Hi Debbie,
Are you referring to an excerpt from Andy Field at http://stats.stackexchange.com/questions/157217/transforming-data-all-variables-or-just-the-non-normal-ones ?
I'm a little uncertain about this myself to be honest. Commenters on the page I linked seem to come to the conclusion that as long as a mixture of transformed and untransformed data (or, in your case, different transformations to deal with different directions of skew) doesn't violate any particular assumptions, it's fine as long as you can justify it.
Hopefully this helps
Ashley
________
From: Research of postgraduate psychologists. [[log in to unmask]] on behalf of Debbie Grimmond [[log in to unmask]]
Sent: 25 July 2016 19:46
To: [log in to unmask]
Subject: Transformations
Dear All
No doubt many of you are preparing for the PsyPag conference or other exciting things. In the vain hope that you are not consumed by getting ready for the excitement- does anyone know anything about transformations?
I have some non normally distributed data, with some variables suffering from being positively skewed, whilst other data is negatively skewed. I know that a number of books etc recommend transforming the data. I vaguely remember reading somewhere that you should transform your data across the board (e.g., if performing a log10 transformation, you should do it for all your variables). I can't find where I have read this. Can any one confirm this as the case?
Many thanks
Debbie
--
Please click here to view our e-mail disclaimer http://www.anglia.ac.uk/email-disclaimer
|