Overview
The Data Quality Lab reports variable proportions, complete and incomplete cases, common binary missingness patterns and a bounded heat-map sample. These are descriptive diagnostics.
Theoretical foundation
Observed missingness can arise under different data-generating mechanisms. Descriptive patterns alone cannot identify MCAR, MAR or MNAR.
Algorithm and computation
Pandas null indicators are aggregated by variable and row. The 20 most frequent binary patterns are reported. Little’s MCAR test is explicitly unavailable. Analysis services apply procedure-specific available, listwise or pairwise deletion; transformations can preview simple constant or mean/median treatments where offered.
Required data
An immutable dataset version. Missing values are those parsed as null by the supported importer and parser.
Assumptions
The researcher distinguishes true nonresponse from planned skips, routing and coded missing values.
Available settings
Diagnostic selection is descriptive. Any treatment must be previewed and approved as a new dataset version.
Worked example
A synthetic survey has 6% missing on one item and 2% on another. Survlytics reports both denominators and the most common pattern; it does not infer a missingness mechanism.
Limitations and non-claims
Little’s MCAR test, multiple imputation, FIML and sensitivity analysis are not implemented. Mean imputation does not preserve uncertainty and should not be treated as a general solution.
Validation notes
Missingness counts and pattern construction are unit-tested. No inferential missingness claim is made.
References and verification status
Little, R. J. A. (1988). A test of missing completely at random for multivariate data with missing values. Journal of the American Statistical Association, 83(404), 1198–1202. https://doi.org/10.1080/01621459.1988.10478722
Unverified draftOpen DOI
Was this article helpful?
Your response helps the methodology team improve this documentation.