1, 62
Skotko,
J. (2004). Experience of a speech pathologist providing clients with
the
SpeechEasyTM device. Stammering
Research, 1, 63-65.
Bartles,
S., & Ramig, P. R. (2004). Clinical Research into use of the
SpeechEasyTM
device. Stammering Research, 1, 66.
Miller,
D. (2004). Effects of using an
Howell,
P. (2004). Response to commentaries on ‘Effects of delayed auditory
feedback
and frequency-shifted feedback on speech control and some potentials
for future
development of prosthetic aids for stammering’. Stammering
Research, 1, 68-77.
Savage and Lieven
Gershkoff-Stowe,
L. (2004). Commentary on ‘Can
the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 101-102.
Skljarov, O. P.
(2004). Commentary
on ‘Can
the Usage-Based Approach to Language Development be Applied to Analysis
of
Developmental Stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 103-105.
Schwartz, M. (2004). Commentary
on ‘Can
the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering?’ by C.Savage
and E.Lieven. Stammering Research, 1, 106.
Anderson,
J. D., & Musolino, J. (2004). How Useful is the Usage Based
Approach to Stuttering Research?
Stammering
Research, 1, 295-296
Demuth, K.
(2004). Production approaches to stuttering. Stammering
Research, 1, 297-298
Savage,
C. (2004). Responses to commentators. Stammering
Research, 1, 107-111.
Savage, C., & Lieven,
E. (2004). Can the Usage-Based Approach to Language Development be
Applied to
Analysis of Developmental Stuttering? Stammering
Research, 1, 299-300
Furnham and Davis
Manning,
W. H. (2004). The social phenomena of stuttering. Stammering
Research, 1, 123-124.
Wade,
J. (2004). Extending the scope of stuttering research and
treatment. Stammering Research, 1, 125-126.
Skljarov, O. P. (2004). Commentary on ‘Involvement of social factors in
stuttering: A review and
assessment of current methodology’ by A. Furnham and
Blood, G. & Blood,
Vanryckeghem,
M. (2004). Commentary
on
"Involvement of social factors in stuttering: A review and assessment
of
current methodology" By A. Furnham and
Davis,
S., & Furnham, A. (2004). Authors response to commentaries on
‘Involvement
of social factors in stuttering: A review and assessment of current
methodology’. Stammering Research, 1, 128-129.
Davis, S., & Furnham,
A. (2004). Involvement of social factors in stuttering: A review and
assessment
of current methodology. Stammering
Research, 1, 307-308
Sage
Stenner,
P. (2004). On reaching the parts that quantitative researchers cannot:
Commentary on Sage, R. (2004) “How interviews with adults who stammer
inform
research directions”. Stammering
Research, 1, 290-292
Research data, software and
analysis section
Howell, P. (2005). The effect
of using time intervals of different length on judgements about
stuttering. Stammering Research, 1, 364-374.
Howell, P., & Huckvale, M.
(2004). Facilities to
assist people to research into stammered speech. Stammering
Research, 1, 130-242
Howell,
P., Davis, S., Bartrip, J. & Wormald, L. (2004). Effectiveness
of frequency
shifted feedback at reducing disfluency for linguistically easy, and
difficult,
sections of speech (original audio recordings included). Stammering
Research, 1, 309-315.
Joukov, S. (2004). Trial
software for frequency shifted and delayed auditory feedback Stammering Research, 1, 316-325.
Marshall,
C. (2005). The impact of word-end phonology and morphology on
stuttering. Stammering Research, 1, 375-391.
Abstracts of
reports from BSA vacation
studentships 2004
Wormald,
L., Davis, S., & Howell, P. (2004). The Family History of
Stuttering in Recovered and Persistent Groups. Stammering
Research, 1, 326
Hussain, K., Khan,
S., Howell, P., & Davis,
S. (2004). Do children who stutter have word finding difficulties? Stammering
Research, 1, 326
Ghazi, S.,
& Lickley, R. (2004). Bilingual
issues in stammering intervention: an exploratory study. Stammering Research, 1,
327
Title
index
A
conversation analytic perspective on stammering: some reflections and
observations.
Acton,
C. (2004). Stammering Research, 1, 249-270
Altered
Recurrent Auditory Information and Stammering: Contributions from
Learning
Theory. Reed, P. (2004). Stammering
Research, 1, 55-57
Are
alerts sufficient to smooth speech? Neumann, K., & Euler, H.
(2004). Stammering Research, 1, 48-50.
Authors
response to commentaries on ‘Involvement of social factors in
stuttering: A
review and assessment of current methodology’. Davis, S., &
Furnham, A.
(2004). Stammering Research, 1,
128-129.
Bilingual issues in
stammering intervention: an
exploratory study. Ghazi, S., & Lickley, R. (2004). Stammering Research, 1,
327
Broadening
the stuttering research base: the possible merits of a conversation
analytic
perspective. Acton, C. (2004). Stammering
Research, 1, 18-20.
Can
altered auditory information affect planning? Evidence from music
performance. Pfordresher,
P. (2004). Stammering Research, 1,
51-53
Can the Usage-Based Approach to
Language Development
be Applied to Analysis of Developmental Stuttering? Savage, C.,
& Lieven,
E. (2004). Stammering Research, 1, 83-100
Can the Usage-Based
Approach to Language Development be Applied to Analysis of
Developmental
Stuttering? Savage, C., & Lieven, E. (2004). Stammering
Research, 1, 299-300
Clinical
Research into use of the SpeechEasyTM device.
Bartles, S., &
Ramig, P. R. (2004). Stammering Research, 1,
66.
Commentary on ‘Can
the Usage-Based Approach to Language Development be Applied to Analysis
of
Developmental Stuttering?’ by C.Savage
and E.Lieven. Gershkoff-Stowe, L. (2004). Stammering
Research, 1, 101-102.
Commentary
on ‘Can
the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering?’ by C.Savage
and E.Lieven.
Schwartz, M. (2004). Stammering
Research,
1, 106.
Commentary
on ‘Can
the
Usage-Based Approach to Language Development be Applied to Analysis of
Developmental Stuttering?’ by C.Savage
and E.Lieven.
Skljarov, O. P. (2004). Stammering
Research,
1, 103-105.
Commentary on ‘Involvement of
social factors in stuttering: A review and assessment of current
methodology’
by A. Furnham and
Commentary on
"Involvement of social factors in stuttering: A review and assessment
of
current methodology" By A. Furnham and
Commentary on Partnerships between
Clinicians,
Researchers, and People Who Stutter in the Evaluation of Stuttering
Treatment
Outcomes by J. Scott Yaruss and Robert W. Quesal. Dahm, B. (2004). Stammering
Research, 1, 16-17.
Commentary
on Partnerships between Clinicians, Researchers and People who Stutter
in the
Evaluation of Stuttering Treatment Outcomes. Everard, R. (2004). Stammering Research, 1,
24-25.
Comments
on the implications of Yaruss and Quesal (2004) on seeking, and
measuring
outcomes of therapy from the perspective of an adult covert stammerer.
Osborne,
V. (2004). Stammering Research, 1, 26-27.
Do children who
stutter have word finding
difficulties? Hussain, K., Khan, S., Howell, P., & Davis, S.
(2004). Stammering
Research, 1, 326
Effectiveness
of frequency shifted feedback at reducing disfluency for linguistically
easy,
and difficult, sections of speech (original audio recordings included).
Howell,
P., Davis, S., Bartrip, J. & Wormald, L. (2004).Stammering
Research, 1, 309-315.
Effects
of delayed auditory feedback and frequency-shifted feedback on speech
control
and some potentials for future development of prosthetic aids for Stammering. Howell, P. (2004). Stammering
Research, 1,
31-46.
Effects
of using an
Elements of statistical treatment
of speech and
hearing science data. Davis,
A., &
Howell, P. (2005). Stammering Research, 1, 333-343
Experience
of a speech pathologist providing clients with the SpeechEasyTM
device. Skotko, J. (2004). Stammering
Research, 1, 63-65.
Extending
the scope of stuttering research and treatment. Wade, J. (2004). Stammering
Research, 1, 125-126.
Facilities to assist people to
research into
stammered speech. Howell, P., & Huckvale, M. (2004). Stammering
Research, 1, 130-242
How interviews with adults who stammer inform
research directions
concerning what we think about our stammer: Participants views from the
Apple
House Fluency Course,
How
Useful is the Usage Based Approach to Stuttering Research?
Anderson, J. D., & Musolino, J. (2004). Stammering
Research, 1, 295-296
Indications
by parents of ‘environmental’ factors that differ between children who
stutter
and controls. Gooding, S., &
Involvement of social factors in stuttering: A
review and assessment
of current methodology. Furnham,
A., & Davis, S. (2004). Stammering Research, 1,112-122
Involvement of social
factors in stuttering: A review and assessment of current methodology.
Davis, S.,
& Furnham, A. (2004). Stammering Research, 1,
307-308
On
reaching the parts that quantitative researchers cannot: Commentary on
Sage, R.
(2004) “How interviews with adults who stammer inform research
directions”.
Stenner,
P. (2004). Stammering Research, 1, 290-292
Partnerships
between Clinicians, Researchers, and People Who Stutter in the
Evaluation of
Stuttering Treatment Outcomes. Yaruss, J. S.,& Quesal, R. W.
(2004). Stammering Research, 1,
1-15.
Production
approaches to stuttering. Demuth, K. (2004). Stammering
Research, 1, 297-298
Programmed
Stuttering Therapy for Children and Adults using ARAI. Ryan, B. (2004).
Stammering
Research, 1, 62
Response
to commentaries on ‘Effects of delayed auditory feedback and
frequency-shifted
feedback on speech control and some potentials for future development
of
prosthetic aids for stammering’. Howell, P. (2004). Stammering
Research, 1, 68-77.
Responses
to commentators. Savage, C. (2004). Stammering Research, 1,
107-111.
Social Factors in Stuttering: A
Call for Rigorous, Accountable
and Timely Research. Blood, G. & Blood,
The
The
effect of using time intervals of different length on judgements about
stuttering.
Howell, P. (2005). Stammering
Research, 1, 364-374.
The
Family History of Stuttering in Recovered and Persistent Groups.
Wormald, L.,
Davis, S., & Howell, P. (2004). Stammering Research, 1, 326.
The
impact of word-end phonology and morphology on stuttering. Marshall, C. (2005). Stammering Research, 1,
375-391.
The
response of patients with Parkinson’s Disease to DAF and FSF. Lowitt,
A., &
Brendel, B. (2004). Stammering Research, 1, 58-61
The
social phenomena of stuttering. Manning, W. H. (2004).
Stammering
Research, 1, 123-124.
The
use of structural equation modeling in stuttering research: Concepts
and directions.
Levine, S. Z., Petrides, K. V., Davis
Trial software for
frequency shifted and delayed auditory feedback. Joukov, S. (2004). Stammering
Research, 1, 316-325.
What
sort of cerebellar processing problem do people who stutter have?
+44 (0)20 7679 2000 - Copyright © 1999-2005 UCL