I have a variable which is based on a count of occurances devided by the total number of potential occurences for each unit. This variable thus varies between zero and one. It should be used as a dependent variable (predicted / outcome) in a regression model. Furthermore, the data are repeated measures (observations nested in individuals).

Do you have advice on which model (family / link function) I should use for this data?

Do you have advice which software in R I should use to model the multilevel structure with the suggested distributional family?

The data look (Q-Q-plot) normal after square-root transformation, but of course they are still censored.

  • $\begingroup$ The obvious family to try would be binomial. The most suitable link would depend on information not given here. $\endgroup$ – Glen_b Nov 27 '14 at 22:31
  • $\begingroup$ Which information do you need? Binomial was also my thought, but logit link seems to require binary input variables. $\endgroup$ – tomka Nov 28 '14 at 0:36
  • $\begingroup$ Binomial family (with or without logit link) can take input in three different forms, one of which is binary. The information required to pin down a link function would depend on your modelling choices/needs or on the data. In the absence of other information I'd probably start with logit. $\endgroup$ – Glen_b Nov 28 '14 at 1:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.