What Is Network Sampling?
Network sampling is a broad class of methods that uses the social connections between people as the basis for building a sample. Instead of drawing names from a list or selecting addresses from a map, you use the relationships within a population, asking sampled individuals to identify others who belong to the target group, or mapping network structures to determine who gets included. The approach covers everything from informal snowball sampling to formal methods like respondent-driven sampling, network scale-up, and multiplicity sampling. What ties them together is the core idea that social networks themselves serve as the sampling frame. When a population is connected through identifiable relationships but doesn't appear on any conventional list, those connections become the infrastructure for reaching people.
Why Network Sampling Matters
Populations that are rare, hidden, stigmatized, or mobile resist standard sampling frames. There's no telephone directory of freelance gig workers, no address database of people living with undiagnosed conditions, and no census of members in underground communities. Network sampling turns social structure into research infrastructure, making it possible to study groups that would otherwise require prohibitively expensive screening or remain entirely unreachable.
How Network Sampling Works
Network sampling methods vary in formality and statistical rigor, but they all exploit the same resource: the fact that members of a population tend to know each other.
Link-Tracing Designs
Link-tracing is the most intuitive network sampling approach. You start with a set of initial respondents and "trace" their connections to find additional participants. Snowball sampling and respondent-driven sampling are both link-tracing methods. The depth of tracing (how many waves of referrals you follow) and the breadth (how many referrals per person) determine the structure and coverage of the resulting sample.
Link-tracing works well when the population is socially connected and members can identify and recruit peers. It struggles when the population is fragmented into disconnected subgroups, if seeds can't reach certain clusters through referral chains, those clusters are excluded entirely.
Multiplicity (Network) Sampling
Multiplicity sampling takes a different approach. Instead of asking respondents to recruit peers, you draw a standard probability sample of the general population and ask each respondent to report on members of the target group in their personal network. "Do you know anyone who [meets the target criteria]?" The target group's prevalence is then estimated from the network reports of the probability sample.
This method was originally developed by Monroe Sirken in the 1970s for estimating the size of rare populations. It extends the effective coverage of a small sample by using each respondent as a sensor for their network. A probability sample of 1,000 people, each reporting on 20 network members, effectively screens 20,000 people for the target characteristic.
The key challenge is multiplicity: individuals who appear in multiple respondents' networks get counted multiple times. Proper estimation requires adjusting for this overlap, which is where the math gets complicated.
Network Scale-Up Method (NSUM)
NSUM estimates the size of hidden populations by asking a random sample of the general population how many people they know in various groups, both groups with known sizes (e.g., postal workers, people named Michael) and the target hidden group. The ratio between known and reported network counts produces a size estimate for the hidden population.
NSUM is valuable for estimating population sizes when direct enumeration is impossible, but it depends on respondents having accurate knowledge of their contacts' behaviors, an assumption that weakens for stigmatized or private characteristics.
Egocentric vs. Sociocentric Designs
Egocentric network studies sample individuals and collect information about each person's immediate social connections (their "ego network"). Sociocentric studies attempt to map the complete network structure of a bounded population, every person and every connection. Egocentric designs are feasible for large populations; sociocentric designs are limited to smaller, bounded groups like organizations, classrooms, or villages.
The choice depends on your research question. If you want to understand how an individual's network position affects their behavior, egocentric data suffices. If you need to understand information flow, diffusion patterns, or community structure, you need sociocentric data.
When to Use Network Sampling
- Reaching hidden, rare, or stigmatized populations that don't appear on sampling frames and can't be screened efficiently from general population samples
- Estimating the size of hard-to-count populations using network scale-up or multiplicity methods
- Studying social influence and peer effects where the network structure itself is a variable of interest
- Community health research where access depends on trust and peer connections within tight-knit groups
- Market research for niche communities connected through shared interests, platforms, or identities
Common Mistakes to Avoid
- Assuming all members of the population are connected. Network sampling only reaches people who are linked to the initial sample through social connections. Isolated individuals or disconnected subgroups will be missed entirely.
- Ignoring the difference between link-tracing and probability-based network methods. Snowball sampling is a convenience method. Multiplicity sampling through a probability base produces valid population estimates. The statistical properties are fundamentally different.
- Treating network-size self-reports as precise measurements. People are notoriously bad at estimating how many others they know. Calibration questions (asking about groups with known sizes) help, but measurement error in network size propagates into all your estimates.
How Quali-Fi Supports Network Sampling
Quali-Fi's survey platform supports network data collection with relationship mapping questions, referral chain tracking, and name generator instruments that capture ego-network data within the standard survey flow. The platform's unique-link system enables link-tracing designs where each respondent generates trackable recruitment URLs for their network contacts.
Frequently Asked Questions
Is network sampling a probability method?
It depends on the specific technique. Multiplicity sampling built on a probability base is a probability method. Snowball sampling is not. Respondent-driven sampling occupies a middle ground, it produces estimates with calculable properties under certain assumptions, but those assumptions are hard to verify in practice.
How do I handle the fact that well-connected people are over-represented?
People with larger networks are more likely to be nominated or recruited in network-based methods. Formal methods like RDS and multiplicity sampling address this through weighting adjustments based on self-reported network size. Without these adjustments, your sample overrepresents the socially connected and underrepresents the more isolated.
Can network sampling work for online communities?
Yes, and it's increasingly common. Online communities have visible network structures (followers, friends, group memberships) that can serve as a sampling frame. Digital link-tracing through social media shares, forum posts, or direct messages follows the same logic as in-person chain-referral methods.
Related Topics
- Respondent-Driven Sampling
- Chain-Referral Sampling
- Adaptive Sampling
- Panel Sampling
- Volunteer Sampling
Use social connections to reach any population. Start a free trial with Quali-Fi and use referral tracking, network mapping questions, and unique participant links to run network-based recruitment.