2
$\begingroup$

I have a dataset where different features were measured for three conditions. Each feature consists of sub-features, and sub-features are nested within features. I'm interested in the contrasts cond1 vs ctrl and cond2 vs ctrl on the feature level. Here is a part of the dataset (the full data contains ~10k features) and here's an overview of the design: enter image description here

Each sub-feature has it's own intercept enter image description here and so I fit a model readout ~ (1|subfeature) + (condition|feature). This gives a boundary (singular) fit warning, likely because a corresponding fixed-effects model readout ~ subfeature + condition:feature has a non full-rank design.

How can I simplify the mixed-effects model to have a full-rank design? In the fixed-effects only case, I would replace the design matrix columns corresponding to the interaction effects by columns corresponding to the contrasts conditioncond{1|2}_vs_conditionctrl:featuref{1|2}.

$\endgroup$
2
$\begingroup$

If the model matrix for subfeature + condition:feature is not of full rank, then I would simply remove condition from the random structure.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Does this answer your question ? If so, please consider marking it as accepted. If not, please let us know why. Thanks ! $\endgroup$ – Robert Long Jul 19 at 4:48

Your Answer

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

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