I have a mixed experimental design with one between-subjects factor and one within-subjects factor and a dichotomous dependent variable. I ran a mixed ANOVA but i need to run a logistic regression as well.
I was wandering how can I test a moderation effect of a third continuous independent variable in the regression. My main issue is the nonlinear transformation of the dependent variable that may leads to interaction coefficients which do not properly reflect moderation effects in the original probabilities. Can someone answers me?
I tested the moderation in the ANOVA using dummy coded variables below and above the median, and in a multiple regression using PROCESS by hayes.
Can someone share his solution?
Must consult the following bibliography and guide you:
Hosmer D, Lemeshow S, Wiley J. Applied Logistic
Regresión. New York. 1989.
Hi, Good Evening
Moderation in a logistic regression:
http://www.cba.uh.edu/jhess/documents/OnTestingModerationEffectsinExperimentsUsingLogisticRegressionAug172009.doc
http://www.sagepub.com/upm-data/47569_ch_15.pdf
Must consult the following bibliography and guide you:
Hosmer D, Lemeshow S, Wiley J. Applied Logistic
Regresión. New York. 1989.
another option is to use log-binomial regression, which models the log of the probablility. if that is still an issue for you, i suppose you can try id-binomial, but that may lead to even more boundary problems (i.e. predicted p must be between 0 and 1) than log-binomial. boundary problems are especially likely with continuous predictors. i suspect that there were boundary problems in your mixed anova, but the procedure you used didn't tell you.
to run a log-binomial in sas:
proc genmod descending; class ;
model = predictors / dist=bin link=log ;
/* note: you can change the dist to poisson or the link to id */
/* if you have repeated measures or use a non-binomial dist, you need the following line */
repeated subject= / type= within=;
/* working correlation type is likely to be ind (independent), exch (exchangeable or compound symmetric). possibly some other, but getting too complex will overburden a small sample, which i assume you have, given that this is an experiment */
using logistic regression is the standard in much medical research, but perhaps not in your field. maybe you need to find out why.
I am not quite sure about the question you are asking, but it seems to me you could run the logistic regression and use the continuous variable as one of the predictors.
Hi brother,
I think you can try GLIMMIX command in SAS for testing moderation of covariate in logistic regression.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA