You've found an article that sounds important. Maybe someone shared it in a parents' group. Maybe it came up in a class, a faculty meeting, or a conversation about a support program. It has an impressive title, an abstract, lots of numbers, and roughly forty words you've never seen before.
You can read it. Research papers aren't written to be mysterious, although they sometimes put in a tremendous amount of effort to look that way. Once you learn what to look for, you can ask useful questions without being a statistician.
First, what kind of article is this?
A published article isn't automatically an experiment, and an experiment isn't automatically good evidence for your particular question.
| What you're reading | What it usually does | What to watch for |
|---|---|---|
| Original study | Collects and analyzes new observations or measurements | Who participated, how they were selected, what was measured, and what was compared |
| Systematic review | Searches for and evaluates a defined body of research | Search strategy, inclusion rules, quality of the included studies, and whether their results are actually comparable |
| Meta-analysis | Statistically combines results across studies, when appropriate | Whether studies were comparable, and how much their results varied |
| Scoping review | Maps what research exists and where the gaps are | Useful for finding the landscape, not necessarily for choosing the most effective intervention |
| Qualitative study | Uses interviews, observations, documents, or other rich evidence to understand experiences and processes | Who was heard, how interpretation was conducted, and what perspectives may be missing |
| Theory or commentary | Develops an argument or explanation | The reasoning and assumptions. It may be valuable without being an effectiveness test |
| Professional guideline | Recommends actions based on a framework, consensus, or evidence review | Who wrote it, what evidence informed it, and whether it's current and appropriate locally |
Qualitative work isn't a failed experiment, and a thoughtful community essay isn't a failed journal article. They're answering different kinds of questions. Don't ask a study of people's experiences to supply a graduation-rate effect size, and don't ask a randomized trial to tell you how every participant felt about belonging.
A useful reading order
You do not have to start at the beginning and read every line to the end. Try this:
- Title and abstract: What question does the paper appear to answer? Write down the claim in your own words.
- Methods: Who was studied, what happened, and what was measured? This is where you find out whether the abstract means what you think it means.
- Results, including tables and figures: What did the researchers actually observe? Look for the numbers, not just the adjectives.
- Limitations: What could have influenced the findings, and who wasn't represented?
- Discussion and conclusion: How do the authors interpret the results? Do they stay close to what they measured?
- Introduction and references: Now go back for the context and earlier arguments, especially if something surprised you.
Reading the abstract twice is fine. The second time, you should know enough to notice which words are doing a lot of work.
Who was actually studied?
When a study says "students," ask which students. A small group of volunteers at one university is not the same thing as students attending Community Colleges across the country. Adults, teenagers, and children aren't interchangeable. Nor are people who requested formal accommodations and people who chose not to disclose a diagnosis.
Look for the sample size (how many people or cases), recruitment (how they were found), inclusion and exclusion criteria, and setting. Then ask who is missing. A study can be well done and still have a narrow answer.
Try it: what does this sample let us claim?
Suppose twelve volunteers at one college liked a new planning app. Is that evidence that the app is a reasonable candidate to test? Sure. Is it evidence that the app increases graduation nationwide? No. We haven't measured graduation, and twelve volunteers at one site can't settle a national question.
What did the researchers measure?
This is where a lot of confident advice falls apart. Researchers might measure student satisfaction, weekly attendance, self-reported stress, task completion, grades, or graduation. Those are all different outcomes.
If a program was designed to help students feel connected, a good belonging measure could be relevant. But a satisfaction survey given after two meetings doesn't establish that the program improved persistence two years later. Likewise, an improvement on a questionnaire isn't necessarily an improvement in everyday functioning. Find the operational definition, the researchers' specific way of turning an idea into something they can observe or measure.
Comparison is useful. Causation is harder.
If students who use tutoring earn better grades, tutoring might be helping. But perhaps the students who seek tutoring were already more motivated, had more time, or were receiving other support. This is called confounding: other explanations could help account for the pattern.
A randomized study tries to make comparison groups more alike by assigning participants by chance. That can strengthen a causal claim, but it does not fix small samples, poor measurements, missing data, or interventions that weren't delivered as planned. Observational and qualitative studies can answer important questions that randomization isn't suited to answer.
A relationship between two measures may be an important clue. It just isn't, by itself, proof that changing one will change the other.
The numbers: a short survival guide
A percentage needs a denominator. If a study says success improved by 50%, ask: from what to what, and among how many people? Going from two successful participants to three is a 50% increase. That sounds less dramatic once you know the count.
A p-value isn't the probability that a theory is true. In common statistical tests, it tells you how unusual the observed data (or something more extreme) would be under a particular statistical model assuming a null hypothesis. A small p-value can be evidence against that model. It doesn't tell you whether the effect is important, universal, or well explained.
An effect size asks how big the difference is. A statistically detectable result may be too small to matter in the classroom. Another study may find a sizeable difference but be too small to estimate it precisely.
A confidence interval expresses uncertainty in an estimate. The interval's exact interpretation depends on the method, but a wide interval is a reminder not to report an estimate as though it were carved in stone.
"Not statistically significant" doesn't mean "the groups are identical." It means the analysis did not find sufficient evidence of a difference by that test and threshold. A study might be underpowered, or the true difference might be very small. Also, finding similar outcomes on one task doesn't establish equivalence across all situations.
Read the axis labels and units. Does the vertical axis start at zero? Are we looking at percentages or counts? Are there error bars? Is it one average hiding very different participants? Look carefully before letting the colorful bars do the thinking for you.
What do the limitations tell you?
A limitations section is not a confession that the research is worthless. Good researchers tell you where the edges of their evidence are. Common issues include small or unusual samples, short follow-up, self-reported outcomes, missing comparison groups, changing conditions, selective participation, and difficulty separating an intervention's effects from everything else happening at the same time.
Don't stop at "small sample." Ask why that matters for the particular claim. A small, careful interview study can tell us a great deal about how a process feels. It probably won't tell us how common an experience is across the whole country.
What would make me change my mind?
This might be the single best question in research literacy. If a finding is surprising, what would you want to see next? A larger replication? A different population? Better measurements? Longer follow-up? A study designed with participants? A comparison with the ordinary service already available?
Another useful question: What would this paper have to show before I changed what I do? Sometimes the sensible response to a study is to change your practice. Sometimes it's to try a small pilot. Sometimes it's to read the next paper.
Read a real study with us
Our first guided reading walks through two linked papers on autistic and non-autistic information sharing. The 2020 study found an apparent disadvantage for mixed groups in a story-passing task. The larger 2025 study did not reproduce that specific difference. Neither paper needs to be declared a disaster for the science to be interesting. You get to examine the methods, results, and interpretations yourself.
Let's read the Crompton studies →
When you want to go a little deeper
The Critical Appraisal Skills Programme (CASP) has free checklists for qualitative studies, reviews, randomized trials, and several other study types. They're a useful next step once you've identified what kind of article you're reading.
The EQUATOR Network collects reporting guidelines such as PRISMA (systematic reviews), CONSORT (randomized trials), and STROBE (observational studies). Those help researchers report their work clearly. Following a reporting checklist doesn't automatically make a study well designed or its conclusions correct.
A few final questions to keep beside you
- What is the actual question, and what kind of evidence could answer it?
- Who was included? Who wasn't?
- What did the researchers measure rather than merely discuss?
- What did they find, and how precise is that finding?
- What else could explain it?
- How far does the conclusion reach beyond the results?
- Does this change what I would do, or does it give me a good next question?
Use the worksheet to collect your answers. And remember, "I don't know yet" is a legitimate scientific conclusion, not a personal failure.
Looking at the same problem from a different perspective?
Understanding neurodivergenceResearch and evidence