0

I have a dataset containing male and female data. I have the response variable metabolic rate, and several predictors (Behaviour, Temperature 1, Temperature 2, Activity, Sex, Body Size, and Body Mass).

First, I conduct the GLMM using an identity link on male and female combined data:

glmer(log(Metabolic_Rate)~ Temperature.1 + Behaviour * Temperature.2 + Sex + Activity + Body_Size + Body_Mass + (1|Week), data= AMRdata, family = Gamma(link = 'identity'))

And then run model simplification. The model works just fine and I have no error messages.

Then, I separate the data by sex and run the model on just female data:


females<-subset(AMRdata,Sex=="F")


glmer(log(Metabolic_Rate)~ Temperature.1 + Behaviour * Temperature.2 + Activity + Body_Size + Body_Mass + (1|Week), data = females, family = Gamma(link = 'identity'))

I am greeted by the error message:

Error in pwrssUpdate(pp, resp, tol = tolPwrss, GQmat = GQmat, compDev = compDev,  :
  (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate

Removal of Temperature.2 takes removes this error, but I really need to keep it in the model.

Does anyone have any suggestions as to how to remove the error?

I'm using lme4 version 1.1.21, and the female model doesn't work if I don't use the log transformation, either. I have tried using a log link, but I get the same error message.

Some data:

structure(list(Metabolic_Rate = c(8.79514591, 16.71840387, 14.1932374, 
10.90741585, 10.7436911, 14.97469781, 19.88267242, 12.43274774, 
15.12038794, 11.84916117, 11.05467852, 19.53495917, 12.14440531, 
12.09564168, 6.78392472, 10.51570692, 8.527792046, 8.731880804, 
10.71404367), Behaviour = c(23L, 17L, 14L, 7L, 99L, 78L, 90L, 
1L, 9L, 29L, 76L, 66L, 43L, 36L, 13L, 4L, 82L, 14L, 59L), Temperature.1 = c(21.9, 
21.7, 18.52, 19.85, 20.45, 20.54, 21.7, 22, 21.32, 21.4, 21.44, 
22.1, 22.22, 22.25, 20.43, 20.9, 21.63, 21.2, 21.52), Temperature.2 = c(17.5, 
15.6, 12.5, 19.8, 16.6, 20.8, 21.4, 21.9, 21, 21.3, 20.5, 22, 
22.1, 22.2, 20.6, 21.2, 21.9, 21.1, 21.5), Activity = c(39.54664352, 
66.75914352, 40.85949074, 44.8505787, 37.20023148, 69.75388889, 
72.43981481, 70.42199074, 20.71481481, 77.27662037, 62.21712963, 
93.22673611, 82.39247685, 89.42141204, 35.35729167, 31.97777778, 
74.65821759, 40.80590278, 54.3755787), Sex = structure(c(1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
2L, 2L), .Label = c("F", "M"), class = "factor"), Body_Size = c(7.6, 
5.8, 7.9, 7.6, 8, 7.5, 7.9, 7.6, 7.2, 7.8, 7.8, 7.4, 7.1, 8.4, 
6, 7.9, 7.2, 7, 8.2), Body_Mass = c(0.747, 0.55, 0.76, 0.673, 
0.691, 0.683, 0.689, 0.789, 0.6, 0.612, 0.637, 0.511, 0.582, 
0.603, 0.408, 0.527, 0.666, 0.483, 0.602), Week = c(1L, 1L, 2L, 
3L, 3L, 3L, 3L, 4L, 5L, 6L, 6L, 8L, 8L, 8L, 8L, 9L, 9L, 9L, 10L
)), class = "data.frame", row.names = c(NA, -19L))
Lizzie Yarwood
  • 53
  • 6
  • Maybe this is helpful: https://stackoverflow.com/questions/38015647/glmer-error-maxstephalfit-pirls-step-halvings-failed-to-reduce-deviance-in-p – Ventrilocus Nov 27 '20 at 13:37
  • A guess: Is your sample size small? Maybe it's too small to run the interaction term? – xilliam Nov 27 '20 at 13:38
  • maybe you could try to replicate the error using some publicly available data, or make up some, so that we can reproduce it – desval Nov 27 '20 at 13:47
  • @xilliam, my sample size is small, so this could be an issue. – Lizzie Yarwood Nov 27 '20 at 14:08
  • @desval, aren't you able to reproduce my error using the data that I have shared? – Lizzie Yarwood Nov 27 '20 at 14:10
  • A better way to share the sample data is to use a format that can be easily copy/pasted to a script. So, in your case, it would be better if you would edit your 'some data' to become the copy/paste of the console output of this code ```dput(AMRdata)``` . – xilliam Nov 27 '20 at 14:24
  • Thanks @xilliam, I didn't know that you could do that! I've edited the info above to include what you've asked for. – Lizzie Yarwood Nov 27 '20 at 14:33

0 Answers0