Feasibility-Constrained Programme Aggregation
Deploying algorithmic recommendation models across dozens of highly specialized public measures causes estimation failure due to tiny sample sizes per measure, empty choice sets for specific client demographics, and operational supply constraints in local training centers.
Picture this
Imagine a restaurant menu listing 43 specific dishes. If a recommendation algorithm tries to predict how much every diner will like each individual dish, it will fail because many dishes are rarely ordered or unavailable. Grouping the menu into 6 broad categories (like pasta, seafood, or salads) ensures every diner gets a reliable recommendation that the kitchen can actually serve.
What the evidence says
Program aggregation eliminated demographic choice-set restrictions (such as native language requirements), increased sample sizes per category to maintain statistical estimation power, and aligned algorithmic choices with operational local course availability.
- Who
- N = 460,442 historical jobseekers evaluated across 43 official Swiss labor market training measures.
- How
- Econometric aggregation of 43 disaggregated administrative training measures into 6 to 8 broad feasibility-consistent categories (e.g., Basic Courses, Language Skills, Computer Skills, Further Training, Employment Programmes, No Programme).
What to do
Aggregate highly disaggregated intervention types into broader functional categories that guarantee valid choice sets for all client demographics while preserving statistical estimation power.
From the source
"By defining a category language skills training which includes German, French and foreign language courses, this category becomes feasible for every jobseeker, and the X_it characteristics (e.g. mother tongue, profession) define which type of language course or further training is appropriate."
Targeting_Labour_Market_Programmes_Results_from_a_Randomized_Experiment.pdf