Melanson
Functional Analysis of Problem Behavior: A 40-Year Review Notes
Abstract
Review of functional analysis literature from June 2012 to May 2022.
Analyzed 1,333 functional analysis outcomes from 326 studies.
Compared characteristics with previous reviews (Beavers et al., 2013; Hanley et al., 2003).
Similarities: child participants, developmental disability diagnosis, line graphs, differentiated response outcomes.
Differences: increase in autistic representation, outpatient settings, supplementary assessments, tangible conditions, multiple function outcomes; decrease in session durations.
Updated participant and methodological characteristics, summarized outcomes, commented on recent trends, and proposed future directions.
Keywords
Functional analysis, literature review, problem behavior.
Introduction
Forty years since Iwata et al. (1982/1994) published study on self-injurious behavior.
Functional analysis used to treat challenging behavior related to social and nonsocial contingencies.
Last extensive review was 10 years ago (Beavers et al., 2013).
Recent focus on safe and efficient ways to identify function of problem behavior (e.g., Saini et al., 2020).
Current paper expands on Beavers et al. (2013) and Hanley et al. (2003) to report trends over the past 10 years.
Method
Searched PsycINFO, ERIC, and Web of Science databases for published research from June 2012 through May 2022.
Used search terms: function, analysis, behavioral assessment, functional analysis, and behavior.
3,269 studies were generated, 906 included for full article screening.
Inclusion and Exclusion Criteria
Identical to Beavers et al. (2013) and Hanley et al. (2003).
Included studies reported data with:
Pretreatment assessment.
Direct observation and measurement of problem behavior.
At least two conditions manipulating environmental variables.
Attempt to demonstrate relation between environmental event and behavior.
Excluded:
Not published in peer-reviewed journal (e.g., theses, dissertations).
Not written in English.
Methodology, Outcome Scoring, and Interrater Agreement
First author coded methodological dimensions and functional analysis outcomes from previous reviews, plus:
Type of functional analysis (precursor, latency, trial-based, brief, extended no-interaction).
Whether authors referred to analysis as interview-informed synthesized contingency analysis, practical functional assessment, or cited Hanley et al. (2014).
Specific functions involved in problem behavior with multiple control.
Second author independently coded 27.6% of the 906 studies screened.
Comparisons made between methodological dimensions and individual functional analysis outcome.
Total agreement divided by total agreement plus disagreement to yield percentage.
Mean interrater agreement:
90.7% (range: 70.8%–100%) for functional analysis method.
97.3% (range: 50%–100%) for individual functional analysis outcomes.
Results
Journals with Published Functional Analysis Studies
326 studies from 48 journals were included.
Journal of Applied Behavior Analysis: 105 studies.
Behavioral Interventions: 49 studies.
Behavior Analysis in Practice: 20 studies.
Percentage of studies from Journal of Applied Behavior Analysis decreased from 46.2% in Beavers et al. (2013) and 64.9% in Hanley et al. (2003) to 32.2% in our review.
Our review included 42.8% of all functional analysis studies published in the past 40 years.
Beavers et al.: 20.8% (Jan 2001-May 2012).
Hanley et al.: 36.4% (before Jan 2001).
Increasing trend in published functional analysis research in the past 10 years compared with the previous 30 years.
Participants and Settings
Most studies included at least one child participant (92.0%).
Beavers et al. (2013): 83.5%.
Hanley et al. (2003): 70.0%.
Child participants:
78.3% included at least one participant between ages 1 and 9.
54.3% included at least one participant between ages 10 and 18.
At least one adult: 17.8%.
Beavers et al. (2013): 24.7%.
Hanley et al. (2003): 37.2%.
Adult participants:
94.8% included participants between 19 and 64.
5.2% included participants over 65.
Combined results:
82.3% of functional analyses included a child participant.
26.3% included an adult participant.
Most studies included participants with a developmental disability (93.3%).
Beavers et al. (2013): 81.6%.
Hanley et al. (2003): 91.3%.
At least one participant with autism: 74.2%.
Beavers et al.: 37.3%.
Hanley et al.: 20.9%.
Participants without a developmental disability: 8.9%.
Beavers et al.: 21.5%.
Hanley et al.: 9.0%.
Combined results:
90.1% included participants with a developmental disability.
47.2% included participants with autism.
11.6% included participants without a disability.
Outpatient clinics: most common setting (50.3%).
Beavers et al. (2013): 21.5%.
Hanley et al. (2003): 7.6%.
School: second-most common setting (29.4%).
Beavers et al.: 44.3%.
Hanley et al.: 31.4%.
Home: 17.2%.
Beavers et al.: 15.8%.
Hanley et al.: 7.6%.
Inpatient hospital units: 8.0%.
Beavers et al.: 57.0%.
Hanley et al.: 32.5%.
Vocational programs: 3.4%.
Beavers et al.: 5.7%.
Hanley et al.: 2.2%.
No functional analyses conducted in an institution.
Beavers et al.: 6.3%.
Hanley et al.: 25.3%.
Other settings (5.2%): adult day center, home for adults with dementia, brain injury rehabilitation day program, local pediatric clinic.
Combined results for settings:
School: 33.2%.
Outpatient clinic: 28.8%.
Inpatient hospital unit: 27.1%.
Home: 13.4%.
Institution: 10.5%.
Vocational program: 3.4%.
Other: 2.2%.
Response Topographies
Aggression: most common (55.2%).
Beavers et al. (2013): 47.5%.
Hanley et al. (2003): 40.8%.
Self-injurious behavior (SIB): 41.7%.
Beavers et al.: 37.3%.
Hanley et al.: 64.6%.
Disruptive behavior: 25.8%.
Beavers et al.: 26.6%.
Hanley et al.: 19.1%.
Vocalizations: 24.8%.
Beavers et al.: 39.9%.
Hanley et al.: 12.6%.
Property destruction: 20.9%.
Beavers et al.: 36.7%.
Hanley et al.: 10.5%.
Stereotypy: 14.1%.
Beavers et al.: 7.6%.
Hanley et al.: 9.0%.
Noncompliance: 11.0%.
Beavers et al.: 8.2%.
Hanley et al.: 4.3%.
Tantrums: 4.9%.
Beavers et al.: 7.6%.
Hanley et al.: 3.6%.
Pica: 3.4%.
Beavers et al.: 3.8%.
Hanley et al.: 2.5%.
Inappropriate mealtime behavior: 3.7%.
Other problem behavior: 22.4%.
Beavers et al.: 25.3%.
Hanley et al.: 3.6%.
Combined results for response topographies:
SIB: 49.1%.
Aggression: 48.4%.
Disruptions and vocalizations: 23.5% each.
Property destruction: 20.4%.
Other topography: 16.2%.
Stereotypy: 10.9%.
Elopement: 8.7%.
Noncompliance: 8.0%.
Tantrums: 5.0%.
Pica: 3.2%.
Inappropriate mealtime behavior: 1.6%.
Summary of Functional Analysis Methodology
Type of functional analysis.
Supplementary assessment.
Condition type.
Number of test conditions.
Assessment duration.
Session duration.
Experimental design and methodology.
Data presentation.
Type of Functional Analysis
ABC (Antecedent-Behavior-Consequence) functional analysis model: 92.0%.
Beavers et al. (2013): 92.4%.
Hanley et al. (2003): 87.0%.
AB (Antecedent-Behavior) functional analysis model: 4.3%.
Beavers et al. (2013): 12.0%.
Hanley et al. (2003): 20.2%.
Both ABC and AB model: 2.5%.
Beavers et al. (4.4%).
Hanley et al. (7.2%).
Combined results:
ABC models: 90.3%.
AB models: 11.7%.
Both ABC and AB models: 5.0%.
20 studies did not explicitly state whether they provided consequences contingent on problem behavior.
Supplementary Assessment
Some supplementary assessment (descriptive or indirect): 25.8%.
Beavers et al. (2013): 18.4%.
Hanley et al. (2003): 10.5%.
Indirect assessments: 22.7%.
Beavers et al.: 20.3%.
Hanley et al.: 4.3%.
Descriptive assessments: 13.8%.
Beavers et al.: 24.7%.
Hanley et al.: 8.3%.
Both descriptive and indirect assessments: 10.7%.
Beavers et al.: 13.3%.
Hanley et al.: 2.5%.
Combined results: outcomes of descriptive and indirect assessments are not often reported.
Either descriptive or indirect: 18.7%.
Only indirect: 15.5%.
Only descriptive: 14.1%.
Both descriptive and indirect: 8.3%.
Condition Type
Social-positive reinforcement: 93.9%.
Beavers et al. (2013): 94.3%.
Hanley et al. (2003): 85.6%.
Attention: 91.7%.
Beavers et al.: 92.4%.
Hanley et al.: 82.7%.
Tangible: 68.7%.
Beavers et al.: 50.6%.
Hanley et al.: 34.7%.
Social-negative reinforcement: 89.3%.
Beavers et al.: 91.8%.
Hanley et al.: 89.2%.
Automatic reinforcement: 52.8%.
Beavers et al.: 49.4%.
Hanley et al.: 59.6%.
Control (or toy play): 95.4%.
Beavers et al.: 94.9%.
Hanley et al.: 87.9%.
Other condition: 4.6%.
Examples: mand compliance, access to ritualistic arrangements.
Combined results:
Social-positive reinforcement: 90.9%.
Social-negative reinforcement: 89.8%.
Automatic reinforcement: 54.5%.
Other test condition: 2.0%.
Of studies testing social-positive reinforcement:
Attention: 88.6%.
Tangibles: 52.6%.
Number of Test Conditions
Multiple test conditions: 89.3%.
Beavers et al. (2013): 92.4%.
Hanley et al. (2003): 89.5%.
Single test condition: 15.6%.
Beavers et al.: 10.1%.
Hanley et al.: 18.4%.
Combined results:
Multiple test conditions: 90.0%.
Single test condition: 15.5%.
Assessment Length
Full assessment (three or more observations across at least two conditions): 90.8%.
Beavers et al. (2013): 86.1%.
Hanley et al. (2003): 82.7%.
Brief assessment: 17.2%.
Beavers et al.: 12.7%.
Hanley et al.: 13.0%.
Combined results:
Full functional analysis: 86.9%.
Brief functional analysis: 14.7%.
Session Duration
5 min: 55.5%.
Beavers et al. (2013): 37.3%.
Hanley et al. (2003): 11.1%.
10 min: 32.8%.
Beavers et al.: 41.8%.
Hanley et al.: 52.0%.
15 min: 2.5%.
Beavers et al.: 7.0%.
Hanley et al.: 28.2%.
Other: 17.2%, almost all (79.7%) were less than 5 min.
Beavers et al.: 12.7%.
Hanley et al.: 4.3%.
The most common session duration other than 5, 10, or 15 min in our review was 2 min (33.9%).
Unknown: 10.7%.
Beavers et al.: 12.0%.
Hanley et al.: 7.9%.
Combined results:
10-min sessions: 41.7%.
5 min: 35.6%.
15 min: 12.7%.
Some other duration: 11.6%.
Unknown: 10.0%.
Trends toward shorter session durations are apparent.
Experimental Design and Methodology
Multielement design: 67.2%.
Beavers et al. (2013): 79.1%.
Hanley et al. (2003): 81.2%.
Pairwise designs: 29.8%.
Beavers et al.: 7.0%.
Hanley et al.: 2.5%.
Most interview-informed synthesized contingency analyses (Hanley et al., 2014) and practical functional assessments (e.g., Coffey et al., 2021) were classified as using a pairwise experimental design.
Reversal design: 13.2%.
Beavers et al.: 12.0%.
Hanley et al.: 15.5%.
Extended no-interaction screening: 12.3%.
Trial-based functional analysis: 10.1%.
Brief functional analysis: 7.4%.
Latency-based functional analysis: 7.4%.
Precursor functional analysis: 2.1%.
Combination of functional analysis types or methodologies: 30.4%.
Beavers et al.: 7.0%.
Hanley et al.: 2.5%.
Unknown: 0.9%.
Beavers et al.: 3.2%.
Hanley et al.: 5.8%.
Combined results:
Multielement: 74.8%.
Combination of two or more different designs: 15.4%.
Pairwise: 15.1%.
Reversal: 13.8%.
Extended no-interaction: 5.3%.
Trial-based: 4.3%.
Brief: 3.2%.
Latency-based: 3.2%.
Unknown: 3.2%.
Precursor: 0.9%.
Data Presentation
Line graph of session values: 86.8%.
Beavers et al. (2013): 89.9%.
Hanley et al. (2003): 75.1%.
Condition means exclusively: 11.0%.
Beavers et al.: 4.4%.
Hanley et al.: 26.7%.
Within-session values: 0.6%.
Beavers et al.: 0.6%.
Hanley et al.: 1.1%.
Combined results:
Session values in line graphs: 83.2%.
Condition means: 15.4%.
Within-session values: 0.8%.
Summary of Functional Analysis Outcomes
Included an outcome only if a graph was included in the study.
Function of each participant’s problem behavior determined by original authors’ interpretation, not our interpretation.
Obtained 1,333 unique outcomes.
Beavers et al. (2013): 445 outcomes.
Hanley et al. (2003): 536 outcomes.
Total: 2,314 individual functional analysis outcomes.
Outcomes from the past 10 years represent over half (57.6%) of the total sample.
Beavers et al.: 19.2%.
Hanley et al.: 23.2%.
Differentiated responding: 91.1%.
Beavers et al.: 91.7%.
Hanley et al.: 95.9%.
Undifferentiated outcomes: 8.9%.
Beavers et al.: 8.3%.
Hanley et al.: 4.1%.
Of the 2,314 combined outcomes across all 40 years, 92.3% of functional analyses were differentiated and 7.7% were undifferentiated.
Problem behavior maintained by a single function: 61.0%.
Beavers et al. (2013): 75.7%.
Hanley et al. (2003): 85.4%.
Single Function Outcomes
Escape: 21.1%.
Beavers et al.: 29.7%.
Hanley et al.: 34.2%.
Automatic reinforcement: 17.1%.
Beavers et al.: 16.9%.
Hanley et al.: 15.8%.
Tangible reinforcement: 12.5%.
Beavers et al.: 12.0%.
Hanley et al.: 10.1%.
Attention: 9.1%.
Beavers et al.: 17.2%.
Hanley et al.: 25.3%.
Other function: 1.2%.
Combined results:
Escape: 25.9%.
Automatic reinforcement: 16.8%.
Attention: 14.5%.
Tangible: 11.8%.
Some other function: 0.7%.
Multiple Control
Multiply maintained problem behavior: 39.0%.
Beavers et al. (2013): 24.3%.
Hanley et al. (2003): 14.6%.
64.7% of multiply maintained outcomes were obtained from individuals who engaged in a combination of multiple topographies.
Combined results: problem behavior is more often maintained by a single function (69.7%) than by multiple functions (30.3%).
Multiple Control Outcomes
Escape: 86.0%.
Tangible reinforcement: 81.8%.
Attention: 55.6%.
Automatic reinforcement: 4.2%.
Other function: 5.3%.
Function combinations:
Escape/Tangible: 41.2%.
Escape/Attention/Tangible: 23.7%.
Escape/Attention: 13.7%.
Attention/Tangible: 11.6%.
Other combination: 9.7%.
Discussion
Best Practice Recommendations
Hanley et al. (2003) and Beavers et al. (2013) presented best practice recommendations for conducting functional analyses.
Limiting response classes to one or a few behavior topographies.
Programming consequences for the occurrence of target behaviors.
Incorporating EO influences before and during assessment.
Including SDs to facilitate discrimination of test conditions.
Conducting relatively brief (10-min) sessions.
Including tests to identify behavior maintained by automatic reinforcement.
Considering relative reinforcement durations when interpreting analysis results.
Testing for functional relations between problem behavior and tangible reinforcement only when preliminary information suggests a relation might exist.
Starting brief and simple (arranging relatively few test conditions) and progressing to more lengthy or complex assessments as needed.
Using other sources of information as adjuncts to structure more complex analyses.
Sustained Recommendations
Increased reliance on ABC functional analysis models.
Brevity of functional analysis sessions.
Over 90% of studies used an ABC model and reported session duration as 10 min or less.
Deviations from Recommendations
Trend in the opposite direction of the original recommendation to limit response classes to one or a few behavior topographies.
46.7% included multiple topographies compared to 27.8% reviewed by Hanley et al.
Assessing multiple topographies may benefit efficiency, safety, and ecological validity.
Separate response topographies may be part of distinct response classes.
Warner et al. (2020) showed full response class membership for all reported topographies.
High prevalence of multiple control outcomes may be a byproduct of combining response topographies.
Further research needed on:
Prevalence of shared response class membership among distinct topographies.
Reliability of topographical reports from caregivers.
Efficiency and integrity of treatment using synthesized vs. isolated reinforcers for distinct topographies.
Combining distinct topographies may produce unknown barriers during treatment.
Separate treatment approaches targeting separate topographies may be warranted in cases where behavior is multiply maintained by social and automatic reinforcers.
Tangible Reinforcement
Prevalence of tangible test and tangible reinforcement functions has continued to increase over the past 20 years.
68.7% of studies included a tangible condition, compared with 34.7% of studies through 2000.
Trend may reflect an increase in access to tangible items in the typical environment.
Participants may also have access to higher quality tangible stimuli than previously existed.
Functional analyses including tangible test condition may be prone to false-positive outcomes.
Caution should be exhibited across all test conditions by carefully matching evocative situations to only those experienced under natural circumstances.
Automatic Reinforcement
Recommendation garnered minimal attention.
51.6% of studies have included a test for automatic reinforcement, decreased from 59.6% in Hanley et al.
Additional measures have been introduced to test for automatic reinforcement more efficiently.
Querim et al. (2013) described a method for prescreening an automatic reinforcement function with a series of brief, no-interaction sessions.
Separating the test for automatic reinforcement from the standard multielement assessment may have minimized its use to assess topographies less likely to be maintained by automatic reinforcement.
Wide use of interview-informed synthesized contingency analysis and practical functional assessment models may have also contributed to the decreasing trend in tests for automatic reinforcement.
Research is needed to establish best practice related to the nature and process of ruling out the need for tests of the automatic reinforcement function.
Simplicity
Hanley et al. (2003) recommended starting brief and simple, progressing to lengthier or more complex assessments as needed.
Researchers have increased use of assessment formats that are relatively brief and have proposed decision-making models for moving from brief to extended analyses.
Simplicity is including no more components than are necessary to inform an effective treatment.
Functional analysis research has also included tests for reinforcers other than attention, tangibles, escape from work, and automatic reinforcement.
The past decade also continued to produce novel antecedent manipulations.
Novel-consequence manipulations included, but were not limited to, physical vs. nonphysical attention, access to stereotypy with a door, and access to chase after elopement.
Outcome Data
Functional analyses were differentiated in 91.1% of outcomes and indicated a single reinforcement function in 61.0% of outcomes.
Consider potential sampling, submission, and publication biases likely to inflate data when interpreting.
In studies of social functions, researchers may prescreen participants for differentiated functional analysis outcomes as a prerequisite to participation in a treatment-efficacy study.
Researchers also may prescreen participants exhibiting a single-function outcome when a treatment is designed for a distinct social function.
Undifferentiated functional analysis outcomes for behavior maintained by automatic reinforcement have been predictive of less effective treatments, resulting in another barrier to publication.
Comparison of results with a clinical sample is recommended.
Stereotypy was virtually always maintained by automatic reinforcement.
An extended no-interaction screening should be considered before manipulating other environmental variables.
Similar results were obtained for individuals engaging in pica.
If the primary topography of concern is aggression, tests for automatic reinforcement may be unnecessary.
Inappropriate Mealtime Behavior
Separate category added for it.
Displayed differentiated results.
Most frequently maintained by escape alone followed by multiple control by escape and attention.
Future research on the assessment and treatment of problem behavior should use consecutive-controlled case series studies to compare other topographies of problem behavior to the results obtained in the current review.
Current Trends and Future Directions
Publication Trends
42.8% of all functional analysis studies were published within the last decade.
Diversity of journals publishing functional analysis outcomes represents a crucial step in effectively disseminating behavioral assessment.
Encourage researchers to continue seeking diverse outlets for publication of functional analysis research to reach as many scientific and clinical communities as possible.
Reviews
Publications returned by our search that were not included in the current review because they were also reviews.
Examples included:
Functional analyses of elopement.
Brief functional analysis methodology.
Caregiver-implemented functional analyses.
Functional analyses of tics.
Precursor functional analysis methodology.
Functional analyses of inappropriate mealtime behavior.
Functional analyses in public school settings.
Methodological variations.
Direct-care staff data collection.
Training.
Functional analyses of verbal behavior.
Trial-based methodology.
Functional analysis efficiency.
Idiosyncratic functional analysis variables.
Participant and Setting Characteristics
Child participants and participants with autism vastly increased.
Factors may be responsible for increase: Private and public insurance coverage for ABA services for children with autism has expanded, diagnostic criteria for autism in the 2013 revision of the DSM-5, and increased attention to lower severity behavior.
Outpatient clinics were reported in more studies.
Tailored Designs
Expanded coding to capture developments in experimental design.
Most design modifications were developed in response to concerns with the efficiency and safety of the multielement functional analysis, but some were also made to accommodate unique topographies of problem behavior.
Enhancements to Assessment Efficiency and Validity
Researchers have improved efficiency in their functional analysis procedures.
Modifications to the dependent measure (e.g., latency vs. rate), independent measure (e.g., synthesis vs. isolation of reinforcers), design (e.g., trial-based vs. multielement), session duration (e.g., 3 min vs. 10 min), assessment duration (e.g., one session vs. one series vs. multiple series), and mode of visual inspection (e.g., within- vs. between-session; structured ongoing visual inspection).
Renewed focus on assessment validity, which may favor relatively lengthier assessments.
Assessment efficiency and validity goals may continue to be in opposition, especially as treatment predictions become more and more precise and sophisticated.
Future research should consider the predictive validity of within-session patterns of responding during relevant test condition(s).
Decreases in session duration may result in better outcomes than decreases in repetition of conditions below three series.
Functional Analysis Safety
Advancements in the detection of safety risk.
At the very least, researchers should provide specific information on injury prevention resources when conducting functional analyses.
These safety options should be explored further using methodology similar to that described by Kahng et al.
Researchers should consider injury to staff as another relevant measure of functional analysis safety.
Advancements in wearable biosensor technology and machine-learning algorithms have the potential to expand research on problem behavior severity and improve the sophistication of measurement of this important variable.
Telehealth
Applications of functional analyses have been conducted in homes, schools, and internationally.
Telehealth is a feasible modality to improve the reach of functional analysis services and research, albeit with variable implementer integrity.
General Conclusion
Extended reviews by obtaining 326 studies, which produced 1,333 separate functional analysis outcomes over the past 10 years.
Functional analysis research has grown. Methodology has shifted to promote:
Efficiency (e.g., shorter session duration).
Safety (e.g., new experimental designs).
Practicality (e.g., synthesized conditions).
Functional analysis researchers have begun to explore avenues to evaluate and improve the predictive validity of functional analysis outcomes.
Presenting an opportunity for another decade of sustained contributions to functional analysis research.