Data source and storage

Signing in requests a new tabular export, including metadata, from the ANU PNG questionnaire in Survey Solutions. Pilot questionnaire version 6 is selected explicitly for both the dashboard and Stata downloads. API credentials remain on the secure server.

Processed data is kept in this tab’s session storage and removed when you sign out. The server keeps session exports in private temporary storage for up to 30 minutes. Session export files expire after 30 minutes and are removed on sign-out.

Variable harmonisation

Regular and _rev fields are merged using their original answer codes. Reversed scales change presentation order, not the meaning of the stored codes.

Q24’s seven answer-order variants are combined. All four Chiefs fields (early, late, regular and reversed) become s2q32. Conflicting nonmissing variants are excluded and reported, rather than silently selecting an answer.

Merge checks will appear after syncing.

Denominators and missing data

The default view includes workflow statuses Completed, Approved by supervisor and Approved by headquarters. You can switch the results tab to all interviews. Enumerator quality always uses completed or approved cases. These are Survey Solutions workflow statuses, separate from the questionnaire’s interview outcome (s3q2), which includes complete with some questions missing (code 7).

Percentages use recorded responses within each group; “don’t know” and “refused” are recorded responses. Missing answers are reported separately and can include routing skips. Unanswered questions in the quality table use the routing-aware diagnostic count.

Interim results are unweighted descriptive summaries, not population estimates. Group sizes below 20 are marked as small samples.

Positivity score

For each respondent, eligible answers are mapped to 0–100, then averaged with equal item weights. Enumerator scores average those respondent scores. “Don’t know”, “refused”, missing and nondirectional questions are excluded. Q15’s “Staying the same” response (code 4) scores 50. Item coverage is shown because respondents can answer different subsets.

This is an exploratory index, not a validated scale or evidence that an enumerator influenced responses. Compare respondent characteristics before interpreting differences. The following mappings are editable in app.py.

Enumerator comparisons

Duration, unanswered questions per interview, and positivity are compared using two-sided Welch t-tests against completed cases from all other enumerators. At least two nonmissing observations in each group and estimable variance are required. Total unanswered counts are not tested because they depend on caseload.

Holm adjustment is applied across all available enumerator-by-metric tests in the current export. Arrows indicate adjusted p < 0.05; hover text provides the raw and adjusted p-values and sample sizes. No arrow is shown when a test is unavailable or not significant. Comparisons do not adjust for respondent characteristics or sampling clusters.

The average row is the arithmetic mean of the enumerator rows, giving each enumerator equal weight. Case counts and unanswered totals in that row are average counts per enumerator. Missing metric means are excluded.