Business Context
A retail bank wants to evaluate the effectiveness of a new email promotion ('Promo') versus no contact ('Control') on customer credit card spending. They need a robust estimation that corrects for confounding factors like Age and Income, providing confidence intervals for the decision makers.
About the Set : causalanalysis
Causal inference analysis and effect estimation.
Discover all actions of causalanalysis Data Preparation
Creation of 5000 customer records with confounding variables (Income, Age), a binary treatment (Promo/Control), and a continuous outcome (Spending).
Copied!
| 1 | |
| 2 | DATA mycas.marketing_campaign; |
| 3 | call streaminit(12345); |
| 4 | DO i = 1 to 5000; |
| 5 | age = rand('NORMAL', 35, 10); |
| 6 | income = rand('LOGNORMAL', 10, 0.5); |
| 7 | prob_promo = logistic(-2 + 0.05*age + 0.00001*income); |
| 8 | IF rand('UNIFORM') < prob_promo THEN treatment = 'Promo'; |
| 9 | ELSE treatment = 'Control'; |
| 10 | spending = 100 + 50*(treatment='Promo') + 2*age + 0.001*income + rand('NORMAL', 0, 50); |
| 11 | OUTPUT; |
| 12 | END; |
| 13 | |
| 14 | RUN; |
| 15 |
Étapes de réalisation
1
Fitting the Outcome Model (Spend) and Propensity Score Model (Treatment assignment).
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | regression.glm TABLE={name='marketing_campaign'}, class={'treatment'}, model={depvar='spending', effects={'treatment', 'age', 'income'}}, store={name='outcome_mdl', replace=true}; |
| 4 | logistic TABLE={name='marketing_campaign'}, class={'treatment'}, model={depvar='treatment', effects={'age', 'income'}}, store={name='ps_model', replace=true}; |
| 5 | logistic.score TABLE={name='marketing_campaign'}, restore={name='ps_model'}, casout={name='scored_data', replace=true}, copyVars={'spending', 'treatment', 'age', 'income'}; |
| 6 | |
| 7 | RUN; |
| 8 |
2
Execution of caEffect using AIPW method with Inference enabled to get ATE (Average Treatment Effect).
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | causalanalysis.caEffect TABLE={name='scored_data'}, method='AIPW', treatVar={name='treatment'}, outcomeVar={name='spending', type='CONTINUOUS'}, outcomeModel={restore={name='outcome_mdl'}, predName='P_spending'}, pom={{trtLev='Promo', trtProb='P_treatmentPromo'}, {trtLev='Control', trtProb='P_treatmentControl'}}, difference={{evtLev='Promo', refLev='Control'}}, inference=true, alpha=0.05; |
| 4 | |
| 5 | RUN; |
| 6 |
Expected Result
The action successfully calculates the Potential Outcome Means (POM) for Promo and Control. It produces the 'CausalEffects' table showing the ATE (Average Treatment Effect) of the promotion on spending, including Standard Errors and 95% Confidence Intervals, confirming the 'Doubly Robust' nature of the estimation.