Statistical Population
A population, in statistics, is the complete set of every individual, event, or item that a study or analysis is actually trying to understand — every customer a company has or could have, every voter in an election, every unit a factory produces. It's rarely feasible or even possible to measure every member of a population directly, which is precisely why a sample — a smaller subset drawn from it — is used instead, and why the entire apparatus of inferential statistics exists: to responsibly generalize from a measured sample back to the full population it was drawn from.
Defining a population precisely, before any sampling happens, is a more consequential step than it might first appear, because a loosely or ambiguously defined population silently changes what any subsequent statistical conclusion is actually about. "All customers" could mean every customer who ever made a single purchase, every currently active subscriber, or every person who has ever visited the website regardless of purchase — three genuinely different populations that would produce three different, equally valid-looking statistics depending on which one a sample was actually drawn from, and a conclusion stated about "customers" in general without specifying which of these is silently making an implicit, and potentially unjustified, generalization beyond whichever specific population the sample actually represented.
A population parameter — a true, fixed numerical characteristic of the entire population, such as its actual mean or actual proportion holding some property — is, by definition, almost always unknown and unknowable directly, since measuring it exactly would require observing every single member of the population, which is rarely practical and sometimes genuinely impossible (a population that's still growing, like "all future customers," can never be fully enumerated no matter how much effort is spent trying). This unknowability is the entire reason a sample statistic — the equivalent measurement computed from a sample rather than the full population — exists as a distinct concept, denoted with different notation specifically to keep the distinction visible: a population mean is conventionally written as μ (mu), while a sample mean is written as x̄ (x-bar), a notational convention that exists specifically to prevent a reader from confusing an estimate for the actual, unobservable truth it's standing in for.
"Population" in this statistical sense should not be confused with its everyday meaning of human population specifically — a statistical population can be any defined set of units at all: every transaction a payment processor handles in a year, every manufactured part coming off an assembly line, every tweet mentioning a particular brand — and the term's generality is deliberate, since the same sampling and inference framework applies identically regardless of whether the units being studied are people, products, or events.
A census — measuring every single member of a population directly, rather than sampling — eliminates sampling uncertainty entirely by definition, but is rarely practical outside of specific cases where the population is small and fully enumerable (a company auditing all of its own employees, for instance) or where a census is mandated and resourced at enormous cost (a national population census), which is exactly why sampling, with its associated uncertainty, remains the overwhelmingly more common approach in practical data analysis.
Read a population as the full, defined universe a statistical conclusion is actually meant to be about, not just a loose approximation of it: precisely defining that universe before sampling begins is what keeps a later conclusion honest about what it actually generalizes to, the distinction between an unknowable population parameter and an observable sample statistic is formalized through deliberately different notation specifically to prevent the two from being confused, and the entire discipline of sampling exists as a practical, necessary compromise in place of a census that's almost never feasible to actually run.