In a paired t-test, why is it important that the order of treatments is randomized when both treatments are applied to the same subjects?
- ATo make the standard deviation of the differences equal to zero
- BTo reduce the effect of lurking variables and order effects on the differencesCorrect
- CTo guarantee the differences are normally distributed
- DTo ensure the sample size is large enough
Why B is right
Randomizing the order of treatments helps control for order effects (such as practice, fatigue, or carryover effects) that could confound the results. Without randomization, systematic differences due to order rather than treatment could bias the paired differences.
Why the other 3 choices miss
- A
- Incorrect because randomization does not reduce variability to zero. It aims to eliminate systematic bias, not eliminate all variability.
- C
- Incorrect because randomization does not guarantee normality of differences. Normality depends on the population distribution and sample size.
- D
- Incorrect because randomization does not affect sample size; it affects the internal validity of the comparison.
In a matched-pairs experiment, 20 subjects each receive both a caffeine pill and a placebo on separate days (in random order). The researcher records reaction time (milliseconds) for each condition. After computing the differences (caffeine minus placebo) for each subject, the researcher finds a mean difference of -15 ms with a standard deviation of 20 ms. The null hypothesis is that caffeine has no effect (mean difference = 0). Which procedure should be used?
- ATwo-proportion z-test comparing proportions who improved under each condition.
- BTwo-sample t-test using the 20 caffeine times and 20 placebo times as independent groups.
- COne-sample t-test on the 20 differences, with H0: μ_d = 0.Correct
- DChi-square goodness of fit test for reaction time distributions.
Why C is right
A matched-pairs design produces one difference score per subject. The appropriate test is a one-sample t-test applied to the n = 20 difference scores, testing H0: μ_d = 0. The t-statistic would be t = (x-bar_d - 0)/(s_d/sqrt(n)) = -15/(20/sqrt(20)) = -15/4.47 ≈ -3.35.
Why the other 3 choices miss
- A
- Incorrect because the data are continuous (reaction times in milliseconds), not proportions. A two-proportion z-test requires categorical outcomes.
- B
- Incorrect because the two sets of observations are not independent — each person provides both measurements. A two-sample t-test assumes independence between groups, which is violated in a matched-pairs design.
- D
- Incorrect because a chi-square goodness of fit test assesses whether observed categorical frequencies match expected frequencies. Continuous reaction time data analyzed in a matched-pairs design call for a t-test on differences.
A paired t-test is used to compare pre-test and post-test scores for 15 students. The mean difference (post - pre) is d-bar = 4.2, and the standard deviation of the differences is s_d = 6.8. The t-statistic for testing H0: mu_d = 0 is
- At = 4.2 * sqrt(15) / 6.8 ≈ 2.39
- Bt = (4.2 - 6.8) / sqrt(15) ≈ -0.67
- Ct = 4.2 / (6.8 / sqrt(15)) = 4.2 / 1.756 ≈ 2.39Correct
- Dt = 4.2 / 6.8 ≈ 0.62
Why C is right
For a paired t-test, t = d-bar / (s_d / sqrt(n)) = 4.2 / (6.8 / sqrt(15)) = 4.2 / (6.8 / 3.873) = 4.2 / 1.756 ≈ 2.39. The test statistic compares the mean of the differences to zero, standardized by the standard error of the differences.
Why the other 3 choices miss
- A
- Incorrect. While this produces the same numerical answer as A (both equal approximately 2.39), it rearranges the formula. Choice A shows the standard form: d-bar divided by the standard error. Both A and D are algebraically equivalent, so the answer is A as it shows the standard formula.
- B
- Incorrect. This incorrectly subtracts s_d from d-bar and divides by sqrt(n). The formula involves dividing d-bar by the standard error, not subtracting s_d.
- D
- Incorrect. This divides d-bar by s_d without accounting for the sample size. The denominator should be the standard error s_d/sqrt(n), not the standard deviation s_d.