A population is the entire group you want to learn about, while a sample is a smaller portion of that group you actually collect data from. In most real-world research, studying every single member of a population is impossible or impractical, so researchers select a sample and use it to draw conclusions about the whole. Understanding this distinction is fundamental to how statistics works, because nearly every study you encounter is based on sample data being used to make broader claims.
Population: The Full Picture
A population in statistics isn’t limited to people. It’s any complete set of items, individuals, or observations that share a characteristic you’re interested in. “All adults in the United States with high blood pressure” is a population. So is “every light bulb produced at a factory in March” or “all transactions on a company’s website last year.”
Populations can be finite or effectively infinite. A national census attempts to reach every person in a country, making the population finite and countable. But a population like “all possible coin flips” has no upper limit. The key idea is that a population includes every member of the group you’re trying to understand, not just the ones convenient to reach.
When researchers define a population, they typically distinguish between two layers. The target population is the broad group they want their findings to apply to (for example, all people with a particular diagnosis worldwide). The study population is the realistic subset available for study, like patients with that diagnosis in a particular city or hospital network. This narrowing is important because conclusions drawn from a study technically only apply to the population the sample was actually pulled from.
Sample: The Practical Shortcut
A sample is any subset of the population selected for actual measurement or observation. If your population is every college student in the country, your sample might be 1,200 students from 30 universities. Researchers work with samples because collecting data from an entire population is usually too expensive, too slow, or logistically impossible.
Consider national election polls. Polling every eligible voter would essentially require running the election itself. Instead, polling organizations survey a sample of a few thousand people and use those responses to estimate how millions will vote. The media then reports these results, often as if they reflect the entire population’s intentions. The accuracy of those reports depends entirely on how well the sample represents the population.
How Researchers Move From Sample to Population
The core purpose of sampling is inference: using what you observe in a small group to make valid statements about the larger group. This leap from sample to population is the engine of most scientific research. A clinical trial tests a drug on a few hundred or thousand patients, then generalizes the results to everyone with that condition. A quality inspector checks 50 items off the assembly line and decides whether the entire batch meets standards.
This inferential leap works because of probability. When a sample is randomly selected, statistical tools can quantify how confident you should be that the sample’s results reflect the population’s true values. A 95% confidence level, for instance, means that if you repeated the same sampling process over and over, the sample average would fall within a specific range of the true population average 95% of the time.
But the leap can also go wrong. Conclusions drawn from a sample only apply to the population that sample was properly selected from. If a health study recruits participants exclusively from one hospital, generalizing those findings to patients everywhere requires caution, because that hospital’s patients may differ from the broader population in meaningful ways.
Parameters vs. Statistics
One practical difference between populations and samples is what we call the numbers that describe them. A measurement that describes a population is called a parameter. A measurement that describes a sample is called a statistic. The terms matter because they signal whether you’re looking at a definitive value or an estimate.
For example, the true average income of every person in a country is a population parameter. The average income calculated from a survey of 5,000 people is a sample statistic. In notation, the population mean is represented by the Greek letter μ (mu), while the sample mean is written as x̄ (called “x-bar”). Similarly, the population proportion is written as p, while the sample proportion is p̂ (“p-hat”). Population standard deviation uses σ (sigma); sample standard deviation uses s.
In practice, you almost never know the true population parameter. The whole point of collecting a sample is to estimate it. The sample statistic is your best approximation, and statistical methods help you measure how close that approximation is likely to be.
What Makes a Sample Reliable
Not all samples are equally useful. A sample is considered representative when it mirrors the key characteristics of the population it came from. The gold standard for achieving this is random sampling, where every member of the population has an equal chance of being selected. When a sample is truly random, the differences between the sample statistic and the population parameter come down to chance alone.
That chance variation is called sampling error, and it’s unavoidable. Even a perfectly designed study will produce a sample mean that’s slightly different from the population mean, simply because you’re looking at a subset rather than the whole. Sampling error shrinks as sample size grows, but it never disappears entirely unless you measure the full population.
Non-sampling error is a separate problem. This includes mistakes in how data is collected, entered, or processed, as well as bias in who ends up in the sample. A survey with confusing questions, a study where certain types of people are more likely to participate, or data entry typos all introduce non-sampling error. Unlike sampling error, these problems don’t improve just by adding more participants. They require better study design.
What Determines Sample Size
A common question is how large a sample needs to be. The answer depends on several factors working together. The most important are your desired margin of error, your confidence level, and how much variability exists in the population.
The margin of error is how much imprecision you’re willing to accept. A poll with a 3% margin of error is more precise than one with a 6% margin, but achieving that tighter range requires a larger sample. The confidence level (commonly 95%) determines how sure you want to be that your results fall within that margin. Higher confidence demands more data.
Population variability also plays a role. If everyone in the population is fairly similar on the trait you’re measuring, a small sample captures the pattern quickly. If there’s wide variation, you need more observations to get an accurate picture. Researchers often look at the standard deviation from previous studies to estimate this variability before deciding on a sample size.
Finally, effect size matters. If you’re trying to detect a small difference between two groups (say, a subtle improvement from a treatment), you need a much larger sample than if the difference is dramatic and obvious.
Side-by-Side Comparison
- Definition: A population is the complete group of interest. A sample is a subset drawn from that group.
- Size: Populations can range from hundreds to billions. Samples are always smaller than the population they come from.
- Measurements: Population measurements are called parameters and are fixed (though usually unknown). Sample measurements are called statistics and vary from sample to sample.
- Feasibility: Collecting data from an entire population (a census) is expensive and time-consuming. Sampling is faster, cheaper, and often the only realistic option.
- Error: Population data has no sampling error because nothing is being estimated. Sample data always carries some degree of sampling error.
- Notation: Population mean is μ; sample mean is x̄. Population standard deviation is σ; sample standard deviation is s.
When Populations Are Actually Measured
There are situations where every member of a population is measured directly. A national census is the most familiar example: countries like the United States and Australia attempt to collect information from every household at regular intervals. Schools sometimes survey every enrolled student rather than sampling. A small business might analyze every transaction from the past year rather than pulling a subset.
These full-population measurements are valuable because they eliminate sampling error entirely. The numbers you get are the true parameters, not estimates. But they come with significant costs in time, money, and logistics, which is why they’re reserved for situations where completeness is worth the investment or where the population is small enough to make it practical.