Wang, Chen, and Atabakhsh suggest an algorithmic approach to revealing criminals’ use of deceptive identities, through analysis of the intentional discrepancies in personal data (name, address, social security number, and so on) criminals use to avoid the linking of their transgressions to their true identities. Unfortunately, the paper does not provide a compelling argument that the proposed algorithm would be effective in any practical application.

First, it is difficult to apply statistical analyses in support of or against a hypothesis without clearly defining the hypothesis being examined. Second, before performing any analysis against a data set, the selection process used in generating the data must be understood. This paper does not disclose whether the 1.3 million records used in the original case study were drawn from a random population or a criminal population (although a selection bias toward the latter seems likely, since the data was provided by the Tucson Police Department). While a comparison of the statistical incidence of errors in criminal record systems with the incidence of errors in a law-abiding population would be interesting, we cannot make such a comparison without additional information on the source data.

In the absence of such information, any conclusions drawn must be suspect. This is compounded by the use of a very small sample to test the authors’ unstated hypothesis (140 record sets were used, all drawn from only 44 known criminals). Two-thirds of the records were used to “train” the algorithm, leaving only 40 records to validate the accuracy of the algorithm.

There is also room for further research within the stated subject area. The authors cite research in criminal profile analysis that includes only one reference on criminal justice data analysis, and fail to cite prominent texts or academic papers on criminal behavior, criminal profiling, or lie detection. The paper seems to focus more on examining errors prevalent in criminal record systems, without differentiating between intentionally deceptive errors and innocent mistakes in data entry. In fact, no clear way to draw that distinction is demonstrated within the paper.

Given the sensitivity of associating an individual with criminal behavior, it would seem prudent for the authors to focus their current research on record linkage toward detecting accidental errors in public and private databases, rather than to claim an algorithmic approach to detecting criminals through demographic data, without having a method to distinguish simple input errors from fraudulent or deceptive identification. This paper seems overly optimistic.