I have some data and since it was binary it was suggested to run a cochran q test on it.
10 binary quesion over 5 treatments. My question is how do I run ine and what do the results tell me. I mean I can find the formula and know that it tests to see if the treatments were effective but what do I look for to indicate significance and how is it different han a matched pair t-test on the data or an ANOVA test?
Cochran’s looks at binary data, t-test and ANOVA do not. It is two-way with infinite levels, unlike t-test and like some iterations of ANOVA. Q is non-parametric unlike t/ANOVA, which means that the latter two are potentially more powerful, but Cochran’s does not require assumptions like normality or homoscedasticity.
When you get the T value (according to Wikipedia, so why’s it not Q?), it seems that you look up the critical value using a chi-squared table, where the column is your alpha, and row/df is your number of treatment levels - 1. If T > χ[sup]2[/sup], then reject, and at least one group is likely different. Run another Q test on two levels of interest if you want more detail after that.
I’ve never run this though. Maybe if there is some example data I could take a stab. The math does not seem that hard.
I don’t have SPSS in front of me, but it might be under “Analyze->Non Parametric->k-related samples.”
Also, it uses T instead of t or F…
Is the data like this example? Notice how each of the subjects in the example do each of the 5 different tasks. If “treatment” means that you have different subjects in each group, you may have gotten some bad advice.
A scenario where you would use an ANOVA would be if you have 5 college classes and you ask the 10 students in each of the classes to respond to the same single Yes or No question. Since the students don’t overlap and the question is the same, you would be doing an ANOVA test of proportions; this would test whether the proportions of “Yes” are the same across the 5 classes. The data layout looks the same as in your scenario but the design is different, making the appropriate test different.
The Cochrane’s Q test requires a very specific experimental design; the design isn’t the most common in Statistics Land. We’d need to know more about your experiment before saying that the Q test is the right test.
I ask a student 10 questions to test where they are on a spectrum e.g
Who do you earn grades for? Myself My parents
Then the treatment occurs and I ask the same 10 questions again and repeat 4 more times. Would Cochran Q work for that? If not why not?
I suspect it is not appropriate because it does not test success/failure and although presented as binary in nature the true answers are somewhere in the middle.
I’m not 100% sure that it’s the best test, but your “Myself My parents” part would mainly affect your interpretation of the results. Yes, they aren’t truly binary, so you couldn’t make any conclusions about doing it for your SO. Regular old Pearson’s chi-squared also might be something to look into.
Missed something in the OP: it’s similar to a paired t-test or repeated measures ANOVA in that each subject is included in each condition, and power is increased because they serve as their own controls. t-test/ANOVA look at continuous data, though.
This sounds like repeated measures logistic to me. I have no idea how to do this in SPSS; in SAS you can do use either proc genmod or catmod.