Longitudinal Analysis

Alan Reifman, Texas Tech University
Return to Main Syllabus

Longitudinal Analysis

Approximating Causal Analysis with Survey (Correlational) Data

Longitudinal Study: Gather data for extended period of time

Panel Study: Specific type of longitudinal study in which the same people are followed up over time

Vacationing and Health

Barriers to True Experiment

Unethical to randomly assign people to different amounts of vacationing, because control group would have to be restricted to no vacationing
Extremely expensive to fund vacations for experimental group(s)

Can Approximate Causal Design with Panel Survey
(Ask about frequency of vacationing, state of health, and income)

2007
Frequency of Vacationing

EXAMINE
CORRELATION
BETWEEN
THESE
VARIABLES

2008
Health

To remove influence of income ("third variable"), look at vacationing-health correlation separately in narrow income categories (e.g., $20,000-30,000, $30,001-40,000, etc.). If positive correlation between earlier vacationing and better health a year later appears in all (or most) of the income groups, this suggests income does not affect the relationship between vacationing and health (i.e., vacationing is correlated with better health, regardless of income).

Seems to satisfy all three criteria for causality:

Correlation between frequency of vacationing and health
Time-ordering -- vacationing came before measurement of health
Third variable -- income -- shown to be unimportant (in this hypothetical example)

One thing prevents a full causal interpretation in survey research: There may be "third variables" that you were unaware of and thus did not measure. Random assignment in true experiment equalizes independent variable groups on all background characteristics, whether experimenter has thought of them or not.

Hendrick, S.S., Hendrick, C., & Adler, N.L. (1988). Romantic relationships: Love, satisfaction, and staying together. Journal of Personality and Social Psychology, 54, 980-988.

Time 1	2 month Interval	Time 2
30 couples complete relationship and individual measures	.	See if couples still together (23 were)

Does having certain characteristics at Time 1 "set the wheels in motion" for break-up by Time 2? (Future-oriented prediction)

Results: Couples who were still together at Time 2 differed from those who had dissolved by Time 2 on many of the Time 1 relationship measures (remember Hendrick & Hendrick measure from earlier in the course).

Those who were still together had:

higher Eros
lower Ludus (game-playing)
higher self-disclosure
higher self-esteem
higher investment
higher commitment
higher relationship satisfaction

"Nun Study" of lifelong factors that may contribute to Alzheimer's Disease

Several hundred nuns have been participating for 20 years or so, providing psychological, physical, and cognitive data. They have also permitted researchers to have access to their medical records, thus adding an archival component. According to the project's FAQ page:

The primary research question in the Nun Study is "What factors in early, mid, and late life increase the risk of Alzheimer's disease and other brain diseases such as stroke?"

Article about the nuns leaving their brains to science, to confirm presence or absence of Alzheimer's

http://edition.cnn.com/2007/HEALTH/12/21/nun.brains.ap/index.html

"One reason the nuns are such a valuable research tool is that as members of the same religious order, they all had decades of similar medical treatment, diets, reproductive histories and preventive care. Almost nine out of 10 had been teachers."

How would you rate the external and internal validity of this study?

Final Notes on Causality

Criterion	True Experiment	Longitudinal/Panel Survey
Correlation between presumed "cause" (IV) and "effect" (DV)	YES (e.g., subjects in high alcohol condition most violent)	YES You get correlation-like statistic
Time ordering -- cause comes before effect (billiard ball analogy)	YES (serve alcohol before playing shock game)	YES Vacationing in 2007 could be correlated with health in 2008
Rule out third variables	YES Random assignment equalizes groups on ALL pre-existing characteristics*	TO SOME DEGREE Statistical techniques let you "hold constant" other variables that are included in your study, but cannot hold constant variables not in your study

*This is where "Quasi-Experimentation" falls short. QE, like panel surveys, thus merely allow you to approximate a causal conclusion.

Additional Resources

Panel Studies (re-contact same people over time, thus examining individual development)

Website for the University of Michigan's Panel Study of Income Dynamics (PSID), a long-term examination of changes in people's economic situation.

A major report using PSID data was released in 2008 by the Economic Mobility Project. This chapter on Education and Economic Mobility is quite informative, in my view (especially see Figure 6).

Two panel studies on viewing of violent television and later violent behavior

From Columbia University:
http://www.cnn.com/2002/HEALTH/parenting/03/28/kids.tv.violence/index.html

From University of Michigan:
http://www.lsa.umich.edu/psych/news/spotlight/spotlight/?id=70

Trend Studies (study each year's group of, e.g, high school seniors; people are different every year, so examining changes in the culture or society over time)

University of Michigan's "Monitoring the Future" annual survey of teens' drug use
(MTF home page; chart on marijuana trends)