Topic 10 Study Guide: Constructing, Deconstructing, and Interpreting Confidence Intervals for Single Means and Mean Differences

Foundations of Confidence Intervals and Multipliers

  • General Structure of a Confidence Interval:

    • Any confidence interval follows the fundamental structural equation:     Confidence Interval=Sample Statistic±(Confidence Multiplier×Standard Error)\text{Confidence Interval} = \text{Sample Statistic} \pm (\text{Confidence Multiplier} \times \text{Standard Error})
    • When estimating a single population mean μ\mu using measurement data, the formula is:     xˉ±(Multiplier×σn)\bar{x} \pm \left(\text{Multiplier} \times \frac{\sigma}{\sqrt{n}}\right)
    • For a sample mean xˉ=16\bar{x} = 16, a 95% confidence level uses a multiplier of 22:     16±(2×σn)16 \pm \left(2 \times \frac{\sigma}{\sqrt{n}}\right)
  • Derivation of the 95% Confidence Multiplier:

    • A 95% confidence interval targets the middle 95% (0.950.95) of the standard normal distribution curve.
    • Area left in the two tails combined equals 100%−95%=5%100\% - 95\% = 5\% (0.050.05).
    • Divided equally between both sides, each individual tail contains 2.5%2.5\% (0.0250.025) of the area.
    • The area below the upper cutoff value is 95%+2.5%=97.5%95\% + 2.5\% = 97.5\% (0.9750.975).
    • Referencing a standard normal distribution table (zz-chart):
    • The zz-score with 97.5%97.5\% of the distribution below it is z=1.96z = 1.96.
    • The zz-score with 2.5%2.5\% of the distribution below it is z=−1.96z = -1.96.
    • Rounding rule for Topic 10: 1.961.96 and −1.96-1.96 are rounded to 22. For Topic 10 examinations, the 95% multiplier is always rounded to 22.
    • Topic 11 transition: In Topic 11, rounding to 22 ceases, and the precise value 1.961.96 is used.
  • Calculating Confidence Multipliers for Alternative Confidence Levels (80% Example):

    • To achieve an 80% confidence level, the target interval encompasses the central 80% (0.800.80) of the curve.
    • The total uncovered area is 100%−80%=20%100\% - 80\% = 20\% (0.200.20), resulting in 10%10\% (0.100.10) in each tail.
    • The cumulative area below the upper boundary is 80%+10%=90%80\% + 10\% = 90\% (0.900.90).
    • From the zz-chart, the zz-score corresponding to a cumulative area of 90%90\% is z=1.28z = 1.28 (with the lower boundary at z=−1.28z = -1.28).
    • The resulting 80% confidence interval formula is:     Statistic±(1.28×Standard Error)\text{Statistic} \pm (1.28 \times \text{Standard Error})

Critical Errors in Constructing Confidence Intervals for Measurement Means

  • Scenario Details:

    • Sample size: n=400n = 400 students.
    • Sample mean salary: xˉ=13,455\bar{x} = 13,455.
    • Standard deviation: σ=1,800\sigma = 1,800.
  • Most Common Error (Directly Multiplying Raw Standard Deviation):

    • Incorrect approach: Identifying the 95% rule as two standard deviations and calculating 13,455±(2×1,800)13,455 \pm (2 \times 1,800).
    • Explanation of error: This produces an interval describing individual data points, not a confidence interval for the population mean.
    • Correction: The population standard deviation must first be converted into the standard error of the mean by dividing by n\sqrt{n}:     SExˉ=σn=1,800400=1,80020=90SE_{\bar{x}} = \frac{\sigma}{\sqrt{n}} = \frac{1,800}{\sqrt{400}} = \frac{1,800}{20} = 90
    • The correct margin of error is then 2×90=1802 \times 90 = 180, yielding the interval 13,455±18013,455 \pm 180.
  • Second Most Common Error (Omitting the Confidence Multiplier):

    • Incorrect approach: Correctly converting σ\sigma to standard error via σn\frac{\sigma}{\sqrt{n}}, but then adding and subtracting it once without multiplying by 22.
    • Explanation of error: Adding and subtracting one standard error (xˉ±1×SE\bar{x} \pm 1 \times SE) yields a 68% confidence interval rather than a 95% confidence interval.

Deconstructing Intervals to Determine Standard Error (Correlation Example)

  • Back-Calculating Standard Errors from Published Intervals:

    • Any reported confidence interval conforming to the general form enables reverse calculation of the underlying standard error (SESE), even for statistics whose formulas have not yet been taught (e.g., in Topic 12).
  • Worked Example: Inter-Reader Reliability of Radiologists:

    • Context: Reliability is assessed using a correlation coefficient.
    • Sample statistic (point estimate): r=0.15r = 0.15.
    • Reported 95% confidence interval: [0.11,0.19][0.11, 0.19].
    • Margin of error calculation:     0.19−0.15=0.040.19 - 0.15 = 0.040.15−0.11=0.040.15 - 0.11 = 0.04
    • Because a 95% interval uses a multiplier of 22, the margin of error represents two standard errors:     2×SE=0.042 \times SE = 0.04
    • Solving for one standard error (or standard deviation of the correlation statistic):     SE=0.042=0.02SE = \frac{0.04}{2} = 0.02

Confidence Intervals for the Difference Between Two Independent Means

  • Motivation and Multi-Sample Context:

    • Real-world scientific inquiries typically compare two or more independent samples rather than assessing a single sample against a known parameter (e.g., treatment group versus placebo group, or tutoring group versus non-tutoring group).
    • Comparing two independent groups introduces two sample means (xˉ1\bar{x}_1 and xˉ2\bar{x}_2) and two sample sizes (n1n_1 and n2n_2).
    • Hypothetical trial: Vaccine group with n1=200n_1 = 200 compared to a placebo group with n2=150n_2 = 150.
    • Five practice problems of this type appear on the WebAssign homework assignments.
  • The Difference Between Means Statistic:

    • Point estimate formula:     Statistic=xˉ1−xˉ2\text{Statistic} = \bar{x}_1 - \bar{x}_2
    • Group labeling convention: Group designations are arbitrary, but labeling must remain consistent throughout calculations. By standard convention, the first group mentioned in the problem description is labeled Group 1, and the second group mentioned is labeled Group 2.
  • Standard Error of the Mean Difference Formula:

    • Standard error of Group 1 mean: SE1=s1n1SE_1 = \frac{s_1}{\sqrt{n_1}}
    • Standard error of Group 2 mean: SE2=s2n2SE_2 = \frac{s_2}{\sqrt{n_2}}
    • Combined Standard Error formula:     SE(xˉ1−xˉ2)=(s1n1)2+(s2n2)2=s12n1+s22n2SE_{(\bar{x}_1 - \bar{x}_2)} = \sqrt{\left(\frac{s_1}{\sqrt{n_1}}\right)^2 + \left(\frac{s_2}{\sqrt{n_2}}\right)^2} = \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}
  • Full 95% Confidence Interval Expression:   (xˉ1−xˉ2)±2×s12n1+s22n2(\bar{x}_1 - \bar{x}_2) \pm 2 \times \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}

  • Step-by-Step Computational Procedure:

    • Step 1: Compute the difference between sample means: xˉ1−xˉ2\bar{x}_1 - \bar{x}_2.
    • Step 2: Calculate the individual standard error for Group 1: s1n1\frac{s_1}{\sqrt{n_1}}.
    • Step 3: Square the Group 1 standard error: (s1n1)2\left(\frac{s_1}{\sqrt{n_1}}\right)^2.
    • Step 4: Calculate the individual standard error for Group 2: s2n2\frac{s_2}{\sqrt{n_2}}.
    • Step 5: Square the Group 2 standard error: (s2n2)2\left(\frac{s_2}{\sqrt{n_2}}\right)^2.
    • Step 6: Add the two squared standard errors together: s12n1+s22n2\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}.
    • Step 7: Take the square root of the sum to find the standard error of the mean difference.
    • Step 8: Multiply the resulting standard error by the multiplier 22.
    • Step 9: Add and subtract this margin of error to the difference between means.
    • Procedural recommendation: Write out every intermediate value explicitly on paper rather than entering the entire equation into a calculator in one step, minimizing order-of-operations mistakes.

Case Study: Lifespan Comparison of Right-Handed vs. Left-Handed Baseball Players

  • Study Parameters:

    • Group 1 (Strong Right-Handers):
    • Sample mean age at death: xˉ1=64.64 years\bar{x}_1 = 64.64\,\text{years}
    • Sample standard deviation: s1=15.5s_1 = 15.5
    • Sample size: n1=1472n_1 = 1472
    • Group 2 (Strong Left-Handers):
    • Sample mean age at death: xˉ2=63.97 years\bar{x}_2 = 63.97\,\text{years}
    • Sample standard deviation: s2=15.4s_2 = 15.4
    • Sample size: n2=236n_2 = 236
  • Calculation Setup:

    • Difference between means:     xˉ1−xˉ2=64.64−63.97=0.67\bar{x}_1 - \bar{x}_2 = 64.64 - 63.97 = 0.67
    • Standard error components:     SE1=15.51472SE_1 = \frac{15.5}{\sqrt{1472}}SE2=15.4236SE_2 = \frac{15.4}{\sqrt{236}}
    • Combined interval equation:     (64.64−63.97)±2×(15.51472)2+(15.4236)2(64.64 - 63.97) \pm 2 \times \sqrt{\left(\frac{15.5}{\sqrt{1472}}\right)^2 + \left(\frac{15.4}{\sqrt{236}}\right)^2}
    • Resulting 95% confidence interval:     [−1.49,2.83][-1.49, 2.83]
  • Formal Interpretation:

    • One can be 95% confident that the true population mean difference in age of death between strong right-handed and strong left-handed baseball players is between −1.49 years-1.49\,\text{years} and 2.83 years2.83\,\text{years}.

Inferential Logic: Interpreting Intervals That Contain Zero

  • The Null Value for Difference Between Means:

    • When comparing two groups, a sample difference of 00 signifies that the groups have identical means (xˉ1−xˉ2=0\bar{x}_1 - \bar{x}_2 = 0).
  • Evaluation of the Baseball Lifespan Interval:

    • Observed point estimate indicates right-handers lived an average of 0.67 years0.67\,\text{years} longer in the sample.
    • The calculated 95% confidence interval is [−1.49,2.83][-1.49, 2.83].
    • Because any value inside the 95% confidence interval represents a plausible value for the true population mean difference:
    • Positive values within the interval (up to 2.832.83) support right-handed individuals living longer.
    • Negative values within the interval (down to −1.49-1.49) support left-handed individuals living longer.
    • Zero is included within the interval, representing the possibility of no difference in lifespan.
    • Conclusion: It cannot be concluded with certainty that right-handed individuals live longer than left-handed individuals.
  • General Inferential Decision Rule for Confidence Intervals:

    • To substantiate an effect or directional difference between two groups, the value representing "no difference" (00) must fall entirely outside the confidence interval.
    • If establishing that Group 1 has a higher mean than Group 2, the entire interval must be strictly positive (>0> 0).
    • If establishing that Group 2 has a higher mean than Group 1, the entire interval must be strictly negative (<0< 0).
    • When the interval straddles zero, the data fails to rule out zero difference between the population means.

In-Class Practice: Calcium Intervention for Premenstrual Syndrome (PMS)

  • Study Design and Methodology:

    • Design: Randomized, double-blind clinical experiment.
    • Target population: Women suffering from premenstrual syndrome (PMS).
    • Total sample size: N=466N = 466 women with a clinical history of PMS.
    • Treatment group protocol: Participants received 1200 mg1200\,\text{mg} of calcium per day.
    • Control group protocol: Participants received a placebo.
    • Outcome measurement: Mean symptom severity rating across 17 symptoms.
  • Sample Statistics:

    • Placebo Group (Group 1):
    • Sample size: n1=235n_1 = 235
    • Mean symptom rating: xˉ1=0.60\bar{x}_1 = 0.60
    • Standard deviation: s1=0.52s_1 = 0.52
    • Calcium Treatment Group (Group 2):
    • Sample size: n2=231n_2 = 231
    • Mean symptom rating: xˉ2=0.43\bar{x}_2 = 0.43
    • Standard deviation: s2=0.40s_2 = 0.40
  • Confidence Interval Construction Setup:

    • Task: Compute a 95% confidence interval for the population mean difference (μplacebo−μcalcium\mu_{\text{placebo}} - \mu_{\text{calcium}}).
    • Point estimate:     xˉ1−xˉ2=0.60−0.43=0.17\bar{x}_1 - \bar{x}_2 = 0.60 - 0.43 = 0.17
    • Standard error computation:     SE(xˉ1−xˉ2)=0.522235+0.402231SE_{(\bar{x}_1 - \bar{x}_2)} = \sqrt{\frac{0.52^2}{235} + \frac{0.40^2}{231}}
    • Full expression for 95% confidence:     0.17±2×0.522235+0.4022310.17 \pm 2 \times \sqrt{\frac{0.52^2}{235} + \frac{0.40^2}{231}}

Questions and Class Discussion

  • Prompt on Alternative Confidence Levels:

    • Query: Request for a student-selected confidence level (preferably an even integer) to illustrate multiplier calculation.
    • Student response: Suggested 80%.
    • Explanation: An 80% level leaves 20% uncovered across both tails, meaning 10% resides in each individual tail. The zz-score with 90% below it is 1.281.28, resulting in a multiplier of 1.281.28.
  • Prompt on Back-Calculating Standard Errors:

    • Query: Given a radiologist inter-reader reliability correlation of 0.150.15 and a 95% confidence interval of [0.11,0.19][0.11, 0.19], what is the standard error (standard deviation of the correlation)?
    • Explanation: The interval spans 0.040.04 on either side of the point estimate (0.19−0.15=0.040.19 - 0.15 = 0.04). Because a 95% interval uses two standard deviations, dividing by 22 yields a standard error of 0.020.02.
  • Prompt on the Value of No Difference:

    • Query: If there is no difference between two groups, what would the mean difference statistic equal?
    • Student response: Zero (00).
    • Explanation: Two identical means subtracted from each other equal zero (xˉ1−xˉ2=0\bar{x}_1 - \bar{x}_2 = 0).
  • Prompt on Inferring Directional Advantage:

    • Query: Given the baseball player lifespan interval of [−1.49,2.83][-1.49, 2.83], can it be concluded with certainty that right-handed individuals live longer than left-handed individuals?
    • Student response: No.
    • Explanation: Because the interval contains zero, it remains possible that right-handers live longer (positive values), left-handers live longer (negative values), or that no true difference exists between the populations (zero).