Direct vs. Indirect Evidence: When Are You Measuring the Thing You Actually Care About?

Are you measuring the phenomenon, or something that merely stands in for it?

You want to study student engagement, so you count learning-platform logins. You want to measure research impact, so you count citations. You want to know whether employees are productive, so you record hours at work. You want to determine whether students read an article, so you check whether they opened the file.

Each measure may contain useful information. Yet each also raises a methodological question: how closely does the evidence correspond to the phenomenon you actually care about?

Research frequently depends on indicators because many constructs cannot be observed directly. That is not inherently a problem. The problem begins when an indicator quietly becomes synonymous with the construct, and conclusions start saying more than the evidence can support.

Look at the inferential distance between the evidence and the claim

Every measure creates a link between what you observe and what you claim

Directness is about the relationship between evidence and the target

Researchers use terms such as direct measure, indirect measure, indicator, and proxy somewhat differently across disciplines. There is therefore no universal classification in which every research variable can be placed permanently into one of two categories.

A more useful way to think about directness is to ask how much inference separates the recorded evidence from the phenomenon named in the research question. The more the researcher must assume that one observable feature stands for another underlying phenomenon, the more indirect the evidentiary relationship becomes.

For example, if the question asks how many messages students posted in an online discussion, a system count of submitted messages may measure the target relatively directly, assuming the records are complete and appropriately defined. If the question asks how intellectually engaged those students were, message count becomes a much more indirect indicator. Posting more often is not identical to engaging more deeply.

The same data can therefore be relatively direct evidence for one question and indirect evidence for another. Directness belongs to the relationship between the evidence and the claim, not to the variable in isolation.

Start by identifying the thing you actually care about

Before deciding whether evidence is direct or indirect, specify the target phenomenon. This may be a behavior, outcome, experience, event, process, physical characteristic, or theoretical construct.

Some targets are comparatively observable. Attendance at a particular class meeting, completion of a specified task, number of published articles, or occurrence of a recorded event may be measured relatively directly if suitable records exist.

Other targets are latent constructs, meaning they cannot be observed directly in the same way. Motivation, anxiety, trust, cognitive load, engagement, organizational climate, and many psychological or social constructs require researchers to infer the underlying construct from observable responses or indicators.

This is why identifying what evidence the research question actually requires should precede the choice of a convenient indicator.

An indicator is not automatically the construct

Suppose a researcher defines online learning engagement using the number of clicks recorded in a learning-management system. Clicks are observable events. Engagement is a broader theoretical construct.

A student might click repeatedly because the interface is confusing. Another might download a reading once and spend an hour studying it offline. A third might leave a page open without attending to it. The platform may record the click accurately, but the inferential problem concerns what that click means.

Researchers should therefore be able to complete two different sentences: “We measured...” and “We interpret this as evidence of...” If those sentences contain different concepts, an inferential step exists and should be justified.

Indirect evidence is often unavoidable

Calling evidence indirect should not be treated as a methodological insult. Much of empirical research depends on measurement models that connect observable indicators to underlying constructs.

You cannot place motivation on a laboratory scale. Researchers may instead use responses to theoretically grounded and appropriately validated items. Similarly, socioeconomic status may be represented using income, education, occupation, or composite measures depending on the conceptualization and purpose of the study.

The methodological issue is therefore not whether an indicator is indirect, but whether the inferential connection is sufficiently defensible for the intended use. Evidence about proxy measures in applied research also shows why this relationship cannot simply be presumed: indirect measures may correspond imperfectly with more direct measures of the behavior they are intended to represent.

Ask what else could produce the same indicator

One practical way to interrogate a proxy is to ask: Could this measure change even if the underlying phenomenon did not?

If the answer is yes, identify why. Citation counts can change because of field size, publication age, visibility, controversy, or citation practices, among other influences, so they should not be treated as a direct measure of research quality. Time spent logged into a platform can increase because a browser tab was left open. Attendance can increase without learning increasing.

Then reverse the question: Could the underlying phenomenon change without this indicator changing? A student could become more deeply engaged while producing the same number of clicks. An employee could become more productive without spending additional hours at work.

These questions do not automatically invalidate the measure. They reveal the assumptions that connect it to the construct.

Direct evidence can still contain measurement error

Directness should not be confused with accuracy. A direct measure can be badly collected, incomplete, unreliable, or systematically biased.

If your question asks how many students attended class, an attendance record may be relatively direct evidence. But a malfunctioning card reader, inconsistent manual recording, or students scanning into a room and immediately leaving can still undermine the measurement.

Conversely, an indirect measure can be carefully validated and highly useful for its intended purpose. The direct-indirect distinction tells you something about the inferential relationship. It does not, by itself, rank the overall quality of the evidence.

Objective does not mean direct

Researchers sometimes assume that automatically recorded data must provide direct evidence because participants did not self-report them. These are separate dimensions.

A digital system may objectively record that a student clicked a video link at 10:03 a.m. Yet that click remains indirect evidence of whether the student watched attentively, understood the video, or learned from it.

Conversely, if the research question concerns perceived difficulty, asking a participant to report perceived difficulty may be relatively direct evidence of the target even though it is self-report.

This is why the choice between self-report and measures obtained independently of participants' reports should be considered separately from the question of directness.

Directness depends on how narrowly the target is defined

Broad constructs often create greater inferential distance. “Learning,” “engagement,” “success,” “well-being,” and “impact” contain multiple possible dimensions. A single indicator is unlikely to represent every dimension equally well.

A final examination score might provide relatively direct evidence of performance on the knowledge and skills sampled by that examination. Calling the same score a complete measure of “learning” requires a broader inference. Calling it a measure of “educational quality” requires a broader inference still.

Careful researchers therefore narrow the wording of claims as the inferential distance increases. Instead of writing that “platform engagement increased,” they may report that “recorded discussion-post activity increased” when that is what was actually measured.

Multiple indicators can help, but they do not automatically solve the problem

Complex constructs are sometimes represented using several indicators. Engagement, for example, might be examined through behavioral activity, self-reported cognitive engagement, participation, and other theoretically relevant measures.

This can provide a richer representation than relying on one convenient proxy. Yet multiple indicators still need justification. Combining several questionable proxies does not automatically produce a valid measure of the construct.

Likewise, collecting the same phenomenon through several sources may reveal agreement, complementarity, or disagreement. Whether this strengthens a conclusion depends on why those sources were selected and what each actually represents. Collecting evidence from more than one source should therefore be purposeful rather than ceremonial.

The same measure can be direct for one question and indirect for another

The data did not change across these questions. What changed was the inferential distance between the recorded event and the phenomenon the researcher wanted to claim had been measured.

Common mistakes when working with indicators and proxies

Make the inferential leap visible before you collect the data

For every important measure in your study, write down the target construct and the actual observation separately. If they are not the same, articulate the reasoning connecting them.

The final case is common in real research. Sometimes the evidence you would ideally collect is not realistically available. A proxy may then be entirely reasonable. The methodological obligation is not perfection. It is transparency about what the proxy captures, why it is defensible, and what remains uncertain.

Before treating an indicator as evidence of your construct

Frequently asked questions about direct and indirect evidence

Measure what you claim, or explain the inference

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.