Description Is Not Interpretation: The Difference That Changes Data Analysis

A dataset can be analysed correctly and still produce a weak piece of research.

The calculations may be accurate.

The charts may be clear.

The statistical tests may have been chosen appropriately.

The qualitative themes may have been coded carefully.

Yet the results section can still leave the reader asking:

So what?

That happens when analysis stops at description.

Description tells us what appears in the data.

Interpretation asks what that evidence means in relation to the research question.

The difference sounds small.

In practice, it separates merely reporting results from actually making sense of them.

Description answers “what?”

Imagine a survey of 500 university students.

You discover that:

62% report using generative AI for academic work.

That is a finding.

It describes something in the dataset.

Or perhaps:

Students who reported greater weekly study time achieved higher average examination scores.

Again, this describes a pattern.

Or in qualitative research:

Participants frequently described uncertainty about whether AI-generated references were genuine.

This too is descriptive.

None of these statements is useless.

Description is necessary.

But none of them, by itself, completes the analysis.

The researcher still has to ask:

What should we understand from this?

Interpretation answers “what does this tell us?”

Take the finding:

62% of students use generative AI for academic work.

Several interpretations might initially seem possible.

Perhaps AI has become normalised within students’ academic practice.

Perhaps students use it only for low-risk tasks such as brainstorming.

Perhaps institutional policies have not kept pace with actual student behaviour.

Perhaps usage is concentrated within particular disciplines.

Perhaps students define “using AI” very differently.

The percentage alone cannot decide among these interpretations.

The researcher has to connect the evidence to:

the research question,

other findings,

the way variables were measured,

the context,

the literature,

and the limitations of the study.

Interpretation is therefore not simply adding an opinion after the statistics.

It is reasoning from evidence.

What are we actually justified in concluding?

That final stage—Bound—is particularly important.

Good interpretation does not only explain what evidence might mean.

It also recognises what the evidence cannot establish.

Numbers do not interpret themselves

Consider this statement:

Average satisfaction increased from 6.1 to 7.4 after the programme.

The numerical difference is clear.

But several questions remain.

Was the change statistically meaningful?

Was it practically important?

Were the same people measured before and after?

Could other events have influenced satisfaction?

Was the scale valid?

Did all groups improve similarly?

Was there a control group?

The number does not answer those questions automatically.

Data becomes evidence only within a research design.

And evidence becomes meaningful only through interpretation.

Statistical significance is not the interpretation

A common version of descriptive reporting looks like this:

There was a statistically significant relationship between X and Y (p < .05).

That tells us something about the statistical result.

It does not tell us everything the reader needs to know.

We might also need to understand:

How strong is the relationship?

In which direction does it operate?

Is the effect large enough to matter practically?

Does the relationship survive alternative explanations?

Does the research design allow causal conclusions?

A very small relationship can be statistically significant in a large dataset.

A potentially important difference can fail to reach significance in a very small sample.

The p-value is part of the evidence.

It is not the meaning of the evidence.

If the answer is no, the discussion may be relying too heavily on statistical vocabulary instead of analytical reasoning.

Association is not explanation

Suppose we find:

Students who spend more time on social media tend to report lower academic performance.

The pattern may be real.

But what explains it?

One possibility is that greater social media use reduces study time.

Another is that students who are already disengaged from their studies use social media more often.

A third variable—such as stress, sleep or socioeconomic conditions—might influence both.

Or the measurement itself may be imperfect.

If the study is observational, the data may show an association without telling us which causal explanation is correct.

Weak interpretation says:

Social media causes poorer academic performance.

Stronger interpretation says:

Higher social media use was associated with lower reported academic performance, but the study design does not establish whether social media use itself produced the difference.

The second conclusion may sound less dramatic.

It is also more defensible.

Good interpretation often requires resisting the strongest claim

Researchers naturally want their findings to matter.

That can create pressure to turn:

“is associated with”

into:

“causes”

or:

“participants reported”

into:

“people experience”

or:

“this sample showed”

into:

“students in general are…”

But a stronger-sounding conclusion is not necessarily a stronger research conclusion.

The quality of interpretation depends partly on whether the claim remains proportional to the evidence.

Sometimes intellectual discipline means saying less.

Context changes what a finding means

Suppose a study finds that only 40% of students regularly attend optional research workshops.

Is that high or low?

Without context, we cannot know.

Perhaps previous participation was 15%.

Perhaps similar universities average 70%.

Perhaps attendance is optional and the workshops occur during examination season.

Perhaps the institution has only recently introduced them.

The same number can support very different interpretations depending on context.

This is why researchers should be cautious with adjectives such as:

high, low, substantial, poor, strong, weak, significant

unless the basis for those judgments is clear.

Qualitative research has the same problem

The distinction between description and interpretation is not limited to statistics.

Consider interview data.

Several participants say:

“I don’t trust references generated by AI.”

A descriptive qualitative finding might be:

Participants expressed concern about AI-generated references.

Useful—but still limited.

Interpretation might go further:

Participants’ concerns appeared to centre less on AI use itself than on their inability to judge when apparently credible outputs were factually unreliable.

Now the researcher is identifying a pattern in meaning.

Or perhaps:

Participants often responded to uncertainty by avoiding AI-generated references entirely rather than developing systematic verification practices.

That is interpretive.

It moves from what people said toward what their responses reveal about behaviour, judgment or experience.

Interpretation is not invention

This is the other danger.

Researchers sometimes believe that interpretation means they are free to speculate.

They are not.

An interpretation must remain anchored to evidence.

Suppose interview participants never discuss institutional policy.

It would be difficult to conclude:

Students’ behaviour is primarily caused by university AI policy.

That may be plausible.

But plausibility is not enough.

The interpretation needs support from the collected evidence or a clearly justified theoretical argument.

Interpretation extends beyond description.

It does not extend beyond what the evidence can reasonably support.

Use another SRS Framework here

Create:

Tables are not the analysis

A results chapter sometimes becomes a sequence of tables followed by sentences that repeat the same numbers.

For example:

Then the text says:

48% agreed, 27% were neutral and 25% disagreed.

The paragraph has technically described the table.

But the reader already had that information.

The researcher should instead ask:

What is noteworthy?

Is opinion clearly divided?

Does the neutral group matter?

Does this pattern differ across subgroups?

How does it compare with expectations or previous research?

What does it contribute to the research question?

Writing should add analytical value rather than simply translate tables into sentences.

The biggest number is not always the most interesting finding

Researchers often focus automatically on the majority response.

But minorities can matter.

Suppose:

78% of respondents report no difficulty accessing an online system.

It may seem natural to conclude that access is not a major problem.

But perhaps the remaining 22% consist disproportionately of rural students, low-income students or people with disabilities.

Now the minority becomes analytically important.

Aggregate numbers can conceal patterns.

Interpretation asks not only:

What is most common?

but also:

For whom does the pattern differ?

Unexpected findings deserve attention

Researchers sometimes treat an unexpected result as a problem to explain away.

But unexpected findings can be analytically valuable.

Suppose your hypothesis predicts a positive relationship and none appears.

Do not immediately assume the research failed.

Perhaps:

the theory does not apply in this context,

the measure captured the concept poorly,

another variable moderates the relationship,

the sample differs from earlier research,

or the assumed relationship simply is not as robust as expected.

Research is not successful only when the data confirms our expectations.

Sometimes the most useful result is the one that forces us to reconsider them.

The literature helps interpretation—but should not overpower the evidence

A common discussion-section formula is:

Our finding agrees with Author A (2022), Author B (2023) and Author C (2024).

Agreement with previous research is useful.

But interpretation should go further.

Why might the findings agree?

Are the populations comparable?

Were the concepts measured similarly?

Does the current study extend the earlier evidence?

Does it reveal a mechanism previous research did not examine?

And if your result differs from previous research, that difference may be even more interesting.

The literature provides context.

It should not dictate what your own evidence is allowed to show.

Theory can deepen interpretation

Suppose students with high levels of peer support are more likely to persist in a difficult academic programme.

Description tells us:

Peer support and persistence are positively associated.

Theory might help us ask:

Does peer support increase belonging?

Does it provide practical assistance?

Does it reduce stress?

Does it reinforce academic identity?

A theoretical framework can transform a statistical relationship into a more developed explanation.

But theory should illuminate the evidence—not be imposed regardless of what the evidence shows.

Interpretation should answer the research question

One useful discipline is to return constantly to the question that produced the study.

If the research question asks:

How do postgraduate students evaluate AI-generated information?

and the analysis reports:

frequency of AI use,

preferred AI tools,

hours spent online,

general attitudes toward technology,

and demographic characteristics,

the researcher may have produced many descriptive results without answering the central question.

Data analysis can easily become distracted by whatever variables happen to be available.

The research question should decide which patterns deserve analytical attention.

Interpretation should separate result from explanation

Researchers should distinguish between:

what the data directly shows

and

why they think that pattern occurred.

For example:

Result:
Students who attended more research-support sessions completed their dissertations sooner.

Possible interpretation:
Research-support sessions may have helped students resolve methodological problems earlier.

That explanation may be sensible.

But other explanations might exist.

Perhaps more motivated students were both more likely to attend support sessions and more likely to finish quickly.

A strong discussion makes this distinction visible.

It does not disguise explanation as observation.

The same evidence can sometimes support competing explanations

Imagine employees working from home report greater job satisfaction.

Possible explanations include:

greater autonomy,

less commuting,

more flexible working hours,

better work-life balance,

or differences in who chooses remote work.

Good interpretation considers alternatives.

This does not mean every discussion must become endlessly uncertain.

It means the researcher demonstrates awareness that findings exist within a network of possible explanations.

Where the research design can rule alternatives out, say so.

Where it cannot, acknowledge them.

Interpretation includes limitations

Limitations are not an embarrassing appendix added at the end of otherwise confident claims.

They are part of interpretation.

If your sample excludes an important population, that affects what the findings mean.

If measurement is imprecise, that affects how strongly relationships can be interpreted.

If data is cross-sectional, that affects conclusions about change or causality.

If interviews were conducted in a particular institutional setting, that affects how widely experiences can be generalised.

Understanding limitations helps determine the boundary of the claim.

A defensible conclusion is better than an impressive conclusion

Compare:

This study proves that AI improves student research ability.

with:

Students who reported more frequent use of AI tools also reported greater confidence in several research tasks, although the cross-sectional design does not establish whether AI use produced that confidence.

The first sounds powerful.

The second tells us much more accurately what the evidence permits.

Research credibility comes from the second kind of sentence.

That question belongs beside almost every analysis.

It forces us to distinguish between:

what we observed,

what we infer,

what we suspect,

and what we can defend.

Analysis is a movement from data to argument

The analytical process can be thought of as a progression.

And at every stage: what uncertainty remains?

That final question matters because research evidence almost never eliminates uncertainty completely.

Its purpose is to reduce uncertainty responsibly.

Data does not speak for itself

Researchers sometimes say:

“Let the data speak.”

But data never quite speaks on its own.

Someone decided:

what to measure,

whom to include,

which questions to ask,

how to code responses,

which comparisons to make,

which model to run,

which themes to identify,

and which findings deserved attention.

Interpretation is unavoidable.

The responsibility of the researcher is therefore not to pretend interpretation does not exist.

It is to make that interpretation transparent, logical and proportionate to the evidence.

The purpose of analysis is understanding

Good data analysis does not end with:

42%.

or:

p < .05.

or:

Theme 3: Lack of institutional support.

Those may be important results.

But the researcher still has work to do.

What does the result contribute to the question?

How does it alter what we understand?

How does it relate to existing evidence?

What explanation is most plausible?

What alternatives remain?

What can we conclude?

What can we not conclude?

That movement—from observation toward defensible understanding—is interpretation.

And without it, even technically perfect analysis remains incomplete.