MIXED EFFECTS (EMPIRICAL BAYES) MODEL

This method is also known as ‘Random Coefficients Model’ or ‘Mixed Effects’ regression.

Empirical Bayes Model, similar to the Pooled regression method, is applied when time-series data is scarce. The only difference is that this method uses Bayesian techniques to leverage information across different sub-groups (stores, products etc.) to generate ‘sub-group level’ estimates of coefficients in addition to ‘overall’ coefficients.

Here's an illustration to demonstrate the power of this approach:

You are trying to evaluate the performance of a program that you ran in 5 test markets for 10 weeks. You would like to know the effectiveness of the program in each of the 5 markets. 10 Weeks doesn't give you a robust enough sample size to run a standard regression model, but with Mixed Effects Modeling you can actually leverage information over the 10 weeks across the 5 markets and actually generate coefficients measuring the impact within each of the 5 markets individually and across the 5 markets in total in a single model.