Thomas K. Burdenski, Jr.
Science Student Council Chair, Quantitative Psychology Representative, Texas
A&M University
(This article was first published in the Winter 2002 issue of
the APAGS Newsletter.)
 |
So, how bright is the future for graduate students in psychology interested
in conducting and publishing longitudinal research? Two of the three
graduate students who attended the 2001 APA Advanced Training Institute in
Longitudinal Methods (LM) shared that this tool enables psychological
researchers to accurately understand underlying processes like how marital
conflict and parenting styles affect children’s social and emotional
development and how various cognitive skills evolve and change over the
lifespan. |
According to Lauren Papp, a 4th
year doctoral student in counseling and developmental psychology and graduate
research assistant in the Family Studies Center in the Psychology
department at the University of Notre Dame, "using LM is giving
investigators better techniques to answer their research questions. In many
areas like developmental, cognitive, clinical, and social psychology, people are
starting to look not at just the relationships that are occurring between
variables but they are starting to ask how and why things happen. Just having
more data from your study is going to help you answer research questions better.
We are getting a better handle on processes at a deeper level."
"LM provides a better understanding of why things are happening and some
of these research questions will help us answer things we haven’t been able to
look at before and will possibly lead into more prevention and intervention
programs which will have a positive influence on many people," she added.
Papp is studying how children develop when exposed to various parenting and
marital conflict styles. "By studying children and families over
time, we can better see how children’s anxiety, depression, internalizing
behaviors, externalizing behaviors, school performance, and how they get along
with their friends at school varies on the basis of their relationships with
their parents and how their parents handle their marital differences," she
said.
"I am interested in longitudinal methods so I can better understand
growth processes and ways to capture changes that occur over time," Papp
added. "It’s a particularly useful way to study children and families who
are constantly changing. Another useful feature is that most of the techniques
used in LM can be used in shorter term studies with much shorter intervals
between measurements, including diary studies."
"The advantage over traditional techniques like repeated measures ANOVA
is that most of the longitudinal methods allow you to consider families who have
missing or incomplete data for one of the years we are studying whereas many of
the traditional methods require complete data to include the family in the
study," Papp said.
Stuart MacDonald, a graduate student in psychology who works with the
Victoria Longitudinal Study in the Psychology department at the University of
Victoria, was drawn to LM for two reasons. "First is the importance of
aging research in an aging society," he said. "Second, I am trained in
lifespan development and my main interest is cognitive aging. The theories of
cognitive aging and the phenomena that they describe specifically hypothesize
age change. I feel it is imperative to actually examine the things we are
interested in longitudinally which permits the assessment of change across
time."
"My ultimate driving force is my interest in age change," said
MacDonald. "I prefer to study actual individual change across multiple
occasions of measurement despite shortcomings of longitudinal designs. The
longitudinal retest interval question depends completely on what you are
studying. If you are examining a relatively labile, transient construct, you
need to have an interval sensitive to that or you may completely miss capturing
the phenomena."
Previously, when researchers wanted to study how IQ or other constructs
varied with age, they used a cross-sectional design that might have tested 100
participants representing each age group in five-year increments from 5–75
years of age (100 five-year olds, 100 ten-year olds, and so on) and then
compared their IQ scores.
Cross-sectional designs like these have been criticized because many
variables are likely to be confounded with age due to cohort effects. For
example, in 2001, members of the 75-year-old cohort are much more likely to be
immigrants and to have served in the military whereas members of the 25-year-old
cohort are likely to have more years of formal education and to be raised,
educated, and socialized differently than their 75-year-old counterparts.
"Cross-sectional designs only simulate change and there are fairly large
assumptions associated with cross-sectional age group differences," said
MacDonald. "If you have a 25 year-old group and an 75 year-old group and
you are examining them cross-sectionally, you assume the only difference between
those groups to be due to age and that’s a big assumption."
In longitudinal designs, the same group of participants is tested repeatedly
over time (i.e., "repeated measures"). In this way, all the confounds
inherent in cross-sectional design are avoided so that, for example, a
researcher could be more confident that age is more responsible for performance
change than confounds in subject variables.
The most common form of longitudinal data collection in psychology is panel
data, which consists of repeated observations of many individual persons,
preferably on more than two intervals referred to as "waves."
Longitudinal studies with just two waves may address some research questions
marginally well, but many others rather poorly.
There are three samples running concurrently in the Victoria study, according
to MacDonald: "the first sample included about 487 participants between 55
to 85 years of age, the second tested 530 and we are still collecting the third
sample. The Victoria study is based on three-year retest intervals because many
of the measures we study reflect relatively enduring trait-like changes
including episodic memory, working memory, perceptual speed, and simple reaction
time," he said. "Every three years we retest individuals. The first
sample began in 1986 so there are five waves or 15 years of available data. For
each wave of testing, there are over 10 hours of data collected on each
participant. With that many waves and measures, LM is really quite flexible in
terms of your approach to examining this rich source of data."
Longitudinal methods are not without disadvantages, however. This type of
research can take 20–30 years to complete and repeatedly testing the same
participants can bias the results unless precautions are taken to ensure that
responses do not reflect previous experiences with the measure.
Also, attrition is a problem when participants die or move away and the rate
of attrition increases as time passes. Pedhazur and Pedhazur-Schmelkin (1991)
reported dropout rates as high as 50% in some studies and MacDonald reported
that the attrition rate is between 20% and 25% between each interval in the
Victoria Study. Moreover, participants do not drop out randomly and there may be
important differences between dropouts and those who remain in the study.
So, how bright is the future for MacDonald? "As a graduate student, it
is great to be involved in a study of this magnitude and to have had all of the
opportunities I wouldn’t have otherwise including access to existing data and
being exposed to the methods," he said. "As soon as students in my
position step out into the post-degree research world, we have to hope we can
forge collaborations with larger scale longitudinal studies until we can get on
our feet and acquire own grant funding. There will be a time away from the
luxury of these data sets."
In MacDonald’s immediate future, cross-sectional studies and conducting
studies over a shorter period of time will fill in some of the gaps in his
research program. "I think there are creative ways to go about this,"
he said. "My supervisor and I have been examining the issue of
intra-individual variability using simple trials data at cross sections. In
effect we still have multiple occasions of measurement in very condensed time
frames."
MacDonald reported that journal editors are quite open to LM manuscripts and
they are excited about the possibilities, but unfortunately some of the
techniques associated with data analysis are very complex and labor intensive.
"Revisions typically require much more effort than typical cross-sectional
studies," he said.
"Some researchers simply refuse to accept some of the longitudinal
approaches and techniques such as Structural Equation Modeling," he added.
"Sometimes this is due to being trained in an experimental model rather
than an individual differences model, although it can also reflect lack of
understanding. Perhaps most frustrating, sometimes you get comments like ‘these
multi-occasion results are quite informative but it would be interesting to see
what happens six years from now.’"
The cost of attending the Advanced Training Institute (ATI) in Longitudinal
Methods, Modeling, and Measurement at the University of Virginia in June of 2002
is only $75 for graduate students who are student affiliates of APA (the
non-affiliate fee is $150). They will be trained in modeling growth and change,
item response modeling, modeling intra-individual variability, and modeling
multivariate and categorical outcomes.
Several of the thirty applicants selected for the 2002 institute will be
either postdocs or graduate students. The week-long intensive training institute
will be held at the University of Virginia and the deadline for application is
February 1, 2002. Applications must be submitted electronically by visiting the
Science Directorate Web site at www.apa.org/science/ati-info.html to
download the application.
Pedhazur, E.J., & Pedhazur-Schmelkin, L. (1991). Measurement, design,
and analysis: An integrated approach. Hillsdale, NJ: Erlbaum.
