Skip to content
AP Statistics · Regression Analysis

Association vs causation

Association vs causation appears in 10 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 Association vs causation is tested

Example 1Medium
A regression of exam score on hours of sleep produces the equation y-hat = 40 + 5x with a 95% confidence interval for the slope of (2.8, 7.2). A student claims that this proves sleeping more hours causes higher exam scores. This claim is not justified because
  1. Athe confidence interval is too wide to draw any conclusions about the true population slope
  2. Bobservational data show association, not causation; lurking variables such as study habits may explain bothCorrect
  3. Cthe confidence interval does not contain zero, so the data establish that more sleep causes higher exam scores
  4. Dthe y-intercept of 40 is too high for the model to be valid

Why B is right

Regression analysis of observational data can establish a statistically significant association but cannot prove causation. Lurking variables (study habits, overall health, motivation, stress levels) may be associated with both sleep duration and exam performance. Without a randomized experiment controlling for confounders, the causal claim is not supported.
Why the other 3 choices miss
A
Incorrect. The confidence interval (2.8, 7.2) does not contain zero, indicating a statistically significant positive relationship. Width alone does not invalidate conclusions about the population slope.
C
Incorrect. A significant positive slope demonstrates association, not causation. Statistical significance answers 'Is there a relationship?' not 'Does one variable cause the other?' It cannot establish that more sleep causes higher scores.
D
Incorrect. The y-intercept of 40 represents the predicted exam score for zero hours of sleep, which is an extrapolation. This does not invalidate the model within the data range or affect the causal argument.
Example 2Medium
A city transportation office surveyed a random sample of 300 adult commuters in the city. Each commuter reported their main commute mode and whether they are satisfied with their commute. The results are shown below. | | Car | Transit | Bike | Total | | --- | --- | --- | --- | --- | | Satisfied | 52 | 38 | 30 | 120 | | Not satisfied | 68 | 82 | 30 | 180 | | Total | 120 | 120 | 60 | 300 | A chi-square test of independence is carried out at the alpha = 0.05 level. All expected counts are at least 24, and the test statistic is chi-square = 6.53. Which of the following gives the correct degrees of freedom, p-value, and conclusion?
  1. Adf = 5 and the p-value is about 0.26; since 0.26 > 0.05, fail to reject H0; there is not convincing evidence of an association.
  2. Bdf = 2 and the p-value is about 0.04; since 0.04 < 0.05, reject H0; there is convincing evidence of an association between commute mode and satisfaction among the city's adult commuters.Correct
  3. Cdf = 2 and the p-value is about 0.04; reject H0; there is convincing evidence that a commuter's mode of travel causes a change in satisfaction.
  4. Ddf = 2 and the p-value is about 0.04; reject H0; there is about a 4% probability that commute mode and satisfaction are independent.

Why B is right

For a two-way table, df = (rows - 1)(columns - 1) = (2 - 1)(3 - 1) = 2. The expected counts are 48, 48, 24 in the Satisfied row and 72, 72, 36 in the Not satisfied row, so the contributions are 0.333 + 2.083 + 1.5 + 0.222 + 1.389 + 1.0, which gives chi-square = 6.53. With df = 2, P(chi-square >= 6.53) is about 0.038. That is less than 0.05, so we reject H0 and conclude that there is convincing evidence of an association between commute mode and satisfaction in the population of the city's adult commuters.
Why the other 3 choices miss
A
Uses df = (number of cells) - 1 = 5 instead of (r - 1)(c - 1) = 2. With df = 5 the p-value would be about 0.26, which leads to the wrong decision.
C
The df, p-value, and decision are right, but the conclusion claims causation. This was an observational survey with no random assignment, so it supports only an association.
D
Misreads the p-value as the probability that H0 is true. The p-value is the probability of getting a chi-square statistic of 6.53 or larger if H0 (independence) were true.

Studied alongside

Other concepts from Regression Analysis.