Interpreting treatment effect size from single case experimental design data: a preliminary analysis of differential effects of treatments designed to increase or decrease behaviour.
Skill-building interventions look stronger on paper than behavior-reduction ones, so use direction-specific benchmarks when you judge effect sizes.
01Research in Context
What this study did
Sutton et al. (2022) looked at hundreds of single-case graphs. They asked one simple question: Do effect sizes look different when we try to make a behavior happen versus when we try to make it stop?
They used within-case Cohen’s d to measure each graph. Then they compared the size and spread of the numbers for increase studies and decrease studies.
What they found
Interventions that increase skills give bigger, jumpier effect sizes. Interventions that decrease problem behavior give smaller, tighter effect sizes.
So a “large” d in one direction is not the same as a “large” d in the other direction.
How this fits with other research
Pichardo et al. (2026) show that caregivers can track feeding-treatment effects almost as well as trained observers. Their data still need the right benchmark; M et al. warn us to pick separate benchmarks for increase and decrease targets.
Mount et al. (2011) found that high variability makes visual inspection harder. M et al. add that part of that variability is baked in when the goal is to increase behavior.
Iivanainen (1998) told us to treat variability as information, not noise. M et al. give a clear example: direction of change explains part of the noise.
Lee et al. (2024) built Rasch cutoffs to judge theory-of-mind scores. Like M et al., they turn raw numbers into rules practitioners can trust.
Why it matters
Next time you read a SCED meta-analysis, check which way the behavior is supposed to move. Expect bigger Cohen’s d values for skill-building programs and smaller ones for reduction programs. Use separate benchmarks when you judge if an intervention is “strong” or “weak,” and share this caution with your team before you drop a useful procedure.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Open your last SCED graph, note if the target behavior was increased or decreased, and compare the effect size to the right directional benchmark before you decide to keep or drop the intervention.
02At a glance
03Original abstract
BACKGROUND: Estimates of treatment effect size from single case experimental design (SCED) data may be impacted by the direction for treatment effects (i.e. ascending or descending slope for the dependent variable). Estimating effect sizes for treatments designed to decrease behaviour are potentially more restricted because the intended direction for treatment is zero (i.e. an absolute basal). Conversely, effect sizes for interventions that increase behaviour are less restricted due to a relatively unconstrained ceiling from a pure measurement standpoint (i.e. no absolute ceiling). That is, treatments that increase behaviour have a broader range of possible effect size values as the ceiling is only limited by demand characteristics and the learners' skills and motivation to exhibit the behaviour. METHOD: The current study represents a preliminary analysis of the mean and range of SCED effect sizes for treatments designed to either increase or decrease target behaviour. A within-case Cohen's d measure that was developed for SCED data was used to estimate treatment effect sizes. RESULTS: Results indicated that the mean and range of effect size values for treatments that increased behaviour were significantly greater compared with treatments that decreased behaviour. CONCLUSIONS: Results are discussed in terms of developing standards, or best practices, specific to interpreting effect size values and meeting quality control requirements for inclusion of the data set in future SCED meta-analytic studies estimating treatment effect size. Specifically, preliminary results suggest that benchmarks for low, medium and high SCED effect size values need to be developed separately for treatments that increase or decrease levels of the dependent variable.
Journal of intellectual disability research : JIDR, 2022 · doi:10.1111/jir.12966