I'm interested in a simple random-intercepts model: $$Y_{ij} = \alpha_0 + \gamma_i + \epsilon_{ij}$$ where $\gamma_i \sim N(0, \sigma_{\gamma}^2)$ independently of $\epsilon_{ij} \sim N(0, \sigma_{\epsilon}^2)$.

In nlme:

l1 <- lme(effort ~ 1, data = ergoStool, random = ~ 1|Subject, method="ML")

and gives estimates of (what I assume are) $\sigma_\gamma$ and $\sigma_\epsilon$:


Random effects:
 Formula: ~1 | Subject
        (Intercept) Residual
StdDev:   0.9090556 2.020727

In gamlss, the same model can be fit, with the random intercepts included in a couple of ways:

t1 <- gamlss(effort ~ 1 + random(Subject), data = ergoStool)
t2 <- gamlss(effort ~ 1 + re(random = ~ 1|Subject), data = ergoStool)

The gamlss models give the same fit, and each has three variance 'components', which agree with one another:

c(sigb = getSmo(t1)$sigb, sige = getSmo(t1)$sige, exp(t1$sigma.coefficients))

       sigb        sige (Intercept) 
  0.9090593   1.0610962   1.9043752


Random effects:
 Formula: ~1 | Subject
        (Intercept) Residual
StdDev:   0.9090594 1.061096



The first is obviously the same $\hat{\sigma}_\gamma$, but I can't work out how the other two (the residual standard deviation from the random intercept part, and the estimate of the standard deviation of the outcome) relate to $\hat{\sigma}_\epsilon$. Is there a relationship between these estimates?


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.