Ethical Issues in Using Standardized Decision-Making Algorithms in Behavior-Analytic Practice

Overview & Clinical Significance

Decision-making algorithms (DMAs) have attracted growing attention in applied behavior analysis as tools that standardize clinical judgment through structured flowcharts and decision trees. When a practitioner works through a DMA, each branching question is intended to channel reasoning toward a defensible, consistent outcome—whether for treatment selection, behavioral function identification, or intervention modification.

Proponents argue this standardization reduces idiosyncratic practice, promotes fidelity, and helps newer clinicians navigate complex decisions with greater confidence.

The clinical appeal is understandable. Behavior analysts routinely face high-stakes decisions under time pressure, with incomplete data and variable staff training.

A well-constructed DMA can serve as a scaffold that structures those decisions. However, the very characteristics that make DMAs attractive also generate significant ethical exposure under the BACB Ethics Code (2022).

The core tension is between algorithmic efficiency and individualized clinical reasoning. The Ethics Code (section 2.01) requires behavior analysts to provide services that are consistent with the current research literature and tailored to each client's unique characteristics.

An algorithm that does not account for individual variation risks reducing evidence-based practice to a mechanical exercise rather than a scientifically informed, person-centered process. Research on functional communication training illustrates how treatment pathways that follow a standard sequence can overlook critical client variables—Dawson et al.

(2026) note that procedures for establishing functional communication responses must account for both the topography of the response and the client's existing mand repertoire, neither of which fits neatly into a one-size algorithm.

Beyond individualization, there are questions of professional accountability. When a clinician follows a DMA and an adverse outcome results, the algorithm's presence may create ambiguity about who bears responsibility for the decision.

This diffusion of accountability is ethically problematic. The BACB Ethics Code is unambiguous: the behavior analyst retains personal responsibility for every clinical decision regardless of the procedural tools used.

Understanding this distinction—between a DMA as decision support versus decision substitute—is foundational for any practitioner considering algorithmic tools in their practice.

Background & Context

The emergence of decision-making algorithms in ABA parallels a broader trend in healthcare and education toward structured clinical decision tools. In medicine, clinical practice guidelines and decision aids have been studied extensively, with evidence showing they can reduce variability and improve adherence to evidence-based recommendations when implemented thoughtfully.

Behavior analysis has drawn on this tradition, with researchers developing DMAs for topics ranging from restrictive procedure authorization to preference assessment selection to reinforcement schedule thinning.

A key finding in this literature is that DMAs developed without adequate empirical validation carry heightened risks. An unvalidated algorithm may embed assumptions that reflect the developer's theoretical commitments or practice context rather than generalizable evidence.

This is particularly relevant given how measurement complexity interacts with decision tools. Van & Kubina (2026) review precision teaching approaches to inner behavior, highlighting that many behaviorally relevant variables—thoughts, feelings, internal states—are not easily captured through the observational data streams that most DMAs rely on.

When an algorithm's input variables exclude relevant private events, the decision output may be systematically biased.

Caregiver data further complicates algorithmic inputs. Pichardo et al.

(2026) examined caregiver report accuracy in pediatric feeding disorder and found systematic discrepancies between caregiver-reported and clinician-observed outcomes, particularly for low-frequency high-intensity behaviors. If a DMA relies on caregiver report as a decision node without accounting for these known measurement limitations, the algorithm will generate unreliable outputs downstream.

The historical development of DMAs in ABA also intersects with questions about who designs these tools and for whom. Algorithms developed in university-based research clinics may not account for resource constraints, staff training levels, or cultural variables present in community practice.

Awareness of this design-to-deployment gap is essential context for any critical evaluation of a DMA's appropriateness for a given practice setting.

Clinical Implications

Practitioners encountering DMAs in their work—whether as tools their organization has adopted, materials distributed at training events, or resources embedded in supervision software—should approach them with structured critical analysis rather than passive adoption.

First, examine the empirical basis of the algorithm. Does the DMA cite peer-reviewed research for each decision node?

Is the research base drawn from populations and settings comparable to those in which you practice? Research on functional assessment approaches provides instructive context: Kaye et al.

(2025) demonstrated that antecedent and functional analyses of echolalia lead to meaningfully different treatment selections depending on which analysis is conducted and how the data are interpreted. A DMA that specifies a single functional assessment pathway without accounting for echolalia topography and contextual variables would truncate exactly the kind of analysis this research shows to be necessary.

Second, consider what the algorithm cannot do. DMAs are bounded by their input variables.

They cannot detect the quality of the therapeutic relationship, the caregiver's current stress level, or the subtleties of a client's motivating operations on a given day. These unmeasured variables are clinically meaningful, and a decision that ignores them may be technically protocol-compliant while being practically inadequate.

Third, evaluate how the DMA handles edge cases. A well-designed algorithm includes explicit guidance on when to exit the standard pathway.

If a DMA lacks clear off-ramps for atypical presentations, it is likely to be misapplied. Research on multilevel intervention outcomes, such as the meta-analysis by Kok et al.

(2026) on externalizing behavior interventions, consistently shows that treatment effects vary substantially across individuals—a pattern that resists algorithmic reduction.

Finally, document your reasoning explicitly when using or departing from a DMA. Clinical documentation should capture not just what decision was made but why, referencing the specific client variables that informed the judgment.

This protects both the client and the practitioner by making individualized reasoning transparent and auditable.

Ethical Considerations

The BACB Ethics Code (2022) establishes multiple standards directly relevant to DMA use. Section 2.01 (Providing Effective Treatment) requires behavior analysts to rely on current research and individualize services.

Section 5.06 (Continuing Education) requires ongoing competence development in areas relevant to practice. Section 1.04 (Practicing Within Competence) prohibits providing services in areas where the practitioner lacks adequate training.

Taken together, these standards create a framework in which algorithmic tools must supplement, not replace, professional competence.

One ethically significant issue concerns the development of DMAs. Brodhead's scholarship examines ethical obligations of algorithm developers: what standards should govern the evidence threshold for including a decision node?

How should uncertainty be communicated to end users? What obligations do developers carry when they become aware that a widely-used DMA is generating systematically poor outcomes?

The masking issue is also ethically relevant. Kaur et al.

(2026) examined a retrospective case series in which protective procedures for self-injurious behavior inadvertently masked the functional reinforcers maintaining the behavior, delaying accurate functional analysis. This pattern—where a procedural tool produces data that obscures rather than clarifies the clinical picture—is directly analogous to how a DMA can generate a plausible-looking decision pathway that channels a clinician away from the investigation the individual client actually requires.

Equity considerations deserve explicit attention as well. If a DMA was developed using data from populations that underrepresent certain racial, linguistic, or diagnostic groups, applying it to those populations without appropriate scrutiny raises genuine fairness concerns.

The behavior analyst's obligation under section 1.07 (Cultural Responsiveness) of the Ethics Code requires active examination of how algorithmic inputs and outputs may interact with client characteristics in ways the algorithm was not designed to detect.

Assessment & Decision-Making

Evaluating whether a DMA is appropriate for use in your setting requires a structured approach that parallels the evidence-based practice standards already familiar to BCBAs.

Begin with a literature review of the DMA's empirical foundation. Identify the studies cited in support of each major decision node and assess their methodological quality.

Single-case research supporting a particular intervention pathway should meet contemporary standards for internal validity and generalization before that pathway is embedded in an algorithm. The scoping review methodology employed by Dawson et al.

(2026) on functional communication procedures offers a model for systematically mapping the evidence base supporting a procedural sequence—a similar approach can be applied to DMA evaluation.

Next, map the DMA's input requirements against the assessment data your practice routinely collects. If the algorithm requires a completed functional behavior assessment but your organization lacks staff capacity to conduct reliable FBAs, the DMA will function on faulty inputs and produce unreliable outputs.

Consider conducting a prospective audit of DMA-guided decisions over a defined period. Track cases where the algorithm produced a recommendation, the degree to which that recommendation was followed, and the outcomes.

Compare outcomes for DMA-guided versus non-DMA-guided decisions where possible. This type of local validation mirrors the single-case research tradition in ABA and generates the kind of practice-specific evidence that can inform ongoing use or modification of the tool.

Finally, establish a formal review cadence for any DMA your organization uses. As new research emerges—such as updated meta-analyses of intervention effectiveness like that of Kok et al.

(2026)—the algorithm's decision nodes may need updating to remain consistent with the current literature. A DMA that is not regularly reviewed against emerging research will gradually become an evidence-practice gap rather than a bridge to one.

What This Means for Your Practice

Behavior analysts do not need to reject decision-making tools to satisfy their ethical obligations—they need to engage with them critically and deploy them deliberately.

Practically, this means treating any DMA you encounter as a starting hypothesis rather than a final answer. Before implementing an algorithm-generated recommendation, ask: does this recommendation fit what I know about this individual client's learning history, behavioral function, and personal preferences?

If the answer is uncertain, additional assessment is warranted before proceeding.

For supervisors and clinical directors, DMA adoption decisions carry organizational responsibility. Staff trained primarily on algorithms without concurrent instruction in the underlying clinical reasoning may be able to follow a flowchart but unable to detect when the flowchart is leading them astray.

Training that explicitly addresses when and why to deviate from a standard pathway—and how to document that deviation—is a prerequisite for safe algorithmic integration.

The scholarship underlying this CEU course directly supports BACB Ethics Code compliance by building the conceptual tools practitioners need to use DMAs responsibly. Understanding the ethical limitations of standardized decision tools is not a reason to avoid them; it is the foundation for using them well.

Critically evaluated and appropriately supervised, a decision algorithm can support more consistent practice. Used uncritically, it can erode the individualized reasoning that defines competent behavior-analytic care.