Cross Validated
2026-11-16 07:48 UTC
By Jade
AI-113-20261116-social-media-33c0ed21
Looking for advice on Random Effects in Generalized Linear Mixed Effects model for Cyclic counterbalanced repeated measures design
I have a cyclic/Latin-square-style counterbalanced repeated-measures design (as shown in image above), and I’m trying to work out the right way to use a Generalized Linear Mixed Effects model to analyse the data for it as I’m new to GLMMs. My aim is to compare how different cell-culture compositions affect the firing rates (expected to be of lognormal or tweedie distribution) of neural cultures while they’re being “trained” to perform tasks. The study uses 100 neural cultures split evenly into two experiments of 50. Experiment 1 uses Culture Type A and Experiment 2 uses Culture Type B (so culture type is nested within experiment, not crossed). Each experiment’s 50 cultures are prepared in two independent batches of 25, and within every batch the cultures are made up with one of three cell-culture compositions (A, B, or C) - giving four batches and all three compositions in both experiments. Each culture is then trained under 9 environmental conditions in a counterbalanced order, performing 2 tasks per environmental condition at 600 trials per task. The GLMM model I currently planning to use is: Firing rate ~ Culture Type + Batch + Cell Culture Composition + Environmental Condition + Task + (1 | Culture ID) However, I feel that using culture ID only oversimplifies the model and I am unsure of what other variables to use as random effect, especially since there’s so many of them. Hence, I would like to seek comments and advice on my current model from those who are experienced…
I have a cyclic/Latin-square-style counterbalanced repeated-measures design (as shown in image above), and I’m trying to work out the right way to use a Generalized Linear Mixed Effects model to analyse the data for it as I’m new to GLMMs. My aim is to compare how different cell-culture compositions affect the firing rates (expected to be of lognormal or tweedie distribution) of neural cultures while they’re being “trained” to perform tasks. The study uses 100 neural cultures split evenly into two experiments of 50. Experiment 1 uses Culture Type A and Experiment 2 uses Culture Type B (so culture type is nested within experiment, not crossed). Each experiment’s 50 cultures are prepared in two independent batches of 25, and within every batch the cultures are made up with one of three cell-culture compositions (A, B, or C) - giving four batches and all three compositions in both experiments. Each culture is then trained under 9 environmental conditions in a counterbalanced order, performing 2 tasks per environmental condition at 600 trials per task. The GLMM model I currently planning to use is: Firing rate ~ Culture Type + Batch + Cell Culture Composition + Environmental Condition + Task + (1 | Culture ID) However, I feel that using culture ID only oversimplifies the model and I am unsure of what other variables to use as random effect, especially since there’s so many of them. Hence, I would like to seek comments and advice on my current model from those who are experienced…
Full article content could not be extracted automatically. Read the original below.
Source:
Cross Validated
· stats.stackexchange.com