Types and Purposes of Criminological Research

Criminology draws on sociology, law, psychology and statistics, and each of those parent disciplines contributes a different way of asking questions about crime. A single research design cannot serve a field that needs to map new phenomena, such as an emerging form of online fraud; count established ones, such as the annual rate of burglary; test why a pattern exists, such as whether unemployment predicts property crime; test whether an intervention works, such as whether a policing tactic reduces repeat victimisation; and interpret what a statute actually requires, such as what conduct a fraud provision covers.

Treating these as one undifferentiated activity called 'crime research' produces studies that answer a narrower question than the one asked, or claim more than their design supports.

The five-way split into exploratory, descriptive, explanatory, experimental and doctrinal research is a working typology, not a rigid taxonomy with sharp edges. A single funded project routinely moves through several types in sequence: an exploratory phase of unstructured interviews with police officers to generate hypotheses about a new offence pattern, a descriptive survey to establish how widespread that pattern is, and an explanatory or experimental test of a proposed response.

Doctrinal analysis of the relevant statute often runs in parallel, because a study cannot describe how an offence is charged without first establishing what the offence legally requires.

The typology also tracks a second, cross-cutting distinction: purpose. A study can be exploratory in design and basic in purpose (testing a theory of victimisation with no policy client in mind), or descriptive in design and applied in purpose (an agency commissions a profile of its own caseload to plan resources).

Design answers the question 'what kind of question is this', while purpose answers 'who is this study for and what will be done with the answer'. Sections 2 through 6 below take the five designs in turn; section 7 returns to purpose.

Exploratory research is used when so little is known about a topic that a structured survey or an experiment would be premature. It is common at the frontier of a field, when a new offence type appears, such as an early study of a novel online scam, or when a well-known offence surfaces in a setting nobody has examined, such as victimisation inside a specific closed institution.

The typical methods are unstructured or semi-structured interviews, small purposive samples, case studies of a handful of incidents, and open review of existing records. The output of an exploratory study is not a tested claim but a set of working hypotheses, a vocabulary for the phenomenon, and a sense of which questions a later, larger study should ask.

Because the sample is small and not randomly drawn, findings do not generalise to a wider population and researchers are careful not to present them as if they did.

Descriptive research takes a phenomenon that is already reasonably well defined and profiles its extent and pattern. It answers who, what, where and how much, using structured instruments applied to a large or representative sample: victimisation surveys, official crime counts, offender demographic breakdowns, or repeated cross-sections that track a rate over time.

The United States' National Crime Victimization Survey and comparable victimisation surveys run in other countries are classic descriptive instruments, because they exist to establish the scale and distribution of victimisation rather than to test why any one respondent was victimised.

A well-designed descriptive study reports confidence intervals or margins of error on its counts, and it distinguishes a genuine change in a rate from a change caused by a shift in how a category is recorded, a recurring problem in cross-year comparisons of official statistics.

The two designs are often chained. An exploratory phase of interviews with people who use a service, or with practitioners handling a new complaint type, generates a working typology of the problem; a descriptive survey then measures how common each type in that typology is across a wider population. The chaining matters because a descriptive instrument built without a prior exploratory phase risks asking the wrong questions, forcing respondents into categories that do not match how the phenomenon actually presents.

Explanatory research, sometimes called analytical research, moves past description to test a proposed relationship between variables. Where a descriptive study reports that burglary rates are higher in a set of neighbourhoods, an explanatory study asks why, and tests a specific candidate explanation, for example whether residential turnover, unemployment or the physical layout of housing predicts the difference.

The dominant method is statistical: regression models that estimate the association between a predictor and an outcome while holding other known predictors constant, longitudinal designs that track the same individuals or areas over time to establish sequence, or structured comparison of groups that differ on the variable of interest.

The core discipline explanatory research imposes is the separation of correlation from causation. Two variables can move together for reasons that have nothing to do with one causing the other: a third factor may drive both, the direction of causation may run the opposite way to the one assumed, or the association may be a statistical artefact of how the sample was drawn.

Robert Merton's strain theory and Edwin Sutherland's differential association theory are both explanatory frameworks in this sense: each proposes a specific mechanism (blocked access to legitimate goals, or exposure to definitions favourable to law-breaking) and generates a testable prediction about who should offend more, which subsequent explanatory studies have supported, qualified or contested using survey and longitudinal data.

Explanatory research using observational data, meaning data collected without the researcher assigning who receives which condition, can never fully rule out an unmeasured confounding variable. A study can show that a proposed cause and an outcome are statistically associated even after controlling for several alternative explanations, but it cannot prove the exhaustive list of alternatives is complete.

This is the limitation that experimental research, covered next, is specifically built to overcome, at the cost of being harder and often more expensive to run in a real-world criminal justice setting.

An experimental design isolates the causal effect of one variable by manipulating it directly and comparing an outcome across a treatment group that receives the manipulation and a control group that does not.

The defining feature is random assignment: when who receives the treatment is decided by a random process rather than by choice, circumstance or need, the treatment and control groups should be statistically equivalent on every other characteristic, known or unknown, before the intervention starts. Any later difference in outcome between the groups can then be attributed to the treatment rather than to a pre-existing difference between the people or places that received it.

The Minneapolis Domestic Violence Experiment, led by Lawrence Sherman and Richard Berk in the early 1980s, is one of the most cited randomised experiments in criminology: police officers responding to misdemeanour domestic violence calls were randomly assigned to arrest, separate, or mediate, and arrest was associated with a lower rate of repeat offending over the following six months in that trial.

The finding briefly reshaped arrest policy in several US jurisdictions before a set of replication experiments in other cities produced mixed results, illustrating a second discipline experimental research imposes: a single trial, however well designed, establishes an effect in one setting and one period, and a policy conclusion needs replication across settings before it can be generalised.

True random assignment is often impossible or unethical in criminal justice research, for example a court cannot randomly assign which convicted defendants receive a harsher sentence purely for research purposes.

Quasi-experimental designs respond to this constraint: they compare groups that differ on the variable of interest for reasons outside the researcher's control, such as a policy change that took effect in one jurisdiction but not a comparable neighbouring one, and use statistical matching or a difference-in-differences approach to approximate what a randomised comparison would have shown. Quasi-experiments trade some causal certainty for feasibility and remain the most common experimental-family design in applied criminal justice evaluation.

Doctrinal research is not empirical in the sense the previous four types are. It does not collect data about behaviour; it analyses primary legal sources, statutes, delegated legislation and the judgments that interpret them, to state what the law currently requires and how courts have applied it to specific facts.

The method is close textual reading and structured legal reasoning: identifying the elements a provision requires, tracing how appellate courts have construed an ambiguous term, and reconciling apparently conflicting judgments into a coherent statement of the current rule. Secondary sources, treatises, law commission reports and academic commentary, support this reading but do not replace the primary text.

Criminology needs doctrinal research because empirical study of an offence is only meaningful once the offence's legal boundaries are established. A descriptive count of 'fraud cases' is uninterpretable without a doctrinal account of what conduct the governing statute actually treats as fraud, and that account can change when the statute itself changes.

India replaced its 1860-era criminal code with the Bharatiya Nyaya Sanhita, 2023, alongside a new code of criminal procedure and evidence act, all of which took effect from 1 July 2024; a doctrinal study of an Indian offence written before that date has to be read against the predecessor provision it analysed, and a study written after it has to cite the current section, not the repealed one.

The same discipline applies to any jurisdiction: doctrinal work has a currency date, and citing a superseded provision as if it were still in force is treated as a basic research error, not a minor slip.

Doctrinal and empirical research are complementary rather than competing. A study of sentencing disparity is explanatory and empirical when it tests whether two defendants with similar records received different sentences for reasons unrelated to the offence, but it is doctrinal when it asks whether the sentencing statute gives the judge lawful discretion to reach either result in the first place.

Serious law-and-society scholarship typically needs both: a doctrinal account of what the law permits and an empirical account of how that discretion is actually exercised in practice, which is precisely the combination the wider comparison between criminology, law and forensic science relies on.

A second, cross-cutting classification sorts studies by purpose rather than by design. Basic research (sometimes called pure research) tests, refines or extends a theoretical claim without an immediate practical client in mind: a study testing whether Travis Hirschi's social control theory predicts self-reported delinquency in a new population is basic research even though it uses a descriptive survey and an explanatory statistical model.

Applied research, by contrast, is commissioned or motivated by a practical question a specific agency or policymaker needs answered, such as which of two patrol strategies a police department should adopt next year. The same statistical techniques can serve either purpose; what differs is the audience and the immediate use the answer is put to.

Evaluation research is a specialised, and in criminal justice the most common, form of applied research. It measures whether an existing programme, policy or intervention achieved the outcome it was designed to achieve, typically by comparing conditions before and after the programme started, or by comparing a group that received it with a comparable group that did not.

Programme evaluation borrows its design directly from the experimental and quasi-experimental toolkit described in section 4: a well-resourced evaluation randomly assigns eligible participants to the programme or a waiting-list control, while a lower-resourced one relies on a matched comparison group and accepts a weaker causal claim in exchange for feasibility.

Government funders in the United States, the United Kingdom, India and Australia increasingly require an evaluation component be built into a new criminal justice programme before it is scaled up nationally, precisely because a programme that looks promising in a pilot description can fail to show any effect once evaluated against a genuine comparison group.

The basic and applied distinction is about motive, not rigour: an applied evaluation can be methodologically rigorous, and a basic study can be poorly designed.

Students new to the field sometimes assume 'applied' means less careful, which is backwards; because an applied evaluation's conclusion is usually acted on directly, funders generally hold it to a higher, not lower, standard of design than a purely theoretical basic study whose conclusion will be debated in the academic literature before anyone acts on it.