Audit a nutrition dataset for unit errors, implausible measurements and inconsistencies (API key)
POST /data-quality/audit · API key required
Send up to 2000 records (a survey or study dataset). Units, ranges, missing codes and variable names are validated with the same rules as POST /research/harmonise (columns, units, missing_codes and date_format work the same way). Children with sex and age (a date of birth with an assessment date, or an age column) get WHO z-scores from the same classifiers as GET /growth/assess, and implausible ones are flagged against the WHO limits (haz -6/+6, waz -6/+5, whz -5/+5, baz -5/+5), or against the survey mean with options.flag_rule survey_mean. Consistency checks: age column against dates, visit before birth, future dates, weight above height (swapped columns), recorded BMI against weight and height, supplied z-scores (supplied_z) against recomputed ones, duplicate records, ids and visits. Survey-level quality: completeness, digit preference in weight, height and MUAC (terminal-digit counts, chi-square p and a chance-corrected heaping index), age heaping, sex ratio and age ratio (exact binomial tests), and the mean, SD, skewness and kurtosis of each z-score; group_by repeats these for each team or cluster and lists those to review. Suspected unit errors are listed with the unit in which the values would be plausible. Nothing is corrected: flagged values are left in the report and set to null in the optional analysis table (options.include_table). Screening aid, not a verdict; SMART/ENA plausibility scores are not reproduced. Nothing is stored.
Request body
Content type: application/json
Responses
200400429