Comprehensive Study Guide to Matched Pairs t-Tests

Foundations of Matched Pairs t-Tests

  • Definition and Experimental Design:

    • A matched pairs design is a special form of a paired comparison experiment where subjects or experimental units are paired up, or each subject performs two tasks under two different conditions.
    • Each subject acts as their own control by completing the experiment twice (e.g., once under condition A and once under condition B).
    • The presentation order of conditions must be randomized for each subject to control for order and learning effects.
    • On examinations, problems do not explicitly state their category; students must independently diagnose whether a problem requires a matched pairs procedure or a two-independent-samples procedure.
  • Establishing Paired Differences:

    • Raw paired data must first be converted into a single sample of differences before conducting inference.
    • The direction of subtraction (e.g., condition 1 minus condition 2) must be explicitly established at the outset.
    • Once differences are calculated, all subsequent inference uses only the differences; the original raw data values are discarded.
    • The selected direction of subtraction must be maintained consistently across the hypotheses, test statistic, p-value interpretation, and confidence interval.

Case Study 1: Instrument Knob Usability (Right vs. Left Hand)

  • Experimental Context and Setup:

    • Research Question: Is it easier for a right-handed person to use an instrument with knobs on the right side or the left side of the device?
    • Subjects and Protocol: 1010 right-handed individuals were asked to perform an identical task using two instrument setups: one with knobs on the right side and one with knobs on the left side.
    • Randomization: The order in which each individual used the two instrument setups was randomized.
    • Variable of Measurement: The response variable is the completion time (measured in seconds). When measuring performance by time, a task being "easier" directly corresponds to taking less time.
  • Step-by-Step Data Transformation (Differences: Right−Left\text{Right} - \text{Left}):

    • Subject 1: 113−137=−24113 - 137 = -24
    • Subject 2: 105−105=0105 - 105 = 0
    • Subject 3: 130−133=−3130 - 133 = -3
    • Subject 4: 101−108=−7101 - 108 = -7
    • Subject 5: 138−115=23138 - 115 = 23
    • Subject 6: 118−170=−52118 - 170 = -52
    • Subject 7: 87−103=−1687 - 103 = -16
    • Subject 8: 116−145=−29116 - 145 = -29
    • Subject 9: 75−78=−375 - 78 = -3
    • Subject 10: 96−107=−1196 - 107 = -11
    • Resulting Sample of Differences (dd): −24,0,−3,−7,23,−52,−16,−29,−3,−11-24, 0, -3, -7, 23, -52, -16, -29, -3, -11
  • Part (a): Checking Conditions:

    • The sample consists of a simple random sample (SRS) of n=10n = 10 pairs from a normal distribution.
    • Note that nn represents the total number of pairs, not the total number of observations.
  • Part (b): Hypotheses Formulation:

    • Because the claim tests if right-handed knobs are easier (i.e., take less time: Right<Left\text{Right} < \text{Left}), subtracting right minus left leads to an expectation of more negative differences than positive differences.
    • Null Hypothesis: H0:μd=0H_0: \mu_d = 0
    • Alternative Hypothesis: Ha:μd<0H_a: \mu_d < 0
    • Here, μd\mu_d denotes the true population mean difference in completion time (Right−Left\text{Right} - \text{Left}).
  • Part (c): Test Statistic Calculation:

    • Formula:         t=xˉ−μ0snt = \frac{\bar{x} - \mu_0}{\frac{s}{\sqrt{n}}}
    • Summary Statistics of Differences:
      • Sample mean difference: xˉ=−12.2\bar{x} = -12.2
      • Null hypothesized difference: μ0=0\mu_0 = 0
      • Sample standard deviation: s=20.0155s = 20.0155 (rounded to four decimal places)
      • Sample size (pairs): n=10n = 10
    • Computation:         t=−12.2−020.015510=−1.9275t = \frac{-12.2 - 0}{\frac{20.0155}{\sqrt{10}}} = -1.9275
    • Degrees of Freedom:         df=n−1=10−1=9df = n - 1 = 10 - 1 = 9
  • Part (d): Determination of the p-Value:

    • Prerequisites for Table C:
      1. Value and sign of test statistic: t=−1.9275t = -1.9275. The test statistic is negative, which matches the direction of the alternative hypothesis (<<). A directional mismatch indicates an error in calculation or hypothesis setup.
      2. Degrees of freedom: df=9df = 9.
      3. Test nature: One-sided test.
    • Table C Look-Up:
      • Locate the row for df=9df = 9 and read across the positive values: 0.70.7, 0.880.88, 1.101.10, 1.381.38, 1.8331.833, 2.2622.262.
      • The absolute value of the test statistic (1.92751.9275) falls between 1.8331.833 and 2.2622.262.
      • Tracing vertically to the one-sided p-value row at the top/bottom of Table C establishes the bounds:             0.025<p<0.050.025 < p < 0.05
      • Bounding values must always be written from smaller to larger (0.0250.025 to 0.050.05).
  • Part (e): Statistical Decision and Conclusion:

    • Decision Criterion: Selecting a representative value inside the interval, such as p=0.03p = 0.03:
      • Comparing to α=0.05\alpha = 0.05: 0.03<0.050.03 < 0.05, which falls in the rejection region.
      • Comparing to α=0.01\alpha = 0.01: 0.03>0.010.03 > 0.01, which does not reject.
    • Formal Decision: Reject the null hypothesis at the α=0.05\alpha = 0.05 significance level.
    • Formal Conclusion: There is sufficient evidence that the mean difference in completion time (calculated as right minus left) is less than zero (μd<0\mu_d < 0).
    • Practical Meaning: Right-handed individuals are significantly quicker when using an instrument equipped with right-handed knobs compared to left-handed knobs.
  • Part (f): 95% Confidence Interval Calculation:

    • Formula:         xˉ±t∗(sn)\bar{x} \pm t^* \left( \frac{s}{\sqrt{n}} \right)
    • Critical Value (t∗t^*): From Table C, matching a 95%95\% confidence level with df=9df = 9 yields t∗=2.262t^* = 2.262.
    • Computation:         −12.2±2.262(20.015510)-12.2 \pm 2.262 \left( \frac{20.0155}{\sqrt{10}} \right)
      • Lower bound: −12.2−2.262(20.015510)=−26.5172-12.2 - 2.262 \left( \frac{20.0155}{\sqrt{10}} \right) = -26.5172
      • Upper bound: −12.2+2.262(20.015510)=2.1172-12.2 + 2.262 \left( \frac{20.0155}{\sqrt{10}} \right) = 2.1172
    • Rounding: Rounding to whole numbers (reflecting original data precision) yields the interval [−27,2][-27, 2].
  • Part (g): Interpretation of the Confidence Interval:

    • We are 95%95\% confident that the true mean difference in time (right minus left) is between −27-27 seconds and 22 seconds.
    • Substantive meaning: On average, the difference in task completion time is expected to range from 2727 seconds faster to 22 seconds slower when using knobs on the right side.

Case Study 2: Football Distance Study (Helium vs. Air)

  • Experimental Context and Design Analysis:

    • Research Question: Does a football filled with helium travel farther than one filled with ordinary air?
    • Subjects and Protocol: 99 novice kickers each punted both footballs (one helium-filled and one air-filled).
    • Design Evaluation and Caveat: Using novice kickers introduces high kicking variability; using experienced kickers who know how to kick consistently would provide a cleaner test of the physical properties of the balls.
    • Controls: The kicking order of the footballs was randomized, and kickers were blinded to which football they were kicking.
    • Response Variable: Distance traveled, measured in yards.
  • Part (a) & (b): Setup and Hypotheses:

    • Direction of Differences: Defined as Helium−Air\text{Helium} - \text{Air}.
    • Hypotheses:
      • H0:μd=0H_0: \mu_d = 0
      • Ha:μd>0H_a: \mu_d > 0
    • If subtraction had been taken as Air−Helium\text{Air} - \text{Helium}, the alternative sign would flip to <0< 0.
  • Part (c), (d) & (e): Inference and Conclusion:

    • Degrees of Freedom:         df=n−1=9−1=8df = n - 1 = 9 - 1 = 8
    • p-Value Interval: Table C gives a one-sided p-value interval:         0.20<p<0.250.20 < p < 0.25
    • Decision: Because p>0.05p > 0.05, fail to reject (do not reject) the null hypothesis at the α=0.05\alpha = 0.05 significance level.
    • Conclusion Statement: There is insufficient evidence that the mean difference in distance traveled (helium minus air) is greater than zero.
  • Part (f) & (g): 90% Confidence Interval:

    • Critical Value: Looking up a 90%90\% confidence level at df=8df = 8 in Table C yields t∗=1.860t^* = 1.860.
    • Interpretation Format: Requires stating the confidence level (90%90\%), the exact parameter name ("mean difference"), the units of measurement ("yards traveled"), the subtraction order ("helium minus air"), and the numerical boundary values.

Matched Pairs vs. Two-Independent-Samples t-Tests

  • Terminology and Conceptual Distinction:

    • Matched Pairs t-Test: Analyzes the "mean difference" (μd\mu_d). Individual differences are calculated pair-by-pair first, and inference is conducted on the resulting single sample of differences.
    • Two-Sample Independent t-Test: Analyzes the "difference in the means" (μ1−μ2\mu_1 - \mu_2). The two samples are independent of each other, their means (xˉ1\bar{x}_1 and xˉ2\bar{x}_2) are computed separately, and their difference is evaluated.
  • Mathematical and Procedural Differences:

    • Two-sample independent procedures utilize data from two entirely distinct groups with no pairing.
    • Formulas for two independent samples are substantially more complex than paired difference formulas, requiring calculator implementation for numeric evaluation.
    • Both test types rely on the tt distribution, but the standard errors, degrees of freedom, and verbal interpretations differ.

Methodological Best Practices, Table C Guidelines, and Examination Protocol

  • Mandatory Use of Table C:

    • All critical values (t∗t^*) and p-value intervals must be obtained directly from Table C.
    • Automated software or external AI tools that compute exact p-values rather than Table C bounding intervals must not be used; non-Table C values are identified and marked incorrect.
  • Precision and Rounding Guidelines:

    • Sample mean difference (xˉ\bar{x}): Maintain exact value if terminating; otherwise, round to four decimal places.
    • Standard deviation (ss): Round to four decimal places (20.015520.0155 in Case Study 1).
    • Test statistic (tt): Round to four decimal places (−1.9275-1.9275 in Case Study 1).
    • Confidence interval endpoints: Can be expressed to four decimal places or rounded to the nearest integer matching the precision of the raw data (e.g., −27-27 to 22 seconds).
  • Reporting Requirements for Conclusions:

    • Whenever stating a conclusion or confidence interval for paired data, two pieces of descriptive information must always be provided:
      1. The physical quantity being measured (including context and units, e.g., "time in seconds" or "distance in yards").
      2. The exact direction of subtraction used to generate the differences (e.g., "right minus left" or "helium minus air").

Questions & Discussion

  • Question: Is it required on homework or exams to write out the preliminary subtraction expression in red (e.g., explicit column subtractions)?

    • Response: No, writing out the step-by-step subtraction explicitly is not required on homework or exams. It is shown during demonstrations to clarify where values originate, though many students prefer writing it out to organize their calculations.
  • Question: In Table C, are the numbers inside the chart only used for confidence intervals, while p-values come from elsewhere?

    • Response: The numbers in the body of the table are critical values. To determine a p-value for a test statistic, locate the test statistic between two numbers within the appropriate degrees-of-freedom row in the chart, then follow those columns to the margin to read the p-value range. To compute a confidence interval, look up the desired confidence level at the bottom and find the corresponding critical value (t∗t^*) inside the table row for the given degrees of freedom.
  • Question: In the interpretation sentence for part (g), what information is strictly required regarding the units?

    • Response: The statement must specify both what the variable describes with its units (e.g., mean yards traveled) and the explicit subtraction direction (e.g., helium minus air) to fully detail the sample statistic.