Why Null model MLM intercept differs from Null OLS intercept?

[Ruslanas Stacevicius]

Hi,

I noticed that Null Mixed Linear Model intercept doesn't match the mean of the response variable and doesn't match Null OLS model intercept too. While in theory I assume they should match.

I ran Null model on some variables from public NHANES dataset and in all cases MLM intercept is slightly lower than OLS and mean

VARIABLE MLM intercept OLS intercept mean BPXSY1 : 125.56 125.63 125.63 BMXBMI : 29.46 29.51 29.51 SDMVSTRA : 126.06 126.21 126.21

Usually differences are small, but it's unclear is there a methodological reason for this or does it happen just because of rounding errors?