The Census Bureau issued on August 18 a preliminary study of non-citizen voting. It said that at least 24,000 persons voted in the November 2020 elections who were not American citizens. This initial finding will evaporate once the Bureau begins to acknowledge how extremely small error rates in data systems more than explain this.  And this allegation will go down in flames like state level searches for illegal voting.

The reality Trump is dealing with is that there are more poeple who want to expose themselves in Grand Central Station than non-citizens who vote.

Early on, the second Trump administration began to document what it viewed as a high rate of non-citizen voting.

I have discussed in the past efforts by numerous states to estimate and verify the extent of voting by non-citizens. These efforts tend to start with comparing driver license declarations of citizenship status to voting records. At this step, a small number of cases are found. Then, when a ballot by ballot review is performed, the number of non-citizens voting shrinks to a vanishing small number.  Perhaps, one out of every 10,000 votes. Here is a summary of seven states. Before taking into account error rates, the Census Bureau is estimating one out of 5,300 voters.

The credibility of the Census Bureau’s study rests to a large extent on how it addresses the risks of errors in data bases. In sum, the study does not do that — yet.

The 24,000 figure arises out of comparing federal databases with voting records collected by a private firm, DataClear.  The Bureau says it was able to match its citizenship status data with 128 million DataClear records. This comes to a rate of non-citizen voting to two hundredths of one percent.

What are the error rates embedded in the databases?  The error rates comparing a name on the voting record, say Carlos Hernández, with a similar name on the federal databases?  The error rates in voting roles is almost certainly much higher.  The Bureau’s study did not acknowledge that it studied error rates.

Say several steps in voting are affected by an error rate of, say, one hundredth of one percent – 0.01%.  The person named Carlos Hernandez is confused with another at the polling booth, or in final edited voting files. (Note this exhaustive and exhausting data verification document from Colorado.)An extremely small transcription error by DataClear.  As in the state level search for non-citizen voting, the vast share of initially positive hits are explained by error rates.  In the Texas case, a citizen working in a local election department was initially flagged as having voted illegally.