Before the Visit
Desk Review Checklist
This page is the working list. For the reasoning, the dimension definitions and the eight-step process, use How to Conduct a Data Quality Assessment. Come back here when you are packing for the visit.
Do these in the office, a week before you travel. Each one saves time at the site.
| Check | What you need in hand |
|---|---|
| Indicators chosen (5 to 8 at most) | Short list with the reason each was picked |
| Written definition for each indicator | One-page indicator reference sheet per indicator |
| Data trail mapped | Source form, tally sheet, entry point, report, dashboard |
| Reported figures for the period | Figures as sent upward, per site |
| Sites picked | At least two points per indicator, one of them a weak performer |
| Records to recount selected | A sample of 10 to 20% of records, chosen before you arrive |
| Site contact informed | Date, who to be present, which registers to leave out |
| Team includes an outsider | At least one person from outside the data chain |
If an indicator has no written definition, stop and write one. You cannot judge validity against a definition that exists only in people's heads. The five data quality dimensions page lists what each dimension tests.
At the Site
Field Checklist: What to Check at Each Site
Work through this list at every site, one copy per indicator. Sixteen checks, grouped by dimension. Ask, then look, then write down what you saw rather than what you were told.
| Dimension | Check | How to check |
|---|---|---|
| Validity | Staff can state the indicator definition | Ask two people separately |
| Validity | Register columns match the definition | Hold the definition next to the register |
| Reliability | Same form version in use at every point | Compare version numbers or layouts |
| Reliability | A written procedure exists and is followed | Ask to see it, then watch a task |
| Reliability | A second person recounting gets the same figure | Have another staff member count |
| Timeliness | Last three reports arrived by the deadline | Check receipt dates, not promises |
| Timeliness | Records entered within the agreed days | Compare collection and entry dates |
| Precision | Disaggregation fields filled in | Scan 20 rows for blanks |
| Precision | Totals equal the sum of their parts | Add up the columns |
| Integrity | Source documents stored and findable | Ask for one from last quarter |
| Integrity | Changes show who made them and when | Look for initials and dates |
| Integrity | No double counting across periods or sites | Look for repeat IDs or names |
| Trace | Reported figure traced back to source | Follow it step by step |
| Trace | Sample recounted | Enter reported, recounted, concordance |
| Close | Findings shared with site staff before you leave | Hold a ten-minute debrief |
| Close | Actions have an owner and a date | Write them on the sheet |
Concordance is recounted divided by reported. The bands and how to read them sit in the how-to page; a data quality scorecard turns your dimension ratings into a summary you can share.
Red Flags and What They Usually Mean
A red flag is not a verdict. It tells you where to look next.
| What you see | What it usually means |
|---|---|
| Staff give three different definitions | Training gap, or the definition was never shared |
| Register counts more than the indicator allows | Register built for another purpose (all visits, not first visits) |
| Reports always arrive on deadline day, never earlier | Numbers are compiled at the last moment, with little review |
| Pencil corrections with no initials | Weak audit trail, an integrity risk |
| Round totals (50, 100, 200) | Estimates entered as counts |
| Perfect concordance at every site | Check whether the site chose the sample |
Treat the pattern across sites as the finding. One site with a problem points to that site; three sites with the same problem point to the system. Where staff enter or tally data in front of you, observation methods help you record what you see rather than what you are told.
Print and Use
The One-Page Printable
The field checklist above is set out on one page, with a header for program, site, indicator and period, space for notes, and a recount line at the bottom. Print one copy per indicator per site.
A Filled Example
The example below follows a fictional clinic network checking "new patients enrolled in the diabetes clinic" at one clinic. Three findings came out of it:
- The register counted every visit, but the indicator counts first visits only (validity).
- Age group was blank on 9 of 60 rows (precision).
- Corrections were made in pencil with no initials (integrity).
The recount matched the report exactly, 41 and 41. That is the point of running the whole list: the number was right, but the register would produce a wrong number the next time staff changed. Concordance alone would have missed it.
Common Mistakes
Printing the generic list and not editing it. Staff do not recognize "source document." They do recognize "the blue tally book." Name the real forms.
Ticking from conversation. If you did not look at the register, the box stays empty.
Choosing the recount sample on site. Staff will point you to the tidy pages. Pick the records in advance.
Leaving the debrief out. A finding nobody hears will not change what anyone does. Ten minutes with the site team is part of the checklist, not an extra.
Treating a passed checklist as proof of quality. It shows the system worked on the day you visited. Repeat on a schedule set in your managing data approach, and see quality assurance for ongoing checks.