I reviewed two articles designed to better illuminate sampling and coding bias in US crime statistics, and thought I'd share. I had always quoted the FBI UCR as gospel, and am now shamed that I had been so ready to accept granular numbers at face value. After all, doesn't 14,467 seem far more precise than 15k?
Read them for yourself, and see if you agree...
In their article "Reporting Crime to the Police, 1973-2005: a Multivariate Analysis of Long-Term Trends in the National Crime Survey (NCS) and National Crime Victimization Survey (NCVS)", Baumer & Lauritsen provide an illuminating overview of the difficulties in providing succinct statements about long-term crime trends in the United States. The authors consulted diverse data sources to provide broad attribution to changes on reporting rates of crime, from improved police/community relations and trust, 911 network improvements and pervasive mobile communications to victimization efforts and reclassification of activity into crimes (e.g. date rape). McGruff the Crime Dog, neighborhood watch programs and even America’s Most Wanted provided reminders to citizens that they should remain alert and to report crimes. However, the authors likewise pointed out negative forces in the reporting of crime, including “anti-snitch” culture shifts, changes in organized crime and decreased belief in the general population that reported incidents would lead to prosecution.
I appreciated that the authors shared the coding information, and normalization assumptions they had made, this was vital to understanding sampling bias. There were various interesting anomalies in coding that I noted where I would have liked further information about the choices made by the researchers – overall, “unknown” was glaringly missing, since offenders’ ages, race and gender are not always known, nor sometimes the victims’, so explanation of assumptions for where “unknown” was mapped would have been helpful.
Likewise, additional demographic coding for analysis was not as granular as I would have expected. Both gender and household coding were binary, apparently not providing transsexual/ambiguous/undeclared/other gender identification to be represented in the information. Marriage was also encoded as a binary value, not representing non-intimate cohabitation, intimate co-habitation or homosexual relationships in the analysis. I found these curious omissions without commentary by the authors. Likewise, the White/Black/Other racial coding was very coarse, though the footnote (6) made this fairly clear why this limitation was imposed. Further exploration of more granular demographics might prove interesting. I would have also liked to have understood why they coded a multiple-offender mixed gender incident as all-male, which seems rather odd.
The dramatic changes over the study in reporting rates for criminality were very surprising, as I had not thought that there would be much, if any, positive change. The incidence of reported crime within the family was in the direction I had thought, but was a dramatic (and rather encouraging) indicator. I must confess that my skills at reading raw statistical regression tables and drawing conclusions kept me from getting as much from this article as I would have liked, but overall it was very eye-opening nonetheless. To truly understand crime, it seems this avenue of research must be encouraged and repeated to capture changes in reporting of crimes caused by broad societal changes.
The second article, “The Use of Official Records to Measure Crime and Delinquency” by Lofton and MacDowall, provides for a stark and shocking coverage of the UCR, and I was floored by the graph showing zero (0) report of criminal homicide for Florida for so many years. What was heartening was the increase in additional sources of information, as well as the raw UCR data, which are made available for more informed evaluation. The article also details the changes in crime reporting due to development and use of the National Incident Based Reporting System (NIBRS), which I had not understood.
It seems that, as the YouTube video “Crime Trends and Measurements” by gscottpleasants indicates, a blended use of reports likely leads to a more normalized view of crime trends in the US. Analyzing the raw data as well can help identify inexplicable gaps in data. By ensuring a blended view of the UCR, UCVS and other surveys, anomalous indicators could more likely be identified to avoid dramatic error. Additionally, the data must be viewed with skepticism in other than very broad trending information. Unless data are viewed with a critical and skeptical eye, errant inputs will create invalid conclusions.
Based on the articles, and a study of the UCR graphs, it seems that crime has been in a steady decline, as a general trend (though I feel conflicted reporting that, with the known errors in the UCR). Based on the Baumer & Lauritsen article that shows a demonstrated increase in the likelihood that crimes will be reported, this gives credibility to the declining graph, since a flat crime rate with a year-over-year increase in reporting of crimes should create an elevated UCR.
I was not surprised by this information, because I had already done the UCR research (though I was ignorant of the various flaws until I read the other articles). Because I am an advocate for Concealed Carry and self-defense, and have done extensive armchair analysis of crime statistics, I was already aware of the trending information. I had also done several hours of analysis cross-correlating gun control information (e.g. Brady Campaign State ScoreCard cross-referenced with state-by-state violent crime statistics).
Please don't mistake this for arrogance, my knowledge is proof that, "In the land of the blind, the one-eyed man is king", and my rumination is of suitable rigor for coffee shop discussions. My research has not had the rigor of peer-reviewed research. Because I have engaged hundreds in conversations on these issues, and largely find advocates for gun control to be ill-informed, I was aware that the crime statistics were broadly misunderstood. However, I had ignored sampling error, bias, discretionary policing and politics in the formation of the UCR data.
No comments:
Post a Comment