Registry Extract Cleaning & Comparison
Registry extracts carry their own conventions: 999 and 888 for missing values, free-text categories and columns whose meaning changed over the years. A comparison run on the raw extract is a comparison run on those conventions.
Teams need the codebook's rules applied and recorded, the categories standardised, and the comparison run with a method they can name, with its interval.
Done by hand, the rules are rewritten for every extract, the sentinel codes slip into means, and the next question on the same registry starts from zero.
Missing codes recoded by rule, every count reported
Declared codes are recoded as missing and categories standardised by operations listed with their counts, approved before the comparison runs; in a 150-record worked example, about 8% of cells were 999.
8 comparison families with 95% intervals
Two-group, one-way, categorical, correlation and covariate-adjusted comparisons from a fixed repertoire, each reported with its confidence interval and a declared multiplicity correction.
Rules that carry to the next extract
The cleaning code and the decisions stay with the study, so the next question on the same registry starts from the documented dataset.
How it works
Read the codebook: declared codes, categories and value ranges are taken from the registry's own documentation.
Apply and record the rules: missing codes, category mappings and exclusions are applied as counted operations and approved at the gate.
Compare: the comparison by area, group or period runs with a repertoire method and a declared multiplicity correction.
Report: results, the rules applied and the counts affected are written up together, with the Missing data section always present.