03 · What You Need to Know
How a Sampling Frame Connects Your Population to Your Sample
A Population Definition Is Not Yet a Sampling Frame
Suppose you define your population as all undergraduate students officially enrolled at a university during the current semester. That definition tells you who belongs to the population, but it does not yet tell you how those students will be identified for selection.
You might obtain an official enrollment database. That database could serve as a sampling frame if its records allow eligible students to be identified and sampled.
For another study, the frame might be an employee roster, household address file, professional registry, school directory, list of establishments, patient registry, administrative database, or geographic listing of units.
The exact form depends on the population and sampling design. What matters is its operational function: it provides the units from which selection occurs.
Target population
The population the research is intended to inform.
Sampling frame
The operational source containing or identifying units available for sample selection.
The two should correspond as closely as feasible, but they are conceptually different. The U.S. Census Bureau defines coverage in terms of the extent to which elements of a target population are represented on a sampling frame and explicitly requires evaluation of frame coverage, timeliness, accuracy, eligibility, and related limitations in its statistical quality standards.
What Would an Ideal Sampling Frame Look Like?
Conceptually, an ideal frame would allow every eligible unit in the target population to be identified correctly, contain no units that fall outside that population, avoid unintended duplicates, and be sufficiently current for the period being studied.
Real frames are rarely perfect.
A student roster may contain students who recently withdrew. A professional registry may omit practitioners who are not registered with that organization. An employee database may retain outdated records. An address frame may lag behind new construction. A list assembled from several sources may contain the same person more than once.
These are not merely clerical inconveniences. They can alter who has an opportunity to enter the sample.
What Is Undercoverage?
Undercoverage occurs when eligible members of the target population are missing from the frame or otherwise not covered by the sampling process.
Suppose a university has 15,000 eligible students, but the email list used for sampling contains only 13,500 because newly enrolled students and students without activated institutional email accounts are absent. Those missing students cannot be selected from that list.
The consequences depend on who is missing. If the omitted students are effectively similar to covered students with respect to the variables being studied, the practical impact may be limited. If they differ systematically on outcomes or characteristics relevant to the study, estimates based on the covered population may be distorted when interpreted as estimates for the full target population.
Watch Out
Random selection does not give a unit a chance of selection if that unit never appears on the frame. A perfectly executed random sample from an incomplete frame cannot directly select people or cases the frame does not cover.
What Is Overcoverage?
Overcoverage occurs when the frame includes units that should not be represented as eligible members of the target population or otherwise includes units erroneously. The Census Bureau's definition of coverage error includes both failure to include appropriate units and erroneous inclusion or duplication.
For example, an employee roster used to study current employees might still contain people who left the organization. A student database might include individuals who are no longer enrolled during the study period.
Some overcoverage can be addressed through eligibility screening before data collection or analysis, but it still affects recruitment efficiency and may complicate the sampling process.
What Happens When the Same Unit Appears More Than Once?
Duplicate records create another frame problem. If the same eligible person appears twice and sampling occurs directly from records, that person may have a greater probability of selection than someone appearing once.
This is why frame construction sometimes requires deduplication. The Census Bureau's statistical quality standards specifically identify combining multiple frames and removing duplicates, or adjusting selection probabilities when units appear in multiple frames, as frame-development considerations.
Duplicates are especially easy to introduce when researchers merge institutional databases, mailing lists, registries, or other sources without a reliable identifier.
An Outdated Frame Can Describe Yesterday's Population
Populations change. People enroll, graduate, move, change jobs, open businesses, close businesses, enter professional registers, or leave them.
A frame that was accurate two years ago may no longer adequately describe today's population. This problem is sometimes described in terms of frame timeliness.
Large statistical systems devote substantial effort to keeping frames current. The Census Bureau, for example, maintains and updates the Master Address File used as a sampling source for major demographic surveys. Its quality standards explicitly identify updating frames and assessing their timeliness as parts of frame development and evaluation.
For a smaller research project, the same principle applies on a different scale. Ask when your list was created, when it was last updated, and whether population changes since that date could matter.
Coverage Error Is Different From Sampling Error
This distinction is important because the remedies are different.
Sampling error
Variation arising because only a sample rather than the entire population is observed under a probability sampling design.
Coverage error
Error associated with mismatch between the population that should be covered and the units actually represented by the frame or data-collection process.
Increasing sample size can reduce sampling variability under appropriate conditions. It does not automatically repair undercoverage. Drawing 5,000 units instead of 500 from the same incomplete frame still gives zero direct selection opportunity to eligible units absent from that frame.
This is one reason a large sample can still be seriously biased.
Is the Sampling Frame the Same as the Accessible Population?
Not necessarily.
Your accessible population concerns the portion of the population you can realistically reach under the study conditions. The frame is the operational source through which units are identified or selected.
Imagine that all employees of an organization are accessible with institutional permission, but the roster provided for sampling accidentally omits recently hired employees. Those employees belong to the accessible population, yet they are missing from the frame.
Conversely, the frame might contain former employees who are no longer members of the target population.
Does Every Study Need a Sampling Frame?
No. The concept is especially central to probability sampling designs that select units from a defined frame, but not every research design uses a conventional list frame.
Some samples are generated through multistage area sampling, where frames exist at different stages. Other studies recruit through networks, purposively select cases, use respondent referrals, recruit volunteers online, or study hard-to-enumerate populations for which no complete list exists.
The absence of a conventional frame does not automatically make research invalid. It changes the available sampling strategies and, importantly, the basis on which population-level claims can be made.
AAPOR's best-practice guidance distinguishes probability surveys, where units are randomly selected from frames covering all or almost all of the population of interest, from non-probability approaches such as opt-in panels or recruitment through social media and personal networks.
What Should You Check Before Using a Frame?
Do not assume that an official-looking spreadsheet is adequate merely because it came from an authoritative office. Evaluate the relationship between the frame and your defined population.
Ask who should appear but may be missing, who appears but should not, whether anyone appears multiple times, whether relevant classifications are accurate, and whether the information is current enough for the study period.
If the frame has important limitations that cannot be corrected, document them. AAPOR's transparency standards call for probability-sample reporting to describe the sampling frame and its population coverage, including segments of the target population that the design does not cover.
