Business Context
A telecom provider analyzes customer churn. They suspect the baseline churn rate differs significantly between geographical regions (North, South, East, West), violating the proportional hazards assumption across regions. Additionally, since tenure is recorded in whole months, there are many tied event times (multiple customers leaving at exactly month 12).
Data Preparation
Create a dataset with categorical regions and discrete tenure times to force ties.
Copied!
| 1 | |
| 2 | DATA mycas.telco_churn; |
| 3 | call streaminit(789); |
| 4 | LENGTH Region $10; |
| 5 | DO i = 1 to 200; |
| 6 | IF rand('uniform') < 0.25 THEN Region='North'; |
| 7 | ELSE IF rand('uniform') < 0.5 THEN Region='South'; |
| 8 | ELSE IF rand('uniform') < 0.75 THEN Region='East'; |
| 9 | ELSE Region='West'; |
| 10 | MonthlyCharge = 50 + floor(rand('uniform')*50); |
| 11 | Tenure = floor(rand('exponential') * 24); |
| 12 | IF Tenure < 1 THEN Tenure = 1; |
| 13 | Churn = (rand('uniform') < 0.3); |
| 14 | OUTPUT; |
| 15 | END; |
| 16 | |
| 17 | RUN; |
| 18 |
Étapes de réalisation
1
Run Stratified Cox Model handling ties with the Efron method.
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | phreg.cox TABLE={name='telco_churn'}, strata='Region', model={depVars={{name='Tenure', event='Churn(1)'}}, effects={{vars={'MonthlyCharge'}}}, ties='EFRON'}; |
| 4 | |
| 5 | RUN; |
| 6 |
Expected Result
The output must contain the 'ModelInfo' table confirming 'Efron' was used for ties. The 'ParameterEstimates' will show the effect of 'MonthlyCharge' on churn, while the 'Stratification' table will confirm that separate baselines were computed for North, South, East, and West regions.