 |
 |
Hi Caroline, I'm not an expert but have recently had an 'un- versus standardised' debate for a paper - these are my thoughts. Unstandardised coefficients are reported in the units of the variable it represents (e.g. mmHg for BP measurements, years of age). If you're interested in the relative importance of variables as predictors, it's hard to compare directly using the unstandardised values as you're comparing apples with oranges. One of the main benefits of standardising is to be able to compare the magnitude of the contribution a predictor makes, because the standardised values represent standard deviations. For the paper I was working on, this was our main focus and we had a number of physiological measurements with different units, so standardised was more appropriate. http://www.jerrydallal.com/LHSP/importnt.htm In your case however, it seems to me (and I'm happy to be shot down by any statistical argument here) that although your questions are measuring different things, the 'units' are all likert points, which greatly reduces the need to standardise to aid your interpretation. You still have to be aware that the variables have differing maximum scores when interpreting them, but there doesn't seem to be a great benefit of having them reported in units of SD. You could try it both ways with one of your regressions and see if it does make it any easier to interpret. You mention you've read that you should 'always' standardise, but it may be worth checking what type of studies and measurements they refer to. Certainly I think you should be consistent, and I'd suggest including a short discussion to justify your approach to show your examiner you've considered the issues. Not the sage-like definite answer, but I hope it helps! Brian On 16/04/2010 04:32, Caroline Wilson wrote: > Hi, I've got creeping doubts over the way I've been writing up my > stats and I hope someone is feeling like they want to be a sage this > morning. For a reason I'm no longer confident about I've been using > unstandardised co-efficients to report multiple regression and > hierarchical multiple regression and I'm now thinking that's maybe wrong. > Here's the background: I've scaled several likert-style questions > (post factor analysis). These had arbitrary negatives so couldn't be > scaled using SPSS but were done manually, and I centred rather than > standardized the scales because I was planning to do moderation > analysis (and if you standardise you risk over-correlation when you > make the interaction terms). > I should mention that some of these variables had differing maximum > scores, due to the number of questions retained after factor analysis > and because some questions had different numbers of answer options. > I've followed advice in Frazier et al (2004) that when reporting > moderation the coefficients aren't properly standardised and are > uninterpretable so you should report the unstandardised coefficients. > For reasons of consistency I thought I'd write up all of my results > using unstandardised coefficients, including the multiple regressions > without interaction terms. I have read in so many other places that > you should 'always' report the standardised co-efficients that I'm > starting to have doubts. > So my first question is - should I be consistent or correct - for my > main effects is unstandardised ok? > I've also drawn an illustration of the effects of interaction for > different groups. The moderator variable was measured on a 10 point > scale and the dependent variable on a 16 point scale (and they're both > centred on 0). My second question is, if I use the unstandardised > coefficients to mark the slopes will it be a true representation or > exaggerated? > I hope that makes sense. Any thoughts or advice that could unfuddle my > brain would be very welcome. > > Reference: FRAZIER, P. A., TIX, A. P. & BARRON, K. E. (2004) Testing > Moderator and Mediator Effects in Counseling Psychology Research. > /Journal of Counseling Psychology,/ 51*,* 115-134. > > Caroline Wilson > PhD Research Student > Institute of Energy and Sustainable Development > De Montfort University > UK >