Every study is built to answer a particular kind of question. Some are good at measuring how common something is. Some can show what tends to come before what. A few can test whether one thing causes another. When a headline overreaches, the mismatch is usually between what the design can do and what the story claims it did.
This guide walks through the designs you'll meet most often in research about children and teenagers, roughly from the weakest kind of evidence for cause and effect to the strongest. "Weakest" here doesn't mean useless. Each design has a job it does well.
Case reports and anecdotes
A case report describes one child or a handful of children in detail. Doctors publish them when they see something unusual, such as a rare side effect or an unexpected recovery.
Case reports are valuable as early warnings. They can't tell you how often something happens, because there is no count of all the children who didn't have the problem. Personal stories in the news work the same way. They can be true and moving, and still tell you nothing about what is typical.
Animal and laboratory studies
Many findings about development start in mice, rats or cells in a dish. These studies allow experiments that would never be acceptable with children, and they help explain mechanisms.
The catch is translation. A rodent's brain develops on a different timeline, in a very different life. Plenty of results from animal studies don't hold up in humans. When a news story says a substance "harms the developing brain", check whether the brain in question belonged to a child.
Cross-sectional surveys
A cross-sectional study measures a group once, at a single point in time. National surveys asking thousands of teenagers about sleep, screen use and mood are typical examples.
They're good at describing how common something is and which things go together. They can't tell you which came first. If teenagers who sleep less also report lower mood, the survey alone can't say whether poor sleep lowers mood, low mood disrupts sleep, or something else, such as stress at home, affects both. The correlation guide goes into this in detail.
Case-control studies
A case-control study starts with children who have a condition (the cases) and compares them with similar children who don't (the controls), looking back for differences in their past.
This design is efficient for rare conditions, because you don't have to follow thousands of children to find a few cases. Its weakness is memory and selection. Parents of a sick child may remember past exposures differently from other parents, and choosing the right control group is hard.
Cohort and longitudinal studies
A cohort study follows the same group of children over time, sometimes from birth into adulthood. Several countries run large national birth cohorts that have followed thousands of children for decades.
Because measurements are taken before outcomes happen, cohort studies can show what comes first. That makes them much stronger than one-off surveys for questions about development. They still can't rule out every alternative explanation. Families who differ in one way, such as how much they read aloud, often differ in others, such as income or parents' education. Good cohort studies measure and adjust for these differences, but they can only adjust for what they measured.
Cohort studies also lose people over time. If the families who drop out differ from those who stay, results can drift.
Randomized controlled trials
In a randomized controlled trial (RCT), children are assigned to groups by chance. One group gets the intervention, such as a reading program, a medicine or a sleep routine, and the other gets something else or nothing.
Random assignment is what makes trials powerful. With enough children, the groups end up similar in every way, including ways nobody measured. So if outcomes differ at the end, the intervention is the most likely reason.
Trials have limits too. They often run for weeks or months, not years. They may enroll children who are easier to recruit than average. And many questions can't be tested this way at all: no one can randomly assign children to poverty, bullying or a parent's illness. In education, trials often randomize whole classrooms or schools rather than individual pupils, which is fine, but the number of schools matters as much as the number of children.
Qualitative studies
Qualitative research uses interviews, focus groups, observation or written accounts. A study might interview 25 teenagers about how they decide what to post online, or observe playground games in four schools.
These studies answer "how" and "why" questions that numbers can't: what an experience feels like, what reasons people give, what researchers hadn't thought to ask. Samples are small by design, and that's appropriate. What qualitative studies can't do is tell you how common a view is in the wider population.
Systematic reviews and meta-analyses
A systematic review gathers all the studies on a question using a search plan written in advance, then assesses and summarizes them. A meta-analysis goes a step further and combines the numbers from several studies into one overall estimate.
Done well, these sit at the top of most evidence rankings, because they show the whole body of research rather than one result. Two cautions apply. First, a review is only as good as the studies it includes: pooling ten weak studies gives you a precise estimate of a weak answer. Second, studies with exciting results are more likely to be published, which can tilt a review toward finding effects. Good reviews check for this.
A quick comparison
| Design | Good for | Main limit |
|---|---|---|
| Case report | Spotting something new | Can't show how often or why |
| Animal or lab | Mechanisms, early testing | May not apply to children |
| Cross-sectional survey | How common, what goes together | Can't show what came first |
| Case-control | Rare conditions | Memory and choice of controls |
| Cohort | What comes first, long-term outcomes | Other differences between families |
| Randomized trial | Testing cause and effect | Short, selective, not always possible |
| Qualitative | Experiences and reasons | Not about how common |
| Systematic review | The overall picture | Depends on the studies included |
Using this on a real study
When you read about a new study, find the design first. Then ask whether the claim fits it. A survey that "shows" an app damages mental health is claiming more than a survey can. A trial that "suggests" a reading program might help is, if anything, underselling itself.
The study checklist starts with the design for this reason. The next guides look at the two questions that come straight after it: correlation and causation, and sample size.
Questions readers ask
- Is a randomized trial always better than an observational study?
- For testing whether something causes an effect, usually yes. But many important questions about children can't be tested in a trial, because it would be unethical or impossible to assign children to, say, poverty or divorce. For those, well-designed cohort studies are the best evidence available.
- What is the difference between a review and a systematic review?
- A narrative review is an expert's summary of a field, and the author chooses which studies to discuss. A systematic review sets out in advance how it will search for studies and which it will include, so readers can check whether anything was left out.
- Where do I find out which design a study used?
- The abstract usually names it in the first or second sentence of the methods part. Look for words like randomized, cohort, longitudinal, cross-sectional, case-control, qualitative, systematic review or meta-analysis.