One aggregated row cannot identify subgroup effects
@prranitarora▲ 11 net votes
A table with one row of population proportions and one total outcome cannot train a supervised model to estimate how subgroups affect that outcome. The data provides only the overall average, not subgroup-specific targets or variation across observations. Collect multiple comparable observations with outcome labels before fitting a regression or tree model.
1 COMMENT
Conclusions about subgroups drawn from aggregated proportions can suffer from aggregation bias. Individual-level records or grouped observations with variation are needed for subgroup effects. Corrections or other cases welcome below.