I ran a regression model with the Y variable (length of stay of patients in days) log10 transformed. I would like to compare the fit of the log10 transformed model with a similar regression where the Y variable is in the raw scale. As such, it would be ideal to compare residuals in the original scale (days). Some papers suggest the back-transformation using Duan's smearing estimate. How do I implement this in R? Thanks,

  • $\begingroup$ (I can attest that the regression using log-transformed LOS will perform reliably, if moderately, better.) Can you summarize the reason for back-transforming, or cite a source? $\endgroup$ – rolando2 Jul 28 '18 at 15:40
  • $\begingroup$ Details of the R implementation are off-topic, but the concept and algorithm are not: those are explained and illustrated in the duplicate threads. $\endgroup$ – whuber Jul 28 '18 at 18:01
  • $\begingroup$ @rolando2 I want to be able to compare MAE and RMSE for several models in their original scale. For example, in Verburg at al. 2014 (10.1371/journal.pone.0109684) $\endgroup$ – FFR Jul 31 '18 at 18:59