A sample supports a claim about a population only when the sampling process justifies that connection. Before calculating a percentage, identify who could be selected, who actually responded, and which population the conclusion concerns.
Parameters and statistics
A parameter describes the population, such as its true proportion supporting a policy. A statistic is calculated from the sample, such as where is the number of supporters among respondents.
The statistic estimates the parameter. Different random samples produce different statistics, so an estimate is not an exact population fact. A census attempts to measure everyone, although measurement errors and nonresponse can still occur.
Example: estimate a count without overstating certainty
A random sample of 250 registered voters contains 145 supporters of a proposal. The sample proportion is In a population of 18,000 registered voters, this suggests about supporters.
This projection concerns registered voters, not all residents, likely voters, or people in another town. Those populations may differ. It also describes support at the time and under the conditions of the survey, not guaranteed voting behavior later.
The numerical calculation is straightforward; the population definition is what makes the interpretation valid.
Random selection and representative coverage
A simple random sample gives each possible sample of a specified size equal selection probability. Random selection helps prevent the researcher from systematically favoring certain respondents.
But selection from an incomplete list can still underrepresent a group. A survey based only on landline telephone numbers may miss many people. Similarly, selected respondents who decline to participate can create nonresponse bias if their opinions differ from respondents' opinions.
Voluntary-response surveys are especially vulnerable to self-selection. A website poll can have thousands of answers without representing the site's readers, much less a broader population.
Example: sample size cannot cure bias
Suppose two surveys ask about public transportation. One surveys 2,000 passengers at a train station; the other randomly selects 400 town residents from a complete address list and achieves a high response rate.
For the opinions of all town residents, the second design is more defensible. The first oversamples people who already use trains. Its larger size may reduce variability around the train-user opinion, but that is not the target population opinion.
Random variation decreases with larger samples under appropriate designs, while systematic bias may remain. These are separate sources of error.
Selection and treatment assignment answer different questions
Random sampling concerns who enters a study and supports generalization to a population. Random assignment concerns which treatment participants receive and supports causal comparison. A study can have one without the other.
A randomly sampled survey can estimate a population association but does not, by itself, establish that one measured variable causes another. Conversely, a randomized experiment on volunteers can support a treatment effect within that study while requiring caution about generalizing to everyone.
A sample frame is the list or mechanism from which selection occurs. Even perfect random selection from an incomplete frame cannot represent omitted members fairly. When evaluating a study, compare the frame with the intended population before considering how many names were selected.
Try it yourself
A university emails all students and receives 5,000 voluntary responses about tuition. Does the large response count establish an unbiased estimate? What additional information matters?
Answer: No. Respondents may differ from nonrespondents. You need to assess participation patterns, population coverage, question wording, and the selection process. Size alone does not establish representativeness.
What to remember
A population estimate requires a defensible sampling method. Distinguish statistics from parameters, sampling variation from bias, and random selection from random assignment.