1 / 32

Chapter 17 Comparing Two Proportions

Chapter 17 Comparing Two Proportions. In Chapter 17:. 17.1 Data 17.2 Proportion Difference (Risk Difference) 17.3 Hypothesis Test 17.4 Proportion Ratio (Risk Ratio) 17.5 Systematic Sources of Error 17.6 Power and Sample Size. §17.1 Data. Two independent groups Binary response

Download Presentation

Chapter 17 Comparing Two Proportions

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Chapter 17Comparing Two Proportions

  2. In Chapter 17: • 17.1 Data • 17.2 Proportion Difference (Risk Difference) • 17.3 Hypothesis Test • 17.4 Proportion Ratio (Risk Ratio) • 17.5 Systematic Sources of Error • 17.6 Power and Sample Size

  3. §17.1 Data • Two independent groups • Binary response • Epidemiologic jargon: Group 1 = “exposed group” and Group 2 = “nonexposed group” • Count “successes” in each group and convert to proportions

  4. Sample Proportions Proportion in the exposed group: Proportion in the nonexposed group:

  5. §17.2 Proportion Difference (Risk Difference) The risk difference is the absolute difference in incidence proportions in the groups.

  6. In large samples, the sampling distribution of the risk difference is approximately Normal

  7. Confidence Interval, Risk Difference Plus-four confidence interval method for a difference in proportions. This method is accurate in samples as small as 5 per group.

  8. 95% CI for p1 – p2, Example WHI data a1 = 751, n1 = 8503, a2 = 623, n2 = 8102

  9. 95% CI for p1 – p2, Example The plus-four method is similar to Wilson’s score method. Here’s output from from WinPepi > Compare2.exe > Program B showing results from the traditional large-sample method and Wilson score CI for the illustrative example.

  10. §17.3 Hypothesis Test • We test the proportions for a significant (“nonrandom”) difference • Two methods are covered in this chapter • z test (large sample) • Fisher’s exact procedure (small samples) • A third method called the chi-square test is covered in the next chapter

  11. z Test A. Hypotheses. H0: p1 = p2 against Ha:p1≠p2 [One-sided: Ha: p1 > p2or Ha: p1 < p2] B. Test statistic. C. P-value. Convert zstat to P-value [Table B or F]

  12. conclude: highly significant

  13. z Test: Notes • z statistic dissection • numerator is observed difference • denominator is standard error when p1 = p2 • A continuity correction can be optionally applied (p. 382) • This z test is equivalent to the chi-square test of association (Chapter 18) • In small samples (fewer than 5 successes expected in either group), avoid the z test and use the exact Fisher or Mid-P procedure

  14. Recall that and Fisher’s Exact Test (2-by-2) Before conducting Fisher’s test, data are rearranged to form a 2-by-2 table :

  15. WHI Data, 2-by-2 Format

  16. Fisher’s Exact Test, Procedure A. Hypotheses. H0: p1 = p2 vs. Ha: p1≠p2 [one sided Ha: p1 > p2or Ha: p1 < p2] B. Test statistic. Observed counts in 2-by-2 table C. P-value. Use computer program (e.g., WinPepi > Compare2.exe > Program B). The mathematical basis of the test is described on pp. 386–7.

  17. Fisher’s Exact Test, Example The incidence of colonic necrosis in an exposed group is 2 of 117. The incidence in a non-exposed group is 0 of 862. Is this difference statistically significant? A. H0: p1 = p2 against Ha: p1≠p2 B. Data.

  18. C. P = 0.014 (WinPepi output shown here). The evidence against H0 is “significant.”

  19. §17.4 Proportion (Risk) Ratio • Let RR refer to an risk ratio or prevalence ratios Interpretation • The RR is a risk multiplier, e.g., an RR of 2 suggests that the exposure doubles risk • When p1 = p2 , RR = 1. This is the “baseline RR,” indicating no association.

  20. RR Example, WHI Data The indicates a positive association; specifically, 15% higher risk (in relative terms) with exposure.

  21. (1– α)100% CI for the RR Note natural log scale of sampling distribution

  22. 90% CI for RR, WHI Example

  23. CI for RR, Computerized Results Output from WinPepi > Compare2.exe > Program B. See prior slide for hand calculations

  24. §17.5 Systematic Error(Advanced Topic) • In observational studies, systematic errors are often more important than random sampling error • Three types of systematic error are considered: • Confounding • Information bias • Selection bias

  25. Confounding • Confounding = the mixing together of the effects of the explanatory variable with the effects of “lurking” variables. Consider this example: • The WHI estrogen experiment found increased morbidity and mortality in estrogen users • Earlier, non-experimental studies found the opposite: lower morbidity and mortality in users • Plausible explanation: In non-experimental studies, estrogen users (self-selected) were more likely to have “lurking” lifestyles factors that contributed to better health, i.e., confounding

  26. Information Bias • Information bias is due to the mismeasurement or misclassification of variables in the study. • Misclassification may be nondifferential (occurs to the same extent in the groups) or differential (one groups experiences a greater degree of misclassification than the other) • Nondifferential misclassification tends to bias results toward the null (or have no effect). • Differential misclassification can bias results in either direction.

  27. Nondifferential & Differential Misclassification - Examples

  28. Selection Bias • Selection bias ≡ systematic error related to the manner in which study participants are selected for study • Example. If we shoot an arrow into the broad side of a barn and later draw a bull’s-eye where it had landed, have we really identified anything worth noting?

  29. 17.6 Power and Sample Size Power and sample sizes analysis for comparing proportions requires us to understand relationships between these factors: r ≡ sample size allocation ratio n1 / n2 1−β ≡ power (type II error) α ≡ significance level (type I error) p1≡ expected proportion, group 1 p2≡ expected proportion in group 2, or some measure of effect size, such as the expected RR

  30. 17.6 Power and Sample Size Because of the complexity of calculations (pp. 396 – 402), use software… Here’s WinPepi’s Compare2 Sample size menu.

More Related