How the High Frequency Checks work

The High Frequency Checks look for problems while fieldwork is still running: interviews that are too fast, GPS that makes no sense, an enumerator unlike the rest of the team. There are twenty-eight, each explained in the check reference. Their defaults follow the practice of field-research teams (IPA's ipacheck and the World Bank's DIME guidance), and every threshold can be changed.

Two ways checks run

• On arrival. Checks about one interview (its duration, its GPS, its answers) run the moment the submission lands. A supervisor learns on Tuesday that something is wrong with Monday's interviews, not when the data is cleaned months later.
• On demand. Checks that compare interviews or enumerators (who is slower than their peers, which answers are copied between interviews, whether the digits follow Benford's law) only have an answer over a whole form. Run HFC checks on Monitor runs the checks you tick over all of a form's data and reports what each found. Checks about one interview can be run this way too.

Days are counted by when each interview happened (its start time), in the workspace's time zone.

Which question holds what

Several checks need to know which questions hold the enumerator, the start and end times, the GPS reading and the stratum. Joripo detects them from the deployed form where it can (the start and end times and the GPS reading from their question types) and shows what it found, with the reason, at the top of the form's quality-check settings: Which question holds what (the form's HFC checks, under its Tools or on Monitor).

If the checks seem to do nothing, look here first. A missing or wrong mapping does not raise an error: the duration check simply never fires, quotas never fill, scorecards stay empty.

Settings per form

Each form has its own quality-check settings: which checks run and their thresholds, built from what each check needs (a question to pick, a number to type). Two forms in one workspace never share settings or data; a form with no settings of its own follows the workspace's defaults. Quality checks in force on Monitor lists, form by form, which checks run and with what thresholds.

Saving the settings makes a new version of them rather than editing in place, so a finding raised last week keeps the thresholds it was raised against. Three checks (straightlining batteries, skip rules and contradictory answers) take structured rules written in a small JSON editor; a visual rule builder for them is not built.

Nothing fails silently

A check that cannot run never looks like a clean survey. Each check run has one of four outcomes, shown on screen and in the Excel report:

In Run HFC checks, a check whose questions were never mapped is shown disabled, naming what it needs, rather than offered as a tick that would run over nothing. An empty findings list says why it is empty: no interviews yet, checks not set up or not run yet, or checks skipped for want of a mapped question. Only when the checks ran and found nothing does it say there is nothing to review.

A flood is folded into one finding

A finding raised on at least half of a form's interviews (and at least 30 of them) points at the form or its mapping, not at the data: "no GPS coordinates" on every interview of a form that collects no GPS. Such a flood is folded into a single finding that names the count. It is collapsed, not dropped: the finding is real and has to be seen once.

An enumerator is compared with the others, not with an average that includes them

Checks that compare enumerators score each one against the rest of the team, never against a team average and spread that include their own values. On a small team, one extreme enumerator inflates the spread enough to hide inside it: with six enumerators, one rounding every answer could pass a two-standard-deviation test against the whole team. When the other enumerators agree exactly, any difference from them is reported as the plain outlier it is.

Tiers

Every check has a tier (basic, standard or advanced), recorded for each finding. Every tier is included in every plan; on the Free plan the checks run on the first 200 submissions a month.