Can we model a bimodal response variable using a mixed effect model?

by M. Beausoleil   Last Updated September 16, 2019 14:19 PM

I have a response variable that is bimodal (basically, 2 normal distributions that are sticked together) and want to model it using a linear mixed effect model.

Here is a quick example (in R):

n1 =500
n2 =500
x = rnorm(n1,mean = 10)
y = rnorm(n2,mean = 15)

hist(c(x,y),breaks =25)

enter image description here


enter image description here

I can run an Expectation-Maximization algorithm for gaussian mixture to get the two distributions (this is a very simple example so the 2 distributions cluster very well)

ores = mixtools::normalmixEM(c(x,y),
                             sigma = NULL, 
                             mean.constr = NULL, 
                             sd.constr = NULL,
                             epsilon = 1e-15, 
                             maxit = 1000,
                             # verb = TRUE, 
                             ECM = FALSE,
                             arbmean = TRUE, 
                             arbvar = TRUE)
plot(ores,whichplots = 2)

enter image description here

My question is:

  1. Is it possible to model this bimodal variable as a response variable in a linear mixed effect model (or a GLMM if there exists a link function for that)?
  2. Should I need to separate the bimodal distribution in 2 distinct unimodal Gaussian distributions and construct 2 identical models but using each distribution in the separate models?
  3. What would be the effect of modelling a bimodal distribution with a linear mixed effect model (with a unimodal residual error)?

Finally, I heard that quantile normalization would be a way to compare the 2 distributions. How can quantile normalization be used to compare the 2 distributions in a linear mixed effect model?

Related Questions

Bi-Modal Tri-Modal data and residuals

Updated August 03, 2018 07:19 AM

Path analysis vs linear model

Updated February 28, 2018 23:19 PM

GLMM with time variable

Updated August 19, 2017 01:19 AM

Linear Mixed-Effects Model Appropriate?

Updated July 30, 2018 02:19 AM