2
$\begingroup$

I am estimating a Cumulative Link Mixed Model (CLMM) in R on a dataset similar to the following:

Likert Age Income Location
Uncomfortable 65 High East
Comfortable 42 Low West
Comfortable 68 Low Center
Neutral 75 High East
... ... ... ...

The code I use to estimate the model is:

m = clmm(Likert ~ 1 + Age + Income + (1|Location), data = d, 
      threshold = "flexible")

My question is what do the fitted.values represent in this model? If I run fitted.values(m) a numeric list with values between 0 and 1 are outputted. How can I interpret these values?

The Ordinal package documentation defines fitted.values as

fitted values evaluated with the random effects at their conditional modes

but I don't understand what this really represents about my dependent variable.

$\endgroup$
2
  • $\begingroup$ Apparently, fitted.values() of a polr object outputs a matrix, with a column for each level of the response. I don't know why CLMM only outputs a list! $\endgroup$
    – Amir H
    Sep 17, 2021 at 0:01
  • $\begingroup$ As an aside what about location makes you treat it as a random effect? $\endgroup$ Oct 27, 2023 at 12:38

1 Answer 1

1
$\begingroup$

Looking through the vignette for the clm function (assuming clmm works similarly), clm_article.pdf, it looks like it's returning the predicted probability of the observed outcome level for each observation. As mentioned by Amir H, polr from MASS returns a matrix with the predicted probabilities of each possible outcome level (not just the observed level for that row).

$\endgroup$
1
  • $\begingroup$ This predicted probably can be used for residuals for diagnostic purposes as in the Li-Shepherd residuals. But the polr output is more useful in practice since it provides the matrix of all predicted probabilities, i.e., you get as columns all values of $y$ with $\Pr(Y=y)$. $\endgroup$ Oct 27, 2023 at 12:37

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.