Confidence Intervals for Proportions If one has a relatively large sample (large enough to use a normal approximation of the binomial parameter p), then one can construct a confidence interval about one’s pq estimate of the population proportion by using the following formula: pˆ Z 2 . For n example, suppose we wish to estimate the proportion of persons who would vote for a guilty verdict in a particular sexual harassment case. We shall use the data from a study by Egbert, Moore, Wuensch, and Castellow (1992, Journal of Social Behavior and Personality, 7: 569-579). Of 160 mock jurors of both sexes, 105 voted guilty and 55 voted not guilty. Our point estimate of the population proportion is simply our sample proportion, 105 / 160 = .656. Is n large enough (given p and q) to use our normal approximation, that is, is np 2 npq (which is essentially a 95% confidence interval for the number of successes) within 0 n ? If we construct a 95% confidence interval for p and the interval is within 01, then the normal approximation is OK. For a 95% confidence interval we compute: .656(.344) .656 1.96 .656 .074 .582 .730 . 160 Suppose we look at the proportions separately for female and male jurors. Among the 80 female jurors 58 (72.5%) voted guilty. For a 95% confidence interval we .725(.275) compute: .725 1.96 .725 .098 .627 .823 . 80 Among the 80 male jurors 47 (58.8%) voted guilty. For a 95% confidence .588(.412) interval we compute: .588 1.96 .588 .108 .480 .696 . Do notice 80 that the confidence interval for the male jurors overlaps the confidence interval for the female jurors. There are several online calculators that will construct a confidence interval around a proportion or percentage. Try the one at http://www.dimensionresearch.com/resources/calculators/conf_prop.html . In Step 1 select the desired degree of confidence (95%). In step two enter the total sample size. In Step 3 enter the number of successes or the percentage of successes. Click Calculate and you get the confidence interval for the percentage. If you prefer a Bayesian approach, try the calculator at http://www.causascientia.org/math_stat/ProportionCI.html . Copyright 2010, Karl L. Wuensch - All rights reserved. Proport.doc Difference Between Two Proportions From Independent Samples We might want to construct a confidence interval for the difference between the pq p q two proportions. The appropriate formula is pˆ1 pˆ 2 Z 2 1 1 2 2 . For our data, n1 n2 a 95% confidence interval is .725 .588 1.96 .725(.275) .588(.412) .137 .146 .009 .283 . Notice that 80 80 this confidence interval includes the value of zero. I have written an PASW macro that will compute such confidence intervals. If you wish to try it, download CI_p1-p2.zip from my PASW/SPSS Programs Page. It is most useful when computing several such confidence intervals. Frequently Asked Questions Copyright 2010, Karl L. Wuensch - All rights reserved.