Generalized Linear Mixed Effects Modeling (GLMM) of Functional Analysis Graphical Construction Elements on Visual Analysis
Stretching the axes on an FA graph can trick even seasoned BCBAs into seeing or missing a function.
01Research in Context
What this study did
The team showed 120 BCBAs the same FA graphs but stretched the x-to-y axes ratio.
Some graphs looked wide and flat. Others looked tall and skinny.
Each BCBA judged if the graph showed a clear function. Their answers were compared to a strict visual-inspection checklist.
What they found
Changing the shape of the graph changed the call. A taller, skinnier plot made function easier to “see.”
Even without the stretch, BCBAs only matched the checklist 60 % of the time. Agreement was low and got worse when the graph was wide.
How this fits with other research
Diller et al. (2016) already showed BCBAs disagree when rating multielement graphs. Dowdy adds that simple axis stretch is one reason why.
Wolfe et al. (2023) found that steep trend and high variability hurt agreement. Dowdy’s GLMM now says axis ratio belongs on that same list.
Wolfe et al. (2018) offered the Conservative Dual-Criterion as an objective helper. Dowdy’s low checklist agreement hints that even rule-based aids can fail if the graph is drawn poorly.
Why it matters
Your eyes are not a meter stick. Squash or stretch an FA graph and you risk calling a false function or missing a real one. Lock the axes so the x unit is twice the y unit before you share the graph. If you must rescale, run a second check with the CDC or another rule to keep the data—and your decision—honest.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Set every FA graph to a 2:1 x-to-y axis ratio before you decide what the data say.
02At a glance
03Original abstract
Multielement designs are the quintessential design tactic to evaluate outcomes of a functional analysis in applied behavior analysis. Protecting the credibility of the data collection, graphing, and visual analysis processes from a functional analysis increases the likelihood that optimal intervention decisions are made for individuals. Time-series graphs and visual analysis are the most prevalent method used to interpret functional analysis data. The current project included two principal aims. First, we tested whether the graphical construction manipulation of the x-to-y axes ratio (i.e., data points per x- axis to y-axis ratio [DPPXYR]) influenced visual analyst’s detection of a function on 32 multielement design graphs displaying functional analyses. Second, we investigated the alignment between board certified behavior analysts (BCBAs; N = 59) visual analysis with the modified visual inspection criteria (Roane et al., Journal of Applied Behavior Analysis, 46, 130-146, 2013). We found that the crossed GLMM that included random slopes, random intercepts, and did not include an interaction effect (AIC = 1406.1, BIC = 1478.2) performed optimally. Second, alignment between BCBAs decisions and the MVI appeared to be low across data sets. We also leveraged current best practices in Open Science for raw data and analysis transparency.
Perspectives on Behavior Science, 2024 · doi:10.1007/s40614-024-00406-4