Your data show behavior changed after the intervention started. Before claiming the intervention caused it, ask: what else could have produced the same change? Each answer to that question is a threat to internal validity.
Why it matters
Internal validity is your confidence that the independent variable, and not something else, changed the dependent variable. When an uncontrolled factor could explain the result, it is a confounding variable, and the functional relation you thought you found is in doubt.
This matters for clients, not only journals. If a program "worked" only because a new medication started that week, the family loses time and the team learns the wrong lesson. Practice spotting the rival explanation in a short case and choosing the design feature that would rule it out.
Key ideas
- Every threat is a rival explanation. Ask, "If the intervention did nothing, could this still have changed the data?"
- Some threats live outside the client. Events in the environment, such as a new teacher or a family move, can change behavior at the same time as your intervention.
- Some threats live inside the client. Growth, development, illness, or fatigue can change behavior with the passage of time.
- Some threats live in the measurement. Observers can drift, and the act of measuring can itself change behavior.
- Designs control threats through replication. A single AB comparison cannot rule out history or maturation. Showing the effect again, by withdrawing and reintroducing the intervention or by introducing it at staggered times, makes rival explanations much less plausible.
- Stable baselines help. A steady baseline predicts what would happen without intervention and keeps you from starting treatment at an unusual high or low point.
Definitions
- History: specific events outside the study that occur while it is running and could affect the behavior, such as a schedule change, a new sibling, or a change in medication.
- Maturation: changes inside the participant that come with the passage of time, such as physical growth, development, or growing tired over a long study.
- Testing: repeated exposure to the measurement itself changes performance, such as improving on a probe through practice or behaving differently because of being observed (reactivity).
- Instrumentation: changes in the measurement system over time, such as an observer gradually shifting how a definition is applied (observer drift) or a switch to a different recording method.
- Statistical regression: extreme measures tend to move back toward a person's typical level on later measurement, whether or not anything was done.
- Multiple-treatment interference: the effect of one treatment is influenced by exposure to another, through order (sequence) effects or carryover between conditions. It also limits external validity, since results may differ when a treatment is used alone.
- Attrition and selection: losing participants partway through, or comparing groups that differed from the start. Mostly group-design concerns, though attrition matters when a single-case study drops tiers or participants.
Examples from a session
Rival explanations in everyday practice
- School: A token system for Jalen, age 8, starts on Monday. That same week the school adds a second adult to his classroom. Fewer call-outs could reflect the tokens, the extra adult (history), or both.
- Early intervention: Over five months of a vocal imitation program, Amara, age 2, gains many new words. Some of that growth could come from typical language development (maturation) rather than the program.
- Clinic: A data collector scores "aggression" for Theo, 14. Over weeks she starts skipping light pushes and counting only forceful hits. The graph trends down even though Theo's behavior has not changed (instrumentation, observer drift).
- Home: Rosa's caregivers ask for help after the worst week of tantrums they can remember. Treatment begins right away, and rates fall back to her usual level within days. Part of that drop may simply be regression toward her typical rate.
- Adult day program: Marcus, 34, is taught with two prompting methods in alternating sessions. If one method's practice carries into the other's sessions, the comparison is muddied (multiple-treatment interference).
- Non-example: In a multiple baseline across three classrooms, Kai's on-task behavior rises only when the intervention reaches each room, at staggered times, with IOA checked in every phase. One outside event or developmental change cannot explain three separately timed changes, so history and maturation are not credible rivals here.
How you respond depends on the threat. For history and maturation, replicate the effect within the design, for example with a reversal (ABAB) or a multiple baseline in which behavior changes only when the intervention reaches each tier. For instrumentation, define behavior precisely, retrain observers, and collect interobserver agreement throughout every phase. For regression, wait for a stable baseline before you intervene. For multiple-treatment interference, use distinct stimuli per condition and counterbalance the order.
Common mistakes / exam traps
- Swapping history and maturation. History is an outside event. Maturation is a change within the person over time. A new teacher is history; growing taller or older is maturation.
- Treating an AB design as proof. A change from A to B shows correlation in time. Without replication, history and maturation remain open.
- Starting treatment at a peak. Intervening right after an extreme baseline point makes any decrease ambiguous because of regression.
- Assuming IOA during one phase is enough. Observer drift can happen at any time, so agreement checks belong in every phase.
Exam trap
If the scenario names an event that happened at the same time as the intervention, think history. If it describes the person getting older, developing, or tiring over time, think maturation. If it describes the observer or the measurement changing, think instrumentation.
Exam trap
Testing and instrumentation are both measurement threats, but they differ in what changed. In testing, the client's behavior really changed because of repeated probes or being observed. In instrumentation, the observer or recording method changed, so the data moved even if behavior did not. Ask: did the person change, or did the ruler?
Quick check
Lena, a BCBA, begins a self-monitoring program for Sam, age 15, who often leaves class without permission. The week the program starts, the school also introduces a new hall-pass policy for all students. Leaving class decreases. Which threat to internal validity is the biggest concern?