Split file regression, or Fixed effects Mixed model

#1
Within a dataset that has a few variables explaining dependent variable A, and influences vary over different categories, it can be obvious to split the file in these categories and run a regression. However, there is also the possibility of running mixed models with fixed factors.

What would be better? I can imagine both methods have its positive and negative aspects but right now I cannot find any pro-mixedmodel arguments over a split file regression.

I'm explaining visitor frequency in over 100 shopping centers, with billions of observations. for those who are interested I could send you the results when finished.

Thanx a lot !!
 
#2
By "mixed models with fixed factors" you probably mean "mixed models with fixed factors". If in the fixed effects approach you add all the interactions of the following form

predictor * 1_{category X}

the approach is equivalent to running linear regression on each category separately. So doing everying at once is better because it allows you to test differences between different categories.... Also, if the number of categories is big, allowing each category to have a random intercept (the random effects option) might lead to a better model.