I've had to put some thought into how Salstat handles tests for 2 or more conditions. The tests we have are analysis of variance (within- and between-subjects), Kruskal-Wallis test (nonparametric equivalent of the between-subjects ANOVA), Friedman test (nonparametric equivalent of the within-subjects ANOVA) and Cochrane's Q test (nonparametric equivalent of a within-subjects ANOVA but for binary responses).
Part of me wonders if putting the tests all together like this is a good idea. Due to different requirements (for example, choosing what constitutes the 2 levels of coding in a variable for Cochrane's Q), I wonder if unique dialogs are a better solution.
Anyway, here are some wireframes of what the dialog can look like.
First of all, the dialog begins with some general help text to orientate users. At the bottom are buttons to cancel or analyse (the latter contingent upon sufficient selections made by the user to actually do analysis).
In the middle, the interface is split into 2 tabs: within-subjects tests and between-subjects tests. Each set of tests requires variables to be chosen in unique ways so it makes sense for users to view a) only one set of tests, and b) all of one set of tests. Tabs are a way to accomplish this.Within-subjects tests
Below is the wireframe for within-subjects tests. After some more useful text about within-subjects tests to orientate users, the left-hand side has a list of tests which the user can select (within-subjects ANOVA, Friedman, Cochrane's Q). On the right, they select the levels of the variable (1 column per level).Between-subjects tests
Below is the wireframe for between-subjects tests. Again, on the left-hand side are the tests (in this case, ANOVA and Kruskal-Wallis). One the right are two drop-downs, one to specify the independent (grouping) variable, the other to specify the dependent variable.
Descriptives and other tests
Descriptive statistics can be retrieved as normal using the descriptives routines. It is always useful to see some so I envisage a small set being automatically generated and displayed. The same goes for tests of homogeneity of variance.Testing
Testing is needed for this but until I can afford this, comments are welcomed.