I'm using the GAMLSS package to model an ex-Gaussian analysis of reaction time data over two time points (repeated measure). When I include random effect for subject (very standard practice for repeated measures analyses as far as I know) the model fit GAIC is better, but the diagnostic Q-Q plot shows heavy tails (and likely inflated results) compared to the model without the random effect of subject - which shows truncated tails and more conservative results. I'm fine with being more conservative, but I don't understand the issue with model fit. This poster gets at the question but in my case it's a much more simple model: Selecting GAM with/without random effects - residual plots vs. AIC

Here's the GAMLSS model with the random effect, maybe I'm doing something wrong?

GAM_Va3 <- gamlss(formula = RT~ valence + random(subID), sigma.formula = ~ valence + random(subID), nu.formula = ~valence + random(subID), family=exGAUS(), data=na.omit(data))

Here's the GAMLSS model without the random effect GAM_Va4 <- gamlss(formula = RT~ valence, sigma.formula = ~ valence, nu.formula = ~valence, family=exGAUS(), data=na.omit(data1))

See Q-Q plots below

Q-Q plot with random effect Q-Q plot without random effect

  • $\begingroup$ Hi there, did you ever get to the bottom of this? $\endgroup$
    – HCAI
    Aug 16, 2018 at 22:09


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.