Does Your Question Require Evidence About Prevalence or Frequency?

Are You Trying to Find Out How Common Something Is or How Often It Happens?

Suppose you want to determine how common generative AI use is among university students. You collect responses from 300 students and discover that 210 report using AI tools for coursework.

Can you conclude that 70% of university students use generative AI?

Not necessarily. You first need to know which students were eligible, how they were selected, what counted as AI use, and the period covered by the question.

Another researcher might ask how often students use these tools. That question requires different information. Knowing whether a student used an AI tool at least once does not reveal whether they used it once during the semester or several times each day.

Research questions about prevalence and frequency may appear straightforward, but their credibility depends on defining precisely what is being counted, among whom, and over what period.

When Does a Research Question Require Prevalence or Frequency Evidence?

What Evidence Is Necessary to Estimate Prevalence or Frequency?

What Does Prevalence Actually Measure?

Prevalence describes the proportion of a specified population that has a particular characteristic or condition at a defined point in time or during a defined period.

In epidemiology, prevalence is commonly used to describe existing cases of a health condition. The underlying logic can also be applied to clearly defined characteristics or behaviors in educational and social research.

Examples include the proportion of teachers currently using AI-assisted lesson planning, the proportion of students reporting academic burnout during a semester, or the proportion of institutions with an established AI-use policy.

A prevalence estimate requires a numerator representing qualifying cases and a denominator representing the relevant population or sample.

The calculation establishes that 30% of the surveyed students met the stated criterion. Whether this estimates prevalence among all students at the university depends on the sampling design, coverage, nonresponse, measurement quality, and any appropriate weighting.

Even a mathematically correct percentage can be a poor population estimate if the denominator does not represent the intended population.

What Is the Difference Between Point Prevalence and Period Prevalence?

Prevalence estimates must have a clear temporal reference.

For example, the proportion of students currently using an AI writing assistant under a defined current-use criterion differs from the proportion who used one at least once during the previous semester.

Period prevalence may include occasional users who are no longer active. Point prevalence reflects a specified moment or status, although its interpretation still depends on the operational definition.

These measures should not be compared without considering their time frames and case definitions.

What Does Frequency Measure?

Frequency describes how often something occurs. It may refer to an absolute count, the number of events per person, or the number of events relative to an exposure or observation period.

Consider three questions:

• How many AI-generated feedback requests were submitted during September?
• How many times did each student use the system during the semester?
• How many feedback requests occurred per 100 student-weeks of observation?

Each question concerns frequency, but the required denominator and interpretation differ.

An absolute count may be appropriate when the total volume of activity is the research target. A per-person frequency is useful when examining how frequently individuals engage in a behavior. An event rate may be preferable when participants contribute different amounts of observation time.

Frequency therefore does not always require a population denominator, but comparisons across groups or periods often become misleading without one.

Why Prevalence and Frequency Are Not Interchangeable

Prevalence generally counts the proportion of individuals meeting a criterion. Event frequency may count repeated occurrences involving the same individual.

Suppose 100 students generate 1,000 AI feedback requests during a semester. Those 1,000 requests do not establish that 1,000 students used the system.

Perhaps only 20 students generated all the requests. In that case, 20% of the 100 students used the system, assuming complete records and a clearly defined use criterion, while the system recorded an average of 10 requests per student across the entire group.

Among the 20 users, the average would be 50 requests per user. These are different descriptive quantities.

A study must therefore specify whether it counts unique individuals, events, episodes, or repeated observations.

How Is Incidence Different From Prevalence?

Incidence concerns the occurrence of new cases during a defined period among individuals initially at risk. Prevalence concerns existing cases or the presence of a characteristic within a population.

In epidemiological research, cumulative incidence represents the proportion of initially at-risk individuals who develop a condition during a specified period, under appropriate follow-up conditions. An incidence rate uses person-time at risk as its denominator.

These distinctions can also be relevant to educational research when studying the onset of a clearly defined event, such as first-time academic withdrawal or first adoption of a technology.

The appropriate measure depends on the event definition and the research question. Recurrent events, changing risk status, and unequal follow-up may require additional analytical considerations.

Why the Denominator Matters as Much as the Numerator

A count becomes interpretable only when the population or opportunity structure is clear.

Suppose University A reports 400 students using an AI learning platform and University B reports 250.

University A might have 20,000 students, while University B has only 1,000. The respective proportions would be 2% and 25%, assuming the counts refer to comparable populations and observation periods.

University A has more users in absolute terms, but University B has a substantially higher proportion.

Neither result is inherently the more appropriate statistic. If the question concerns total service demand, counts may matter. If it concerns the extent of adoption across institutions, proportions may be more informative.

Researchers should specify the denominator before collecting data rather than selecting one afterward to make the findings appear more striking.

How Should You Define a Case or Event?

A prevalence or frequency estimate depends on a reproducible operational definition.

For AI use, possible definitions include having ever tried a tool, using it during the previous 30 days, using it at least weekly, or submitting at least one recorded request during the semester.

These definitions are not equivalent.

For example, a student who experimented with an AI tool once six months ago may qualify as an ever-user but not as a current weekly user.

A credible study should specify the criterion, its rationale, the reference period, and how qualifying cases or events are identified.

Where validated instruments or established definitions exist, researchers should consider them. Where new definitions are necessary, their implications for interpretation should be made explicit.

Why Sampling Determines Whether a Prevalence Estimate Can Be Generalized

Prevalence questions often concern a population larger than the individuals directly observed.

If the intended population is all undergraduate students at a university, surveying only students enrolled in an elective AI course may produce a distorted estimate of AI adoption.

Students in that course may be unusually interested in technology. Even if every enrolled student responds, the resulting percentage does not automatically represent the wider university.

Probability sampling can support design-based population estimation when properly implemented, although coverage errors, nonresponse, and measurement problems may remain.

Nonprobability samples may also provide useful descriptive information, but broader population claims generally require additional assumptions that should be stated and evaluated.

Sample size affects precision, but increasing sample size does not by itself eliminate systematic selection bias.

These issues are central when designing a prevalence study intended to estimate population characteristics accurately.

What Evidence Is Needed When Frequency Changes Over Time?

A question about frequency during one period differs from a question about how frequency changes.

If you want to know how many students used an AI tool during a semester, one appropriately defined observation period may be sufficient.

If you want to determine whether weekly usage increased throughout the semester, you need comparable observations across relevant time points.

Differences in measurement procedures, platform availability, academic schedules, or population composition may complicate interpretation.

Repeated observations can support evidence about change over time, but the design must distinguish genuine change from shifts in measurement or sample composition.

What About Uncertainty in Prevalence Estimates?

A sample prevalence estimate is generally subject to uncertainty when used to infer a wider population proportion.

Researchers may report confidence intervals or other appropriate uncertainty measures, taking account of the sampling design and assumptions.

For a simple random sample, the familiar binomial-based standard error depends on the estimated proportion and sample size. Complex sampling, clustering, weighting, finite population corrections, and other design features may require different methods.

Confidence intervals describe sampling uncertainty under a specified statistical framework. They do not automatically account for nonresponse bias, misclassification, or other systematic errors.

Consequently, a narrow interval should not be interpreted as proof that the estimate is accurate.

When Is Prevalence Evidence Insufficient?

Prevalence evidence is appropriate for estimating how common a characteristic is. It does not automatically explain why that characteristic occurs or whether an intervention caused it.

For example, finding that AI use is more prevalent among high-achieving students does not establish that AI use improves achievement.

The difference could reflect prior academic performance, access to technology, motivation, or other factors.

Questions about relationships or causal effects require additional evidence consistent with their intended inference.

This distinction reflects the broader need to match the type of evidence to the research question.

Estimating Student AI Adoption and Usage Frequency

The example demonstrates why prevalence and frequency should be calculated and interpreted separately. Both describe use, but they answer different questions.

Common Errors in Prevalence and Frequency Research

How to Decide Whether Your Study Needs Prevalence or Frequency Evidence

First, determine whether your question concerns the proportion of people with a characteristic or the number of times an event occurs.

If it concerns prevalence, define the eligible population, qualifying characteristic, reference period, and denominator.

If it concerns frequency, define the event, decide whether repeated occurrences count, and determine whether absolute counts or rates are more appropriate.

For a population-level study, pay particular attention to how accurately your design represents the target population.

For repeated observations, consider whether the research question requires a single period estimate or a design capable of describing patterns across time.

The essential principle is to decide what quantity you want to estimate before selecting the sampling strategy, instrument, or analytical procedure.

Before Estimating Prevalence or Frequency, Verify These Requirements

Frequently Asked Questions About Prevalence and Frequency Evidence

Prevalence Measures How Common a Characteristic Is; Frequency Measures How Often Events Occur

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.