Quantitative and Qualitative Research Methods in Criminology

The split between quantitative and qualitative research in criminology traces back to a much older argument in the social sciences about what counts as knowledge. Auguste Comte argued in the 1830s that society could be studied with the same logic used in physics or chemistry, through observation, measurement and the search for general laws, a position now called positivism.

Emile Durkheim applied that logic directly to a topic close to criminology in his 1897 study Suicide, using official statistics across regions and religious groups to argue that a private act had social causes that could be measured at population level. Durkheim's choice of suicide, an act often assumed to be purely individual and psychological, was deliberate, since showing that even this act followed stable social patterns made the strongest possible case for a positivist social science.

A different starting point came from Max Weber, who argued that social science needed Verstehen, interpretive understanding, because human action is meaningful in a way that falling objects are not. A person who steals for survival and a person who steals for thrill can produce identical statistics, a stolen item, yet the two acts are not the same social fact. Weber's line of thought fed into an interpretivist tradition that treats meaning, not measurement, as the proper object of study.

The Chicago School of sociology brought this interpretive stance into the study of deviance directly. W. I. Thomas and Florian Znaniecki's multi-volume study The Polish Peasant in Europe and America (1918 to 1920) used personal letters and life histories rather than counts, and later Chicago-trained researchers extended the same closeness to the subject into studies of gangs, hobos and street corner life. Criminology therefore inherited both traditions almost from its founding, one counting crime from the outside, the other describing it from the inside.

Neither paradigm has won the argument outright, because they answer different questions. A positivist study can tell you that burglary rates rise when unemployment rises in a given city over a given decade.

An interpretivist study can tell you what an act of burglary means to the person who commits it, and why that meaning makes the act feel justified to them. The two answers are not competing versions of the same fact, they are answers to different questions, and criminology needs both.

This division still shapes how criminology journals review submissions today. A quantitative paper is expected to justify its sampling frame and report effect sizes with confidence intervals, while a qualitative paper is expected to justify its sampling logic and demonstrate immersion in the setting, and reviewers trained mainly in one tradition sometimes misjudge work from the other by applying the wrong standard to it.

Quantitative criminology relies on three main sources of data. The first is official statistics, crime counts recorded by police and courts and published by national agencies, such as the FBI's Uniform Crime Reporting Program in the United States, now migrating fully to the incident-level National Incident-Based Reporting System, or the National Crime Records Bureau's annual Crime in India report.

Official statistics are cheap to obtain and cover an entire jurisdiction, but they only record crime that is reported and then recorded, so they miss the dark figure of crime, the gap between offences that occur and offences that appear in the official count.

The second source is the victimisation survey, which asks a representative sample of the public directly whether they have experienced crime in a given period, regardless of whether they reported it to police.

The Crime Survey for England and Wales, run by the Office for National Statistics and known as the British Crime Survey before its 2012 rename, and the National Crime Victimization Survey run by the US Bureau of Justice Statistics are the two most cited examples, alongside the Australian Bureau of Statistics' Crime Victimisation Survey.

These surveys reveal a great deal of unreported crime, but they depend on memory, on willingness to disclose sensitive experiences, and they generally exclude some serious categories such as homicide by definition, since the victim cannot respond. Cost and design questions for this instrument are covered separately in Cost of Crime and Victimisation Surveys.

The third source is the experiment, including randomised controlled trials, where researchers assign cases to a treatment and a control condition and compare outcomes, for instance testing whether a policing tactic reduces repeat calls to a hotspot compared with ordinary patrol.

Experiments give the strongest basis for causal claims because random assignment rules out most alternative explanations, but true experiments are hard to run in criminal justice for ethical and practical reasons, so quasi-experimental designs that approximate random assignment are common instead.

Across all three sources, the researcher has to operationalise each concept before counting it, deciding, for example, whether gang membership means self-report, police record or both. That decision shapes every number the study later produces, which is why quantitative reports state their operational definitions explicitly rather than leaving the term to common sense.

Two teams studying the same phenomenon with different operational definitions can produce numbers that look contradictory even though neither team made an error, which is why a careful reader always checks the definitions section of a quantitative crime study before comparing its figures against another one.

Qualitative criminology gathers depth rather than breadth, typically from a small number of people or one setting studied closely. The semi-structured interview is the most common tool, using a loose guide of open questions rather than a fixed questionnaire, which lets an offender, a victim or a practitioner explain their own reasoning in their own words and lets the researcher follow up on anything unexpected that emerges.

Interviews can be conducted once or repeated with the same person over time, and a skilled interviewer treats an unexpected answer as a lead to pursue rather than a deviation from the script to correct.

Ethnography and participant observation go further, placing the researcher inside the setting itself for an extended period. William Foote Whyte's Street Corner Society (1943) is an early and still widely taught example, built from years spent living alongside a Boston street corner gang, and Howard Becker's Outsiders (1963) used fieldwork among jazz musicians and marijuana users to build his labeling account of deviance.

The method produces the kind of contextual, textured record that Clifford Geertz called thick description in The Interpretation of Cultures (1973), an account rich enough that a reader who was never present can grasp what an act meant to the people who did it.

The case study method examines one bounded instance, a single prison, a single gang, a single policy rollout, in depth, often triangulating interviews, documents and observation within that one case.

A related technique, grounded theory, developed by Barney Glaser and Anselm Strauss in The Discovery of Grounded Theory (1967), builds explanatory concepts directly out of the data through a cycle of coding and comparison, rather than testing a theory decided in advance, which suits topics where existing theory does not fit the setting well.

Both techniques require the researcher to keep detailed field notes and memos as the study proceeds, since the analysis in grounded theory is built inductively from those notes rather than applied afterward to a fixed set of variables decided before data collection began.

These methods share a limitation that quantitative critics raise consistently: findings from one gang, one prison wing or one interview sample cannot be assumed to hold anywhere else without further work, and the researcher's own presence can change what a setting looks like.

Qualitative researchers accept this trade-off deliberately, because the depth it buys cannot be recovered from a survey instrument built for scale. Ethnographers manage this trade-off by staying in a setting long enough for their presence to become unremarkable to participants, and by reporting their own position, for instance whether they were seen as an insider or an outsider, so a reader can judge how that position might have shaped what was shared with them.

Quantitative research is judged chiefly on two properties, validity and reliability. Validity asks whether a measure or a study captures what it claims to capture. Internal validity asks whether the study design actually rules out alternative explanations for an observed effect, which is why a randomised experiment scores higher on internal validity than a simple before-and-after comparison with no control group.

External validity asks whether a finding generalises beyond the specific sample and setting studied, for instance whether a hotspot policing result from one city would hold in a city with a different policing culture. A quasi-experimental design that lacks random assignment, comparing two similar but not identical cities before and after one adopts a policy, sits between these two extremes and has to argue its internal validity case rather than assume it.

A separate concept, construct validity, asks whether the operational measure genuinely reflects the underlying idea. A study that measures fear of crime purely by asking whether someone avoids a particular street at night may be picking up general caution rather than fear specifically tied to crime, which would weaken construct validity even if the survey itself is well administered.

Researchers usually address this by testing a measure against an established instrument before relying on it, or by asking several independent experts whether the survey items plausibly reflect the concept the study claims to be measuring.

Reliability is a separate question from validity, a measure can be reliable without being valid. Test-retest reliability checks whether the same respondent gives the same answer on separate occasions. Inter-rater reliability checks whether two independent coders classify the same case, for example the same police incident report, in the same way, which matters heavily whenever official crime categories depend on an officer's judgement call at the point of recording.

A third form, internal consistency, checks whether multiple survey items meant to tap the same underlying concept, several statements about feeling unsafe, for example, correlate with each other as expected, and is commonly reported using a statistic such as Cronbach's alpha.

These standards matter because official crime statistics are routinely criticised on exactly these grounds. Recording practices differ across police forces and over time, so a change in the counted crime rate can reflect a change in recording rules rather than a change in real offending, a reliability problem, not a validity one.

Distinguishing the two lets a researcher diagnose what actually went wrong with a disputed number, instead of dismissing the whole data source. A well-run quantitative study reports both properties openly rather than assuming them, stating how a key measure was validated and how consistently it was applied, so a reader can judge the numbers on their own merits.

Applying validity and reliability directly to interview or ethnographic work does not fit well, since qualitative research does not claim to measure a fixed variable the same way twice.

Yvonna Lincoln and Egon Guba proposed a parallel set of standards in Naturalistic Inquiry (1985): credibility, transferability, dependability and confirmability, built specifically for research grounded in participants' own meanings. These four standards do not ask a qualitative study to imitate a quantitative one, they ask a different but equally rigorous question of it, whether the account can be trusted on its own terms.

Credibility, the qualitative counterpart to internal validity, asks whether the account is a believable representation of participants' reality. Researchers build credibility through triangulation, checking one source of data against another, such as comparing what an offender says in an interview with what the case file records, and through member checking, taking a draft interpretation back to participants to see whether it matches their own understanding of events.

A study that relies on a single interview transcript with no corroborating source is harder to trust than one where the researcher has cross-checked an account against case records, observation notes or a second participant's version of the same event.

Transferability, the counterpart to external validity, asks whether the findings could plausibly apply to another setting. Because a qualitative sample is small and rarely random, the researcher does not claim statistical generalisability. Instead, thick description does the work, a reader with knowledge of a different prison or a different gang can judge for themselves whether the mechanisms described are likely to travel there, based on how much contextual detail the researcher has supplied.

A reader assessing transferability looks for the amount of contextual detail supplied about the setting, the participants and the researcher's own role, since thin description gives no basis for that judgement while thick description gives enough to make it responsibly.

Dependability, roughly parallel to reliability, asks whether the research process was documented clearly enough that another researcher could follow the same trail of decisions, and confirmability asks whether the conclusions are traceable back to the data rather than to the researcher's own assumptions, often checked through an audit trail of field notes, coding memos and decision logs kept alongside the interview transcripts themselves.

Together these four standards do the same job for a qualitative study that a methods section and a statistical appendix do for a quantitative one, they let another researcher assess how much weight the findings deserve, rather than asking a reader to simply trust the author's interpretation.

Quantitative methods give breadth, statistical power and a basis for causal claims that generalise across a jurisdiction, which makes them the right tool for questions such as whether a sentencing reform changed reoffending rates nationally, or whether a demographic group is disproportionately stopped by police across a large dataset of recorded stops.

Their weakness is that a number can hide the reasoning behind it, two burglaries counted as identical statistics can have entirely different causes and meanings for the people involved. This is also why national crime commissions and research councils in several countries now fund large linked programmes that pair a statistical arm with a qualitative arm rather than commissioning either in isolation.

Qualitative methods give depth, context and access to meaning that a closed survey question cannot reach, which makes them the right tool for questions such as how young people in a specific neighbourhood come to justify carrying a weapon, or how a specific court actually processes a plea negotiation in practice rather than on paper.

Their weakness is limited generalisability and a heavier dependence on the individual researcher's interpretation, which is why the audit trail and member checking described above matter so much to the method's credibility.

A growing share of criminological work does not choose one tradition exclusively but combines them, a mixed methods design. A common sequence runs a large survey first to establish a pattern, for instance that reported fear of crime is higher in a particular district than the recorded crime rate would predict, and then follows with interviews in that same district to explain the gap between the statistic and the lived experience behind it. Used well, the qualitative phase explains the mechanism the quantitative phase could only describe as a correlation.

The practical rule for a student choosing a design is to start from the question, not from a preferred method. A question phrased as how many, how much or does X cause Y across a population points toward quantitative tools. A question phrased as how or why does this happen for these people in this setting points toward qualitative tools.

When a single project needs both the scale of the first answer and the mechanism of the second, mixed methods is the honest choice, provided the added cost and complexity are justified by the question rather than added for its own sake.

Funding bodies and supervisors increasingly expect this justification to be stated explicitly in a research proposal, since a mismatch between the question asked and the method chosen is one of the most common weaknesses examiners flag in student dissertations.