Tom K. WongVerified account

I’ve been looking through the “Undocumented immigrants, U.S. Citizens, and Convicted Criminals in Arizona” report, as well as the underlying data I received from the Arizona Department of Corrections (ADOC), and there’s reason to be concerned about the report’s findings…

In Table 1, they’ve counted each row in the data as a unique individual. This is incorrect, as this means that the same person is counted multiple times (one can confirm this by a simple tabulation by race without removing duplicates)…

And looking at the number of times each “IDNO” number appears by “commit_number_for_inmate” shows that ADOC is cleanly counting individuals in the data who they encounter multiple times, however…

The “citizen” data are less clean. Per the codebook: “1=US CITIZEN, 2=non-US citizen 3=non-US citizen, deportable 4=error, should be recoded into a 3 5= NON-CITIZEN, not deportable 6=Criminal Alien Deportable Status Uncertain 9) UNKNOWN”…

Based on these categories, there’s a lot of noise in the report's “Undocumented immigrants: Non-US citizen, Not Legal Permanent Resident” and “Documented immigrants: Non-US citizen, Legal Permanent Resident” categories. @AlexNowrasteh describes the implications of this noise

But if we take the study on its face, use the same coding as the report does to assume citizen v. likely documented v. likely undocumented (which is easily replicable based on the report’s assumptions), and address duplicates, key findings in the report disappear…

For example, likely undocumented immigrants are less likely to be first degree murderers than are citizens per the ADOC data, which is just the opposite of what the report claims...

As another example, likely undocumented immigrants are less likely to be gang members than are citizens per the ADOC data, also contra the report...