3
$\begingroup$

I have a data set which I think requires an linear mixed model to analyze, but am unsure about some details. Below is a brief description of the experiment:

I have data from ~ 30 mice half of which are male and half of which are female. I have a behavioral measure that I'd like to predict based on activated cells in the brain. I performed a cluster analysis on cells counts in 485 regions and a SVD on each cluster, which leaves me with 9 repeated measures (I am only using the first singular value for each cluster) and sex as a between groups variable. I would think that the model should be:

lmer(Behavior ~ sex + cluster + (sex * cluster) + (cluster | mouse)

My only concern is that my DV is not a repeated measure. My question is then if this is the correct approach. See the data below:

enter image description here

edit: Reposting data for a single mouse in response to the first comment

`Behavior   Cluster Sex Mouse
28  -0.038136592    0   1
28  -0.023112313    0   1
28  -0.222644927    0   1
28  -0.208861993    0   1
28  -0.269723875    0   1
28  -0.062751677    0   1
28  -0.191682896    0   1
28  -0.245295508    0   1
28  -0.190930967    0   1`
Improve this question
$\endgroup$
7
  • $\begingroup$ Is each row in the dataset one mouse ? Where is the variable "cluster" in the dataset and what is it exactly ? How do the 9 "repeated measures" which I assume are the columns in the image with decimals enter the model Behavior ~ sex + cluster + (sex * cluster) + (cluster | mouse). And the outcome variable Bahaviour (is it Attack Durati in the picture. or something else) ? $\endgroup$
    – Robert Long
    Commented May 24, 2021 at 16:05
  • $\begingroup$ Yes each row is a mouse The data re-arranged for a single mouse is now in the original post. $\endgroup$
    – AA1989
    Commented May 24, 2021 at 18:15
  • $\begingroup$ Ok but I asked 4 questions and you've only answered the first. $\endgroup$
    – Robert Long
    Commented May 24, 2021 at 18:49
  • $\begingroup$ The outcome variable is Behavior, as indicated in the edited post. The variable cluster in the dataset is the column which says "cluster", as indicated in the edited post. Cluster is the first singular value of a module that was identified by using a clustering algorithm on the cell count data. There were 9 modules/cluster detected, and each value is the first singular value for said module/cluster. $\endgroup$
    – AA1989
    Commented May 24, 2021 at 18:59
  • $\begingroup$ Ok thanks. I think I get the picture now. I don't think a mixed model is appropriate here since you don't have repeated measures. The 9 clusters are really just 9 variables for the same mouse so a model with fixed effects for those 9 variables plus sex would be appropriate but with only 30 mice, you will need to use some kind of penalization. $\endgroup$
    – Robert Long
    Commented May 24, 2021 at 19:16

1 Answer 1

Reset to default
3
$\begingroup$

As noted in the comments a mixed model is not appropriate here, since there are no repeated measures of the outcome.

The 9 clusters are really just 9 variables for the same mouse so a model with fixed effects for those 9 variables plus sex would be appropriate but with only 30 mice, you will need to use some kind of penalization such as LASSO.

Also if the clustering algorithm used to find the 9 clusters results in orthogonal clusters - is the 9 cluster variables are uncorelated with each other (or their sample correlations are low) then you could just fit a model with, say, the 3 clusters with the highest correlation with the response.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thanks for the answer. The clusters aren't orthogonal. Three of the nine are correlated (r =.6-.7) however the others aren't (r < .4) $\endgroup$
    – AA1989
    Commented May 25, 2021 at 12:14

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.