Data Core · Data Collection · Module

Data quality

Whether the data is complete, accurate, timely and consistent — measured rather than assumed, since the assumption is always that it is fine.

The idea

How it works

Data quality is discovered rather than measured in most organisations: someone acts on a number, the result is wrong, and an investigation finds the cause. Measuring it in advance is considerably cheaper.

Completeness is the dimension that most often fails and the easiest to check. A field populated in sixty per cent of records is producing analysis about the sixty per cent.

Working with it

In practice

  1. 01

    Measure completeness per field

    What proportion of records have it. The result is routinely worse than assumed and takes minutes to obtain.

  2. 02

    Check accuracy against reality

    Sample records and verify them against something external. Internal consistency is not accuracy.

  3. 03

    Fix at source, not in the report

    A cleaning step in a report fixes one output. Fixing collection fixes all of them.

  4. 04

    Publish the quality figures

    A number reported alongside its completeness is used more carefully than one reported alone.

One level in

The components of data quality

A component is something that exists afterwards which did not exist before — a deliverable or a mechanism, not an intention.

  1. The measures

    Completeness, accuracy, timeliness and consistency, per data set.

    Learn
  2. The causes

    Where quality problems originate, which is usually one step in collection.

    Learn
  3. The fixes

    What was changed at source, rather than corrected downstream.

    Learn

Report a figure alongside its completeness. A number known to rest on sixty per cent of records is used more carefully.