AP Stats Ch. 24 (Matched Pairs t-test and confidence intervals)
Matched Pairs Overview
Definition: Matched pairs involve pairing up data to minimize variability between groups.
Purpose: Used in hypothesis testing and confidence intervals for data that is naturally related.
Examples of Matched Pair Designs
Observations collected in pairs, such as:
Same subjects measured before and after treatment.
Twin studies comparing twins' characteristics.
Comparing left and right shoes on the same person.
Motivation: Eliminate variability and improve the accuracy of testing.
Differences and Pairwise Analysis
When data is paired, the analysis focuses on the pairwise differences rather than the original datasets.
Calculation: Differences are calculated from paired data, and these differences are analyzed using a one-sample t-test.
Example: For a left foot/right foot dataset, calculate a third column based on the differences.
Paired T-Test Mechanics
Key Components:
Null Hypothesis (H0): The mean of the differences (μ_d) equals zero (no effect).
Alternative Hypothesis (H1): μ_d is greater than zero (indicates an effect).
Formula:
[ t = \frac{\bar{d} - \mu_0}{\text{SE}} ] Where ( \bar{d} ) = average of the differences, ( \mu_0 ) is typically zero, and ( \text{SE} ) is the standard error of the differences.
Degrees of Freedom: ( n - 1 ), where n is the number of pairs.
Obtain p-value using the t-distribution.
Assumptions for Paired T-Test
Data must be paired: Observations are related.
Independence assumption: Differences must be independent.
Randomized condition: Randomization in sample selection.
Sample size condition: At least ten samples for normal approximation.
Normal population assumption: Population of differences should follow a normal distribution.
Nearly normal condition: Check with histogram or probability plot of the differences.
Confidence Interval Calculations
Formula: ( CI = \bar{d} \pm t_{} \cdot \text{SE} ) Where ( t_{} ) is the critical value from the t-distribution.
Common Pitfalls
Avoid using a two-sample t-test for paired data: This increases variability, negating the purpose of pairing.
Use paired t-test only when data are actually paired: If not paired, use a two-sample t-test.
Watch for outliers: Outliers affect the differences, impacting the test results.
Do not use box plots for comparing means of paired groups; this may reintroduce variation that the matching aimed to eliminate.
Example Application (Medicine vs. Placebo)
Input data into TI Nspire:
Analyze the placebo and medicine data along with the differences.
Setup:
Null Hypothesis: mean difference is zero (no effect).
Alternative Hypothesis: mean difference is greater than zero (indicating effectiveness of medicine).
Perform paired t-test using the differences column:
If p-value < alpha (0.05), reject the null hypothesis, providing evidence that the medicine works.
Summary
The concepts of matched pairs are crucial for reducing variability and accurately detecting effects in comparisons between groups.
Bottom Line:
Matched pair designs provide a robust framework for making valid inferences about the effects of treatments, by focusing on the differences within paired observations and minimizing external variability.