|
Help (wait for the button to appear above) |
|
||||
The Two sample item on the Proportions (P statistics) submenu on the Stat menu provides confidence intervals and/or hypothesis tests for the difference of two population proportions using data from two independent samples. The procedure is a Z test using the normal approximation to the binomial. The results are displayed in the Results panel.
To use the procedure, select the item and a dialog will pop up. The dialog will prompt the user to enter the number of "successes" and the total number of observations for each sample. Then inference may be done on the difference of proportions (the first sample minus the second).
For example, suppose a poll is taken in which 150 randomly sampled registered voters are asked whether they support a particular issue. Several weeks later, another sample is taken of 120 voters and they are asked the same question. It is of interest to test whether the level of support for the issue has changed between the times the polls were taken. Suppose that 74 in the first sample were in favor and 58 in the second sample supported the issue.
Clicking the Next -> button will allow the user to select either a confidence interval (specifying confidence level) or a hypothesis test (specifying null difference in the proportions and alternative hypothesis). To test whether the proportion supporting the issue has changed, the null difference should be zero, and the alternative should be "not equal".
Click on the Calculate button and the output is displayed in the Results panel. The values input (number of successes and number of observations for each sample) are displayed in a table. Below that, the hypothesis testing output consists of the estimated difference in proportions, the standard error of the estimate, the associated Z statistic, and the P value. In the confidence interval setting, the output includes the estimated difference, its standard error, and the lower and upper confidence limits.
