Why Cohort Entry Needs a Date
A study population is not always a fixed list of entities selected once at the beginning of observation. An entity may become eligible only after meeting an age requirement, accumulating enough account tenure, or having sufficient predictor history available. The study design must therefore specify not only who may enter, but when each entity first qualifies.
Eligibility is assessed at an index date: the date on which the prediction is defined. The earlier material on prediction index dates and predictor availability provides the context for that date. Here the focus is translating requirements into a rule that can be applied consistently to each entity and each candidate index date.
The entity might be a policy, account, customer, or other unit specified by the study. Eligibility is not a permanent label unless the rules make it one: a condition such as “account is active on the index date” can change over time. Accordingly, the rule must say which conditions are date-specific and which are thresholds that, once reached, remain satisfied.
Writing Eligibility as a Rule
A precise eligibility specification names the unit, the conditions, the date on which each condition is checked, and the schedule of candidate index dates. For example, a hypothetical study might require that an account be active on the index date, that its owner be at least 21 years old, and that the account have at least 90 elapsed days of tenure. If the prediction also requires a minimum amount of predictor history, that availability requirement must be stated as well.
For date-based thresholds that, once reached, remain satisfied, it is useful to calculate the first date each threshold is met. The first date on which all such conditions are met is the latest of those individual dates. But the actual cohort entry is the first candidate index date on or after that date. A monthly study, for example, may not permit entry on the day an entity meets the threshold if predictions are made only on the first of each month.
Worked Example: Combining Age and Tenure Dates
A hypothetical daily study observes accounts during June 2025. An account is eligible only if its owner is at least 21 years old on the index date, the account is active, and at least 90 elapsed days have passed since coverage began. Required predictor fields are available throughout June. The owner was born on June 15, 2004; coverage began on March 20, 2025; and the account remains active all month. Find the cohort-entry date.
Method. Find the first date each persistent threshold is met, then check the status condition on the resulting date.
Age condition. The owner turns 21 on June 15, 2025. The age requirement is not met on June 14 and is met beginning June 15.
Tenure condition. Count elapsed days from March 20 to June 18 using successive date boundaries: March 20 to April 1 is 12 days; April 1 to May 1 is 30 days; May 1 to June 1 is 31 days; and June 1 to June 18 is 17 days. Thus the elapsed time is \(12+30+31+17=90\) days, so the tenure requirement is first met on June 18.
The later threshold date is June 18. On that date the owner is at least 21, the account is active, and 90 elapsed days of coverage have passed. The predictor fields are also available. Therefore, the cohort-entry date is June 18, 2025. On June 17, tenure is only 89 elapsed days, so the account is not yet eligible.
Common mistake avoided. Counting March 20 to June 18 as 11 days, 30 days, 31 days, and 18 days mixes date-boundary conventions. In particular, June 1 to June 18 is 17 elapsed days, not 18. Using consistent elapsed-day intervals gives 90 days and the correct entry date.
Scheduled Index Dates and Available History
A threshold date and an entry date are not always the same. If a study issues predictions only on monthly index dates, eligibility is tested on those dates rather than continuously. An entity that meets a condition between two monthly indexes generally enters at the next scheduled index, provided every condition still holds then.
The same logic applies to historical data requirements. If a predictor requires a defined history before the index date, the entity cannot enter until that history is available. The history requirement is evaluated relative to the candidate index; having data arrive later does not make it available at an earlier index.
Worked Example: Finding Entry on a Monthly Schedule
A hypothetical account study has index dates on the first day of each month from April through June 2025. An account qualifies if it is active, has at least 60 elapsed days of tenure, and has at least 30 elapsed days of predictor data from the system’s start date, March 1, 2025. The owner is already old enough to satisfy any age requirement. The account opened on February 10, 2025 and remains active. Find its first eligible monthly index.
Method. Calculate each threshold, then test the candidate monthly dates in order.
Tenure threshold. From February 10 to March 1 is 19 elapsed days; March 1 to April 1 is 31 days. Therefore, by April 1 the account has \(19+31=50\) elapsed days of tenure, which is short of 60. From April 1 to April 11 is another 10 days, so 60 days are reached on April 11.
Predictor-history threshold. March 1 to April 1 is 31 elapsed days, so the 30-day data-history requirement is satisfied by the April 1 index.
Check the scheduled indexes. April 1 fails because tenure is only 50 days. May 1 is 80 elapsed days after February 10: \(19+31+30=80\). The account is active, has at least 60 days of tenure, and has more than 30 days of predictor data. Thus the first eligible monthly index is May 1, 2025.
The threshold date for tenure is April 11, but that is not a candidate index. The scheduled-entry rule moves entry to the first eligible index after that date, May 1.
From Individual Rules to a Cohort
A practical cohort specification should be applied entity by entity. For each entity, determine which candidate index dates are possible, evaluate every requirement at each relevant date, and record the first date at which all requirements hold. An entity with no qualifying date by the study’s end has no entry during that study window.
State whether one row represents an account, policy, customer, or another entity, and list the dates on which eligibility may be assessed.
Specify the threshold, the relevant date, and whether the condition must hold only at the index or has to be accumulated beforehand.
For duration requirements, use elapsed time between date boundaries. Do not mix inclusive calendar-day counts with elapsed-day counts.
At each candidate index, verify all requirements, including any changing status and availability of required predictor history.
Keep the first qualifying index as the cohort-entry date. If no candidate date qualifies within the study window, record that there is no entry in that window.
The “latest threshold date” shortcut is valid only for persistent thresholds. For example, age and accumulated tenure normally remain satisfied after they are reached. Account activity may not: an account can become inactive. Therefore, an account that reaches the required tenure must still be checked for activity on the scheduled index date. The shortcut also cannot create a scheduled index that does not exist.
Worked Example: Applying One Rule to Several Entities
A hypothetical study has monthly index dates on May 1, June 1, and July 1, 2025. To qualify, an entity must be at least 25 years old, have at least 45 elapsed days of account tenure, be active on the index date, and have 30 elapsed days of predictor data. The data system began recording the required fields on April 1, so the history requirement is met by May 1 and thereafter. All three accounts remain active. Find each account’s first cohort-entry date, if any.
| Entity | Date of birth | Account opened | Study ends |
|---|---|---|---|
| A | April 20, 2000 | March 10, 2025 | July 1, 2025 |
| B | May 10, 2000 | March 1, 2025 | July 1, 2025 |
| C | January 1, 1990 | June 20, 2025 | July 1, 2025 |
Entity A. A is 25 by May 1 because the 25th birthday was April 20. Tenure from March 10 to April 1 is 22 elapsed days, and April 1 to May 1 is 30 days, giving \(22+30=52\) days by May 1. The data-history requirement is met, and A is active. A enters on May 1.
Entity B. B is not yet 25 on May 1 because the birthday is May 10. B is 25 on June 1. Tenure from March 1 to June 1 is \(31+30+31=92\) elapsed days, so the 45-day threshold is met by then. The data-history and activity conditions are also met. B enters on June 1.
Entity C. C is old enough, and the data-history requirement is met on the scheduled dates. But by July 1, only 11 elapsed days have passed since June 20, which is less than 45. The 45-day threshold is not reached until August 4: June 20 to July 1 is 11 days, July 1 to August 1 is 31 days, and August 1 to August 4 is 3 days, totaling \(11+31+3=45\). There is no candidate index on or after August 4 within the study window, so C has no cohort-entry date in this study.
Check. A enters at the first listed index when every condition holds; B waits until the first index after the birthday threshold; and C does not meet tenure by the last index. The results are consistent with the scheduled-index rule.
Common Eligibility Traps
- Confusing eligibility with presence in the source data. A record existing in a table does not by itself establish that the entity meets the study’s eligibility requirements at an index date.
- Using a threshold date as the entry date when indexes are scheduled. If the threshold is met on April 11 but indexes occur monthly on the first, the next possible entry may be May 1, not April 11.
- Counting dates inconsistently. “At least 90 elapsed days” is not the same as counting both the start and end dates as full days. State and apply one convention.
- Checking only one condition. The first date an entity reaches the age or tenure threshold is not necessarily its cohort-entry date; every requirement must hold on the candidate index.
- Assuming entry is possible for everyone. An entity may not satisfy the rules before the study window closes. Record that result rather than assigning an earlier, ineligible index.
Precise cohort entry makes the study population reproducible: two analysts applying the same rule to the same dates should identify the same first qualifying index for each entity. It also makes disagreements easier to diagnose, because each requirement and its date can be checked separately. This topic is aligned with the SOA Exam PA (Predictive Analytics) syllabus.
Check Your Understanding
For each problem, apply the stated elapsed-day convention and use only the candidate index dates given.
- A hypothetical daily study requires an account to have at least 40 elapsed days of tenure. It opened on August 5, 2025, remains active, and all other conditions are met. What is its first eligible date? (Answer: September 14, 2025)
- A study has monthly indexes on October 1, November 1, and December 1, 2025. An active account opened on September 10 and needs 45 elapsed days of tenure. All other requirements are met. What is its first eligible index? (Answer: November 1, 2025)
- A daily study requires age at least 30 and at least 30 elapsed days of account tenure. A person born on December 2, 1995 has an active account opened on November 20, 2025. What is the first date on which both conditions hold? (Answer: December 20, 2025)
- Monthly indexes occur on January 1, February 1, and March 1, 2026. An entity must have at least 60 elapsed days of tenure and at least 20 elapsed days of predictor history. Its account opened November 25, 2025; the required data became available on December 15, 2025; it is active throughout. Find its cohort-entry date. (Answer: February 1, 2026)
- A study has indexes on May 1, June 1, and July 1, 2026, and requires at least 50 elapsed days of tenure. An active account opened on June 20, 2026, and all other conditions are already satisfied. Does it enter during the study window? (Answer: No; it reaches 50 days on August 9, after the last candidate index.)