My research interests center around the development and
application of advanced statistical methods to improve the measurement
and testing of psychological phenomenon. Studying quantitative psychology
has afforded me the ability to work in many subfields of psychology and
this is one of the reasons I am pursuing a degree in quantitative psychology.
I plan on continuing my research and actively pursuing outside support
to fund future research, which will involve both undergraduate and graduate
student assistants.
Applications of Structural Equation Modeling
My experience with structural equation modeling began with my involvement
in research studying a series of U.S. high-school based studies on the
relationship between perceptions of intergroup contact and prejudice reduction
among adolescents. This research is based on an adaptation of Berry, Trimble,
and Olmedo's (1986) acculturation theory, which emphasizes immigrants'
dilemma of interacting with the host culture while valuing the traditions
of their culture of origin. We adapted a mutual acculturation approach,
which adapts both dimensions (which we call outgroup orientation and ethnic
identity, respectively), to the study of intergroup prejudice. In multiple
studies, we have demonstrated that outgroup orientation is a consistent
mediator of the intergroup contact - prejudice relationship, while ethnic
identity mediates prejudice for some samples but not others.
My master's thesis at California State University, Northridge compared
four of the major ethnic groups studied (i.e. African Americans, Asian
Americans, Euro Americans and Latino Americans), in a multi-group path
analysis performed through structural equation modeling software (i.e.
EQS). I found that the four groups did not significantly differ in terms
of the model and this was counter to what was expected. Further studies
have looked at comparing the mutual acculturation model to a model based
on common ingroup identity (Gaertner & Dovidio, 2000). The results
of this study are presented in Wittig, Molina, Giang, and Ainsworth (in
press). In another study we separated ethnic identity into separate components
of ethnic identity exploration and ethnic identity commitment in order
to investigate how each component affects outgroup orientation and prejudice
(Whitehead, Wittig & Ainsworth, under review). Additionally, I am
working on a study that is an extention of my master's thesis that compares
ethnic status groups in a full measurement, multi-group, mean-structure
analysis in two different high school intervention programs (Ainsworth,
Wittig & Rabinowitz, in preparation).
In a separate line of research I compared two different estimation procedures
in utilizing two-level structural equation models. Research subjects are
often sampled within existing groups and it is known that this type of
sampling needs to be analyzed by methods that take the grouped nature
of the data into account. Multilevel structural equation modeling is one
of many methods that addresses clustered or hierarchical data designs
and like any method in structural equation modeling there is always a
question of what estimation procedure to utilize. The Muthèn (1989)
approximation to maximum likelihood estimation (MUML) has been a popular
estimation method in multi-level SEM because of its simplicity but is
not a true maximum likelihood estimator when groups have unequal sizes.
My study compares of the accuracy of the Bentler and Liang (2003) full
information maximum likelihood estimator (BLML), which is a true maximum
likelihood estimator in the unbalanced case, in capturing true parameter
estimates when compared to the popular MUML method of estimation. Results
favor the BLML in accuracy and efficiency in capturing parameter estimates
as well as in the percent of admissible solutions and the percent of rejected
models.
I am also involved in two grants that utilize structural equation modeling
techniques. One of them is an R01grant applying structural equation modeling
to test a mediational model of the utilization of an AIDS vaccine dissemination
program. The other is an R34 grant that applies structural equation modeling
to test a mediational model adapting group self-management programs, developed
and proven effective for persons with chronic medical illness, for patients
who also have depression. Both of these grants are in the data collection
phase and I have already assisted in the developmental stages as a structural
equation modeling expert. Future research will also be tied to these two
grants pending the completion of data collection.
Application of Item Response Theory Models
Item response theory (IRT) is an increasingly popular approach to the
development, evaluation, and administration of psychological measures.
In Reise, Ainsworth, and Haviland (2005) we introduced item response models
to a general psychological audience and illustrated why item response
models should increase in utilization by psychologists in the future.
In the paper we first introduced three IRT fundamentals: (a) the item
response function, (b) information functions, and (c) the invariance property.
We next illustrated how IRT modeling can improve the quality of psychological
measurement. We then proceed to supply evidence to suggest that the differences
between IRT and traditional psychometric methods are not trivial; since
IRT applications can improve the precision and validity of psychological
research across a wide range of subjects. We are currently working on
a similar, yet more extensive invited chapter that will detail the utility
of IRT models in using and creating personality measures (Reise, Morizot
& Ainsworth, in preparation).
In Reise, Meijer, Ainsworth, Morales, and Hays (in press) we applied
group-level parametric and non-parametric item response theory models
to the Consumer Assessment of Health Plans Survey (CAHPS®) 2.0 core
items in a large sample (35,000+) of Medicaid recipients nested within
over 100 health plans. Results indicated that CAHPS® responses are
dominated by within health plan variation, and only weakly influenced
by between health plan variation. In other words patient views of their
health plan vary more within each health plan than they do across health
plans; meaning that people are as happy (or unhappy) about their own health
plan when compared to members of other health plans. Thus, although the
CAHPS® 2.0 survey has acceptable psychometric properties when analyzed
at the individual level, large sample sizes are needed to reliably differentiate
among health plans. These results illustrate why it is important to study
evaluations of health care, such as CAHPS®, at multiple levels of
analyses.
I was also recently hired by Telesage to perform IRT analyses for a
grant to develop new personality and psychopathology scales (e.g. depression,
anxiety, occupational functioning, interpersonal functioning, physical
functioning, etc.) intended for use with both "normal" and psychopathological
respondents. During phase 1 of the grant I was responsible for performing
basic item response functions to establish each scales measurement properties
(e.g. unidimensionality, item scale relationship, item information, etc.),
which included identifying items from each scale with poor measurement
properties. My results were then used to apply for funding for the second
phase of the grant in which I performed further tests on just the depression
and anxiety scales. In addition to the establishing basic scale properties
I performed differential item functioning comparing male and female respondents
in order to identify items that are interpreted and utilized in the same
way by both genders.
Currently, I am also working on a project whose purpose is to first
identify unidimensional subscales of the MMPI-2 adult survey and MMPI-A
adolescent survey by utilizing item response models. Secondly, when unidimensional
scales are identified and the scales overlap in both the 2 and A versions
we will perform differential item functioning analysis in order to identify
whether scale items are interpreted and utilized in the same manner in
both age groups. It is important to identify differential item functioning
so that we can identify whether a scale can be used to track changes in
trait level over time or at least cross-sectionally.
Future Research
One line of research I plan to pursue in the near future is testing IRT
model assumptions when applying these models to psychological data. Although
much is known about IRT models when applied to aptitude measurement, these
models have not been systematically investigated outside this domain.
The ultimate goal of this line of research is to inform and improve the
quality of measurement and outcomes tracking in the applied world of personality,
psychopathology, an psychological assessment. To accomplish these objectives,
this research program will use real psychopathogy data to assess the degree
to which IRT model assumptions are violated. This will in turn inform
Monte Carlo simulations based on the outcome of using the real data sets.
Specifically, analysis of real data sets suggests that issues such as
the effects of multidimensionality, non-normal distributions, and violations
of local independence need further exploration. Moreover, the basic findings
derived from real data and Monte Carlo investigations will lead to systematic
improvements when IRT is applied to psychological data. Specifically,
the proposed study will lead to further applications of IRT methods to
analyze the psychometric functioning of existing scales, person-fit assessment,
linking scales, differential item functioning, and computerized adaptive
assessments within the psychological domain.
Additionally, I have plans for more research involving the mutual acculturation
model and prejudice. One line of research will try and address the recent
debates concerning the relationship between ethnic identity and prejudice
in recent literature. Results from my previous research concerning the
mutual acculturation model have pointed out certain discrepancies in terms
of the role of ethnic identity in mediating the relationship between contact
and prejudice. Other studies have shown that high levels of ethnic identity
lead to more prejudice and while still others show that prejudice would
be reduced. I have plans for a line of research that is designed to try
and uncover moderating and mediating variables that can be contributing
to the seemingly contradictory findings. Advanced latent variable models
will by utilized to test for 1) latent profiles within a sample of high
school respondents and 2) moderated mediation, as further explanations
for the previous discrepant results. Another line of research will utilize
growth curve analyses to try and 1) track the predictors of change in
prejudice over time (e.g. does change in conditions of contact predict
change in prejudice while mediated by a change in outgroup orientation)
and 2) identify mixtures of responses within a sample in order to identify
the types of students for which a prejudice reduction intervention is
successful.
|