Left Truncation
This section moves to another topic, which is the accurate calculation of hazard probabilities. As noted in the previous chapter, the customers in the Subscribers table data have an unexpected property: Customers who stopped before some date (that depends on the market) are excluded from the table. This phenomenon, where customers are excluded based on their stop date, is called left truncation, and naïve hazard calculations on left-truncated data produce incorrect results. In the previous chapter, the problem of left truncation was handled by introducing LeftTruncationDate into all the queries. This section presents a more flexible method, based on an idea called time windows.
Left truncation is a problem because hazard estimates on left truncated data are simply incorrect. The solution to left truncation is to calculate the hazards using only a “time window” of activity—the calculation only uses information from customers active during the time...