Statistics in Criminological Research (Unit VI)

Criminological questions are either descriptive (how much crime, of what kind, where) or explanatory (does a given factor raise or lower offending). Descriptive statistics answer the first kind; tests of association, regression and time-series methods answer the second.

Which regression family applies depends mechanically on the outcome variable: a continuous outcome such as sentence length calls for linear regression, a binary outcome such as reconvicted or not calls for logistic regression, and a count outcome such as burglaries per ward per month needs its own count models, covered further below.

A statistical claim is only as strong as the research design behind it. An association shown by a regression on cross-sectional data does not establish which variable came first, or whether an unmeasured third factor produced both.

The level at which data is collected also matters: a per-offender dataset supports individual-level modelling, while an aggregate dataset such as a monthly city count supports only area-level or time-series analysis, and treating the two as interchangeable is the ecological fallacy covered in section 4.

The most basic descriptive tool is the crime rate, the count of a given offence divided by the population at risk, conventionally per 100,000 people. Standardising by population lets researchers compare places or years of different sizes on a common footing, which is why agencies such as India's National Crime Records Bureau, the FBI's Uniform Crime Reporting programme, and the Crime Survey for England and Wales all publish rates rather than raw counts.

Beyond a simple rate, a trend index rebases a rate to 100 in a chosen base year so later years read as a percentage change, stripping out the scale of the original numbers to show direction and pace of change.

Every descriptive measure built this way inherits the limits of the underlying count: a rise in a recorded rate can reflect more offending, or it can reflect a change in reporting or recording practice, the gap known as the dark figure of crime. An age-standardised rate is used when the groups being compared have different age structures, since offending is not spread evenly across age groups.

When two variables are categorical, such as offence type and area type, the first step is a cross-tabulation, and the chi-square test checks whether the categories are associated more than chance would predict. Building that table and judging sample size are inference fundamentals shared across quantitative research generally, not specific to crime data.

What is specific to crime research is picking the regression family once the design moves beyond a single two-way table. A continuous outcome such as sentence length calls for linear regression. A binary outcome such as whether a released prisoner is reconvicted within two years calls for logistic regression, whose coefficients are read as odds ratios rather than a raw change in the outcome. A crime count fits neither family and needs its own count models, covered next.

Many crime outcomes are counts, such as burglaries per ward per month, and Poisson regression is the starting model, assuming the variance of the counts equals their mean. Crime counts are rarely collected over identical windows, so an offset, commonly the log of the population, converts predictions into a rate per unit of exposure.

The Poisson assumption is frequently wrong for crime data, a pattern called overdispersion, and negative binomial regression adds a dispersion parameter to produce wider, more honest standard errors.

Time-series analysis studies a sequence of measurements taken at regular intervals, such as a monthly robbery count, for trend and seasonality. Criminologists reach for it to evaluate an intervention: the interrupted time-series design compares the trend before an intervention date against the trend after it, looking for a change in level and a change in slope. Its weakness is that any other event coinciding with the intervention date can be mistaken for an effect of that intervention unless ruled out.

Meta-analysis combines the results of many separate studies of the same question into a single, more precise estimate, typically weighting each study by its sample size or precision.

Its main hazard is publication bias, since studies finding a significant effect are more likely to be published than studies finding none, which analysts check for with a funnel plot. Two errors recur across all of this: the ecological fallacy, where a group-level correlation is wrongly assumed to hold for individuals, and spurious correlation, where two variables move together without either causing the other.