how to perform t test or anova test?

[TomiKong]

If only mean values,variance and sample size are known,  are there procedures or macros which could perform t test or anova test without raw data?

Thanks in advance.

[art297]

SAS can do a t-test on aggregated data and Michael Friendly has a freely available macro then can disaggregate data so that you can analyze it in SAS with whatever statistical proc you need.

A link that includes a link to Michael's macro, as well as a discussion of the issues you should be concerned with in using aggregate data that way can be found at: SAS-L archives -- November 2004, week 5 (#74)</title><style type="text/css"><!--BODY { font-family: ...

[plf515]

Yes indeed Art. But, as the link carefully notes, Michael Friendly's program assumes the original data are normally distributed.

It might be interesting to experiment and see how far off results are with that macro by aggregating data, disaggregating it with the macro and then comparing results to t-tests (or whatever) on the original data.

[art297]

If one doesn't know how their data are distributed they have no way of knowing what statistic to use.  The results will be the same as if one had the raw data but, even with the raw data, one can't rely on the resuts of a t-test if the data doesn't meet the test's assumptions.

The problem with using aggregate data, if one doesn't already know what the underlying distribution looks like, is that they have no way of knowing it any particular test's assumptions have been met.

[plf515]

Art - Exactly. I just wanted to make it clear in this thread that that was the case.

[SteveDenham]

And to expand even more on Art's point, are the mean and standard deviation even best estimators, if you don't know the underlying distribution.  I mean, you can appeal to the Central Limit Theorem, but sometimes appeals are denied, due to sample size and just plain misspecification.

Steve Denham