lucasmaystre/crime-comparisons

A machine-readable version of comparison data in Thurstone's 1927 paper.

Crime comparisons dataset

This repository provides the data used in Louis L. Thurstone's seminal 1927 paper, "The Method of Paired Comparisons for Social Values" (link to paywall or open-access HTML version) in machine-readable format.

Thurstone's study aimed at showing that pairwise-comparison models could be used to investigate interesting questions beyond psychophysics. To that end, he set out to create a "crowdsourced" scale of the seriousness of criminal offences by asking 266 students questions of the type:

Which crime is worse: A or B ?

where A and B are chosen among a list of 19 criminal offenses. Using his "Case V" model of pairwise-comparison outcomes (see this excellent tutorial for more information), he was able to produce a scale that looks roughly as follows.

How to analyze the data

If you are using Python, I recommend that you try out choix, a library that I've developed to learn models based on comparion data. You can see an example in this notebook.

Errors in the original data

The contents of ccdata-original.csv faithfully reproduce the data in Table 1 of Thurstone's paper (page 399). However, upon close investigation, it seems that there are a number of small mistakes and typos in this table.

• "Assault and Battery" vs. "Adultery": ratios sum only to 0.999
• "Assault and Battery" vs. "Homicide": ratios sum to 0.100
• "Arson" vs. "Homicide": ratios sum to 0.100
• "Arson" vs. "Libel: ratios sum to 0.999
• "Perjury" vs. "Counterfeiting": ratios sum to 0.999

I decided to fix those (according to my own subjective interpretation of what should have been there), and the result is given in ccdata-fixed.csv.

The labels of the crimes corresponding to the rows and columns of the table are provided in labels.txt.