Internal validity is about whether your independent variable actually caused the change in your dependent variable, or whether something else could explain the results. The MR SMITH ID mnemonic covers nine common threats that can undermine this causal claim: Maturation, Regression to the mean, Selection, Mortality, Instrumentation, Testing, History, Interactions, and Diffusion. If you can spot which threat applies to a given scenario, you can evaluate whether a study's conclusions are trustworthy.
Difficulty: Introductory to Intermediate | Prerequisites: Basic understanding of experimental design (independent vs. dependent variables, control vs. experimental groups).
In any research methods or experimental psychology course, you will learn that a well-designed experiment aims to isolate the effect of the independent variable on the dependent variable. Internal validity is the degree to which a study achieves that isolation. When something other than the IV could plausibly explain the results, the study has an internal validity threat. The MR SMITH ID mnemonic gives you a structured checklist of nine such threats. This material typically appears in units on research design, and it connects directly to understanding confounds, experimental controls, and how to read published studies critically. You should already be comfortable with the difference between experimental and control conditions, random assignment, and what it means for a variable to be "confounded."
Maturation
Changes within participants over time that affect the dependent variable, regardless of the experimental treatment. These are natural, internal processes: fatigue, hunger, ageing, boredom, or simply the passage of time altering how people respond.
Think of it as: participants changed on their own between measurements, and that change got mixed up with the treatment effect.
Regression to the mean
When participants are selected for extreme scores on a first measurement, their scores tend to move closer to the average on a second measurement, purely due to natural variability. This is a statistical phenomenon, not a real change in the underlying trait.
Think of it as: if you pick the top scorers and test them again, they will probably score a bit lower the second time, even if nothing happened to them.
Selection
Systematic differences between groups that exist before the study begins, usually because participants were not randomly assigned. If one group differs from the other in a way that is relevant to the outcome, any difference in results could be due to who was in each group rather than the treatment.
Think of it as: the groups were already different from each other before anything happened.
Mortality (attrition)
Participants dropping out of a study at different rates or for different reasons across conditions. If the people who leave one group are systematically different from those who stay, the remaining sample is no longer comparable to the other group.
Think of it as: people quit, and the quitters were not random, so the groups became unequal.
Instrumentation
Changes in the measurement instrument or procedure over the course of the study. This can include a calibration drift in equipment, interviewers becoming more skilled or more relaxed, or a shift in how questions are asked between time points or conditions.
Think of it as: the "ruler" you are measuring with changed between measurements, so the scores shifted for reasons that have nothing to do with the treatment.
Testing
The act of being measured or tested at one point in time affects performance on a later measurement. Simply taking a pretest can sensitise participants to the topic, prime certain responses, or make them more confident on the posttest.
Think of it as: the first test taught participants something or changed how they approached the second test.
History
An external event, unrelated to the treatment, that occurs between measurements and affects the outcome. This is something outside the study that both groups (or all participants) experience, and it provides an alternative explanation for any observed change.
Think of it as: something happened in the outside world during the study that muddied the results.
Interactions (selection interactions)
A combination of selection with another threat (typically maturation or history). The groups are already different in some way, and that pre-existing difference interacts with another threat to produce a differential effect. The key idea: the other threat would not be a problem if the groups had been equivalent to begin with.
Think of it as: two threats teaming up, where selection is always one of them.
Diffusion (diffusion of treatment)
Information about the treatment or the study leaks from one group to another, contaminating the control condition. When control participants learn what the experimental participants experienced, the distinction between conditions breaks down.
Think of it as: the experimental "secret" got out, and the control group was no longer truly naive.
The trick to these questions is to read the scenario and ask yourself: what changed, and why? Each threat has a distinctive fingerprint.
Maturation: Look for a time delay between measurements. Did participants have time to change naturally (get tired, get bored, forget things, grow older)? The passage of time is the clue.
Regression to the mean: Look for participants selected on the basis of extreme scores. If someone picked the highest or lowest scorers for follow-up, scores will drift toward the average on retest.
Selection: Look for non-random assignment. Were groups formed on the basis of a characteristic (all females in one group, all males in another)? If so, the groups differ before the study even begins.
Mortality: Look for participants dropping out, especially when the dropout rate or reason differs between conditions.
Instrumentation: Look for changes in how the study was administered over time. Did the confederates get better at acting? Did the equipment change? Did the questions shift?
Testing: Look for participants being measured twice on the same or a similar measure. The first test can affect performance on the second.
History: Look for an external event that happened during the study and could affect the outcome. A news event, a campus training, a policy change.
Interactions: Look for two threats at once, where one of them is selection. Typically, the groups were already different (selection), and then something else (maturation, history) hit them differently because of that pre-existing difference.
Diffusion: Look for information leaking between conditions. Did control participants learn about the treatment? Did word spread on campus about the study?
Some threats can look similar on first reading. When scenarios overlap, choose the threat that is the best (most specific) fit.
History vs. maturation: history is an external event; maturation is an internal change in the participant. If a hurricane shuts down campus, that is history. If participants simply had three weeks to forget, that is maturation.
Selection vs. interactions: if the groups were formed non-randomly and that is the only problem, it is selection. If the non-random groups also experienced a second threat differently (e.g., selection + maturation), it is interactions.
Testing vs. instrumentation: testing is about the effect of being tested on the participant. Instrumentation is about the measurement tool itself changing.
Regression to the mean vs. testing: regression is about selecting extreme scorers and seeing their scores drift back. Testing is about the act of taking the test affecting a later score. Regression requires selection on extreme values.
Students often confuse history with maturation. History is an outside event (a news story, a campus-wide training). Maturation is an internal, time-based change in the participant (fatigue, forgetting, ageing). Both involve time passing, but the cause is different.
Students assume that any time participants are tested twice, the threat is testing. Testing specifically refers to the first test influencing performance on the second. If the measurement instrument itself changed, that is instrumentation.
Students sometimes label a scenario as selection when it is actually interactions. Pure selection means the groups were different before the study and that alone explains the result. Interactions means the pre-existing group difference combined with another threat (often maturation or history) to produce the outcome.
Students forget that regression to the mean requires selecting participants based on extreme scores. If participants were not chosen for being extreme, regression to the mean does not apply.
The MR SMITH ID mnemonic is a favourite exam format: you will be given a study description and asked to match scenarios to threats. Practise doing this until it is automatic.
You will almost certainly be asked to distinguish between threats that look similar (history vs. maturation, selection vs. interactions, testing vs. instrumentation). Know the distinguishing features cold.
Some exams include a "no threat" option. A scenario where everything went according to plan and the results are as expected has no validity threat.
Look out for scenarios where the answer choices overlap. The instruction is always to choose the best (most specific) threat.
True or false: If a hurricane delays a study by two weeks and both groups' memories fade equally, the most likely threat is history.
Answer: False. That sounds more like maturation (natural memory decay over time). History would be an external event that differentially affects the outcome.
Fill in the blank: When participants are selected for extreme scores and their scores move toward the average on retest, this is called ______.
Answer: Regression to the mean.
True or false: Diffusion occurs when participants in the control group learn about the experimental treatment.
Answer: True.
Fill in the blank: When the measurement tool or procedure changes over the course of a study, this is a threat called ______.
Answer: Instrumentation.
True or false: Interactions always involve selection combined with at least one other threat.
Answer: True.
Q: A researcher assigns all morning participants to the control group and all afternoon participants to the experimental group. Afternoon participants perform worse. What is the most likely validity threat?
A: Selection. The groups were formed based on time of day, not randomly, so pre-existing differences (e.g., fatigue, motivation) between morning and afternoon people could explain the result.
Q: A study measures anxiety before and after a therapy programme. Between measurements, a major earthquake strikes the region. Both groups show increased anxiety. What threat is present?
A: History. The earthquake is an external event unrelated to the treatment that affected outcomes.
Q: A researcher gives participants a memory pretest, then a treatment, then the same memory test as a posttest. Scores improve for both groups. What threat is most likely?
A: Testing. Participants may have performed better on the posttest because they had already seen the same questions, not because of the treatment.
Q: In a drug trial, participants who experience side effects drop out of the treatment group. The remaining treatment participants show strong improvement. What threat is this?
A: Mortality (attrition). The dropout was not random; those who left may have been the ones for whom the drug was least effective, inflating the treatment group's results.
Q: A study recruits the 20 students with the lowest exam scores for a tutoring programme. After tutoring, their scores improve. Is this necessarily due to tutoring?
A: Not necessarily. Regression to the mean is a strong alternative explanation. Because these students were selected for extreme (low) scores, their scores would likely have improved somewhat on retest even without intervention.
This material connects to confounding variables, because each validity threat is essentially a type of confound, a variable that covaries with the IV and provides an alternative explanation. It also connects to external validity (whether results generalise beyond the study), which is a separate set of concerns but part of the same broader question of how much you can trust a study's conclusions. Understanding internal validity threats is foundational for critically evaluating published research, which you will do throughout the course and in any field that relies on empirical evidence.
Internal validity, threats to internal validity, MR SMITH ID, maturation threat, regression to the mean, selection bias, attrition, mortality threat, instrumentation, testing effect, history threat, selection interactions, diffusion of treatment, confounding variables, experimental design, control group, random assignment, Campbell and Stanley, Cook and Campbell, validity threats psychology, research methods PSY 101