 |
 |
Hello again. Thanks to all those who took the time to help overcome my confusion regarding the McNemar test. From the communications I received a few things emerged. Practically all of you agreed that the McNemar test should have an associated test value in association to the p one. The confusion arises from the fact that when the sample size is small the distribution is not a good approximation of the standard distribution and therefore it is unsafe to base your analysis on an approximation of the chi-square test. Under such circumstances the best approach is to use a form of the binomial sign test which calculates the exact probability of a specific set of binary results given n trials. It would appear that SPSS calculates the binomial test when asked to calculate the McNemar and it actually states this in the data output. Although, when you put in McNemar as a search term on the SPSS home page the resulting explanation suggests that the Chi-squared equivalent should be calculated such that a test statistic value and a the accompanying p-value should be given. I will have to seek clarification on this issue from SPSS themselves. In addition I appreciated the comments that were made with respect to experimental design, confidence limits, and appropriate statistical packages. Thanks again, Michael Gormley. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%