Apart from how others point out this is a study on toilets, not rape relief centres, or sports, or prisons, or single sex wards in hospitals, nor short lists or awards etc etc etc, their own list of limitations (my italics) include;
Limitations of this study include issues inherent with the data source. For example, the data used to represent safety and privacy violations in public restrooms were police records of criminal incidents. While these records should have a relatively high level of reliability in their objective accuracy in recording the existence of such incidents, they fail to include any incidents that were not reported to local law enforcement. For example, it is estimated that only 30 to 35% of rapes and sexual assaults are reported to the police (Truman & Langton, 2014). Nevertheless, by assessing trends over time and using a matched pairs analysis, the authors sought to control for any issues related to unreported incidents. There is no reason to assume that incidents are more or less likely to be reported in a locality with a GIPANDO than in a matched locality.
The crime reports also were not recorded in a way that allows a reviewer to distinguish between incidents involving cisgender people and transgender people. Police departments generally do not distinguish between sex assigned at birth and gender identity. Therefore, there is no way to identify if there were any incidents that involved transgender people being attacked in public restrooms because of their externally perceived gender. A 2008 survey of 93 transgender people in the Washington, DC metropolitan area found that 9% reported experiencing physical assault in a public restroom (Herman, 2013). There was also no way to identify if there were incidents of transgender people or people pretending to be transgender accessing restrooms with intent to harm others. Among the incidents that had notes attached providing more detail, there was no evidence of transgender people being either victims or perpetrators of crimes or of people pretending to be transgender in order to harm others in public restrooms.
It is also important to note that violent and other privacy-related crimes in public restrooms, locker rooms, and changing rooms are exceedingly rare. As a point of comparison, our findings indicated that reports of privacy or safety violations in these public spaces occurred annually at most at a rate of 4.5 per 100,000 population in the jurisdictions we studied; in the Commonwealth of Massachusetts in 2015, violent crimes were reported at a rate of 390.1 per 100,000 population, and rapes were reported at a rate of 32.6 per 100,000 (Federal Bureau of Investigation, 2016). While this may be comforting to those who have safety and privacy related concerns about those spaces, the rarity of such incidents may act as a limitation to this analysis. Nevertheless, the matched pairs design was used intentionally to compensate for limited data.
The data were requested from 15 different police departments of different sizes and geographies. Each had its own individual record keeping system, policy for responding to public records requests, and records clerks. Some departments responded by sending extra data and allowing the researchers to search through to find the relevant incidents, while others sent tables with dates and criminal codes. Some appeared to have the ability to search electronically while others had to search manually. Therefore, we are unable to determine whether every single search was equally thorough and turned up every single incident that matched the researchers’ search criteria. For example, the locality that showed the highest number of restroom incidents was the locality in which the police department sent their full criminal logs to the researchers and allowed the researchers to review the records to find incidents that met their search criteria. The higher number of incidents might be more likely to indicate that the researchers performed a more detailed and exhaustive search than the other searches performed within police departments, rather than that there were actually more incidents in that locality. That locality was a matched pair locality, so this may have contributed to the greater number of incidents reported in matched pair localities, as compared to GIPANDO localities. However, the difference-in-difference approach would account for any such bias because we do not rely on the numbers of individual incidents reported for the analyses, but instead rely on the differences within jurisdictions before and after passage of GIPANDOs. We can assume that data collection efforts were consistent within each jurisdiction, and therefore, our calculations produce differences that are comparable across jurisdictions.
Finally, though all of the requests were worded and followed up upon in the same manner, the depth of the results may have varied. Three localities were unable to provide complete incident data, which may decrease the internal validity of the current study. Cases where there was missing data from a matched locality led to the exclusion of the locality with a GIPANDO from the analysis because of the lack of comparable data, which may impact the external validity of the current study.