0

I have a dataset, the distribution looks like this, before and after log-transformation. The data tends to separate itself towards two extremes, thus motivating me to classify them as binary and fit logistic regression. On the other hand, fitting a mixed-model on the log-transformed data might be a simpler way to go about this.

I have tried to do both, and the results are not that different. How would you qualify which fit is more appropriate? I have been using R in my analysis and fit them using glm() and lmer().

Appreciate the feedback, Thanks! enter image description here

Todd Burus
  • 632
  • 2
  • 12
Joshua Budi
  • 1
  • 1

1 Answers1

0

Sometimes the search for the right transformation is not going to help materially. From your graphs I'm guessing that many values would have fallen below 0 and some above 100 if the measurement approach had allowed. In this circumstance the most effective technique is probably censored regression, also called tobit regression. The UCLA data-analysis site will probably be helpful. Probably also this worked example.

rolando2
  • 11,645
  • 1
  • 39
  • 60