It's just a special case of omitted variables with categorial variables. So instead of parameter estimates being biased up or down x amount (to the extent covariates are correlated with error terms), with Simpsons's paradox the mean effect is completely wrong due to improper grouping. This often leads to flipping signs on estimated parameters -- 'surprising' results that gets papers published.

My favourite explanation: http://vudlab.com/simpsons/