Sampling Methods in Criminological Research (UGC-NET Unit VI)

A sampling frame is the list a researcher actually draws names from, an electoral roll, a hospital register, a prison admissions list. Survey methods assume such a list exists and is reasonably complete. Criminology often cannot make that assumption: there is no register of active burglars, no directory of drug dealers, and no membership roll for a street gang.

Official records give only a partial and distorted substitute. Arrest, prison and conviction data capture only the offenders who were caught, not a random slice of everyone who offends. This gap between offending and officially recorded offending is one reason fieldwork-based sampling exists as its own tradition inside criminological research, alongside survey and records-based designs.

Remember the key exam point: because no frame exists, a researcher cannot claim a known, equal probability of selection for the final sample, so classical statistical inference does not apply cleanly to snowball or purposive samples of hidden populations.

Where a usable frame exists, probability sampling applies: simple random sampling gives every listed unit an equal chance; stratified sampling samples separately within subgroups so small groups are not lost by chance; cluster sampling selects whole groups, such as police stations, and studies everyone eligible within them. These designs support a stated margin of error and are the basis of prison censuses and court-file audits.

Where no frame exists, which is the ordinary condition for studying active, uncaught offenders, researchers use non-probability sampling instead: purposive sampling selects respondents deliberately for a trait the study needs, quota sampling fills target subgroup counts, and convenience sampling recruits whoever is easiest to reach. None of these support a statistical population estimate, and each still assumes a researcher can locate eligible people directly, which fails outright once a population actively conceals itself.

Snowball sampling recruits an initial small group of seeds who fit the study's criteria, then asks each to refer others who also fit those criteria, growing the sample outward through personal networks. It was formalised as a statistical method by Leo Goodman in 1961, and its value for stigmatised and criminal populations was worked out by Patrick Biernacki and Dan Waldorf in a widely cited 1981 paper on chain-referral sampling of heroin users.

Respondent-driven sampling, developed by Douglas Heckathorn in 1997, keeps the chain-referral mechanic but adds structure: numbered referral coupons, recorded recruitment links, and each respondent's reported network size, which together let the final sample be statistically weighted toward a population estimate.

The weighting rests on assumptions, that recruitment behaves like a Markov chain and that respondents report network size accurately, and both can fail, so it is used where a rough prevalence estimate is genuinely useful, for example sizing a hidden population for a harm-reduction service.

Neither method removes bias entirely. Referral chains stay largely inside existing social clusters, so isolated members of a hidden population remain structurally hard to reach no matter how many referral waves run.

For NET purposes, keep the two names paired but distinct: plain snowball sampling describes mechanism and decision-making, while respondent-driven sampling is the version to reach for when a rough prevalence estimate is genuinely needed, for example sizing a population for a public-health or outreach programme rather than a purely criminological question.

Reaching a hidden population usually runs through a gatekeeper: a prison superintendent who approves a research protocol, a police commander who permits officers to speak to a researcher, or an informal community figure whose introduction vouches for the researcher. This access is rarely a one-time approval; it is a relationship that must be maintained and can be withdrawn.

For NET purposes, the honest limit on generalisability is the key takeaway. Chain-referral and purposive samples describe the people the network actually reached and who agreed to talk, not a statistically representative slice of every offender in that category.

A gatekeeper also filters the sample before referral even starts, by choosing who they are willing to introduce, which compounds the network-clustering bias already built into chain referral. Findings from this kind of study travel best as evidence about mechanisms and decision-making, not as a precise population rate.

This is also why a probability sample and a chain-referral sample answer different exam-style questions. A probability sample of prison records can estimate how many people were convicted of an offence in a given year. Only fieldwork with active, uncaught offenders can explain how an offence is actually planned or how risk is weighed in the moment, a question a records-based sample cannot reach because it never includes anyone who was not caught.