How do you measure data accuracy? Follow

Our primary form of accuracy measurements is via call downs in which we use operators to call a sample of our data records to verify that the information is correct. For example, they would call 206-555-1234 to verify that John Smith is linked to that phone number and lives at 456 Main Street. We do this on statistically significant samples so that we can take the learnings from these call downs and apply changes to our entity resolution algorithms to improve the accuracy of our links.

Of course, there is certain data that cannot be measured with call downs, such as an email/name link. In these situations, we will use a variety of other mechanisms to ensure we are producing accurate data. Examples include using truth sets or double or triple corroboration from other sources.