How Do You Design a Study to Describe Patterns Over Time?

How Can You Design a Study That Describes Change Over Time Accurately?

Suppose a university wants to understand how students' use of generative artificial intelligence has changed over several academic years. The institution has survey results from different periods, but the questionnaires were administered to different students, some questions changed, and participation rates varied.

Can these observations establish a meaningful trend? Or might the apparent increase in AI use reflect differences in measurement, respondents, or the composition of the student population?

Describing patterns over time requires more than arranging percentages chronologically. Researchers must decide what kind of change they want to observe, which units or populations should be measured, how frequently observations should occur, and whether the measurements remain comparable.

The central methodological problem is distinguishing changes in the phenomenon from changes introduced by the research process itself.

Use Repeated, Comparable Measurements Matched to the Type of Change You Want to Describe

Designing Descriptive Research That Captures Meaningful Temporal Patterns

First, Define What You Mean by a Pattern Over Time

A temporal pattern is a systematic or observable feature of how a measured characteristic varies across time. It may involve an increase, decrease, fluctuation, period of stability, seasonal variation, or an apparent change in the direction or rate of change.

For example, researchers may want to describe the annual proportion of students using generative AI, monthly variation in learning management system activity, changes in teachers' digital competencies, or the distribution of research topics across publication years.

These questions share a temporal dimension, but they do not necessarily require the same research design.

A study examining annual institutional enrollment counts describes a population-level series. A study following individual students' digital competencies across four semesters describes within-person development. A study comparing successive cohorts of newly hired teachers describes changes across cohorts rather than the development of the same individuals.

Before selecting the design, determine the kind of evidence needed to establish change over time.

Distinguish Population Trends From Individual Change

One of the most consequential decisions is whether the research question concerns changes in a population's overall characteristics or changes experienced by particular individuals or units.

Suppose the proportion of students reporting weekly AI use rises from 40% to 65% over three years. This may represent a genuine population-level increase, assuming comparable population definitions, measurement, and estimation procedures.

However, the increase does not show that any particular student changed from nonuse to use. The students observed in the first and final years may be different individuals.

Conversely, following the same students can reveal transitions in their behavior, but the resulting cohort may no longer represent the entire university population as new students enroll and others graduate.

Neither perspective is inherently superior. Their usefulness depends on the question being asked.

Choose an Appropriate Temporal Research Design

Several designs can support descriptive research over time. They differ in which units are observed and what kinds of change can be identified.

The labels may overlap. For example, an aggregate time series may be constructed from repeated cross-sectional surveys or administrative records. A cohort can also be followed longitudinally.

The essential question is not which label sounds most sophisticated, but whether the observations support the intended description. The distinction between repeated cross-sectional and longitudinal studies becomes particularly important when researchers intend to interpret individual change.

How Many Time Points Are Needed to Describe a Trend?

Two comparable observations can establish an observed difference between two periods. They cannot adequately characterize more complex temporal behavior between those periods.

With three or more observations, researchers can begin examining whether change appears consistent, irregular, or nonlinear. However, no universal minimum number of observations guarantees a reliable trend analysis.

The number of measurements should reflect the phenomenon's expected rate of change, the length of the observation period, and the intended analytical resolution.

For example, annual measurements may be suitable for describing changes in institutional enrollment. They may be too infrequent to characterize weekly learning engagement or short-lived fluctuations in technology use.

Seasonal patterns generally require observations across repeated seasonal cycles. Detecting abrupt changes or fitting complex time-series models can require substantially more observations than simply describing annual differences.

Researchers should avoid claiming that a stable trend exists from a small number of observations when the intervening pattern remains unknown.

Choose Observation Intervals Based on the Phenomenon

The spacing of measurements influences which temporal patterns can be observed.

Short intervals can capture rapid fluctuations but may increase data collection costs and measurement burden. Long intervals reduce collection frequency but may conceal temporary changes, reversals, or seasonal variation.

Consider a study of learning management system activity. Daily measurements might reveal weekday and weekend patterns. Monthly averages could conceal those differences while remaining useful for describing broader institutional activity.

Observation intervals should therefore be selected according to the research question and the temporal dynamics of the characteristic.

The timing of data collection also matters because academic calendars, examination periods, holidays, and institutional events may influence the observed values.

Researchers should distinguish the measurement date from the reference period. A survey administered in November may ask about behavior during the preceding seven days, the current semester, or the entire academic year. These definitions should remain consistent when results are compared.

Maintain Comparable Measurements Across Periods

Temporal comparisons require measurements that represent sufficiently similar constructs across observations.

If a university asks students in one year whether they have ever used AI, but asks students in another year whether they used AI during the preceding week, the resulting percentages do not measure the same characteristic.

Comparability can also be affected by changes in response categories, survey administration mode, scoring procedures, population eligibility, or the meaning of the construct itself.

When established scales are used repeatedly, longitudinal measurement invariance may be relevant. It concerns whether a measure operates sufficiently consistently across time to support the intended comparisons.

For rapidly changing technologies, identical wording does not necessarily guarantee identical meaning. A question about AI use may capture different tools or practices as the technology develops.

Researchers should preserve a stable core of measures where possible, document revisions, and evaluate whether any changes compromise comparisons. If a measurement break cannot be reconciled, the affected observations should not be presented as a continuous equivalent series without qualification.

Account for Changes in Population Composition

Population-level trends can be influenced by changes in who belongs to the population.

Suppose postgraduate students report higher AI adoption than undergraduates. If the proportion of postgraduate students increases over time, the university-wide adoption percentage may rise even if usage within each student category remains unchanged.

Researchers should therefore consider whether population composition has changed across measurement periods.

Stratified reporting can reveal whether a trend occurs within relevant subgroups. Standardization or appropriately constructed weights may help distinguish compositional change from changes in subgroup-specific characteristics.

These adjustments require reliable population information and explicit assumptions. They should not be treated as automatic corrections for every source of temporal bias.

For repeated surveys, the principles of accurate population description apply separately to every observation period.

What Happens When Participants Drop Out of a Longitudinal Study?

Longitudinal studies introduce the possibility of attrition, meaning that some participants stop contributing observations over time.

If participants who withdraw differ systematically from those who remain, the observed trajectories may not represent the original cohort.

For example, students experiencing severe academic difficulties might be less likely to complete later surveys. An apparent improvement in average academic engagement could partly reflect the selective loss of participants with lower engagement.

Researchers should document retention, examine available characteristics associated with missing observations, and choose analytical procedures appropriate to the missing-data assumptions.

Mixed-effects models, multiple imputation, or inverse-probability weighting may be useful in particular circumstances, but none eliminates bias without defensible assumptions.

Attrition should be considered during study planning, including decisions about recruitment, follow-up procedures, participant burden, and the required number of initial observations.

How Should Temporal Patterns Be Summarized?

Descriptive temporal analysis should begin with the actual values observed at each period. Depending on the characteristic, researchers may report counts, proportions, rates, means, medians, distributions, or other appropriate summaries.

Absolute change and relative change communicate different information.

For multiple periods, researchers may examine period-to-period changes, cumulative changes, or estimated trends. The chosen summary should correspond to the measurement scale and substantive question.

Counts may need to be converted into rates or proportions when the size of the population changes. An increase in the number of students using a service may reflect enrollment growth rather than an increase in the proportion of students using it.

When Are Trend Lines and Statistical Models Appropriate?

Graphs can help researchers examine temporal patterns, but the visual presentation should not imply more certainty than the data support.

A line graph may be suitable for ordered observations, provided the time axis is correctly scaled and the plotted values represent comparable measurements.

Statistical trend models can summarize patterns, estimate rates of change, or quantify uncertainty. However, a fitted linear trend assumes a particular functional form that may not describe abrupt shifts, nonlinear behavior, or seasonal variation adequately.

Repeated observations may also be correlated. Time-series methods should account for relevant serial dependence when statistical inference is intended.

Smoothing techniques, moving averages, and flexible trend models can help reveal patterns, but they may conceal short-term variation or create misleading impressions when applied without justification.

Researchers should distinguish observed data points from fitted or smoothed values and explain the analytical assumptions used.

Can a Descriptive Trend Establish the Effect of an Intervention?

Not by itself. A change occurring after an intervention does not demonstrate that the intervention caused the change.

Suppose student engagement increases after a university introduces a new learning platform. The increase might reflect the platform, but it could also reflect changes in course requirements, student composition, assessment schedules, or other concurrent developments.

A descriptive study can document the timing and magnitude of the observed change. Estimating the intervention's causal effect requires an appropriate identification strategy and consideration of what would have happened without the intervention.

Interrupted time-series designs, comparison series, or other quasi-experimental approaches may be suitable in particular settings, but their validity depends on assumptions that extend beyond descriptive trend analysis.

The distinction between describing whether change occurred and explaining why it occurred should remain explicit.

How Should Researchers Report Temporal Limitations?

Transparent reporting should identify the observation periods, population definitions, sampling or data sources, measurement instruments, missing observations, and analytical procedures.

Researchers should disclose changes in instruments, eligibility criteria, data coverage, or institutional context that could affect comparability.

For repeated cross-sectional surveys, the reporting should clarify that observations concern successive population samples rather than individual trajectories. For longitudinal panels, the report should address attrition and the handling of repeated observations.

Uncertainty estimates should reflect the sampling design and dependence structure where relevant. If the data represent a complete administrative series, sampling uncertainty may be less relevant than measurement completeness and the stability of definitions.

The purpose is to make clear which temporal conclusions the observations support and which remain uncertain.

Tracking Generative AI Adoption Across Four Academic Years

Had the university instead followed the same students over four years, it could have examined individual transitions and trajectories. That design would answer a different question and introduce additional concerns about attrition and cohort representativeness.

Common Errors in Descriptive Studies of Temporal Patterns

How to Choose the Right Design for Describing Change Over Time

Begin by identifying whether the study concerns population distributions, individual trajectories, cohort differences, or aggregate institutional indicators. The design should preserve the information necessary for the intended temporal comparison.

Next, determine how frequently meaningful change could occur and whether reliable measurements can be collected at those intervals. Consistency of definitions may be more valuable than increasing measurement frequency without a clear purpose.

When planning a new study, it may be useful to prepare a temporal measurement protocol specifying observation dates, reference periods, instruments, population definitions, and analytical summaries.

For existing datasets, begin with a comparability audit. A visually compelling trend is of limited value if its observations do not represent sufficiently similar constructs or populations.

Checklist for Designing Descriptive Studies Over Time

Frequently Asked Questions About Describing Patterns Over Time

Meaningful Temporal Patterns Require Comparable Evidence

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.