#ContentAnalysisofABAonTwitter/X: Finding Behavior Analysis’ Heart on Social Media

Abstract

Collecting data on social topics by analyzing verbal behavior on social media is a modern approach to assessing social validity. Given the evolving public perception related to applied behavior analysis (ABA), we sought to gain an objective understanding of online discourse related to the field. In an earlier project, we explored the relevance and sentiment (i.e., positive, negative, neutral) of a stratified sample of Twitter/X posts related to ABA (e.g., #ABA, #BehaviorAnalysis, #appliedbehaviouranalysis) across 11 years (2012–2022). Tweets with negative sentiment garnered engagement scores that were about three times higher than those of positive and neutral tweets. However, the nature of the discourse was unknown. In the present study, we extended the results by conducting a content analysis of a subsample of tweets categorized as indicating either a bias toward or against ABA. Tweets supporting ABA focused on promotion, positive effects on recipients, effective procedures, financial support, and ethical practices. Tweets against ABA highlighted harmful effects on recipients, criticisms of specific procedures, ableist assumptions, and general anti-ABA sentiments. The results were interpreted according to social validity and invalidity, and disability studies perspectives. Our findings have meaningful social implications for ABA practitioners, those who access our services, and the general public.

At the core of behavior analysis is the emphasis on changing environmental circumstances that influence behavior, rather than blaming people for their circumstances—applied behavior analysis (ABA) focuses on changing those circumstances (see Friman, 2021). Consistent with its pragmatic roots, core ethical principles entrenched in behavior analysts’ code of conduct include benefiting others, treating others with compassion, dignity, and respect, behaving with integrity, and ensuring competence (BACB®, 2022). ABA is an expansive scientific discipline with the potential to serve as a general-purpose science relevant to the entire human experience (Heward et al., 2022). However, like other applied professions, the history of ABA includes both favorable and unfavorable examples of applications of our science in practice. There are extensive examples of ABA positively affecting the lives of people with and without disabilities (Heward et al., 2022). Recently, behavioral researchers and practitioners have emphasized compassionate care (e.g., Gatzunis et al., 2023; LeBlanc et al., 2020; Rohrer et al., 2021; Taylor et al., 2019), cultural responsiveness (e.g., Gatzunis et al., 2023; Jimenez‐Gomez & Beaulieu, 2022; Beaulieu & Jimenez‐Gomez, 2022), and trauma-informed intervention approaches (e.g., Houck & Dracobly, 2022; Rajaraman et al., 2022).

Alternatively, some applications of behavior analysis reflect an unfortunate history. These examples include conversion therapy (e.g., Rekers & Lovaas, 19741) and the inappropriate use of contingent skin shock (e.g., articles and videos of the Judge Rottenberg Center on American Broadcasting Company [Powell, 2018, July 1], Cable News Network [CNN, 2012, May 25], National Broadcasting Company [NBC, Glorioso, 2012, April 12]). For progress in any field, including ABA, we need to understand and learn from these events and refocus on the foundations of our science and the goal of making effective and meaningful changes in the pursuit of people’s quality of life (Schwartz & Kelly, 2021). Over 40 years ago, Wolf (1978) introduced the notion of social validity and underscored the importance of considering the perceived acceptability and value of our goals, procedures, and outcomes of the people we support. Relatedly, Baer (1987) also introduced the concepts of social invalidity. Social invalidity refers to the actions of consumers who not only disapprove of certain aspects of the ongoing program but are also motivated to act on their disapproval. This action may involve withdrawing from the program, urging others to do likewise, and lodging complaints with community leaders and the media. Baer et al. argued that as the field of ABA evolves, social validity assessments must evolve to keep up with these developments. Baer et al. hoped that social validity measures would have made significant progress 20 years after they described the field’s seven dimensions.

In a review of social validity assessment in behavioral research, Huntington et al. (2022) found that 47% of intervention studies between 2010 and 2020 included social validity assessments, with a sharp increase in 2019 and 2020. This upward trend is, in fact, a substantial improvement from 13% of intervention studies in a similar review conducted by Carr et al. (1999) and 12% of intervention studies from 1999–2016 identified in the replication review by Ferguson et al. (2019), who focused specifically on the flagship journal of the field at the time–the Journal of Applied Behavior Analysis. Though the proportion of approximately half the relatively recent research incorporating social validity measures is less than ideal, the upward trend is promising. Schwartz and Baer (1991) described four groups of people who could be involved in evaluating the social validity of behavioral services and supports. These groups were (a) direct consumers or recipients, (b) indirect consumers, others who are affected by the intervention, (c) the immediate community, and (d) the extended community.

When further exploring who is routinely included in the social validity assessment identified by Huntington et al. (2022), Huntington et al. (2024) found that authors routinely conducted a single assessment with indirect consumers at the end of the intervention. Direct consumers were routinely (65%) excluded from social validity assessments, and when included, demographic information was inconsistent and often lacked relevant dimensions of diversity (e.g., sex, gender, ethnicity, and race). Thus, while increased reporting of social validity is promising, researchers should also seek to engage with direct consumers meaningfully (Huntington et al., 2024). The latter communities may learn from direct consumers of ABA about the approaches used and outcomes.

Additionally, there are many contexts in which we can evaluate the acceptability of our practices, and innovative methods are available for behavior analysts to assess social validity. Barton et al. (2018) outlined objective methods, including participant observation, choice-making, and preference assessments to establish normative comparisons, blind ratings, maintenance or sustained use, and participant preference regarding conditions (see Hanley, 2010). These methods are particularly suitable for assessing feedback from individuals directly receiving support or interventions and are essential for informing individualized programs and strategies. However, as the “discipline is moving into large-scale dissemination, valid social validity assessment will soon become crucial to survival” (Bear et al., 1987, p. 323). It is important to consider others who are influenced by our practices, including those who are indirectly affected (e.g., family members, supporters, and others within one’s close social network), as well as members of the immediate and extended community (Schwartz & Baer, 1991).

Gathering social validity data is possible using numerous methodological approaches.2 Content analysis is a qualitative approach to organizing and analyzing written text into patterned and meaningful units of analysis (Hsieh & Shannon, 2005; Krippendorff, 2018). Although qualitative approaches are relatively less common in behavior analysis, similar to quantitative approaches, qualitative research is expected to be conducted with a high degree of methodological rigor (Brantlinger et al., 2005; Snodgrass et al., 2022). Content analyses are recommended for synthesizing information (Levac et al., 2010). They have been employed in the field of behavior analysis and have been used by behavior analysts. For example, Vollmer et al. (1992) conducted a content analysis as part of a program evaluation to evaluate how critical elements related to best practices were included in behavioral programs for addressing self-injurious, aggressive, or harmful behaviors. The content checklists were further validated by experts in the field and became one of the only empirically derived systems to evaluate the content of behavior support plans (Williams & Vollmer, 2015). These examples highlight how qualitative approaches may be used to contextualize quantitative data and, most importantly, apply qualitative methods in soliciting feedback on the social validity of our intervention goals, procedures, and outcomes. Relatedly, Heinicke et al. (2021) employed a qualitative approach to code interview data on public speaking in an attempt to improve’ behavior analysts’ dissemination of our science. Further, some behavior analysts have made calls to rethink the place of qualitative methods in behavior analysis to align practices with the needs and priorities of service recipients and other relevant consumers (e.g., Burney et al., 2023, 2024).

Schwartz and Baer (1991) questioned whether social validity and invalidity assessments could be used to identify important and relevant social goals. At that time, Schwartz and Baer noted that this could be problematic due to societal changes and that obtaining a substantial random sample of perspectives from society members would be challenging and costly. Indeed, societal attitudes, beliefs, and values regarding a specific issue or group evolve, often influenced by shifts in cultural norms, power dynamics, and social movements. Today, the content shared on social media may act as a significant indicator of current social validity on a large scale.

Qualitative methods may help advance our understanding of the general acceptance of public-interest topics (Snodgrass et al., 2022). Large-scale social change may be possible thanks to rapid technological advancement and the interconnected nature of social media. Online discourse has demonstrated significant power to influence real-world social events. Countries with oppressive governments have been toppled (e.g., the Arab Spring), predatory sexual practices have been spotlighted (e.g., the #MeToo movement), and large-scale protests and movements for social justice and change have been organized (e.g., Black Lives Matter in Fisher & Rouse, 2022). The above examples represent an aggregate of verbal behavior in response to cultural phenomena, potentially influencing overt actions. The study of online social opinion appears to be a sound approach to measuring social validation.

Negative sentiment toward ABA has been observed and documented by members of the behavior analytic community (e.g., Gorycki et al., 2020; Leaf et al., 2022; Vyse, 2022). These accounts suggest that negative perceptions of ABA are present and possibly increasing (Malkin et al., 2024). Negative concerns are coming to light in recent disability studies research that explores Autistic adults’ reflections on their ABA experiences (e.g., Anderson, 2023; McGill & Robinson, 2020). Some members of the disability community disapprove of ABA and are advocating against ABA and highlighting reports of problematic outcomes (e.g., abuse and ableism; Dettmering & Hodzic, 2024). From a disability studies lens, ableism is a form of oppression that stems from the tendency in society to value people and abilities that conform to what society considers the “normal” or “standard” (Goodley, 2014). It is reflective of the often-unrecognized underlying assumption of the dominant medicalizing of disability and the ideology (i.e., hegemony) of normalcy (Davis, 2014). It is this assumption that presupposes that Autistic people require a cure, treatment, or rehabilitation to pass as “normal” (Rosqvist, 2012). It is expressed in the language and approach that treating autistic people is a problem to be fixed or cured (Anderson, 2023). Repeatedly, this research emphasizes the need to actively listen and respond to the concerns expressed by the autism community (e.g., Anderson, 2023; McGill & Robinson, 2020).

The methods employed and populations studied within disability studies are relevant to behavior analytic practice if considered from a social validity perspective. Baer et al.’s (1987) advice has never been more relevant given today’s social climate; using varied research methods may move behavior analysts closer to the never-ending pursuit of finding our hearts, as Wolf (1978) suggested almost 45 years ago. We also see increased awareness of the importance of avoiding ableist language and assumptions within autism research (Bottema-Beutel et al., 2021). These are hopeful patterns; however, to our knowledge, the impact of such innovations on ABA discourse in the broader online community has yet to be measured. Understanding the narratives with a bias for and against the field can encourage researchers and practitioners to engage in meaningful dialogue with interested parties and use this information to inform research and practices.

Social media outlets are well-suited to examine public perceptions and general uptake of ABA. One such platform is Twitter/X.3 Despite its recent rebranding, it has been used to disseminate information quickly and effectively due to its broad reach (Nau et al., 2022). A tweet consists of limited characters (i.e., text), photos, a GIF, and/or video visible to the public (X, n.d.). Tweets can receive engagement through re-tweets, replies, likes, and quote-tweets, and users can also mention (@) other users or use hashtags (#) to cross-reference content for a topic or theme (X, n.d.).

Malkin et al. (2024) identified over 100,000 tweets using search terms related to ABA, and among 11,000 randomly sampled tweets, 5408 were deemed relevant to ABA. If this proportion holds for the remaining tweets, there is an abundance of online content about ABA. While most of the tweets relevant to ABA were positive, there has been a greater increase in tweets with negative sentiment across the years, and these tweets received engagement scores approximately three times higher than those of positive and neutral tweets. This suggests concerns related to the social validity and invalidity of ABA. However, the nature of the discourse remains unclear. Some concerning hashtags appear in tweets related to ABA (e.g., “#abaharms; #ptsd; #abaisabuse; #abuse; #trauma”; Malkin et al., 2024, p.5). Nevertheless, hashtags provide an incomplete picture of the content since they only indicate relevance to a topic rather than serving as a summary of the post itself. Consequently, there remains a gap in understanding the actual content of those tweets.

Social media is increasingly becoming a primary platform for accessing news online, and many people use the internet to access health information (Martin, 2018). It is important to recognize that the information people find can influence their decisions to seek evidence-based services and can also inform policy decisions. Given that these media are visible to all, including different levels of consumers and policymakers, it is essential to scrutinize them. We sought to gain an understanding of online discourse suggesting the social validity and invalidity of ABA. We conducted a content analysis of the tweets garnering engagement that demonstrated a bias toward and against ABA on Twitter/X.

Method

Content analysis typically involves seven systematic steps: (1) form the research questions; (2) select the sample; (3) define categories; (4) conduct training; (5) conduct coding; (6) determine trustworthiness; and (7) analyze the results (Kaid, 1989, as cited in Hsieh & Shannon, 2005). Kim and Kuljis (2010) proposed a similar process for applying content analysis to web-based content, except they incorporated pilot training and replicated this procedure after all the coding was completed. Steps 5 and 6 were repeated in our study, given the large sample size.

Positionality

Qualitative researchers, being aware of their susceptibility to influence from social, cultural, historical, racial, and sexual factors, must acknowledge how their own experiences and perspectives may impact their interpretations throughout the research process (Creswell & Creswell, 2018). This self-awareness is a critical aspect of qualitative research, as it enables us to accurately represent the perspectives of marginalized groups (Green, 2016) by recognizing and addressing our biases. The first three authors are behavior analysts, each with over 20 years of experience in the field. Collectively, we also engage in a diverse range of behavior analytic and disability studies research. The first author identifies as part of the disability community and primarily conducts mixed methods and participatory research. The second author primarily conducts quantitative behavior analytic research. The third author uses qualitative, mixed methods, and quantitative research approaches (group and single case designs). The fourth and fifth authors are graduate students specializing in ABA. Members of the research team also identify as neurodivergent and have lived experiences of autism, attention-deficit/hyperactivity disorder, and dyslexia.

Sample

We used the Malkin et al. (2024) data set for the current analysis. The data set is available via the Open Science Framework (OSF) platform (https://osf.io/58yse/). Specifically, we downloaded 11 years of data and conducted a preliminary analysis to categorize the data for the current analysis. We assessed verbal behavior in tweets to determine whether the content was relevant to ABA and whether tweets contained a bias toward, away from, or no bias related to ABA. Data were also gathered on user engagement with each tweet (calculated by summing the re-tweets, replies, likes, and quote-tweets, per tweet; Fang et al., 2022). Specifically, mean engagement was calculated by summing all interactions (retweets, replies, likes, and quote-tweets) within each sentiment category (positive, neutral, and negative) and dividing by the total number of tweets in that respective category. For instance, a tweet’s engagement score of two could result from any combination of two interactions, such as one like and one reply or one retweet and one quote-tweet (c.f., Malkin et al., 2024).

Data Gathering, Sampling, and Categorization

As stated above, the data in the current analysis were first gathered and analyzed by Malkin et al. (2024). The original sample of 119,911 downloaded tweets included tweets posted between January 1, 2012, and January 1, 2023. At the time of data gathering (January of 2023), the investigators received approval to access the Twitter API (Academic Track; this provides access to the full historical Twitter archive4), using the “Tweepy scraping library” in the Python programming language (Python code and the entire data set are available for download via OSF). Tweets were included if they were in English and contained any of the following hashtags: #ABA, #BehaviorAnalysis, #BehaviourAnalysis, #AppliedBehaviorAnalysis, and #AppliedBehaviourAnalysis, or the words included in these hashtags appeared in search results using these search terms. Re-tweets were excluded from the data set to avoid counting the same content more than once. Following extraction, to ensure a representative sample, Malkin et al. used a stratified sampling approach; the sample was divided by year. That is, 1000 tweets per year were included for a total of 11,000 tweets. Following training on coding procedures, coders manually scored the stratified sample according to the definitions found in Table 1 of Malkin et al. Briefly, tweets were scored according to their relevance (i.e., related to ABA) and sentiment (i.e., positive—a bias toward ABA; negative—a bias against ABA; or neutral—content that did not indicate a bias). Malkin et al. found 5408 tweets relevant to ABA out of 11,000 tweets, with 2780 (51.41%) coded as neutral, 2369 (43.81%) coded as positive, and 259 (4.79%) coded as having negative sentiment toward ABA. For the content analysis, we narrowed the sample of our content analysis to the tweets with any engagement (engagement scores greater than 0; n = 1250), given that tweets with no engagement produced no discourse among users. See Fig. 1 for a visual depiction of the data screening process.

Table 1.

Sentiment definitions reproduced from citation removed for anonymized review

Sentiment	Description
Positive	Tweet includes words that indicate a bias toward ABA (e.g., include positive words about ABA and the field; tweets that refer to the positive impact of ABA on people's lives; tweet in opposition to statements against ABA (e.g., opinion, people or government; tweet opposed to government restrictions on ABA services); if the tweet explicitly includes words like “effective,” “science-based,” “scientific.” Could also include events, workshops, training, resources, and materials stated WITH emotive emphasis (e.g., positive words – “exciting,” “amazing” or happy emojis or exclamation points, such as “Register for our amazing upcoming workshop!!!”; “I’m excited to give this talk at ABAI”; Include tweets referencing or including ABA memes)
Negative	If the tweet includes words that indicate a bias against ABA; typically negative tweets include hashtags like #ABAisabuse or #ABAharms; sometimes tweets that mention PTSD related to ABA; describe the harms of ABA
Neutral	A statement that does not indicate a bias about ABA (e.g., that does not include any positive or negative words); includes tweets related to job postings, events, workshops, training, resources, and materials stated WITHOUT emotive emphasis (e.g., positive words—“exciting”, “amazing” or happy emojis or exclamation points; “Register for our upcoming workshop.”; “I’m doing a talk at ABAI.”)

Reprinted from “#ExploratoryAnalysisOfSentimentTowardABAonTwitter,” by A. Malkin, P.B. Riosa, L. Mullins, K. Thompson, and A. Kretschmer, 2024, Behavior Analysis in Practice, 56(5), p. 765. Copyright 2024 by the Association for Behavior Analysis International

Fig. 1.

Data set: gathering, sampling, and categorization

Developed Categories

To develop the coding system, we further identified a sub-sample of tweets with the most engagement (i.e., a total > 4 engagement score; n = 365). In an attempt to ensure the coding system comprehensively represented the full ranges of the discourse within the sample, the first author reviewed all the tweets coded as biased toward ABA (n = 276) and against ABA (n = 89) and inductively identified potential codes based on recurring patterns of issues or comments. Inductive coding is a process that explores and discovers patterns and concepts without predetermined categories. This process began when the first author read and re-read each tweet and generated a list of codes describing that data with a single word or short sequence of words. These codes were then categorized and labeled on the basis of common forms of tweets (e.g., promotion, dissemination, Memes, and GIFs) and considering the dimensions of social validity (i.e., goals, procedures, and outcomes) and relevant disability studies concepts (e.g., hegemonic normalcy). We used these categories to develop a codebook with category labels, descriptions, and examples (Guest et al., 2014). The codebook was used during training to enhance coding consistency among different coders.

Training and Coding

Two master’s level graduate students specializing in ABA were trained to code tweets (the fourth and fifth authors). We reviewed the codebook and discussed multiple examples. The observers independently coded small samples of tweets (n = 40 [20 + /20 −]) and then met with the first author to review coding (i.e., compare results, address questions, and resolve disagreements based on consensus). This process was repeated (n = 80 [40 + /40 −]) until acceptable Cohen’s kappa values were attained (k = 0.72–0.81; Banerjee et al., 1999).

The sub-sample was then divided between the two observers, with 39% of the tweets coded by both observers (k = 0.86). The team met to review the coding (as identified above). To further confirm the trustworthiness of the coding system, the first author reviewed all the tweets in each category to determine whether the category labels and descriptions were representative of the data. After this review, minor revisions to the category labels and descriptions were made to the codebook, and recurring hashtags that were unique and representative of the category were identified.

The two observers received booster training with a revised codebook and conducted a sample coding (n = 100 [50 + /50 −]; k = 0.83). The remaining data set (n = 1024) was then divided equally between observers, with an additional 33% of tweets coded by both observers (k = 0.95). After coding, the team met again to review coding. We conducted a final trustworthiness assessment by dividing the categories between the observers who reviewed all tweets in each category to determine if category names and descriptions were representative.

Results

Bias Toward ABA

Six categories representing a bias toward ABA were identified. The frequency of tweets representing a bias toward ABA (n = 1046) per category is presented in Table 2. Table 2 also includes examples and recurring hashtags indicating common categories for a bias toward ABA. The dissemination and promotion category was the most common (n = 678). It included tweets that promote events (e.g., conferences, training, podcasts), jobs, services, organizations, and achievements (e.g., passing the BCBA exam, new employment, earning awards). The following most common category, highlighting effective procedures (n = 120), included references to the effectiveness of the ABA procedures and approaches (e.g., citing research, talking about data, describing effective interventions or procedures). The positive effects on ABA recipients (n = 96) category refer to improvements or achievements that could affect the quality of life for people who received ABA (e.g., independence, doing new things, learning skills, and reducing challenging behaviors).

Table 2.

Frequency and recurring #s of tweets representing a bias toward ABA

Bias Toward ABA	Recurring #s	Sample Tweets	n
Dissemination and Promotion	#BCBA #RBT	No more waiting for ABA therapy for your child on the Autism spectrum. Contact us today to get started for immediate services. #ABA #ABAtherapy #autism #autismspectrumdisorder #ASD #behaviortherapy #oldham #louisville #shelby #jefferson #autismtherapy #nowaiting #nowaitlist We are very proud to announce that BlueSprig’s Learning and Performance Development team were awarded the 2022 Silver #Learning in Practice #Award in the category of Talent Management from Chief Learning Officer. #appliedbehavioranalysis #behaviortherapy #bcba #autismadvocates	678
Highlighting Effective Procedures	#Research #Data	People are often shocked at how well positive beh. strategies work. #ABA is real science that people go to school for, not made-up flimflam Give them attention for expected behavior, build relationships, pair yourself with reinforcing things (like caring/kind attention) so that your presences become reinforcing! #RelationshipsMatter #AppliedBehaviorAnalysis #LetFreedomRoar Using the 7 dimensions of #ABA, behavior analysts ensure that your child's treatment plan is evidence based, socially significant, and individualized!	120
Positive Effects on ABA Recipients	#Independent #Lifeskills	Increases Life Satisfaction ABA provides autistic children and their families with a priceless gift: the capacity to enjoy life when children gain independence, social skills, and the ability to cope with frustration. #ABA #LifeSatisfaction Farley said Greyson, who once was non-verbal and had difficulty understanding words, has made major strides after receiving ABA therapy for a little over six months and gradually increasing the time with his Autism therapist. #JerseyShore #Autism #ABA #TodayInABA my four boys who were all non verbal and severe at diagnosis, 16 years later are now studying their chosen passions, art, media, photography, music and of course my star runner #autism #ABA	96
Behavior Analysis Memes	#ABA #Autism #Opinion	Like, seriously! 😂 😂 😂 #aba #appliedbehavioranalysis #appliedbehavior #behavior #abameme #abamemes #abajoke #abahumor #learnaba #psychology #bcba #bcaba #rbt #beardedbehaviorist #learnaba #OhBehavePodcast #schedulesofreinforcement #compoundschedulesofreinforcement	88
Financial Support	#AutismDoesntEndAtFord #OAP	I want to teach you how to swim and ride a bike and fly a kite and play guitar and read a book. But first we need to teach how to use your words, to hold a spoon, to sit on a toilet, to hear your name and to stay safe. #NeedsBasedTherapy #AutismDoesntEndAtFord #ABA It is CLEAR non-medical professionals will be deciding/speaking over TRUE medical need. Bureaucracy and well funded "gate keepers" have always drained and corrupted the OAP. #MedIcareForAutism #SevereAutism #Nonverbal #aba/IBI Families cheer passage of #autism coverage bill #ABA #insurance #AutismAwareness	80
Promoting DEI and Ethical Practices of ABA	#EqualityIsNotEquity #Ethics	One, cannot blame an entire practice for the actions of a few. #ABA As behavior analysts we are always learning and always striving to be better. Practicing compassionate #ABA can mean a lot of things. These are just some of the ways we can show up to serve our clients with compassionate care	47

Behavior analysis memes (n = 88) included tweets with amusing or interesting statements (including quotes by predominate figures in the field, images with text, or short single-scene repeating videos) that are relevant to those with the behavior analytic community (including references to positive representations of autism) and often made references to popular culture (e.g., The Office, Seinfeld, or the Muppets). This category did not include photos that celebrate accomplishments (e.g., graduation, team photos, training sessions), flyers for materials, articles, podcasts, training events, or images with lots of text or infographics.

The next category referenced the need for the financial support of ABA (n = 80). This category included references to the importance of funding ABA, including advocating for increased funding, creating awareness about insurance changes, and advocating against funding cuts. Of particular note, this category included the hashtags #AutismDoesntEndAtFord and #OAP in response to the 2019 changes to the Ontario Autism Program (OAP) by the Ontario Provincial Government in February 2019 (OAP Advisory Panel, 2019).

The last category included tweets about promoting diversity, equity, and inclusion (DEI) and ethical practices of ABA (n = 47). This category referenced the importance of ensuring the ethical application of ABA; promoting diversity, equity, and inclusion in ABA; and expressing a values-based perspective of care, cultural humility, and compassion. There were also references regarding promoting the regulation or licensure of the field of behavior analysis.

Bias Against ABA

Table 3 presents the frequency of tweets representing a bias against ABA (n = 204) per category. Table 3 also includes examples and recurring hashtags indicating common categories demonstrating a bias against ABA. The harmful effects on ABA recipients (n = 99) category included tweets referencing the potential negative and harmful effects or outcomes ABA has on service recipients, including specific concerns related to negative long-term outcomes and impact on mental health. This category also included references to ABA producing trauma, torture, harm, and abuse. The hegemonic normalcy (n = 93) category referenced ableism within ABA. That is, the idea that ABA attempts to make people conform to societal norms. This category included critiques around the tendency to target behaviors that the Autistic community believed to be problematic (e.g., stimming, eye contact, quiet hands, compliance) and efforts to make Autistics behave like their “non-autistic,” “typically developing,” or “neurotypical” “peers” and “classmates.” This category also included tweets referencing concerns that Autistic experiences are devalued and ignored. There were some references that ABA can be dehumanizing or used to change a person’s personality or identity (e.g., including conversion therapy).

Table 3.

Frequency and Recuring #s of Tweets Representing a Bias Against ABA

Bias Against ABA	Recurring #s	Sample Tweets	n
Harmful Effects on ABA Recipients	#ABAisAbuse #ABAHarms	Ah, how I wish there were something like #ShinigamiEyes but for marking people/organizations that are pushing harmful things like #ABA or #MMS for #autism * #ABA "therapy" gives Autistics a staggering #PTSD rate: https://t.co/idUzhr9wyY #YesAllABA is abuse. #BanABA You may feel the urge to tell me that many parents support #ABA, and say that ABA works. I'm aware that they say so. Many men also feel that 'it works' to beat their wives, and that it makes them more compliant. 'It works' does not mean 'it's right'	99
Hegemonic Normalcy	#ActuallyAutistic	#ABA cannot cure #ASD. With that being said, a cure for autism is not something that we’re seeking in the first place ** A huge disconnect between what #ABA industry thinks is an important outcome (what you can physically do for yourself, compliance and how non autistic you seem) and what autistic people are saying is important for Quality of Life. https://t.co/2y8Cj3QYfo #ABAresearch #Autism No other disability seems to have such a thriving circus spinning around it, touting cures / recovery / reversal etc. Being threatened for speaking out (also against quack cures & treatments) has brought this experience up a few notches though. #Autism #ABA #ActuallyAutistic	93
Anti-ABA	#StopABA #PBSFails #ABAcult #ABASham #BetterwaysthanABA	Before you waste money on becoming an #BCBA or an #RBT, talk to the #ActuallyAutistic and #actuallyadhd communities about how we feel about #ABA, #pbs. Because you are wasting money An important article discussing research and evidence showing that #ABA / #PBS does not work and explaining why. it also discusses #BetterWaysThanABA such as Collaborative & Proactive Solutions. #PBSisABA Teaching blah blah non-scientific autism blah blah non-scientific problem behavior blah blah non-scientific blah blah #non-scientific #ABA	88
Harmful Procedures and Approaches	#StoptheSkock #ConversionTherapy	Watched Birds of Prey this week. There was a scene involving torture with a cattle prod. I felt the need to point out that the JRC does this to children every day for longer and at higher strengths for stimming. #stoptheshock #aba 2nd employee of #autism center backs up allegations that nonverbal boy was locked in room at #Bierman #ABA Centre While we celebrate the ban of 2SLGBTQ + conversion therapy coming into effect here in Canada, we want to remind all our followers that conversion therapy still lives on in #ABA—and all of us should invest in the movements out there to ban this awful practice	28

^* Tweet included a link to Kupferstine (2008). ** Tweet included a link to McConachie et al. (2020)

The anti-ABA statements (n = 88) category included statements that reject the practice of ABA because it is reported to be unethical (e.g., of unethical allegations) or statements that the evidence used to support ABA is inaccurate (i.e., rejecting the science of ABA). Some more specific examples included statements that ABA is a financially driven field and other statements reflecting dismissive attempts to positively rebrand ABA. Overall, this category included many general references against ABA.

The smallest category from our analysis was harmful or ineffective procedures and approaches (n = 28). This category included references to specific procedures perceived or labeled as problematic. The specific procedures that were criticized included punishment and aversive techniques in general, and more specifically, restraints and contingent electric shock, and other procedures, including planned ignoring and ABA being used as a form of conversion therapy.

Discussion

This study sought to understand the nature of the discourse on ABA on Twitter/X by analyzing a subsample of tweets over 11 years. Six categories of tweets reflected a bias toward ABA, including (1) dissemination and promotion, (2) positive effects on ABA recipients, (3) highlighting effective procedures and approaches, (4) behavior analysis memes, (5) financial support for ABA, and (6) promoting ethical practices in ABA. Support for ABA spanned a range of topics, such as promoting events, jobs, services, and achievements, emphasizing the effectiveness of ABA procedures, advocating for increased funding for ABA, promoting ethical practices, and referencing positive framing of autism. In comparison, four categories of tweets represented a bias against ABA, including (1) harmful effects on recipients of ABA, (2) harmful procedures and approaches, (3) the applications of ABA based on ableist assumptions related to hegemonic normalcy, and (4) general anti-ABA sentiments. These biases against ABA reflected concerns about the ethical application of ABA, apprehensions regarding the prioritization of financial gain, and criticisms of specific procedures perceived as harmful or ineffective.

Unsurprisingly, the most common types of tweets involve the promotion of dissemination, marketing, and behavior analysis memes, as social media is often used to raise awareness, advertise services, connect with potential customers, and promote products or services. However, other “toward” and “against” categories appear to reflect the commonly understood dimensions of social validity and invalidity (i.e., goals, procedures, and outcomes; Wolf, 1978). The “toward” and “against” categories included references to effects recipients, which may reflect desired or harmful goals and outcomes, and identified interventions that may be effective or considered problematic. While we appreciate effectiveness is not the same as acceptability (Schwartz & Bear, 1991), references to these interventions may reflect the social significance or acceptability, or lack thereof, of these to society. When demonstrating a bias toward ABA, tweets endorsed its effectiveness and positive impact on recipients while enhancing DEI and ethical applications. Those who demonstrated a bias against ABA highlighted unethical applications and problematic interventions that may present concerns about being socially invalid. Further, we must reconsider the underlying ableist assumptions that may inform how we assess goals and outcomes and focus on more neuroaffirming approaches and ways of being. The results support that these dimensions are indicators of social significance and have implications for research and practice.

Both categories included additional areas for professionals to consider when designing, implementing, evaluating, and disseminating ABA in practice and research. Within the bias toward ABA category, there were examples of language highlighted as problematic within the bias against ABA category. For example, the term therapy was often used in the Dissemination and Promotion category without further context (e.g., “No more waiting for ABA therapy [emphasis added] for your child on the Autism spectrum” (#3895, Fusion Autism Center, 2022), which can perpetuate the assumption that Autistic people must be fixed, treated, or cured. Further, the Behavior Analysis Memes category included repeated examples of sentiments that incorporate a condescending tone, such as “All behavior is functional #ABA. My face when I hear “their behavior came out of nowhere” (Tweet# 26,261, NEISD SE BEHAVIOR, 2019), or “The 5th #dimension of #ABA is #Effective: We #collect #data on objective measures of behavior and make changes to… http://t.co/gS0vlGMvGx (Tweet #88,275, Omar Rogue, 2014), which could imply that opinions without objective data may be dismissed and have little validity. These messages might be tongue-in-cheek, good-natured fun, or ribbing at people’s views in the field. However, as with all humor, it might be a case of the fact that “it’s funny ‘cause it’s true”—Homer Simpson. Behavior analysts have long been taught that data-based decision-making trumps all other sources of information. This is especially true of data presented in single-case research designs (e.g., Baer et al., 1968; Cooper et al., 2020; Johnston & Pennypacker, 1993). An over-reliance on this type of data to inform practices in the field has been argued to have hampered the growth of the field and the ability to answer novel research questions; group designs and qualitative research methods have recently been suggested to remediate these shortcomings (see Burney et al., 2023; Critchfield & Reed, 2017; Dixon et al., 2023). Similarly, behavior analysis has a complex history of apprehension with thoughts and feelings (i.e., private events; cf. Baum, 2011; Dougher, 2013; Friman et al., 1998). A focus on objective behavior in research and practice while ignoring sentiment toward that research and practice may have contributed to some of the current sentiments toward the field. Without thoughtful consideration of our use of language in both private and, in this case, public contexts, behavior analysts are at risk of losing ownership of their own narratives.

Alternatively, there were also examples within the bias against category that perpetuated misinformation about the evidence against ABA. For example, a tweet in the bias against the ABA category referenced Kupferstein’s (2018) study, which reported a link between elevated levels of PTSD symptoms among Autistic people who experienced ABA interventions despite the methodological and conceptual limitations of this research (see Leaf et al., 2018). These areas are particularly relevant in the context of increased advocacy from the Autism community about ableism. Nevertheless, behavior analysts often meet these reflective endeavors with attempts to correct or clarify misconceptions (see Arthur et al., 2023; Cox et al., 2018). Though such responses are certainly warranted to inform scientific discourse further, these commentaries may, however, fail to fully appreciate the underlying concerns associated with a medicalized model of disability. This perspective, raised by the disability community more than 30 years ago (Oliver, 1990), does not reject all attempts to address impairments through intervention but rather the assumption that disability is seen as a problem that must be fixed, treated, or cured. Indeed, seminal works in behavior analysis touted results associated with providing treatment to children on the autism spectrum such that they are “indistinguishable from their normal friends” (Lovaas, 1987, p. 8). Although examples of “today’s behavior analysis” (e.g., Hanley, 2001; Rajaraman et al., 2022) would likely reject the approaches used in Lovaas’ early work, common hashtags, such as “#autismtherapy” illustrate why the Autism community may reject the support of the field. It is worth noting that given the negativity bias or the tendency for negative information to receive more attention online (Qahri-Saremi & Montazemi, 2023), our sample may not fully account for the current sentiment regarding ABA in society. However, given that this negative valence verbal behavior has greater levels of engagement and can influence the behavior of important parties with a vested interest, systematic approaches to increase the ability of behavior analysts to communicate with the public are warranted and are essential for the continued acceptability of the field overall.

To move forward, the field may use the current study as a data point indicating that some improvement might be required in effective science communication. This type of work has been undertaken in other fields with great social importance (climate change, pandemic mitigation, etc.) and is a field of study unto itself (cf. Jamieson, 2017). Effective science communication empowers stakeholders by informing them about the benefits, risks, and other costs of their decisions related to that science. This enables them to make well-informed choices and sound judgments (Fischhoff, 2012). Behavior analysts may benefit from the study of communicating behavior analysis. Researchers in the field are also well-suited to carry out this research, given decades of research on verbal behavior (e.g., Hayes et al., 2001; Skinner, 1957; Sidman, 1971; and the resultant lines of applied research). A tutorial on effective science communication is out of the scope of this manuscript. However, we suggest that behavior analysts make contact with some online resources from outside of the field (e.g., the American Association for the Advancement of Science AAAS Communication Toolkit) and from resources facilitated by the field (e.g., O’Donnell, 2022).

Methodological Rigor

Throughout this research, we engaged in numerous activities to enhance methodological rigor and trustworthiness (akin to internal validity and reliability; see Lincoln, 1995) in our qualitative inquiry. In alignment with research ethics guidelines and practices, our investigation utilized exclusively publicly accessible data from Twitter/X. This methodological decision reflects established research standards wherein data collected from public social media forums constitutes publicly available information for research purposes. In contrast, content shared within private online communities or closed groups necessitates explicit consent and additional ethical considerations due to inherent expectations of privacy and confidentiality among their members.

We acknowledged the researchers’ positionality and worked collaboratively through the research, engaged in rigorous data analysis, and maintained a detailed audit trail to ensure transparency and accountability. This diverse experience with ABA and lived experiences of disablement was particularly relevant to the study as it allowed the team to strategically integrate rigorous methods (e.g., inter-coder agreement and bracketing) while being keenly aware of the more nuanced experiences of ableist language and hemogenic assumptions in the codebook development, coding, and interpretation. We maintained acceptable Cohen’s Kappa values during coding (Banerjee et al., 1999). Further, when coding, observers also identified tweets that warranted further discussion by other research team members. We aimed to provide a comprehensive and reliable overview, and we took great care to ensure that our methods and findings were consistent with best practices in the field (Brantlinger et al., 2005). For example, we included several sample tweet categories to enhance the authenticity by allowing readers to determine how representative the category is. That said, several limitations are worth noting.

Identifying the appropriate sample can be challenging when conducting a content analysis of web-based data (Kim &Kuljis, 2010) and represents a limitation. This was certainly the case herein, given the amount of data initially downloaded. As we were unaware of any standardized methods to determine appropriate sample sizes, we elected for a large yet feasible sample size. Our process involved an iterative sampling procedure that attempted to achieve a representative sample of the tweets that received engagement across time. However, although the total engagement score is a commonly used measure in the literature (Fang et al., 2022), the cutoff score we used may have influenced the current study's findings. It was observed that tweets with 1–4 engagement scores were more often flagged by the researchers as warranting further discussion as they may have been less explicit (the content was more ambiguous, or bias was less apparent). Given that we used a stratified sampling procedure, this is the most unbiased method; however, we cannot determine whether our data set represents the entirety of the sample. A further limitation related to sampling was the rebranding of Twitter/X, which occurred in July 2023. During our initial coding for sentiment, we included both the content of the tweet and any relevant links. Over time, and particularly following the rebranding of Twitter to X, we encountered numerous broken or restricted links, particularly within the Dissemination and Promotion and Behavior Analysis Memes categories. For example, during the final trustworthiness assessment of all the tweets in each category, more than 200 tweets previously coded as Behavior Analysis Memes could not be confirmed (i.e., cannot be found online). This loss of content may be due to several reasons, and it may not be possible to identify which reason affected which tweet. For example, changes in Twitter/X algorithms for posting content may have occurred and were not explicitly communicated to the public. Further, the loss of content in this manner is a common issue internet-wide and has been termed “link rot”; links break due to issues such as content being moved, renamed, deleted, or sites disappearing. Additionally, changes in URL structure, like the shift from “http:” to “https:” or the omission of “www,” can also cause links to break (Kille, 2015).

A final limitation inherent in our results reflects a manifest level of analysis, as tweets were coded on the basis of the words and images included and without confirmation of authorship. We avoided making assumptions about authors and whether their tweets reflected direct experience with the field or lived experience (e.g., lived experience of autism or parents of children on the spectrum). This approach was an attempt to be more objective and transparent but may not reflect the tweet author’s perspective bias. That is, it is possible that behavior analysts who support the use of ABA may have created tweets that were satirical or an attempt at humor in nature, which may have been coded as biased against ABA depending on their choice of terms.

Future research could involve a deeper latent analysis of the tweets that might reveal hidden meanings and linguistic functions for more nuanced discourse analysis. Such analysis could benefit from drawing critical disability studies and critical autism studies.

Consideration for Practice

In light of these methodological implications, the following next steps for the field might be the long-advocated changes in how behavior analysts communicate, collaborate, and perceive the role of the verbal of others (e.g., Boivin et al., 2021; Doughty et al., 2012; Foxx, 1996; Henderson et al., 2023; Hineline, 2018). This online discourse might influence behavioral services’ selection (or lack thereof) owing to perceptions related to social validity and invalidity. Further, these types of analyses may guide future strategic initiatives, such as the development and dissemination of targeted, evidence-based information to better support parties with a vested interest in their efforts to gather information online (Whisenhunt Saar et al., 2023). Rejection by the autistic community can have significant detrimental outcomes for both the state of research and the practice of ABA. The influence of tweets with a bias against ABA may have cumulative effects. At the individual level, people who see these posts may think twice before contacting an ABA practitioner to help their child. The number of people with this thought is likely proportional to the number of people those posts reach. However, those with the potential to influence greater numbers of people may also see and be influenced by the same posts. These perceptions can affect systems that support ABA. Policymakers may take notice of the perceptions of the field and reduce or eliminate funding; reviewers for grant funding agencies may also hesitate to fund research perceived to be unacceptable. For example, the frequency of the recurring #AutismDoesntEndAtFord and #OAP was likely due to public disapproval in response to proposed changes in funding eligibility for services through the Ontario Autism Program (OAP Advisory Panel, 2019). Alternatively, poor perceptions of the field pose risks to the ability to provide much-needed services and can halt progress due to a lack of support for large-scale research.

A way forward may be to increase the practice of participatory and inclusive research methods, whereby researchers collaborate with the Autism community to better understand their concerns and experiences. Additionally, future research would benefit from a content analysis of tweets by behavior analysts that may include ableist language. There have been attempts to raise awareness about problematic and ableist applications within ABA (e.g., Bottema-Beutel et al., 2021; Shyman, 2016, 2019; Veneziano & Shea, 2023). In addition, #ActuallyAutistics could be invited to observe and report on actual applications of ABA approaches to promote the “televisibility” (Hanley, 2021) of our field and align the preferences of different relevant groups of people with practices in the field (Cox & Javed, 2023).

Conclusions

Ultimately, technological progress in how behavior analysis is practiced is critical, but this progress must be coupled with efforts to achieve social acceptability. In the words of Baer et al. (1987), “If an intervention is socially invalid, it can hardly be effective, even if it changes its target behaviors thoroughly and with an otherwise excellent cost–benefit ratio; social validity is not sufficient for effectiveness but is necessary to effectiveness” (Baer et al., 1987, p.323). Once both effectiveness and social validity are achieved, perhaps behavior analysis can flourish at an even greater scale.

Author Contribution

Laura Mullins: conceptualization, methodology, investigation, data curation, formal analysis, visualization, supervision, project administration, & writing – original draft, review & editing.

Albert Malkin: conceptualization, methodology, formal analysis, writing – original draft, review & editing.

Priscilla Burnham Riosa: conceptualization, methodology, formal analysis, writing – original draft & review.

Allison Kretschmer: investigation, formal analysis, visualization.

Jess Walker: investigation, formal analysis, visualization.

Funding

The authors declare that this research received no external funding.

Data Availability

The Python code to download Tweets from X/Twitter and the datasets generated and analyzed during the current study are available in the Open Science Framework repository, https://osf.io/58yse/; https://doi.org/https://doi.org/10.17605/OSF.IO/58YSE.

Declarations

Competing Interests

The authors declare that they have no conflict of interest.