Mixed-Methods Research in Crime and Justice

A mixed-methods study is not two separate studies bound into one report. The defining feature is integration: the researcher plans, from the design stage, how the quantitative and qualitative strands will speak to each other, and builds the analysis so that one strand's output can inform the other's interpretation.

Abbas Tashakkori and Charles Teddlie, in their 2003 Handbook of Mixed Methods in Social and Behavioral Research, framed this as a third methodological tradition sitting alongside, not underneath, the quantitative and qualitative traditions it draws from.

Criminology has particular reasons to reach for this toolkit. Official crime statistics, whether police-recorded figures, court disposals or prison reconviction data, describe patterns at scale but say little about the decisions, relationships or institutional pressures that produced them.

Ethnographic and interview-based work does the opposite: it can trace how a probation officer's judgement shapes a case outcome, or how a young person narrates their own drift into offending, but a single site or a small purposive sample cannot tell a policymaker how often that pattern occurs elsewhere.

Jennifer Greene, Valerie Caracelli and Wendy Graham's 1989 article on mixed-method evaluation set out five reasons researchers combine methods: triangulation (checking one finding against another), complementarity (using each method to illuminate a different facet of the same phenomenon), development (using one method's results to build the next phase), initiation (using contradictions between methods to open new questions) and expansion (using different methods for different parts of a study to broaden its scope). Criminological studies typically draw on more than one of these purposes at once.

The trade-off is cost. A mixed-methods project needs expertise, time and often a research team spanning both traditions, and journals and funders have historically been more comfortable evaluating a single-method study. Understanding when integration earns its cost, rather than treating it as automatically superior, is the practical skill this topic builds toward.

The demand for this kind of evidence is not confined to one system. The United States National Institute of Justice funds programme evaluations that pair administrative outcome data with participant interviews, the United Kingdom's Ministry of Justice publishes reconviction statistics that researchers routinely follow up with qualitative fieldwork, Correctional Service Canada commissions mixed-methods reviews of its rehabilitation programming, and India's prison and probation research, drawing on National Crime Records Bureau data alongside case-level fieldwork, faces the same gap between what official counts show and what practitioners and offenders describe on the ground.

In a convergent design, sometimes still called concurrent triangulation design in older literature, the researcher collects quantitative and qualitative data during the same general period, analyses each strand on its own terms, then brings the two sets of results together at the interpretation stage.

Neither strand depends on the other's findings to proceed. The strength of this arrangement is speed: because the strands run in parallel rather than one waiting for the other to finish, a convergent study can be completed in a single fieldwork window.

A criminological example makes the logic concrete. A researcher studying fear of crime in a city neighbourhood might run a household survey asking residents to rate their perceived safety on a numeric scale, while at the same time conducting open interviews asking residents to describe, in their own words, what makes a street feel unsafe.

The survey establishes how widespread high fear scores are across the sample. The interviews explain what respondents mean by unsafe, which can include factors, such as poor lighting or the visible presence of derelict buildings, that a numeric scale never asked about directly.

The analytic challenge is merging two datasets that were built independently and may not map onto each other cleanly. A common technique is data transformation, where qualitative themes are counted and turned into simple frequencies so they can sit next to the survey results in the same table, or where the interview quotes are grouped by whether they support, extend or contradict the survey pattern.

Whichever merging technique is used, the researcher has to justify it in the write-up rather than simply presenting both datasets and leaving the reader to reconcile them.

Convergent designs work best when the research question genuinely needs both a prevalence answer and a meaning answer at the same time, and when the researcher has the capacity to run two data collection efforts concurrently. They work poorly when the qualitative sample is too small or too selectively recruited to say anything about the pattern the survey describes, which is a common criticism levelled at rushed convergent studies.

An explanatory sequential design runs in two clearly ordered phases. The quantitative phase comes first and produces a result, often a pattern, an unexpected difference between groups, or an outlier that the numbers alone cannot account for. The qualitative phase follows and is built specifically to explain that result, usually by returning to a purposively selected subset of the same population rather than a fresh sample.

Consider a study of a community sentencing programme that finds, from administrative reconviction data, that participants who complete the full programme reoffend at a noticeably lower rate than those who drop out early, even after the researchers statistically control for age, prior record and offence type.

That quantitative finding raises an obvious follow-up question: what is it about completion, rather than the underlying differences between people who complete and people who drop out, that is doing the work? A second, qualitative phase interviewing both completers and early leavers about their experience of the programme, their relationship with supervising staff and the barriers they faced, is built to answer exactly that question.

The sequencing matters because the qualitative sample is drawn deliberately from the quantitative results, a strategy researchers call purposive follow-up sampling. Interviewing people who fit the pattern the numbers revealed, and sometimes deliberately interviewing outliers who defy it, gives the qualitative phase an analytic edge that a randomly drawn qualitative sample would not have.

The limitation is time. Because the second phase cannot be designed until the first phase's results are in hand, an explanatory sequential study typically takes longer to complete than a convergent design, and the researcher has to keep access to the original population open across both phases, which is not always possible when working with a criminal justice agency on a fixed data-sharing agreement.

An exploratory sequential design reverses the order. The study opens with a qualitative phase, usually because so little is known about the phenomenon that a survey or coding scheme would be premature, and uses what that phase finds to build a quantitative instrument that is then tested on a larger sample.

This design suits genuinely under-studied or newly emerging areas of criminology particularly well. A researcher starting from scratch on how young people experience online grooming, for instance, cannot write a meaningful survey scale without first understanding the vocabulary victims themselves use, the stages they describe passing through and the warning signs they identify in hindsight.

An initial round of interviews or focus groups surfaces those themes, which the researcher then turns into closed survey items, checks for internal consistency, and administers to a much larger sample to establish how common each theme is.

The instrument-building step is the part most often done badly in practice. Turning a rich interview theme into a single survey item risks flattening a complex experience into a checkbox that no longer captures what the interview described. Careful exploratory sequential work pilots the new instrument, checks that respondents interpret each item the way the qualitative phase intended, and revises items that do not survive that check before the full quantitative rollout.

Exploratory sequential designs are common in scale development for concepts such as procedural justice perceptions or institutional trust, where a validated measurement tool did not previously exist and the qualitative phase does the foundational work of defining what should be measured before anyone tries to measure it at scale.

Triangulation is often used loosely to mean any study that uses more than one method, but the term describes a specific analytic move: placing an independent quantitative finding next to an independent qualitative finding on the same question and asking how they relate. Three outcomes are possible. Convergence means the two strands point to the same conclusion, which strengthens confidence in that conclusion beyond what either strand could support alone.

Complementarity means the two strands address different facets of the phenomenon and together produce a fuller picture than either gives on its own. Divergence means the two strands appear to contradict each other, which is not a failure of the study but a finding in its own right, since it usually signals that the phenomenon is more complicated than either method captured.

A useful illustration of divergence comes from comparing how crime is discussed in media coverage with what victimisation survey data shows about actual risk. Reported fear of a particular crime type can run far ahead of that crime's recorded frequency, and a mixed-methods researcher treats that gap itself as data worth explaining, rather than dismissing either the survey or the qualitative fear narratives as simply wrong.

Integrating the two strands formally, rather than just discussing them one after the other in a results section, is usually done through a joint display, a table that lines up a quantitative result in one column against the qualitative data on the same theme in an adjacent column.

Guetterman, Fetters and Creswell's 2015 methodological work on joint displays in health and social research popularised this format as the clearest way to show a reader exactly how a mixed-methods conclusion was reached, rather than leaving the reader to trust the researcher's narrative summary.

The conclusion a researcher draws only after this integration step, one that neither the quantitative nor the qualitative strand could have supported alone, is called a meta-inference. A meta-inference is the payoff of a mixed-methods design: it is a claim the study earns specifically because it combined the two forms of evidence, and a well-written mixed-methods paper should be able to point to exactly which joint display or comparison produced each meta-inference it reports.

The philosophical objection to mixed-methods research is older than the design vocabulary used to describe it. Quantitative traditions in criminology generally sit within a post-positivist stance, treating crime patterns as measurable phenomena that exist independently of the researcher observing them.

Much qualitative tradition, particularly interpretivist and constructivist work, treats meaning as something negotiated between researcher and participant rather than simply discovered. Critics of mixed-methods designs argue that combining the two within one study papers over this incompatibility rather than resolving it, a position sometimes called the incompatibility thesis.

Tashakkori and Teddlie's pragmatist response, echoed by Creswell and Plano Clark, is that a researcher can use whichever combination of methods best answers the practical question at hand without first resolving the underlying philosophy of science debate, so long as the study is transparent about what each strand can and cannot claim. This pragmatist position is now the dominant working stance in criminological mixed-methods practice, even though the incompatibility thesis remains a live debate in methodology seminars.

Beyond philosophy, the practical costs are real. Mixed-methods studies generally cost more and take longer than single-method studies of comparable scope, because they require expertise in two analytic traditions, often a larger research team, and enough time to run both a quantitative and a qualitative phase properly rather than treating one as an afterthought bolted onto the other.

Funders and ethics boards reviewing criminal justice research also need to see a clear integration plan up front, since a proposal that simply lists two unconnected data collection activities is not a mixed-methods design.

Because integration is the feature reviewers most often find missing, reporting standards for the field now ask authors to state their design type explicitly by name, show at least one joint display or equivalent integration device, and state their meta-inferences separately from the quantitative and qualitative findings that fed into them. A reader should be able to trace each integrated conclusion back to the specific comparison that produced it.