Hi everyone,
I'm a PhD student currently struggling with some regressions and this problem has been irritating me for days. Basically, I'm looking at how a model of work stress (the job demands-control model) predicts mental health on a questionnaire. Due to problems of multicollinearity, I need to mean centre the variables but my problem is that the mean-centred interaction term bears no correlation with the original non-centred version. Surely the 2 should correlate, but getting a high score on the original (ie high demands x low control) does not equate to a high score when the two variables are mean-centred. So really I'm asking for help - is mean centering the two variables (demands and control) before multiplying them the correct method? I guess that the problem is due to the fact that after centering there are some minus and positive values which cancel each other out.
Any help gratefully received! Please feel free to email me personally at [log in to unmask]
Thanks in advance,
Tom Willis
Health Psychology group
Institute of Psychological Sciences
University of Leeds
Leeds
LS2 9JT
Tel: 0113 343 6694
Fax: 0113 343 5749
E-mail: [log in to unmask]
|