18.5 Data dictionary and quality assessment

The dictionary records the dataset, variable, type, definition, allowed values or unit, and known quality issue. It also distinguishes identifiers from measurements and explicitly defines Censored as a renewal outcome that cannot yet be observed.

The raw visit data contain 64,131 rows but only 63,734 distinct visit identifiers. There are 397 duplicated visit identifiers, 1,163 missing durations, and 117 durations recorded as 720 minutes. The raw member file also contains missing postal FSAs, unknown service areas, and inconsistent capitalization in membership types.

An appropriate cleaning procedure is to:

1. preserve the raw files unchanged;
2. remove exact repeated visit identifiers after confirming the duplicate mechanism;
3. convert implausible durations above the defined operational limit to missing rather than inventing replacements;
4. standardize category labels through a documented mapping;
5. retain unknown geography as missing or a separate category when relevant; and
6. report how every change affects row counts and downstream analysis.

Duration should not be imputed automatically. It is unnecessary for visit counts, and an imputed value could create false precision in analyses of time spent at a facility.