Denis Aydin
2009-Aug-12 12:27 UTC
[R] Generating logistic regression data for specific ORs
Dear R-users I want to generate data for a logistic regression for an epidemiological simulation. First, I created a "disease-vector" containing a "1" if a subject is a cases (i.e. has the disease) and a "0" if a subject is a control. E.g.: > disease <- as.factor( c(rep(1, n.cases), rep(0, n.controls)) ) Then, I want to generate two lognormally distributed exposure vectors, one for cases and one for controls. The parameters of the distributions should be chosen in a way that a logistic regression model has a specific OR (or beta1) for the exposure. Something like that: > exp.cases <- lnorm(n.cases, mean.cases, sd.cases) > exp.contr <- lnorm(n.controls, mean.controls, sd.controls) > exposure <- c(exp.cases, exp.controls) > model <- glm(disease ~ exposure, family = binomial) Unfortunately, I don't know how to generate the exposure vectors in a way that the logistic regression has a specific beta1 or OR. In particular, I want the control over the parameters of the exposure distributions of cases and controls. Could anyone help me on that? Any help is appreciated. Denis -- Denis Aydin Institute of Social and Preventive Medicine at Swiss Tropical Institute Basel Associated Institute of the University of Basel Steinengraben 49 ? 4051 Basel ? Switzerland Phone: +41 (0)61 270 22 04 Fax: +41 (0)61 270 22 25 denis.aydin at unibas.ch www.ispm-unibasel.ch