ANOVA and t tests with data entered as mean, N and SD (or SEM).

You must enter raw data in order to perform a paired t test, repeated measures ANOVA, or the nonparameteric tests.  But raw data are not needed for the unpaired t tests or ordinary (not repeated measures) ANOVA. Prism can do these tests from data entered as mean, sample size (N), and SD or SEM.

t test and one-way ANOVA

Prism (since version 4) can do unpaired t tests and regular (not repeated measures) one-way ANOVA with data entered as mean, N and SD or SEM. To do this:

1. Create a "Grouped" data table ("two grouping variable' was the term used by Prism 4),
2. Choose to format the table with subcolumns labeled for entry of mean, SD and N or mean, SEM and N.
3. Enter data on only the first row. 
4. Click Analyze and choose t-tests or one-way  ANOVA. These choices are in the Column set of analyses.

 You'll get exactly the same results after entering raw data that you would have gotten using mean, SD and N.

What if you have entered (or pasted) all the data into one data set, rather than into one row? Use the Transpose analysis to create a results table with the data all in one row.

Two-way ANOVA

1. Create a "Grouped" data table ("two grouping variable' is the term used by Prism 4),
2. Choose to format the table with subcolumns labeled for entry of mean, SD and N or mean, SEM and N.
3. Enter data. 
4. Click Analyze and choose two-way  ANOVA from the Grouped set of analyses.

With two-way ANOVA, the results will be identical if N is the same for every cell or if N varies systematically (one column has all one N, and the other column has a different N). If the N values vary in an erratic fashion, you'll get more accurate two-way ANOVA results by entering raw data. If you enter averaged data, the results will be only a little bit off if the N values differ by only a small amount.

What if you want to compare best-fit values from regression?

Usually the t test, and one-way ANOVA, compare means. What if you want to compare best-fit values of slopes or EC50 or some other parameter fit by linear or nonlinear regression? 

The regression program reports the best-fit value of the parameter, the standard error of that parameter, and the number of degrees of freedom of the fit. Enter the best-fit value as the "mean" and the standard error of that parameter as the "SEM". But what you should you enter as "N"? The t test (and ANOVA) work by subtracting one from the entered value of N, and uses that difference as the number of degrees of freedom. When comparing results from regression, the concept of "N" doesn't quite fit. But you know the number of degrees of freedom, as that is reported by the regression. Add one to that df, and enter that value as "N". Prism will then subtract N-1 and use the difference as the number of degrees of freedom, which will be correct.