Field monitoring projects

Field monitoring turns a form's data into the numbers a field team is run by. It sits under Monitor, one row per workspace; reviewers and up open it, and owners and managers set it up. A monitoring project reads one source:

• a Joripo form of the workspace, read as submissions arrive, with the workspace's privacy rules applied;
• a SurveyCTO form, pulled from a SurveyCTO server on a schedule or with Sync (the password is sealed on the server);
• a CSV, Excel or Stata file you import, replacing or adding to what is there (rows missing from a new file are marked, never deleted).

The quickest start is a form's Dashboard button on Design: the first time a manager presses it, Joripo makes a monitoring project of the form and a dashboard built from its variables. See Dashboards.

Setting a project up

Owners and managers work through these, in order:

• Variables: what each field is for: its label, roles, which one identifies a household or a respondent, value labels, which are hidden or not exported. Bulk edits are possible.
• Derived variables: new values computed with the XLSForm expression language, such as ${age} >= 18 and selected(${services}, 'anc'), typed, ordered by what they depend on, with loops refused and a preview.
• Indicators: a count, distinct count, sum, mean, median, minimum, maximum, percentage or ratio, each with who is counted (the universe), what qualifies (the numerator) and what happens when an answer is missing. A missing answer is never counted as zero. Build conditions with the visual builder or write them; the preview shows the value on today's data and where every record went. An indicator can be computed per submission, per row of a repeat group (household members), or over a linked monitoring project.
• Rules: what to look at: a duplicate household ID, an interview too short, an answer out of range, an outlier, poor GPS accuracy, an enumerator whose rate is out of range. A rule is tested on the data before it can be switched on. What it finds goes to Review, where a reviewer confirms a finding as a real problem, marks it not a problem or corrected, with a note; every decision is kept. A finding is something to check, not a verdict on anyone.
• Productivity: enumerators down, days across, the interviews each did each day as the project counts them (every submission, completed interviews, or unique households), with totals, working days, low days and the share of a daily target. Days off, leave and not-yet-started are never counted as zeros, and every cell opens the records behind it.
• Targets: target sets per dimension, with an overall target, a fieldwork window, a pace and an expected line.

Cleaning lines

The Cleaning page keeps the team's Stata corrections, keyed on the submission's KEY (replace a06 = 3 if key == "uuid:…", drop if key == "uuid:…"), and applies them every time the project's data is read: every dashboard, table, rule, tracksheet and Sheets tab shows the cleaned data, and a dashboard says it is "after the cleaning lines". The submissions themselves are never changed, so editing or removing a line undoes it. replace, drop if and keep if are applied, with Stata's conditions and abbreviations; any other command is listed as not applied, so a whole do-file can be pasted. Check against the data shows what each line would change before it is saved.

These lines change only what field monitoring shows. To correct the data itself, for exports and everyone else, use Review and correct.

Google Sheets

A project can write tables to a Google spreadsheet, from seven templates: manually, after each sync, hourly or daily. It uses a Google service account whose key you upload (sealed on the server), and only ever writes the tabs it made: a tab someone else made in the same spreadsheet is never touched.

What is not proven yet

Not yet run against live services. The SurveyCTO and Google Sheets connections are tested against faithful copies of those services' protocols, not against a real SurveyCTO server or Google account. Watch the first real use. Encrypted SurveyCTO forms are not supported.

Weighted results are point estimates only: no standard errors, design effects or confidence intervals. Calculation happens in memory per project, which is comfortable at pilot scale (tens of thousands of records).