External validity is the extent to which a study's results hold beyond the study itself: for other people, in other settings and at other times. Internal validity asks a narrower question that comes first: did the thing the researchers changed actually cause the result they measured? A strong research claim needs both, and the design choices that improve one often weaken the other.
Below you'll find both definitions, a comparison table, the main threats to each with fixes, two real research cases and a worked example you can reuse when you critique a study or write a lab report's limitations section.
External validity is how well a study's findings generalize beyond the specific sample, setting and moment in which the study was run. Donald Campbell and Julian Stanley, in their 1963 survey Experimental and Quasi-Experimental Designs for Research, framed it as the question of generalizability: to what populations, settings, treatment variables and measurement variables can this effect be generalized?
William Trochim's Research Methods Knowledge Base entry on external validity breaks that question into people, places and times. When you judge a study's external validity, ask whether the result would hold:
Trochim also describes two ways researchers justify a generalization. The sampling model starts by defining the target population and drawing a fair sample from it, so the results can be extended to that population. The proximal similarity model is the fallback when a fair sample isn't possible: you generalize to people, places and times that closely resemble the ones you studied, and you become more cautious as contexts become less similar.
Population validity is the part of external validity that concerns people: can you generalize from your sample to the wider population you care about? It depends on who the target population is and how the sample was drawn. A random sample of all US undergraduates supports much broader claims than 80 volunteers from one psychology department's participant pool.
Ecological validity is how closely a study's tasks, materials and setting resemble the real-life situation you want to explain. Memorizing a list of unrelated words on a lab computer is a weak model of studying for a real exam. Reaction times measured in a driving simulator tell you less about real driving than reaction times measured on a road.
Ecological validity and external validity overlap, but they aren't the same thing. A realistic field study at one company may not generalize to other companies, and an artificial lab task can produce a finding that replicates across many populations. A related distinction separates mundane realism (does the study look like everyday life?) from psychological realism (does it trigger the same thoughts and motives as everyday life?). Many methodologists argue that psychological realism matters more for generalization.
Qualitative researchers usually talk about transferability instead of external validity. Rather than claiming statistical generalization, they describe the setting and participants in enough detail that readers can judge whether the findings would carry over to their own context.
Internal validity is the degree to which a study shows that the independent variable, and not something else, caused the change in the dependent variable. Campbell and Stanley called internal validity "the basic minimum without which any experiment is uninterpretable." If a study can't show the effect is real within its own sample, there's nothing to generalize.
Trochim's guide to establishing cause and effect lists three conditions a causal claim has to meet:
The third condition is where most studies struggle, and it's what the threats to internal validity below are about. A randomized experiment handles it best because random assignment spreads differences between people across groups by chance, including differences nobody thought to measure. Correlational and observational studies can show the first two conditions but usually can't fully rule out confounding variables.
Trochim treats internal validity as one of four validity types, alongside conclusion validity (is there really a relationship?), construct validity (did you measure the right concepts?) and external validity. Don't confuse any of them with reliability in research design, which is a separate idea: reliability is about consistent measurement, while validity is about whether conclusions are justified.
Internal vs external validity comes down to two different questions: internal validity asks whether the effect is real in this study, and external validity asks how far that effect travels. The table compares them point by point.
| Feature | Internal Validity | External Validity |
|---|---|---|
| Core question | Did X cause Y in this study? | Would X cause Y for other people, places and times? |
| Main concern | Alternative explanations and confounding variables | How representative and realistic the study is |
| Key design tool | Random assignment to conditions | Random (probability) sampling from the target population |
| Classic threats | History, maturation, testing, instrumentation, regression, selection, attrition | Unrepresentative samples, artificial settings, pretest sensitization, reactions to being studied |
| Usually strongest in | Controlled lab experiments | Field studies and large probability-sample surveys |
| Main ways to improve | Control groups, blinding, standard protocols, low attrition | Diverse samples, field settings, replication across contexts |
| What it can't tell you | Whether the effect applies outside the study | Whether the effect was causal in the first place |
Random sampling and random assignment sound alike but protect different kinds of validity. Random sampling (also called random selection) decides who gets into your study, so it supports external validity. Random assignment decides which group each participant joins, so it supports internal validity. Trochim's explainer on random selection and assignment draws the same line, and Campbell and Stanley noted back in 1963 that the main purpose of randomization in lab experiments is internal validity, not external.
A study can use one, both or neither. A typical psychology experiment randomly assigns volunteers from one university's participant pool, which gives it good internal validity but leaves its population validity open to question. A national survey that randomly samples adults has no random assignment at all, so it describes the population well but can't prove cause and effect by itself.
Threats to internal validity are alternative explanations that could produce your result even if the treatment did nothing. Campbell and Stanley listed eight classic ones, and later textbooks added a few more. The table uses one running example: a university offers six weeks of peer tutoring to students who scored in the bottom 10% on the first statistics exam, then compares their final exam scores with their first exam scores.
| Threat | What It Means | How It Could Fool You in the Tutoring Study | How to Reduce It |
|---|---|---|---|
| History | An outside event between measurements affects the outcome | The instructor adds a review session before the final | A control group that experiences the same events |
| Maturation | Natural change over time, such as growing older, more tired or more experienced | Students adjust to college and study better by the end of term anyway | A control group measured over the same period |
| Testing | Taking one test changes scores on a later test | Students learn the instructor's question style from the first exam | A control group, or a design with no pretest |
| Instrumentation | The measuring tool, or the people scoring it, changes | The final is easier, or a new grader scores more leniently | Equivalent test forms and trained, blind scorers |
| Statistical regression | Extreme scores tend to move toward the average on retesting | Students picked for very low scores were likely to score higher next time, partly because bad luck pulled the first score down | Compare with a control group selected the same way |
| Selection | Groups differ before the treatment starts | If tutoring is optional, the students who sign up may be more motivated | Random assignment |
| Attrition (experimental mortality) | Participants drop out, especially unevenly across groups | The weakest students withdraw from the course, raising the average of those left | Track dropouts and analyze everyone originally assigned |
| Selection-maturation interaction | Groups that differ at the start also change at different rates | Tutored first-years are compared with sophomores who are past their first-term adjustment | Random assignment, or closely matched groups |
| Experimenter bias | Researchers' expectations shape how they treat participants or score results | Tutors who know the study's goal grade practice work more generously | Blind scorers and standard scripts |
| Demand characteristics | Participants guess the hypothesis and act to fit it | Tutored students report studying more because they think that's expected | Objective outcome measures and neutral instructions |
| Diffusion of treatment | The control group receives some of the treatment | Tutored students share their tutoring notes with friends in the control group | Separate the groups by time or location |
Notice how many of these threats a simple control group would handle. The tutoring study as described is a one-group pretest-posttest design, which Campbell and Stanley classed as "pre-experimental." Their summary table marks history, maturation, testing and instrumentation as definite weaknesses of this design, and when participants are picked for extreme scores, as they are here, regression joins the list. Trochim's page on single-group threats walks through the same problem with a first-grade math program.
Confounding variables cut across this whole list. A confounder is a third variable linked to both the treatment and the outcome, such as motivation in the optional-tutoring version above. Random assignment is the most reliable fix because it balances confounders you haven't thought of, not just the ones you measured.
Threats to external validity are reasons a result that's real in your study might not hold for other people, settings or times. Trochim describes each threat as an explanation of how a generalization could be wrong. Campbell and Stanley named four, and the last two rows below come from later frameworks such as Trochim's people, places and times.
| Threat | What It Means | Example | How to Reduce It |
|---|---|---|---|
| Interaction of selection and treatment | The effect depends on traits of the people sampled | A stress app helps psychology majors who volunteered out of curiosity but not working adults | Probability sampling, more diverse samples, checking the effect within subgroups |
| Reactive or interaction effect of testing | A pretest makes participants more or less sensitive to the treatment | A pretest on recycling attitudes makes people pay closer attention to a recycling campaign | A posttest-only design, or a Solomon four-group design that measures the pretest's effect |
| Reactive effects of experimental arrangements | People act differently because they know they're in a study | Employees work harder while researchers are watching | Unobtrusive measures, longer observation, natural settings |
| Multiple-treatment interference | Earlier treatments carry over into later ones | Participants who already tried two memory techniques are tested on a third | One treatment per person, or counterbalanced order |
| Setting (situation) effects | Location, time of day, the researcher or other conditions shape the result | A study skills workshop that works in a quiet lab fails in a noisy dorm lounge | Replication in several settings |
| Time (history) effects | The result depends on when the study ran | A smoking study run right after major public health news, Trochim's example | Replication at different times |
The third row is often called the Hawthorne effect, after productivity studies at Western Electric's Hawthorne plant near Chicago between 1924 and 1933. Use the label with care. A 2014 systematic review of the Hawthorne effect found that taking part in research does change behavior in many studies, but the evidence was too mixed to say when, how or by how much. Its authors suggested the broader term "research participation effects."
Psychology's best-known external validity problem is its reliance on a narrow slice of humanity. In "The Weirdest People in the World?" (2010), Joseph Henrich, Steven Heine and Ara Norenzayan used the acronym WEIRD (Western, Educated, Industrialized, Rich and Democratic) for the societies most studies sample. Drawing on Jeffrey Arnett's 2008 review of top journals in six psychology subfields from 2003 to 2007, they reported that 96% of subjects came from Western industrialized countries, which hold about 12% of the world's population. In the Journal of Personality and Social Psychology, 67% of American samples were made up entirely of psychology undergraduates.
The same paper describes cross-cultural tests of the Müller-Lyer illusion, the classic pair of equal lines with arrow-shaped ends. Across 16 societies, American undergraduates showed the strongest illusion: on average, one line had to be about a fifth longer before they judged the two as equal. San foragers of the Kalahari showed no measurable illusion at all. Even a basic perceptual effect failed to generalize across populations.
The marshmallow test shows how a famous finding can shrink when it's tested on a broader sample. Follow-up studies such as Shoda, Mischel and Peake (1990) linked preschoolers' ability to wait for a second treat with later achievement, but the children came from a preschool on the Stanford University campus and the follow-up samples were small. In a 2018 conceptual replication, Watts, Duncan and Quan used data on 918 children recruited at 10 US sites, focusing on 552 children whose mothers had not completed college. The link between waiting time and achievement at age 15 was about half the size originally reported, and it shrank by about two-thirds once they controlled for family background, early cognitive ability and the home environment.
That second result is an internal validity lesson too. Much of what looked like an effect of early self-control overlapped with differences in children's backgrounds, which is why the two kinds of validity are worth checking together.
Internal and external validity often trade off because the controls that rule out alternative explanations also make a study less like real life. Campbell and Stanley described the two criteria as "frequently at odds in that features increasing one may jeopardize the other."
A tightly controlled lab experiment holds the room, instructions, timing and materials constant. That isolates the cause well, but the setting is artificial and the participants are usually whoever was easiest to recruit. A field study in real classrooms or workplaces looks more like life, but you can't control everything that happens to people during it, so alternative explanations creep back in.
The trade-off isn't absolute. A field experiment that randomly assigns people in a natural setting can do well on both, and a lab task can generalize widely if it engages the same psychological processes as real life. Many research programs handle the tension in stages:
Not every study needs broad external validity, either. A lab experiment built to test whether a theory's prediction holds can be valuable even if its exact setup never occurs in daily life. What matters is that the authors don't claim more generality than the design supports.
Replication matters more than many students expect. When the Open Science Collaboration repeated 100 published psychology studies for a 2015 paper in Science, 97% of the original studies had reported statistically significant results, but only 36% of the replications did, and replication effects averaged about half the size of the originals. Treat any single study, however well designed, as a starting point.
You improve internal validity mainly through control over the design, and external validity mainly through choices about who, where and when you study. These are the main tools for each.
The quickest way to learn these concepts is to rate one study on both. The study below is invented, but it's described the way a research methods assignment might present one.
The study: A researcher recruits 120 students from an introductory psychology course at one university. Each student is randomly assigned to read a textbook passage for 20 minutes in a campus lab, with their phone either in another room or face-down on the desk. A research assistant then gives a 10-question quiz on the passage. The phone-in-another-room group scores higher on average, and the difference is statistically significant.
Here's how that analysis might read in the limitations section of a lab report:
Because participants were randomly assigned to conditions and completed identical tasks, the design supports a causal interpretation of the phone-location effect within this sample. However, all participants were introductory psychology students at a single university, and the 20-minute laboratory task differs from typical self-directed study. The findings may therefore have limited population and ecological validity. Future research should test the effect with more diverse student samples and in naturalistic study settings over several weeks.
When you write about validity, name the specific threat, explain how it could change the result, and say what design feature would fix it. A vague line like "the sample was small, so the study lacks validity" loses marks because it doesn't say which kind of validity is at risk or why. Sample size mostly affects statistical power and precision, and a large sample can still be unrepresentative.
Use these questions as a checklist when you read a methods section:
If you're designing your own study, the same questions shape your methods. Start with a focused research question, decide whether you need an experimental or nonexperimental design, and plan your data collection using the basics of empirical research. The difference between descriptive and inferential statistics also explains why generalizing from a sample takes more than describing it.
If you'd like a second pair of eyes, a tutor can go through a study critique or lab report you've drafted and explain where your validity arguments hold up and where they need more support. You can request tutoring or feedback on your own draft, or get help with research methods coursework through our psychology homework help.
A common example is asking whether a lab finding holds in real life. Suppose an experiment shows that intro psychology students remember more of a passage when their phone is in another room. External validity is the question of whether that result holds for other students, in real study sessions, over a whole semester. Population validity covers the people part, and ecological validity covers the setting and task.
Yes, and it's common. A well-controlled lab experiment with random assignment can show a clear cause-and-effect relationship in its sample, yet rely on volunteers from one university and an artificial task that looks nothing like daily life. The reverse happens too: a large national survey can describe a population well without establishing causation. Strong research programs combine designs to cover both.
Internal validity usually comes first. Campbell and Stanley called it the "basic minimum" for interpreting an experiment, because if the effect isn't real in your own study, there's nothing to generalize. After that, priorities depend on the purpose. A study testing whether a cause works at all leans on internal validity, while a study informing policy for a specific population needs strong external validity as well.
No. Ecological validity is about how realistic a study's setting, tasks and materials are. External validity is the broader question of whether results generalize to other people, settings and times. The two often move together, but not always. A realistic study at one workplace may not generalize to other workplaces, while an artificial lab task can replicate across many different populations.
Not directly. Random assignment decides which group each participant joins, so it protects internal validity by balancing differences between groups. External validity depends more on random sampling, which decides who gets into the study in the first place. An experiment can randomly assign 100 volunteers from one campus and still have limited population validity. The two techniques solve different problems.
Most textbooks name population validity and ecological validity. Population validity asks whether results from your sample apply to the wider population you care about. Ecological validity asks whether results from your study's setting and tasks apply to real-world situations. Some frameworks, such as Trochim's people, places and times, also treat time as its own dimension: would the findings hold if the study ran in a different year?
Reliability is about consistency: whether a measure gives similar results when it's repeated under the same conditions. Validity is about accuracy: whether a measure captures what it claims to and whether a study's conclusions are justified. A measure can be reliable without being valid, like a bathroom scale that always reads two pounds heavy. Internal and external validity concern a study's conclusions, not a single measure.
Post your assignment, compare tutor bids and work with an expert who explains each step.
Psychology Homework Help for Every Course
Get affordable and top-notch help for your essays and homework services from our expert tutors. Ace your homework, boost your grades, and shine in online classes—all with just a click away!
Fast, secure, and handled by vetted experts.
Arguments can be divided into two main categories: Inductive and deductive. use this post to get insight on inductive vs deductive argument and examples for each type.
Read More
BA vs BS explained: how a Bachelor of Arts and Bachelor of Science differ in courses, credits and skills, real examples, and which fits your career goals.
Read More
Do you have a descriptive or inferential statistics task? Get help from professionals with experience in the field of statistics. Reach out to Acemyhomework today to get professional assistance with your tasks.
Read More