Innovation Advisory · Framework 17
6 questions · during fieldworkThe High-Frequency Check
Six questions for the checks you run while a survey is still in the field, which is the only time most data problems can still be fixed. The canvas comes from the field operations code in FieldStack.
Is each enumerator compared with others who worked in the same area?
Enumerators are almost never assigned to places at random, so an enumerator whose numbers look unusual may simply have been sent somewhere unusual. Compare each one within the same block or village cluster, and keep that caveat attached to every outlier list you send.
Are back-check questions sorted by type, with an error rate for each?
Type 1 questions catch fabrication, type 2 questions test the enumerator’s skill, and type 3 questions test whether the instrument is stable. Report each rate separately. One pooled rate makes an honest enumerator asking hard recall questions look worse than a careless one asking easy ones.
Are timestamps read in the local time of the interview?
ODK records time in UTC, so an ordinary 09:00 interview in Bihar is stamped 03:30. A check for interviews outside working hours that reads UTC flags the whole survey, and a field team that receives one useless report stops reading the next one. FieldStack defaults to Asia/Kolkata for this reason.
Does any check change the data?
Flag, send back and record. A check that quietly repairs a value hides the repair from the analysis and teaches the enumerator nothing. Every correction should come from the field, with a note saying who made it and why.
Who reads tomorrow’s report, and what will they do with each flag?
Write the report for the supervisor who will read it on a phone before the day starts. Name the enumerator, the household and the question, and give one action for each flag. A long report with no actions is skipped.
Will the checks run where the team works?
FieldStack’s field operations code uses base R and no packages, because it has to run in a district office on a connection that will not install anything. Test the checks on the machine and the connection the team will use.
What the six questions are doing
High-frequency checks fail in two ways. They flag the wrong people, which happens when an enumerator is compared with colleagues working somewhere else, when back-check questions of different kinds are pooled into one error rate, or when timestamps are read in the wrong time zone. Or they are right and nobody acts on them, because the report is long, generic or arrives where it cannot be opened. The six questions cover both.
Use it in the week before fieldwork starts, and again after the first three days of data arrive.