UCR Classification Error

***

Advanced Research Methods

Week 3

Discussion 1

***

Nolan, Haas, and Napier focus on the issue of classification error, although they do mention the “error structure” in reporting, and summarize errors in ”missing” data (not noticed by the public, not reported by the public, not filed by the police, and errors in compilation), they reiterate that the issue of focus is the misclassification of crime that is reported to the UCR. They begin by discussing possible sources of classification error; mistakes by officers, bad report writing habits, deliberate downgrading of particular crimes to reduce the crime rate , and automation problems (Nolan, Haas, & Napier, 2011, p.500). Nolan, Haas, and Napier move on to illuminate how to measure classification error by the use of record accuracy and statistical accuracy. “Record accuracy refers to the estimate of classification error in specific crime types viewed alone”. (Nolan et al, p.500) Conversely, “Statistical accuracy refers to the accuracy of the crime totals after all crime types have been examined and offsetting misclassifications have been considered.” (Nolan et al, p.501). They discuss the mathematics involved in the methodology and provide examples of their work; specific error classifications that they found included burglary versus larceny counts, the classification of domestic violence reports as simple assault versus aggravated assault, simple assault versus aggravated assault in general, robbery, and in general found that violent crime was under counted. Finally, they found that 4.17% of the crimes reported to the UCR and included in their study were misclassified. (Nolan et al, p.517)

Two solutions to resolve these problems are provided by Nolan, Haas, and Napier .The first solution is training for LEO that would aid in reporting to“classify crimes according to UCR definitions” (Nolan et al, p.518). The second solution would be to “statistically adjust” reported crime by manipulating reported numbers “on the magnitude and variation of known error in individual crime types as well as aggregate totals.” (Nolan et al, p.518).

Although the first solution should improve accuracy in reporting based upon the efficiency and range of training and the willingness of LEO management to follow up on a continual basis, the second solution of statistical adjustment suffers from the circumstances which overall cause the UCR to suffer from potentially unreliable data.

There are many such circumstances; UCR data is based on a voluntary response sample, which itself is subject to incentive in the form of grant money; “From a statistical point of view, such a sample is fundamentally flawed and should not be used for making general statements about a larger population” (Triola, 2014, p.7). Nolan, Haas, and Napier recognize their use of nonprobabilty sampling; “Although UCR is not produced by sampling, the program is considered statistical because adjustments are made for missing or erroneous data”(Nolan et al, p.499). Another issue is the use of the hierarchy rule; “An incident where a suspect broke into a dwelling, stole property, raped and murdered the inhabitant, burned the structure to destroy evidence, and escaped in the victim's car was counted as one crime, a homicide due to the UCR's hierarchy rule.” (South University Online, 2014, para.3). The hierarchy rule itself is violated by the reporting of arson, which is always reported to the FBI. Other issues involve the failure to include child abuse as a violent crime, the limitation of rape to female victims ( and not accounting for same-sex rape), and racial classifications. Nolan, Haas, and Napier do briefly discuss the issue of false reporting by police, so that consideration was addressed. They defend their findings in that they “believe that the limited context for our study is of less importance in terms of the generalizability of the findings than it would be if we were dealing with other aspects of UCR error” Nolan et al, p.517)

Nolan, J.., Haas, S., & Napier, J. (2011). Estimating the impact of classification error on the "statistical accuracy" of Uniform Crime Reports. Journal of Quantitative Criminology, 27(4), 497–519. Retrieved July 16, 2014 from http://search.proquest.com.southuniversity.libproxy.edmc.edu/docview/901188160

South University Online. (2014). MCJ5100 : Advanced Research Methods and Analysis I : Week 3: Week 3 - The UCR. Retrieved July 16, 2014 from myeclassonline.com

Triola, M. (2014). Elementary Statistics, 12^th ed. Pearson. 

***

Followup

***

I might have to reread the study, but I think you hit a point that the Nolan gang missed, which would be the potential difference in defining crimes uniformly across different jurisdictions.

After reading the Va. legal code, I go back to wondering if the purpose of the law is not to protect the citizenry from each other but to generate income for lawyers... 

***

You have to dance with the one you brought.  If all the data you have is the UCR, then that's what you have to work with.  In the text, they dicuss the concept that you don't always have access to full data, and have to work with what's available.  Which is why you see the researchers in this week's assignments discuss "statistical adjustment" and "imputation" of data.

The problems with the UCR are well-known, and could be fixed, but then you'd have to deal with the accuracy of historical comparisons because your standards have been changed.

There is also the NIBRS

***

Nolan et al briefly touch on the "dark figure of crime" (although they don't refer to it as such) in their discussion of the "error structure" of crime reporting.  Most of their discussion was focused on the misclassification of crime in the UCR. .They summarize the ”missing” data by category (not noticed by the public, not reported by the public, not filed by the police, and errors in compilation)

***