0
$\begingroup$

I have been trying to perform beta regression modeling with random effects. I have sex ratio (0.561, 0.765 etc) as the response variable, and climatic variables + years (1970-2021) as predictor variables. Different farms were used as random effects. I used glmmTMB + beta family + logit for modeling. All explanatory variables were standardised.

But the residual plot shows significant quantile enter image description heredeviation. May I know why it is happening?

  1. The predictor variables were standardized
  2. No transformation of response variable
Improve this question
$\endgroup$
7
  • $\begingroup$ The first and obvious issue is that beta regression isn't suitable for modelling ratios. Please clarify: Your dependent variable is indeed a ratio and not a proportion? Also, "beta regression" is not sufficient to describe the model. What is your link-function? Maybe try a different a link function. Consider if you might have missing predictors. If you use time as a predictor, maybe that relationship isn't linear. How did you model the precision parameter? ... $\endgroup$
    – Roland
    Commented Apr 19 at 5:11
  • $\begingroup$ Dependent variable is ratio (0.543, 0.843 etc). Link function is logit. I have years as a predictior (but removing it didn’t help). $\endgroup$
    – Rahul
    Commented Apr 19 at 12:47
  • $\begingroup$ DV is a sex ratio. $\endgroup$
    – Rahul
    Commented Apr 19 at 13:05
  • $\begingroup$ I used glmmTMB with beta family and spatial groups as random effect $\endgroup$
    – Rahul
    Commented Apr 19 at 13:11
  • $\begingroup$ Do you have the counts or just the ratios? If you have the counts, you could use a logistic regression with random effects. $\endgroup$
    – COOLSerdash
    Commented Apr 19 at 16:12

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.