Skip to content
AP Statistics · Regression Analysis

Correlation vs causation

Correlation vs causation appears in 9 reviewed AP Statistics questions and sits in Regression Analysis. Below are worked examples showing exactly how the exam tests it, with the answer marked and every distractor explained.

How Correlation vs causation is tested

Example 1Medium
A set of data shows a strong negative association between the number of firefighters sent to a fire and the amount of property damage (in dollars). A student concludes that sending more firefighters causes more property damage. What is the most likely explanation for this association?
  1. AThe relationship is an example of extrapolation beyond the range of the data
  2. BThere is likely a confounding variable, the severity of the fire, associated with both the number of firefighters and the damageCorrect
  3. CThe correlation must be positive, not negative, so the student has misread the data
  4. DThe student's conclusion is correct because the data show a strong association between the number of firefighters and the amount of damage

Why B is right

The severity of the fire is a confounding (lurking) variable. Larger, more severe fires require more firefighters AND cause more property damage. The association between firefighters and damage does not mean firefighters cause damage; both are driven by fire severity.
Why the other 3 choices miss
A
Incorrect because extrapolation involves making predictions outside the observed range of data, which is not the issue here.
C
Incorrect because the problem states the association is strong and negative (more firefighters, more damage), which is plausible due to confounding — there is no reason to doubt the sign of the correlation.
D
Incorrect because a strong association does not establish causation, especially in an observational context where confounding variables likely exist.
Example 2Harder
A school district examines reading scores across 10 elementary schools. For each school, the percentage of students receiving free lunch and the mean reading score are recorded. The regression yields r-squared = 0.78. A district official states: "This proves that free lunch programs lower reading scores." Which of the following is the most valid criticism?
  1. ATen schools is always too few observations to perform regression
  2. BReading scores should be measured using the median, not the mean
  3. CThe r-squared value is not high enough to support any conclusion
  4. DThe analysis uses aggregated school-level data rather than individual student data, and the correlation between aggregated variables can overstate the individual-level relationshipCorrect

Why D is right

This is an example of the ecological fallacy — drawing conclusions about individuals from aggregated (group-level) data. Correlations using group averages tend to be higher than correlations at the individual level. Additionally, the free lunch percentage is a proxy for socioeconomic status, and causal language is inappropriate for an observational study.
Why the other 3 choices miss
A
Incorrect because while 10 observations is small, regression can be performed with as few as 3 observations; the critical issues are the ecological fallacy and the causal claim.
B
Incorrect because the choice between mean and median is a secondary concern; the primary issue is the ecological fallacy and the inappropriate causal inference.
C
Incorrect because r-squared = 0.78 is actually quite high; the problem is the level of analysis and the causal claim, not the strength of the correlation.

Studied alongside

Other concepts from Regression Analysis.