What was Special about Edith Abbott's Crime Statistics Reports? Part 2
Part 1 of this series described Edith Abbott’s unusually strong academic and practical background in statistics, which gave her the experience and training to write her 1915 report Statistics Relating to Crime in Chicago. In this part, I’ll discuss some of the features that make that report, and a follow-up she published in 1922 titled Recent statistics relating to crime in Chicago, resonate with a statistician in 2019.
Abbott studied topics in crime statistics that were less common in other reports, such as homicides by police, and persons detained because they could not pay bail. What struck me about her work, however, is how modern her approach to statistics was. Anyone can calculate an average or a percentage from a data set. But the essence of the discipline of statistics is providing an assessment of the accuracy and applicability of those statistics. How valid are they, and to what populations do they apply? Abbott focused on those aspects of the statistics, and her approach to collecting and evaluating statistics presaged current topics of statistical research.
Abbott’s Approach to Statistical Investigation
Abbott assembled published statistics about crime from the police department, the municipal court system, the adult probation office, and the House of Correction (city jail). She also gathered unpublished statistics on criminal complaints — which were supposed to include crimes for which no one was arrested as well as crimes resulting in an arrest — from the police department.
She investigated the quality of the data sources by comparing the statistics from one data source to those from others. She found, for example, that some types of crime had five times as many arrests as criminal complaints, when one would expect fewer arrests than criminal complaints because for many crimes no one is arrested. She therefore concluded that the data on crimes known to the police were unreliable.
Abbott knew that even reliable statistics on criminal complaints would not provide a complete picture of crime because “much crime is undetected,” but argued that records of crime known to the police would provide better information than the numbers of arrests and convictions. An increase in the number of arrests did not necessarily mean that crime was increasing but “may merely indicate greater activity on the part of the police” or may be “merely pseudo-activity resulting in the arrest of large numbers of innocent persons.”
Because she had no trustworthy information on crimes known to the police, she was forced to rely on statistics about arrests. But she repeatedly emphasized that one could not necessarily conclude that crime had increased from an increased number of arrests:
From the statistics that have been given, it appears that there was in the year 1913 … an unmistakably large increase in the number of arrests. If the number of arrests indicates the extent of crime, then there was obviously a very marked increase in crime in the year 1913. If the figures as to the relation between arrests and population are to be trusted, the year 1913 would popularly be called a serious “crime year” that put our crime-rate back more than a decade. It is very important therefore to note that the number of arrests is not synonymous with the number of crimes … (Abbott, 1915, p. 22).
Collecting statistics on the number of crimes known to law enforcement agencies became, in the 1920s, a primary reason for launching the Uniform Crime Reporting System, which today is one of the major sources of information about crime in the United States. Despite her experience in conducting surveys of Chicago residents, Abbott did not appear to suggest what was to be the next development for measuring crime: asking people directly about victimizations (both known and unknown to the police) they have experienced. Then again, no one was using surveys for government statistics on any subject at that time so it is not surprising that Abbott did not suggest a crime survey; the United States government did not start measuring crime through surveys until 1973.
Abbott didn’t just take the data that were given to her and unquestioningly tabulate the results. She evaluated the fitness of each source of data for answering questions about crime, and if no adequate data source existed (which was usually the case), she described what kind of data collection should be done. For example, Abbott discovered that Chicago in 1915 did not keep track of prior convictions. She recommended that all persons convicted of crimes be fingerprinted and photographed (not just the persons who could not post bail) so that judges could distinguish first-time from repeat offenders.
Because of her comprehensive and multiple-data-source approach to the statistics, Abbott identified numerous areas in which police resources could be used more efficiently. She found, for example, that fewer than 15% of arrests were for felonies. The remainder were for crimes such as disorderly conduct and petty offenses. At the same time, more than 60% of the felony charges were discharged in preliminary hearings. Abbott concluded that either a large number of innocent people are arrested, or “a large number of persons who are legitimately arrested and who should be convicted are being released because of some defect in our prosecuting machinery. Whether this defect is to be attributed to the police, the courts, the Grand Jury, or the State’s Attorney’s office, is not within the province of this discussion” (p. 31). She also reported that more than 80% of the commitments to the House of Correction were for non-payment of fines: Chicago was paying 46 cents per day to incarcerate persons who typically owed less than 20 dollars. Abbott suggested it would be more cost-effective to allow persons owing small fines to pay by installment while on probation. The City Council Committee concluded from Abbott’s statistics that the “present machinery catches poor, petty and occasional criminals, and punishes them severely, but fails signally to suppress the professional criminal.”
Statistical Reasoning Principles in Abbott’s Crime Statistics Reports
For each statistical table, Abbott carefully documented (1) where the data came from, (2) how the statistics were calculated, (3) how accurate those statistics were likely to be, and (4) how the system could be changed to give better statistics.
The statistical issues and principles she addressed continue to be relevant today.
Counting rules and definitions matter. Abbott compared Chicago’s felony arrest rates to those of New York and London, but said that comparisons for many crimes were not valid because the definitions of those crimes differed. She asked “When is a murder a murder?”
Crime definitions and counting rules differ among sources of crime data in the United States (the FBI’s Uniform Crime Reports, local police statistics, homicide information from death certificates, data from surveys such as the National Crime Victimization Survey). All of these differ from definitions used in other countries and those proposed by the United Nations. This is a major challenge for comparisons.
Always give the source of a statistic. How many times have you read a statistic in a newspaper or on a website, and wondered “Where did that come from?” So did Abbott. She commented in her 1922 report on the murder statistics reported by the Chicago “Crime Commission,” (Abbott made her opinion of the organization clear by putting its name in quotes), which had claimed that its activities were responsible for a decrease in crime from 1919 to 1920. But Abbott noted that the murder statistics cited by the Commision were wildly out of line with those from the police department and wrote: “The source of these extraordinary statistics is not given by the ‘Commission.’”
Multiple data sources are needed to study crime. In addition to using the multiple data sources to evaluate the quality of different statistics, Abbott wove the sources together to form a mosaic picture of crime in Chicago. Combining information from different data sources is a focus for statistical research in 2019. More data about more topics are available than ever before, but some data sources are more reliable for a particular purpose than others.
Use an appropriate statistic to answer a question. This seems obvious, but choosing an appropriate statistic can be the most challenging part of an investigation, and often a “convenient” statistic does not answer the question of interest.
One of the “hot topics” for research in the 1900s was the relationship between immigration and crime. Newspaper reports often ascribed “crime waves” to recent immigrant groups in a city. Volume 36 of the 1911 Congressional Dillingham Commission report on immigration asked: “Is the volume of crime in the United States augmented by the presence among us of the immigrant and his offspring?” Chapter 9 of this report presented statistics from 1905-1908 Chicago police data on the percentage of arrests for different types of offenses by nativity and within each nationality group. They reported, for example, that 4.6% of all arrests of native-born persons were for violent offenses, and that 7.4% of all arrests of foreign-born persons were for violent offenses. But these statistics, even if accurate and even if arrests were proportional to the number of crimes committed, did not answer the question posed by the commission. The foreign-born might have had a higher percentage of arrests for violent offenses and yet lower rates of crime for all crime categories. It depends on how many native- and foreign-born persons were living in Chicago at the time.
Abbott compared the percentage of arrests and convictions for native- and foreign-born men from the Chicago police data with the population percentages from the 1910 U.S. census. By relying on multiple data sources, she was able to calculate that while the foreign-born accounted for 54% of Chicago men age 21 and over (according to the census), they accounted for only 36% of the arrests and 35% of the convictions, and concluded that the “various foreign groups show almost uniformly a smaller percentage of convictions than their proportion of the population entitles them to have.”
Although Abbott would have known about the “advanced” statistical methods of regression and correlation, she did not use them in her report. And, for the most part, she did not need to. Her statistical tables gave the requisite information and were easy for her readers to understand. There is no reason to use a complicated analysis when a simpler one will answer the questions.
Missing data and measurement errors affect all statistics, even those from a census. Abbott documented deficiencies of the data she had obtained, and all of her conclusions were conditional on the quality of the available data. Her conclusions about immigration, for example, acknowledged the poor quality of the police department’s information on nativity and nationality. She wrote: “When the police are asked, ‘What nationality is the prisoner?’ in order that the right kind of interpreter may be sent, the answer is likely to be, ‘We don't know what nationality he is. He can't speak anything that anyone here can understand.’ It is safe to say that in such cases the ‘nationality’ of the immigrant is not likely to get into the record correctly.”
In other statistical work, Abbott stated that her conclusions depended on assumptions made about the nature of missing data. In Women in Industry, for example, she wrote that even though many establishments failed to report the number of women employees, one could still estimate the percentage of women among employees by assuming that the percentage of women is “much the same in the establishments that do not report as in those that do.” In fact, she argued that estimates of the percentage of employees that are women were likely to be too low because the employees of companies that failed to provide breakdowns by gender would “invariably be entered as ‘men employed’.”
The Uniform Crime Reports have missing data (from nonreporting law enforcement agencies as well as from crimes that are not reported to or recorded by the police) and measurement error (from misclassifying types of crime or misrecording characteristics of victims or offenders). How much do these affect the statistics?
Collect data in such a way that the analysis will be easy and clear. Throughout her report, Abbott suggested improvements for the data collection methods. If you have good procedures for collecting data, you often don’t need hugely complicated statistical methods to obtain the results. She also, by the example of her crime report and all her statistical work, emphasized that each step of the process needs to be transparent and defensible. She presented the tables of statistics, told where they came from and how she did the calculations, interpreted them, and then wove them into a memorable narrative.
Next: Why isn’t Abbott better known today for her statistical contributions?
Copyright (c) 2019 Sharon L. Lohr