I have a study wherein we enrolled about 40 subjects and from each subject we have collected repeated images of burns on subject's body. Approx. 2 burn locations on the body are selected initially and then repeatedly measured (imaged). Therefore, I have repeated measures of burn areas that are nested within a subject.

From within each image there are $TP$ correct results from $n=TP+FN$ trials, so I have an aggregated binomial likelihood.

I would like to model the intercept as burns nested in subjects, while modeling a random slope for the repeated measure (time) for each burn area.

From this post, I believe this is the correct model definition in lme4,

se.lme4_1 <- glmer(cbind(TP, FN) ~ 1 + time_delta_days + (1|subject_id:burn_id) + (1+time_delta_days|burn_id),
                   data = d,
                   family = binomial(link = "logit"))

the question I have is how is this model formula represented?

Here is my initial try:

$$TP \sim Binomial(n_i, p_i)$$ $$logit(p_i) = \alpha + \alpha_{subject[i]} + \alpha_{burn[i]} + \beta_{burn[i]}*time$$ $$\alpha_{subject[i]} \sim Normal(0, \sigma_{subject})$$ $$\sigma_{subject} \sim HalfCauchy(0,1)$$ $$\begin{bmatrix}\alpha_{burn}\\\beta_{burn}\end{bmatrix} \sim MVNormal(\begin{bmatrix}intercept_{burn}\\slope_{burn}\end{bmatrix},S)$$ $$S = \begin{pmatrix} \sigma_{slope_{burn}} & 0 \\ 0 & \sigma_{intercept_{burn}} \end{pmatrix}*R* \begin{pmatrix} \sigma_{slope_{burn}} & 0 \\ 0 & \sigma_{intercept_{burn}} \end{pmatrix}$$

$$intercept_{burn} \sim Normal(0, 10)$$ $$slope_{burn} \sim Normal(0, 10)$$

$$\sigma_{slope_{burn}} \sim HalfCauchy(0,1)$$ $$\sigma_{intercept_{burn}} \sim HalfCauchy(0,1)$$

$$R \sim LKJcorr(2)$$

I think the part of the model that is not making sense to me is how to deal with the nesting (i.e., the subject[i]).

Thanks in advance!



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.