Ggovire
GlossaryData

Entity resolution

2 min read
Short answer
Entity resolution is the process of deciding that records from different sources describe the same property or the same person. In property data it means matching across county parcel numbers, address formats, and owner names that appear differently in every source. It is the step that makes multi-signal analysis possible and the step most often done badly.

Entity resolution is deciding that two records refer to the same thing. In property data it is the difference between a pile of unconnected filings and a description of what is happening to a property.

The property side#

Parcel numbers are not enough. Minnesota counties assign them independently. The same number exists in multiple counties, so only the combination of county and PIN identifies a parcel.

Formats vary. Dashes, leading zeros, section-township-range encodings, sequential numbering. The same county's own systems may present the same number differently.

Numbers get reassigned. A split or combination retires an identifier and creates new ones, breaking the link between a property's history and its current record.

Addresses are worse. Abbreviations, directionals, unit designations, addresses that changed when a street was renumbered, and vacant parcels with no address at all.

The owner side#

Harder still.

The same person appears as Robert J Anderson, Bob Anderson, Anderson, Robert, and Robert Anderson and Mary Anderson across four sources.

Entities multiply the problem. Smith Properties LLC, Smith Properties, L.L.C. and Smith Props LLC may be the same registration or three different ones, and the filings will not say.

Trusts, estates and personal representatives introduce their own naming conventions. And an owner deliberately holding property across several entities is doing something entirely lawful that looks, in the data, like several unrelated owners.

The two failure modes#

Under-matching. Records that belong together stay apart. The picture is incomplete, which is bad and at least visible — you can tell you are missing things.

Over-matching. Records that do not belong together get merged. The system produces a confident, coherent description of a property or an owner that does not exist as described.

The second is far more dangerous, because nothing announces it. A join that succeeds looks the same whether it was right or wrong.

Which is why the default should be caution#

Where a match is uncertain, leaving records separate produces an incomplete picture. Merging them produces a wrong one.

An incomplete picture can be improved. A wrong one gets acted on.

Why it matters beyond the technique#

Every claim about a property's situation depends on this step having been done correctly.

A statement that a parcel has a foreclosure, three code violations and a tax delinquency is only true if all four records genuinely describe the same parcel. Get the resolution wrong and every downstream number — rates, outcomes, predictions — is measuring something that is not there.

Common questions

Why can't you just match on the parcel number?
Because Minnesota parcel numbers are not unique across counties, formats differ by county, and numbers are reassigned when parcels split or combine. Matching on PIN alone silently merges unrelated properties in different counties.
Why are owner names so hard to match?
Because sources record them differently — with and without middle initials, married and maiden names, entity suffixes abbreviated inconsistently, and trusts and LLCs named in ways that vary between filings. Common names compound all of it.
What happens when it goes wrong?
Two kinds of error. Records that belong together stay separate, so the picture is incomplete. Or records that do not belong together get merged, producing a confident description of a property that does not exist as described. The second is worse because nothing flags it.
Keep reading