Questions: Correlational, Longitudinal, and Observational Research

Question 1

A researcher wants to study how growing up in severe poverty affects adult health outcomes. They cannot use a randomized experiment. Why not?

• A. Randomized experiments require larger samples than are practically available for poverty studies
• B. You cannot ethically or practically assign children to grow up in poverty — randomization is impossible here
• C. Observational designs always produce more externally valid results than experiments
• D. Correlational designs have already established that poverty causes poor health, so an experiment is unnecessary

Answer: (B)

Randomization is impossible when the independent variable is an existing life condition rather than something experimenters can assign. This is exactly the class of question — the most important questions in psychology — where correlational and longitudinal designs are not a fallback but the appropriate tool. Option A misunderstands why experiments fail here; option C overstates the case for observational designs; option D confuses correlation with established causation.

Question 2

A study finds that ice cream sales and drowning rates are strongly positively correlated across months of the year. A student concludes that eating ice cream increases drowning risk. What is the most likely explanation for the correlation?

• A. The correlation is purely spurious and has no causal explanation
• B. Drowning incidents cause communities to seek comfort food, driving up ice cream sales
• C. A third variable — hot weather — independently increases both ice cream consumption and swimming activity
• D. The sample size was too small, producing a misleading correlation coefficient

Answer: (C)

This is the classic third-variable problem. A significant correlation is consistent with three causal structures: X causes Y, Y causes X, or a third variable Z causes both. Here, summer heat increases both swimming (and thus drowning risk) and ice cream sales. The correlation is real but the causal interpretation is wrong. This is why correlation does not imply causation — but note that option A is also wrong: the correlation is real and has a real explanation, just not the one stated.

Question 3

Longitudinal designs establish causality because they measure variables over time and can demonstrate that one variable preceded another.

• T. True
• F. False

Answer: False

Temporal precedence (X occurs before Y) is a necessary but not sufficient condition for causation. Longitudinal designs can establish that X preceded Y, which is stronger than a cross-sectional correlation — but unmeasured third variables that precede both X and Y remain possible confounders. Cross-lagged panel models strengthen the causal argument further, but none of these techniques definitively rule out all confounding. Only randomization, by distributing confounders equally between conditions, establishes causation.

Question 4

Observational research designs often have higher ecological validity than laboratory experiments because they study behavior in its natural context rather than a controlled, artificial setting.

• T. True
• F. False

Answer: True

Ecological validity refers to how well findings generalize to real-world conditions. By definition, observational designs study behavior where it naturally occurs — parent-infant interaction at home, peer conflict on a playground — which means what is measured is the actual phenomenon, not a laboratory approximation of it. Experiments gain internal validity (causal conclusions) at the cost of this external validity. The two designs trade off different strengths.

Question 5

Why is a significant correlation between two variables necessary but not sufficient evidence for a causal relationship between them?

Model answer: A correlation tells you that X and Y vary together, but three causal stories are all consistent with that fact: X causes Y, Y causes X, or a third variable Z causes both. A significant correlation rules out the possibility that X and Y are completely unrelated, but it cannot by itself distinguish among these three explanations.

The necessity part: if X truly causes Y, we should observe a correlation between them — its absence would be strong evidence against causation. The insufficiency part: correlation alone cannot tell us the direction or source of the relationship. Establishing causation typically requires additional design features (temporal precedence from longitudinal data, random assignment, statistical controls for confounders) that eliminate the alternative explanations.