Paired Data and Matched Pairs Design: Comprehensive Guide

Master paired data analysis and matched pairs design: analyzing dependent samples with detailed step-by-step solutions through these 5 exercises.

Paired Data and Matched Pairs Fundamentals
d̄ = Σdi/n, sd = √[Σ(di - d̄)²/(n-1)], t = d̄/(sd/√n)
Matched Pairs Statistics
Key Definitions:

Paired Data: Two measurements on the same subject or matched subjects

Differences (di): di = xi - yi for each pair

Mean Difference (d̄): Average of all paired differences

Standard Deviation of Differences (sd): Variability of paired differences

Matched Pairs Design: Experimental design where subjects are paired based on characteristics

Dependent Samples: Samples where observations are related/correlated

5-Step Procedure:
  1. Calculate differences: di = xi - yi for each pair
  2. Compute summary statistics: d̄ and sd
  3. State hypotheses: About mean difference μd
  4. Calculate test statistic: t = d̄/(sd/√n)
  5. Make decision: Compare to critical value or p-value
Tip 1: Always subtract in consistent order (before - after, method A - method B)
Tip 2: Paired tests have higher power than two-sample tests when data is correlated
Tip 3: Use paired tests for before-after, twin, or matched subjects
When to Use: Dependent samples, repeated measures, matched subjects
Advantages: Removes individual variation, increases power, controls for confounding
Solution: Exercises 1 to 3
1 Before-After Weight Loss
Exercise 1
A weight loss program is tested on 10 participants. Weights (in lbs) before and after the program are given. Test if the program is effective at α = 0.05.
Data:

Before: 180, 175, 190, 165, 185, 170, 195, 160, 178, 182

After: 175, 170, 185, 162, 180, 168, 190, 158, 175, 179

Analysis Approach:

Matched pairs t-test: Before - After (positive difference = weight loss)

Step 1: Calculate differences

d = [5, 5, 5, 3, 5, 2, 5, 2, 3, 3]

Step 2: Calculate mean difference

d̄ = (5+5+5+3+5+2+5+2+3+3)/10 = 38/10 = 3.8 lbs

Step 3: Calculate standard deviation of differences

Deviations: [1.2, 1.2, 1.2, -0.8, 1.2, -1.8, 1.2, -1.8, -0.8, -0.8]

Sum of squared deviations: 1.44 + 1.44 + 1.44 + 0.64 + 1.44 + 3.24 + 1.44 + 3.24 + 0.64 + 0.64 = 15.6

sd = √[15.6/(10-1)] = √1.733 = 1.317

Step 4: State hypotheses

H₀: μd = 0 (no weight loss)

H₁: μd > 0 (weight loss occurred)

Step 5: Calculate test statistic

t = d̄/(sd/√n) = 3.8/(1.317/√10) = 3.8/0.417 = 9.12

Step 6: Find critical value

df = n-1 = 9, α = 0.05, right-tailed: t₀.₀₅ = 1.833

Step 7: Make decision

Since t = 9.12 > 1.833, reject H₀

Step 8: Conclusion

There is significant evidence that the weight loss program is effective.

Final Answer:

t = 9.12, df = 9, p-value < 0.001

Reject H₀: Program is effective (average weight loss = 3.8 lbs)

Applied Rules:

• Paired Analysis: Convert to one-sample t-test on differences

• Right-tailed Test: Testing for positive effect (weight loss)

• High t-statistic: Strong evidence against H₀

2 Memory Training Study
Exercise 2
A memory training program is tested on 8 subjects. Scores before and after training are recorded. Test if training improves scores at α = 0.01.
Data:

Before: 70, 75, 80, 65, 85, 72, 78, 68

After: 75, 80, 85, 70, 88, 78, 82, 72

Step 1: Calculate differences

d = [5, 5, 5, 5, 3, 6, 4, 4]

Step 2: Calculate mean difference

d̄ = (5+5+5+5+3+6+4+4)/8 = 37/8 = 4.625

Step 3: Calculate standard deviation

Deviations: [0.375, 0.375, 0.375, 0.375, -1.625, 1.375, -0.625, -0.625]

Sum of squared deviations: 0.141 + 0.141 + 0.141 + 0.141 + 2.641 + 1.891 + 0.391 + 0.391 = 5.878

sd = √[5.878/7] = √0.8397 = 0.916

Step 4: State hypotheses

H₀: μd ≤ 0 (training doesn't improve scores)

H₁: μd > 0 (training improves scores)

Step 5: Calculate test statistic

t = 4.625/(0.916/√8) = 4.625/0.324 = 14.28

Step 6: Find critical value

df = 7, α = 0.01, right-tailed: t₀.₀₁ = 2.998

Step 7: Make decision

Since t = 14.28 > 2.998, reject H₀

Step 8: Conclusion

At α = 0.01, there is significant evidence that training improves memory scores.

Final Answer:

t = 14.28, df = 7, p-value < 0.001

Reject H₀: Training significantly improves scores (avg improvement = 4.63 points)

Applied Rules:

• Small Sample: Use t-distribution

• Conservative Alpha: α = 0.01 requires stronger evidence

• High Power: Large t-statistic indicates strong effect

3 Blood Pressure Study
Exercise 3
Blood pressure measurements (systolic) for 6 patients before and after medication. Test if medication significantly reduces BP at α = 0.05.
Data:

Before: 140, 145, 150, 135, 142, 148

After: 135, 140, 145, 130, 138, 142

Step 1: Calculate differences

d = [5, 5, 5, 5, 4, 6]

Step 2: Calculate mean difference

d̄ = (5+5+5+5+4+6)/6 = 30/6 = 5.0 mmHg

Step 3: Calculate standard deviation

Deviations: [0, 0, 0, 0, -1, 1]

Sum of squared deviations: 0 + 0 + 0 + 0 + 1 + 1 = 2

sd = √[2/5] = √0.4 = 0.632

Step 4: State hypotheses

H₀: μd = 0 (no BP change)

H₁: μd > 0 (BP reduction)

Step 5: Calculate test statistic

t = 5.0/(0.632/√6) = 5.0/0.258 = 19.38

Step 6: Find critical value

df = 5, α = 0.05, right-tailed: t₀.₀₅ = 2.015

Step 7: Make decision

Since t = 19.38 > 2.015, reject H₀

Step 8: Conclusion

There is significant evidence that medication reduces blood pressure.

Final Answer:

t = 19.38, df = 5, p-value < 0.001

Reject H₀: Medication significantly reduces BP (avg reduction = 5.0 mmHg)

Applied Rules:

• Very Small Sample: df = 5, t-distribution

• Large Effect Size: High t-statistic despite small n

• Medical Context: Significant BP reduction is clinically important

Solution: Exercises 4 to 5
4 Software Performance
Exercise 4
Processing times (seconds) for 7 programs using old vs new software. Test if new software is faster at α = 0.05.
Data:

Old Software: 12, 15, 10, 18, 14, 16, 13

New Software: 10, 12, 9, 15, 12, 14, 11

Step 1: Calculate differences

d = [2, 3, 1, 3, 2, 2, 2] (Old - New = time saved)

Step 2: Calculate mean difference

d̄ = (2+3+1+3+2+2+2)/7 = 15/7 = 2.14 seconds

Step 3: Calculate standard deviation

Deviations: [-0.14, 0.86, -1.14, 0.86, -0.14, -0.14, -0.14]

Sum of squared deviations: 0.02 + 0.74 + 1.30 + 0.74 + 0.02 + 0.02 + 0.02 = 2.86

sd = √[2.86/6] = √0.477 = 0.691

Step 4: State hypotheses

H₀: μd ≤ 0 (new software not faster)

H₁: μd > 0 (new software is faster)

Step 5: Calculate test statistic

t = 2.14/(0.691/√7) = 2.14/0.261 = 8.20

Step 6: Find critical value

df = 6, α = 0.05, right-tailed: t₀.₀₅ = 1.943

Step 7: Make decision

Since t = 8.20 > 1.943, reject H₀

Step 8: Conclusion

There is significant evidence that the new software is faster.

Final Answer:

t = 8.20, df = 6, p-value < 0.001

Reject H₀: New software is significantly faster (avg improvement = 2.14 seconds)

Applied Rules:

• Technology Testing: Matched pairs for same programs

• Performance Metrics: Time reduction is beneficial

• Consistent Direction: All differences positive (faster)

5 Educational Intervention
Exercise 5
Pre- and post-test scores for 9 students using a new teaching method. Test if there's any significant change at α = 0.05.
Data:

Pre-test: 72, 68, 75, 70, 65, 78, 73, 69, 71

Post-test: 78, 72, 79, 75, 70, 82, 77, 74, 75

Step 1: Calculate differences

d = [6, 4, 4, 5, 5, 4, 4, 5, 4] (Post - Pre = improvement)

Step 2: Calculate mean difference

d̄ = (6+4+4+5+5+4+4+5+4)/9 = 41/9 = 4.56

Step 3: Calculate standard deviation

Deviations: [1.44, -0.56, -0.56, 0.44, 0.44, -0.56, -0.56, 0.44, -0.56]

Sum of squared deviations: 2.07 + 0.31 + 0.31 + 0.19 + 0.19 + 0.31 + 0.31 + 0.19 + 0.31 = 4.22

sd = √[4.22/8] = √0.528 = 0.727

Step 4: State hypotheses

H₀: μd = 0 (no change)

H₁: μd ≠ 0 (change occurred)

Step 5: Calculate test statistic

t = 4.56/(0.727/√9) = 4.56/0.242 = 18.84

Step 6: Find critical value

df = 8, α = 0.05, two-tailed: t₀.₀₂₅ = ±2.306

Step 7: Make decision

Since |t| = 18.84 > 2.306, reject H₀

Step 8: Conclusion

There is significant evidence of a change in test scores.

Final Answer:

t = 18.84, df = 8, p-value < 0.001

Reject H₀: Significant improvement in scores (avg gain = 4.56 points)

Applied Rules:

• Two-tailed Test: Testing for any change (increase or decrease)

• All Positive Differences: Consistent improvement pattern

• Highly Significant: Very strong evidence against H₀

Comprehensive Paired Data Guide
d̄ = Σdi/n, sd = √[Σ(di - d̄)²/(n-1)], t = d̄/(sd/√n)
Matched Pairs Formula
Mean Difference
d̄ = Σdi/n
Center of distribution
Std Dev of Differences
sd = √[Σ(di - d̄)²/(n-1)]
Variability measure
Test Statistic
t = d̄/(sd/√n)
Standardized difference
Key Paired Data Concepts:

Within-Subject Design: Same subjects measured twice

Blocking Variable: Characteristics used to match subjects

Correlation Advantage: Paired design removes individual variation

Reduced Variance: sd is typically smaller than pooled variance

Increased Power: Higher probability of detecting true effects

When to Use Paired Design:

Before-After Studies: Same subjects measured pre/post intervention

Matched Subjects: Similar individuals paired based on characteristics

Cross-Over Trials: Subjects receive both treatments

Twin Studies: Genetically similar pairs

Repeated Measures: Multiple measurements on same unit

Memory Tip: "Paired = Related, Independent = Separate"
Memory Tip: "Subtract in logical order (After - Before)"
Memory Tip: "Paired tests have more power than independent tests"
Common Mistakes: Treating paired as independent, wrong subtraction order, forgetting to check conditions
Best Practices: Verify pairing, check normality of differences, interpret in context

Questions & Answers

Question: How do I know when to use a paired t-test versus a two-sample t-test? What's the key difference?

Answer: The key is determining if your samples are dependent or independent:

Use Paired T-Test When:

  • Same subjects measured twice: Before/after, pre/post
  • Matched pairs: Twins, matched on characteristics
  • Related subjects: Spouses, siblings, same experimental unit
  • Within-subject design: Each subject receives both treatments

Use Two-Sample T-Test When:

  • Independent groups: Completely different subjects
  • Between-subject design: Different subjects receive different treatments
  • Random assignment: Subjects assigned to groups independently

Key Question: Can you pair up the observations in a meaningful way? If yes, use paired test. If no, use two-sample test.

Advantage of Paired Test: Removes individual variation, leading to smaller standard error and higher power.

Question: Why do paired t-tests have higher power than two-sample t-tests? What's the statistical reasoning?

Answer: Paired t-tests have higher power due to reduced variability:

Two-Sample T-Test Standard Error:

SE = √[(s₁²/n₁) + (s₂²/n₂)]

This includes both within-group and between-subject variability.

Paired T-Test Standard Error:

SE = sd/√n

Where sd is the standard deviation of the differences.

Why sd is Typically Smaller:

  • Removes individual variation: Subtracts out baseline differences between subjects
  • Focuses on treatment effect: Only measures variation in differences
  • Controls for confounders: Same subjects eliminate subject-specific factors

Mathematical Reasoning:

Var(d) = Var(X - Y) = Var(X) + Var(Y) - 2Cov(X,Y)

When X and Y are positively correlated (as in paired data), Cov(X,Y) > 0, making Var(d) smaller.

Result: Smaller standard error → larger t-statistic → higher probability of detecting true effects.

Question: What happens if I ignore the pairing and treat paired data as independent? How does this affect my results?

Answer: Treating paired data as independent has several negative consequences:

1. Reduced Power:

  • Larger standard error due to including individual variation
  • Smaller test statistic
  • Higher probability of Type II error (missing true effects)

2. Inflated Variance:

  • Between-subject variability added to treatment effect
  • Makes it harder to detect true differences

3. Incorrect Inference:

  • Violates independence assumption of two-sample test
  • Invalidates p-values and confidence intervals
  • Incorrect degrees of freedom

Example: If paired data has sd = 2 but individual measurements have s₁ = s₂ = 8, the paired test SE would be 2/√n while the two-sample test SE would be √[(64/n) + (64/n)] = √(128/n). The two-sample test has much larger standard error!

Bottom Line: Always respect the study design. If data is paired, analyze it as paired data. This maximizes your ability to detect true effects while maintaining statistical validity.