Haphazard sampling is a non-probability sampling method where the person selecting the sample tries to choose items without any conscious pattern or bias, but without using a formal randomization tool like a random number generator. The selector attempts to mimic randomness through personal judgment, picking items that seem arbitrary. It shows up most often in auditing and basic research contexts where true random sampling feels impractical or too costly.
The key thing to understand: haphazard sampling looks random but isn’t. That distinction has real consequences for what you can and can’t conclude from the results.
How Haphazard Sampling Works
In practice, haphazard sampling means a person looks at a population of items and picks ones “at random” using their own judgment. An auditor might flip through a filing cabinet and pull invoices without any particular system. A researcher might approach people on a street corner, choosing whoever seems available. There’s no numbered list, no computer-generated selection, no systematic interval. The selector simply tries to avoid obvious patterns.
The intent is to approximate randomness cheaply and quickly. Instead of assigning every item a number and using software to select from that list, the person just grabs what feels representative. This makes it fast to execute and requires almost no planning infrastructure, which is why it remains popular despite its limitations.
Why It Differs From Random Sampling
True random sampling (also called probability sampling) gives every item in a population a known, calculable chance of being selected. This is what makes statistical inference possible. You can calculate margins of error, confidence intervals, and the probability that your sample reflects the whole population, because the math depends on every unit having a defined selection probability.
Haphazard sampling can’t deliver any of that. Because a human is making the selections, there’s no way to calculate the probability that any given item ended up in the sample. The selection odds are unknown and almost certainly unequal. This means results from haphazard samples have known generalizability only to the sample itself, not to the broader population the sample was drawn from. You can describe what you found, but you can’t rigorously project those findings outward.
Haphazard Sampling vs. Convenience Sampling
These two terms overlap significantly, and many researchers treat them as near-synonyms. Both are non-probability methods where subjects or items are selected based on accessibility rather than statistical design. The subtle distinction, when one is drawn, is about intent: convenience sampling explicitly prioritizes ease of access (you sample whoever is nearby or willing), while haphazard sampling implies the selector is at least trying to avoid systematic patterns even though no formal randomization exists.
In practice, the difference is often negligible. A researcher approaching people in a park is doing both: selecting whoever is conveniently available while trying to pick people who seem varied. The biases that affect one method affect the other in nearly identical ways.
Where Human Bias Creeps In
The core problem with haphazard sampling is that humans are genuinely bad at being random. Even when trying to select without a pattern, people introduce unconscious biases that skew the sample in predictable ways.
- Selection bias: The probability of choosing a given item ends up related to some characteristic that also affects the thing you’re measuring. An auditor might unconsciously gravitate toward invoices that are easier to read or in more accessible locations.
- Demographic bias: When sampling people, field staff tend to approach individuals who look similar to themselves in age or ethnicity, and potential respondents are less likely to engage with recruiters whose demographic profile differs from their own.
- Physical location bias: Easily accessible participants get overrepresented. Items at the front of a file, people near an entrance, records at the top of a stack all have a higher practical chance of selection.
- Avoidance of extremes: People attempting to pick “randomly” tend to avoid items that seem unusual or that stand out, which can systematically exclude the very outliers that matter most.
These biases aren’t hypothetical. They introduce enough variation to create statistical noise in analyses while simultaneously preventing the sample from being representative enough to control for that noise.
Haphazard Sampling in Auditing
Auditing is the field where haphazard sampling has the most formal recognition. Both AICPA and PCAOB auditing standards explicitly permit haphazard selection as a valid method for choosing items to test. Auditors use it when examining internal controls, verifying transactions, or testing account balances.
There’s an important catch, though. Using haphazard selection automatically classifies a sampling application as nonstatistical. Statistical sampling requires a statistically acceptable selection method, meaning random selection, not haphazard selection. This limits what auditors can do with the results. Some audit firms penalize haphazard samples by requiring larger sample sizes to compensate for the lack of true randomness. In a survey of large audit firms, three firms reported requiring higher sample sizes when auditors use haphazard selection for substantive testing.
Most firms prefer and encourage random or systematic selection methods but allow haphazard selection as a practical alternative. Audit guidance suggests that auditors using haphazard methods may need additional training to increase the likelihood of selecting representative samples. When errors are found in haphazard samples, projecting those error rates to the full population requires careful judgment, and AICPA guidance outlines situations where it may be appropriate not to project an error at all.
Advantages and Limitations
The advantages are straightforward: haphazard sampling is the easiest, least time-intensive, and least expensive sampling strategy to implement. It requires no sampling frame (a complete list of every item in the population), no software, and minimal planning. When you need a quick, rough look at a dataset or population and formal generalization isn’t the goal, it can be perfectly adequate.
The limitations are substantial. Results lack generalizability to any identifiable target population or subpopulation beyond the sample itself. You cannot calculate meaningful margins of error. Differences between subgroups within the sample may be unreliable because underrepresented groups introduce modest amounts of variation, enough to create noise but not enough to analyze properly. In research contexts, these scientific disadvantages generally outweigh the practical advantages when the goal is to draw conclusions about a larger population.
When Haphazard Sampling Makes Sense
Haphazard sampling is most defensible in situations where the stakes of imperfect representation are low or where the population is relatively uniform. If you’re an auditor spot-checking a batch of invoices from a single vendor processed in the same month, the items are similar enough that haphazard selection is unlikely to miss something systematic. If you’re running a pilot study to test whether a survey instrument works before committing to a full probability sample, haphazard selection is efficient and the results aren’t meant to be generalized anyway.
It becomes problematic when you need results that hold up to scrutiny, when the population has meaningful subgroups you need to capture proportionally, or when the findings will drive high-stakes decisions. In those cases, the time and cost savings don’t offset the inability to make reliable inferences. A truly random sample, even a small one, gives you a statistical foundation that no amount of careful haphazard picking can replicate.