1

I have encountered a problem in R that seems fairly odd to me to say the least. I fitted a robust regression framework using rlm from the MASS package so the specification is like this:

regression <-  rlm(y ~ x1 + x2 + log(x3) , 
               data = df,
               na.action = na.exclude, method = "MM")

however everytime I run this command and print the summary, R delivers slightly different estimates, standard errors and thus t-, and p- values. The discrepancies are not huge, but it´s enough to render 10% with p=0.07 to no significance with p=0.11. Does anyone know where this is comming from? I want to make clear that in between running nothing was changed in the dataset. There were no commands run inbetween or anything. I´d be relly thankfull for help here as this makes me really insecure about my results. Thanks in advance

philipp.kn_98
  • 139
  • 9
  • 2
    `rlm()` regression in MASS is an iterative approach which has a random component. So your results may vary slightly. If you want less variation, set a lower `acc` parameter (e.g. `acc=1e-6`). This narrows down the output. – Martin Wettstein Aug 18 '20 at 18:55
  • Or you can set the `seed` parameter fixed. – Nuclear03020704 Aug 20 '20 at 09:59

0 Answers0