Hello Reeba,

My hunch is that data really don't need to be normally distributed for you to be able to conduct satisfactory statistical analyses on them. A useful, down-to-earth article about this was written by Geoffrey Norman about 6 years ago:

DOI 10.1007/s10459-010-9222-y

Apparently concerns should arise, however, if there are outliers in the data if you're using parametric tests - but that's a completely different kettle of fish.

I hope that's helpful.

Robt.

It appears to me, without much information I admit, you want to consider longitudinal models perhaps with Box-Cox transformations to deal with normality and variance issues. Here is my answer to a similar question:

First, you say the transformations didn't work. Did you try an automated Box-Cox procedure? These work a large percentage of the time. Such procedures are available in R, SAS and most other packages.. For SPSS see:

https://www.google.com/search?q=box+cox+transformation+spss+youtube&rlz=1C1CHBF_enUS847US847&oq=Box+Cox+transformation+SPSS&aqs=chrome.3.0l4.41438j0j8&sourceid=chrome&ie=UTF-8

In general, this topic comes under the general heading of longitudinal models, which include standard repeated measures, GEE and the rest. An excellent set of notes on these models can be found here:

https://www4.stat.ncsu.edu/~davidian/st732/notes.html

I would be very surprised if these resources don't solve the problem.

Best wishes, David Booth

For repeated measures I would recommend a multilevel linear model. The family of Generalized Linear Models gives you the choice of a response distribution. Transformation of response variables is unnecessary. Both approachzes cxan be combined, as I explain in my online book:

https://www.researchgate.net/project/Book-New-statistics-for-the-design-researcher

Thank you everyone. I really appreciate the inputs.