The 2020 Census is less accurate for small rural areas and minority populations than anyone expected, according to researchers who analyzed the published data.

Here is what happened. The Census Bureau added random noise — fake variation — to population counts before releasing the data. The goal was to prevent anyone from using the published numbers to figure out who specific people were. This is called differential privacy.

The problem is straightforward: random noise barely matters when you have a large number to begin with. Imagine you have 800,000 people in a county and add a small amount of noise — the percentage of error is tiny. But if you have 400 people in a small township and add the same amount of noise, the error becomes huge. The privacy protection works, but it creates counting errors in exactly the places where the count matters most for those communities.

Why does this matter. Census data determines how many federal dollars your town gets, how districts are drawn for elections, and what researchers use when they study health, poverty, or other issues. If the starting numbers are wrong, everything built on top of those numbers gets wrong too. And because most people do not see the noise that was added, they do not know the data is less reliable.

The unfairness here is worth naming. Small, scattered communities — often rural or minority populations — are the ones hit hardest. These are already the groups with less voice in policy discussions. Now they also have less accurate data about themselves in the official record.

The Census Bureau had to make a choice: stronger privacy or more accurate counts. They chose stronger privacy because researchers had shown that older Census methods could be broken — that people's individual data could be reconstructed using computer tricks. So the Bureau said, we need more protection.

But choosing stronger privacy meant choosing less accurate counts, especially in small populations. This is a real engineering problem with no perfect answer. The question now is whether the Bureau can do better by 2030 — either by adjusting their approach to privacy, or by using different techniques entirely.

Right now, anyone using 2020 Census data for small areas needs to know that these numbers carry error from the privacy process. Researchers are starting to document this, so people understand what they are working with.