Course · Intermediate · Data Preparation
Data Quality Assessment
Run a routine data quality assessment the way a donor auditor will — verification factors, consistency over time, and a report that names the fix as well as the fault.
What you will be able to do
- Score a dataset against the five quality dimensions with a measure for each, rather than calling it "good" or "poor"
- Compute reporting completeness and timeliness as indicators in their own right, with their own denominators
- Recount a source register against what was reported upward and turn the difference into a verification factor with a tolerance band
- Choose which facilities to verify when you cannot visit them all, and state what your sample can and cannot conclude
- Detect digit preference, heaping and implausibly stable series, and calibrate the check against what chance produces
- Write a DQA report whose every finding carries a root cause, a corrective action, an owner and a deadline
Standards and methodologies
Everyone in this sector has been in the meeting where somebody says the data quality is poor. It is never a useful sentence. Poor how — incomplete, late, inconsistent with the register, or internally contradictory? Poor everywhere, or in four facilities out of thirty-eight? Poor enough to change the decision, or poor in a way that moves the number by half a point?
A data quality assessment is the discipline that replaces that sentence with numbers. It has a shape the sector has agreed on: five dimensions, each with a measure; a recount against the source; a sample you can defend; and a report where every finding has an owner and a date. Donors run it on you, and running it on yourself first is the difference between a finding and a surprise.
This course completes the module. Data Cleaning and Validation taught you to find and fix defects in a file. Joining and Reshaping taught you to put files together and reconcile them. This one turns both into a routine somebody else can run on a schedule — and into a document that changes what happens next quarter, which is the only reason to do any of it.
Every technique is shown in Python and in R. The worked examples run against the DHIS2-shaped vaccination extract, where 642 of 2,736 facility-months carry no report at all, and against the SMART survey, where one measurement team’s heights end in .0 or .5 sixty-nine percent of the time.
One warning about tone, because it decides whether this work is useful. A DQA that reads as an accusation produces defensive reporting, and defensive reporting is worse data. The finding is about a system, not about a person, and the course keeps returning to how that is written.
Practice
Reading the lessons is not the same as having done the work. Each of these applies the course to a dataset it did not teach on.
- Lab · 120 minThe defect that was not the problemRun a five-dimension assessment on a food security survey, quantify what every defect costs the headline, and discover that the largest single driver of the published figure is a threshold choice nobody wrote down.
- Exercise · 45 minTriage a draft DQA reportFive findings from a colleague's draft. Two are real, two are chance dressed as evidence, and one is a district failure written up as twenty-seven facility failures. Test each against the data and rewrite the survivors.
Progress
Take it offline
The whole course as a typeset PDF — every lesson, every code example, the data dictionary and the indicator definitions. Generated from the same source as this page.
The LaTeX source ships alongside each PDF, so an organisation can rebrand the handout or fold a lesson into its own training pack.