Number of cases is lower for fixed eeffects than for between effects in multilevel on panel data

[Maria Gardener]

Hi!

I'm estimating between and fixed effects on a sample of around 1300 participants, with three waves (as a level). Since there is quite a lot of attrition in my sample, I expected the N of the fixed effects to be lower than the N of the between effects, but the exact opposite is the case. There is probably a very logical explanation to this that I'm not seeing, but could somebody maybe tell me why this is the case?

Thanks in advance,

Maria

[Andrew Musau]

Can you show the exact output? The initial sample is the same for both, but the effective number of observations should be a lot lower for the between estimator relative to the fixed effects estimator as the former averages over time and estimates a cross-sectional equation (see, e.g., #5: https://www.statalist.org/forums/for...tor-regression). It will be difficult to justify the use of the between estimator with panel data as it does not eliminate the heterogeneity and thus remains biased as pooled OLS. The estimator is for the most part useful in considering the random effects model.