Questions: Longitudinal Designs: Methods for Studying Change

Question 1

A 30-year longitudinal study of cognitive aging finds that participants who complete all waves score significantly higher on cognitive tests at baseline than those who drop out early. What threat to validity does this represent?

• A. Practice effects — participants improve because they are repeatedly tested
• B. Cohort effects — participants born in the same year share historical experiences
• C. Selective attrition — participants who remain in the study are systematically different from those who drop out
• D. Regression to the mean — extreme scorers at baseline move toward average over time

Answer: (C)

This is a textbook example of selective attrition: participants who remain in the study are healthier and more cognitively intact than those who drop out, making the surviving sample increasingly unrepresentative of the original population. This 'healthy survivor bias' makes cognitive decline appear less steep than it actually is. Selective attrition is especially dangerous because it is not random — the reason for dropout is correlated with the outcome being measured.

Question 2

A researcher wants to determine whether social media use in adolescence causes anxiety in young adulthood. Which design feature of a longitudinal study makes it stronger than a cross-sectional study for this question?

• A. Longitudinal studies use larger samples, increasing statistical power
• B. Longitudinal studies measure the same individuals over time, establishing that social media use preceded anxiety
• C. Longitudinal studies eliminate all confounding variables through repeated measurement
• D. Longitudinal studies avoid the need for a control group

Answer: (B)

The key advantage is temporal precedence: measuring the same individuals at multiple time points lets you verify that social media use (measured first) predicts later anxiety (measured after), ruling out the reverse causal direction. Cross-sectional studies measure variables at one moment and cannot establish which came first. Note that temporal precedence alone doesn't prove causation — confounds remain — but it eliminates one major alternative explanation unavailable in cross-sectional designs.

Question 3

At minimum three measurement waves are required in a longitudinal study to distinguish linear from non-linear change trajectories.

• T. True
• F. False

Answer: True

Two waves can only describe a straight line between two points — you cannot determine whether the trajectory is curved, accelerating, decelerating, or U-shaped with only two measurements. Three or more waves allow you to model non-linear trajectories. For example, a cognitive ability measure taken at ages 10, 15, and 20 can reveal whether growth is constant, accelerating, or plateauing — information hidden when only two time points exist.

Question 4

A longitudinal design automatically establishes a causal relationship between variables measured at different time points.

• T. True
• F. False

Answer: False

Longitudinal designs establish temporal precedence — one of three conditions for causal inference — but not causation by themselves. The other two conditions are covariation (variables must be correlated) and elimination of alternative explanations (ruling out confounds). Without experimental manipulation or very strong theoretical and statistical controls, a longitudinal correlation between Variable A and later Variable B may still reflect a third variable causing both. The topic's own misconceptions section is explicit: longitudinal designs do not automatically imply causality.

Question 5

Why is selective attrition more damaging to a longitudinal study's validity than random attrition, even if the same number of participants drop out in both cases?

Model answer: Random attrition reduces sample size and statistical power but does not bias the remaining sample — the participants who stay are representative of those who left. Selective attrition, by contrast, systematically removes participants who differ from completers on the very outcomes being studied, making the surviving sample unrepresentative of the original population. For example, if the least healthy participants drop out of an aging study, the average health trajectory in the remaining sample looks more positive than it actually is in the population. This bias cannot be corrected simply by having a larger original sample — it affects the direction of conclusions, not just their precision.

This distinction matters for understanding when attrition is a statistical annoyance versus a fatal threat to validity. Researchers must compare completers and dropouts on baseline characteristics to diagnose whether attrition is random or selective.