Dear all,
I have run a logistic regression with several predictors: age; gender (coded "1" for women and "0" for men); dimension 1 of phenomenon X; dimension 2 of phenomenon X; gender*dim1; and gender*dim2.
N=267 (n men = 70; n women = 197).
All variables in the regression model were standardized.
When gender is coded as follow: women "1" and men "0", here are the significant predictors:
Dimension 1 (b+); and gender*dim1 (b-).
Then, I just tried to run the regression in another way, in order to fully understand the significant interaction effect. So, I coded men "1" and women "0". Technically, I expected to find same significant predictors (dim1 & dim1*gender) with same b for dim 1 and opposite b for dim1*gender. However, I found the following significant predictors: dim2 (b-) and gender*dim1 (b+). To me, the significant predictor "dim2" is totally unexpected.
Therefore, I performed two other analyses. I have selected only women (filter) and run the logistic regression with the following predictors: age, dim1, and dim2. The significant predictor was dim2 (b-).
Then, I have selected only men and run the same logistic regression and I obtained dim1 (b+) as a significant predictor.
Can someone help me solve this problem?
Many thanks in advance,
Check the following link
http://www.restore.ac.uk/srme/www/fac/soc/wie/research-new/srme/modules/mod4/11/index.html
I would suggest doing an interaction plot with your different variables( see:
https://www.google.com/search?rlz=1C1CHBF_enUS874US874&biw=1600&bih=827&ei=LJr9XbTDBYGntQbNgKjgAg&q=montgomery+design+and+analysis+of+experiments&oq=montgomery&gs_l=psy-ab.1.0.0i67l2j0l8.4302873.4312901..4317448...1.2..3.127.5158.62j1......0....1..gws-wiz.....6..0i71j33i10j0i362i308i154i357j0i273j0i131.-ipHefJQ87E
and/or the below attachment) I believe this should help you to see what is going on. remember that your model has a complicated RHS. Best wishes, D. Booth
It is difficult to tell the exact specification of the models so you need all main effects before you do interactions so Y = b0 + b1Gender + b2X1 + b3Gender*X1 and the best way to understand is make conditional predictions and plot these against X1 distinguishing the genders. In this model the effetc of Gender will change as the values of X1 change , that is the point of the interaction, and b1 will be the effect of Gender when X1 is zero.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics