A recent study of the communication technique rapid prompting method (RPM; a.k.a. spelling to communicate)1 in a prestigious journal bears the bold title “Eye-Tracking Reveals Agency in Assisted Autistic Communication.” Unfortunately, the study does nothing of the kind. It does, however, reveal the lengths to which proponents of this unsubstantiated communication method are willing to go to avoid the simple tests that would actually prove it works.
Details of the study will follow, but for now let me explain the problem in a nutshell. Questions of agency are questions of cause. An agent causes things to happen, and for hundreds of years, scientists have known that the best way to demonstrate a cause and effect relationship: identify the causal variable by systematically eliminating other possible causes. Scientists use control groups, placebos, or other experimental controls to rule out potential alternative hypotheses. This is also known as manipulating the independent variable.
The authors of this study chose not to do that. Instead they engaged in a highly detailed analysis of the dependent variable—the autistic participants’ eye movements and pointing—and concluded that, given the shape of their behavior, the participants must have been the agents (authors) of the words written. This kind of argument is reminiscent of supporters of intelligent design who claim that, due to the great intricacy of the human eye, human eyes could not be the product of evolution and must have been the work of an intelligent designer. In science, we don’t merely observe a phenomenon, throw up our hands, and say, “Well, it must have been caused by X.” We test for X and all the other possible causes.
The authors could be forgiven if they were astronomers rather than psychologists. Astronomers are limited to observation. They cannot conduct experiments on stars and planets to test hypotheses. They can only use sophisticated methods of observation and do their best to speculate about the workings of the universe. But the authors of this study were not limited in this way. The general methods of testing for a cause-and-effect relationship in this case are widely known and easy to administer. They just chose not to use them.
The study in question was published in the open access journal Nature Scientific Reports on May 12, 2020, and was written by Vikram K. Jaswal, Allison Wayne, and Hudson Golino. Rapid prompting method (RPM), the technique studied in the article, involves an assistant holding a letterboard in the air in front of the autistic person, who points at the letters on the board to spell out words and sentences (see Figure 1). The glaring alternative explanation for the origin of the words tapped out is created by the involvement of the assistant. Given that the assistant is not touching the participant, the supporters of RPM have never given an adequate explanation for why the board cannot simply be placed on a table or mounted on an easel and must instead be held—often being moved around in the air—by another person. In addition, why can’t the letterboard be replaced by a touch-sensitive tablet, eliminating the need for the assistant to call out the letters as the participant points to them on the board? (I am getting a little ahead of myself here but bear with me.)
These questions are quite relevant because, as readers of this column will recall, RPM has been described by its critics as an alternative version of the thoroughly discredited technique facilitated communication (FC). FC also involves the help of an assistant—a facilitator—who holds the hand, elbow, or shoulder of a nonspeaking person as they type at a keyboard. Many double-blind tests of FC in which the facilitator did not know what was being asked of the typist have shown that in a Ouija board–like phenomenon, the facilitators were the unconscious authors of the messages on the keyboard—not the nonspeaking person with autism. Correct responses were obtained only if the facilitator knew the answer (Hemsley et al. 2018; Schlosser et al. 2014).
The findings with FC were so devastating that several professional and advocacy associations issued policy statements against its use (Behavior Analysis Association of Michigan N.d.). More recently, the American Speech-Language-Hearing Association issued a separate statement against the use of RPM, citing the lack of scientific evidence for RPM and its similarity to FC. They concluded that “information obtained through the use of RPM should not be assumed to be the communication of the person with a disability” (American Speech-Language-Hearing Association 2018).
Despite the lack of evidence for these techniques and the professional positions against their use, both FC and RPM remain popular with parents of children who have the nonspeaking form of autism. The reasons for this are easy to understand: these methods offer the hope that nonspeaking children do not suffer from a cognitive deficit at all. Rather, they are intellectually typical—even superior—and their disability is merely a physical problem that can be overcome with the help of a facilitator or assistant. As the Jaswal et al. (2020) article points out, some people have, with the help of their facilitators or RPM assistants, gone on “to graduate from college, to write acclaimed poetry and essays, and to publish a best-selling memoir” (1). Those are truly wonderful accomplishments, but nagging questions remain: Who wrote those college papers and exams? Who penned the poetry and essays? Who wrote the memoir?
To further articulate the stunning nature of the claims made by FC and RPM supporters, consider how we usually learn to speak, read, and write. Typically, a child first learns to talk and carry on a conversation, then to read by decoding written words and sentences, and finally to write sentences. The claim of both FC and RPM is that individuals with autism have skipped over the first two steps. Merely through passive exposure to printed and spoken words in their environment, they have acquired the necessary skills to begin to type out words and sentences. Such an unusual claim is perhaps not impossible, but it would require convincing evidence. So far, the evidence is lacking for both FC and RPM, and the recent article by Jaswal et al. does not provide much additional clarity.
The study was designed to track the eye movements of people with the nonspeaking form of autism as they tapped out words on a letterboard. The participants were nine people (eight males and one female) who “were recruited from a centre that specializes in supporting children and adults with limited speech to access other forms of communication. … One of the communication methods taught at the centre involves the letterboard that is the subject of this study” (Jaswal et al. 2020, 7). The same assistant held the letterboard for all nine participants during the test sessions, which were all conducted at the center where the students received RPM instruction. The assistant is described as someone who had been providing instruction in the use of the letterboard for over four years and had “interacted regularly with each [participant] for at least two years” (Jaswal et al. 2020, 8). So, the participants and the assistant were very familiar with each other and had been conversing via the letterboard for at least two years. The participants had been instructed in the use of the letterboard for an average of 3.24 years.
According to the description provided, the data were collected during a lesson. The assistant read a text to the participants that was broken into seven sections, each two to six sentences in length. After each section, the assistant asked the participant between one and six questions. The questions were of varying complexity, ranging from “Spell ‘decade’” to “What’s something you’ve had to wait for?” Although the example video clips provided with the article show only the more complex, open ended questions, presumably all the questions asked were used as data for the study. None of the participants had been previously exposed to the lesson or the questions.
During the study, participants wore a sophisticated piece of equipment built into a pair of glasses. The glasses contained two tiny video cameras—one pointed at the participant’s eye, and another, the scene camera, pointed out in the direction the participant was looking. Computer software tracked the movement of the participant’s pupil and was able to determine where the participant was looking at any given moment.
The goal of the experiment—which was achieved—appears to have been to demonstrate that the participants’ eye movements anticipated the next letter in a word. As depicted in Figure 3, the authors aimed to measure both the finger movements and eye movements of the participants to show the following sequence: fixation of both eye and finger on letter n-1 (“L”), fixation of eye on letter n (“I”) with finger in the air, and fixation of both eye and finger on letter n (“I”). The videos produced by the eye movement software were the raw data that was subsequently scored by two coders who recorded finger taps and the location of the eye-gaze based on the movement of the red dot.
The coding data were then submitted to several sophisticated statistical analyses and presented in data tables and beautiful multi-colored graphs.
The anticipatory gaze at the upcoming letter, combined with the speed and accuracy of spelling, were offered as evidence of agency and against any influence or cueing by the assistant. In addition, the authors point to two pieces of evidence in particular that they say was typical of non-autistic typists: (1) the participants paused longer on average between words than between letters within a word, and (2) the participants were quicker to tap the second letter of a pair of letters within a word if that pairing was more common in English usage. For reasons I will explain below, the first of these two findings is trivial, and the second seems to be exactly what you would expect from practicing this technique—whether you were being cued or not.
To their credit, the authors suggest that their results are limited to the nine individuals tested, and they make no claim that they are indicative of the abilities of all nonspeaking people with autism. They do, however, suggest that based on their results, the cognitive abilities of nonspeaking people with autism have been “significantly underestimated” (Jaswal et al. 2020, 7).
Critical Analysis of Jaswal et al.
To fully understand RPM and how it works, you need to watch the video clips provided with the article. They represent a small fraction of what must have been the raw data of this study, but they reveal much more than the elaborate statistical analysis provided in the article.
If you only read the published article, you would be surprised to discover that the assistant was much more active than the description in the methods section suggests. The authors reported that the assistant read the lesson and asked the questions, but they didn’t describe the behavior of the assistant while the participant was responding. In most cases (with one notable exception), the assistant was calling out each letter as it is tapped, and when a word was completed the assistant read the full word out loud. For example, at the beginning of the video for Participant 2, the assistant can be heard saying, “T-H-A-T,” “THAT,” “I-S,” “IS,” “H-A-R-D,” “HARD.” It seems likely that this feedback from the assistant serves as a reinforcement for pointing to certain letters, and that the spoken word serves as reinforcement for the creation of a correctly spelled word. Now imagine an average of 3.24 years of this kind of direct feedback. The student might eventually build up a collection of motor sequences that will satisfy the assistant and be rewarded by her verbal responses and the eventual completion of the assignment.
This kind of active involvement of the teacher or assistant seems to be a defining feature of RPM. For a dramatic example, see the beginning of the film produced by the Washington Post below. In this case, the assistant must repeatedly cajole and coax the young man to keep going. To me, he looks decidedly unhappy as he rocks, grunts, and occasionally bites his wrist before getting through the sentence. In the Washington Post video, because the board is constantly moving around in the air and is facing away from us, we would not know what the man is typing but for the letters and words the assistant calls out.
Jaswal et al. concede that the assistant in their study sometimes “redirected participants who seemed to lose their train of thought …, requested clarification, interrupted, and said aloud a word before the participant finished spelling it” (emphasis added; 6), but they defended these actions by suggesting that they are a common feature of conversation among speaking people. However, if you look at the videos you will see something that bears little resemblance to typical conversation or—more to the point—typical writing or typing. This is particularly puzzling in a world filled with keyboards and computer tablets that could eliminate the need for an assistant altogether. The experimenters are hoping to show the agency of the participants, but after watching the videos, I am struck by the agency of the assistant.
A video produced by the Washington Post. The first few minutes show an example of RPM in action. (Source: YouTube.)
Holding a letterboard in the air creates a problem for the claim that the participant is doing the typing. Jaswal et al. said their coders could recognize a tap to the board because it bent under the pressure of the participant’s finger. I have watched the videos, and in several cases, I cannot tell whether the finger was meeting the letterboard or the board was hitting the finger. Both actions would produce a similar bend of the board, but the agency would be quite different.
As to the pointing and eye movement data, the authors suggest that longer pauses between words are typical of non-autistic typists, but what is not typical of non-autistic typists is to have someone else speak each word as you type it. If one does not watch the videos, this alternative explanation might not come to mind, but it is quite possible the longer pause is simply produced by the participant stopping momentarily to listen to the assistant before starting the next word. The assistant has informed the participant by her actions that a word has been completed, and the participant pauses briefly before starting another. As a result, the pause between words is an artifact of the RPM procedure. There could be other explanations as well, but this hypothesis seems obvious.
As to the quicker movement to common letter pairs versus less common letter pairs, this seems expected as well. If, after an average of over three years of practice, the participant has learned sequences of letters that will result in the assistant saying a word and moving the lesson along, then it is quite plausible that the participant has built up a store of such responses. Assuming the words taught appear in relation to their natural frequency, then one might assume that the N-to-T movement would happen more quickly than T-to-N, because the N-to-T sequence is more practiced. But even if we grant that the participants have learned the sequence of letters that will be reported as a word by the assistant, does that constitute proof of independent authorship? Who is choosing the words?
Questions about the Methodology
In my opinion, we don’t have enough information to evaluate this study. Here are some of the questions that arise from the report as presented.
- Who served as coders? Jaswal et al. don’t indicate who the coders were or what they knew about the goals of the study. Neither do they reveal what, if any, relationship the coders might have had to the participants, the center where the study was conducted, or the authors. Given that they provided the data for the study, it would be helpful to know more about them.
- Was the audio on while the coders were scoring the data? If the coders heard the assistant calling out the letters, words, and sentences, this would be a potential influence on their judgments.
- Were the participants nonspeaking individuals? The authors describe this as a test of the use of a letterboard by “nonspeaking autistic people” (Jaswal et al. 2020, 7), and yet Participant 9 spoke quite clearly. Based on the video clip and the authors’ own description of the video clip, Participant 9 recited the letters out loud as she tapped them out on the board and then spoke each word when it was completed. This seems like a relatively high level of ability for someone who is described as nonspeaking. As a result, her behavior may not be typical of those with more severe language deficits—such as the young man in the Washington Post video. Participant 9’s standardized test results were similar to those of the other participants in the study, but it is not clear how many participants were capable of speaking letters and words. Neither videos nor information about the verbal abilities of the other six participants were provided.
- Were the participants’ answers correct, and did they make sense? The dependent measures reported were correct and incorrect spellings, for which the authors employed a somewhat lenient criteria (e.g., words with extra leading or trailing letters were marked as correct). We have video clips of the open-ended question responses for three of the participants, and they do make sense. But one of them clearly had some level of verbal ability, and the answers for the 212 other questions asked in the study are not reported.
- Can the participants read written text? The authors claim their participants were expressing their own thoughts, tapped out in words that were spelled correctly, and yet the readings and questions in the lessons were read aloud to the participants by the assistant. The participants are nonspeaking, but they presumably write sentences. Are the participants capable of reading sentences? Could the questions they were asked have been presented in written form? As mentioned above, for most of us, learning to read is a prerequisite for learning to write. If these participants have learned to write, doesn’t it stand to reason that they might also be capable of responding to written text? Admittedly, these questions go beyond the boundaries of the present study, but they follow from it. Getting answers would help to further illuminate the abilities of these individuals.
Finally, to return to the original point of this article, this is a descriptive study. It did not test cause and effect. As described above, the assistant was very active throughout the process of gathering participant responses, creating the usual suspicions about the influence of the assistant. As is always the case, the burden of proof rests squarely on the shoulders of those who, like Jaswal et al., make the claim that the data from this study provide “compelling reason to believe that participants in our study were spelling their own thoughts” (7). In my opinion, the evidence they offer is not compelling. As a result, they have not met the burden of proof incumbent on them.
Let me dwell on this last point a bit longer. As described above, the claim of RPM is very dramatic. If true, it would deviate from the process most of us used to learn to write and challenge common understandings of autism. Knowing this, most scientists would strive to support their claims by creating the strongest test possible. For example, over thirty-five years ago, Allen and Beatrix Gardner (1984) made a similarly bold claim about language ability in a different species. They adopted a chimpanzee named Washoe and taught her American Science Language (ASL), which they claimed she used appropriately and creatively. Knowing that their results would fly in the face of prevailing notions of animal intelligence, they devised a stringent double-blind test that could not have been more different from what Jaswal et al. have done.
The Gardners built an apparatus that allowed an observer to watch the signs that Washoe produced without being able to see the stimulus that Washoe was looking at. In addition, for some of the tests, they recruited observers who were recent graduates of Gallaudet College (now University), were fluent in ASL, and did not know Washoe. As stringent as their methods were, the Gardners’ research and the many animal language studies that followed remain controversial (Lyn 2012), but the Gardners knew their findings would not be taken seriously unless they devised a strong test. Why have the proponents of RPM behaved so differently? RPM has been used for almost two decades in the United States, and yet its proponents have assiduously avoided submitting RPM to the kinds of tests that could validate it.
What the Authors Might Do Instead
Understanding that after working hard to get a paper published in a good journal it can be quite annoying to be told what you should have done, I will press on with the following suggestions.
- Fully embrace the principles of Open Science, whose primary advocate is Brian Nosek, who is Vikram Jaswal’s Psychology Department colleague at the University of Virginia. This would mean publicly pre-registering the study, which involves describing the planned study in detail before the process of data collection begins, making it possible for other researchers to comment on the plan at the start. Furthermore, if the plan is subsequently changed on the fly, those changes must be reported. Finally, the data should be made publicly available so that they can be examined and reanalyzed by anyone who wishes to do so.
To their credit, Jaswal et al. did post some of the data on the Open Science Framework website, as well as the code used to perform the statistical analyses, but they failed to make available the most important data of all: the videos of the sessions. As a result, it is impossible to observe the assistant’s behavior and make an independent assessment of the participants’ responses.
- Conduct a simple test of authorship rather than mere agency in finger tapping. There are many ways this could be done, but here is one possibility. Blindfold the assistant, and place noise-cancelling headphones on both the participant and the assistant. Pump white noise into the assistant’s headphones and ask the questions through the participants’ headphones. The assistant can hold the letterboard, as usual, but would not be able hear the questions or see the letterboard. Videotape the session. Such a test would give much stronger evidence of authorship than an elaborate analysis of eye movements and the speed of tapping. Rather than some kind of vague “agency,” the researchers could address the central question of who is responsible for the typing by making assistant influence more difficult—if not impossible—and thereby controlling for the most important alternative hypothesis.
- Collaborate with your critics. The question of authorship has become remarkably contentious, and as I will outline more fully below, there is much at stake. In my opinion, one of the great advances of the Open Science movement has been to crowdsource science and make it a more transparent and collaborative enterprise. Perhaps the truth can be found by working together rather than at cross purposes.
I hate writing articles such as this one. During the early years of my career, I worked with children and adults with profound autism, people who were nonspeaking and often self-injurious and aggressive. I also got to know many of their parents, and although I have not experienced it myself, I gained a sense of what it is like to have so many of your expectations and hopes for parenthood unfulfilled. No one wants to bring more disappointment to these families. I genuinely hope RPM works, and despite the odds against it, perhaps empirical tests will reveal that it does. What a wonderful outcome that would be. But at the moment, we just don’t have enough evidence.
It is unpleasant to always be in the role of casting stones, and I am tempted to just let it all go. If parents want to believe their children can communicate with these techniques, why should I and the other critics of FC and RPM speak up? In a free society, parents are allowed wide latitude in how they raise their children, so why not remain quiet and respect their choices to use FC and RPM? It would be easy to do so, but in my opinion, there is too much at stake—for the children and for public policy. The history of FC is marked by false claims of sexual abuse that have devastated innocent families (Lillienfeld 2007) and at least one famous case of actual sexual misconduct (Moriarty 2018). Even setting aside these salacious examples, there is much at stake for everyone involved with these techniques.
Most importantly, there are the nonspeaking children and adults at the center of these debates. If RPM is a valid communication technique, then all is well, and it should be widely promulgated. But if it doesn’t work, then hundreds (thousands?) of children and adults have been submitted to many hours and years of wasted time. The opportunity costs for these individuals are enormous. They could have been educated with more validated methods or simply had more free time to do whatever makes them happy. They have been controlled rather than liberated, and their potential has been limited rather than advanced. For this reason alone, it seems very important to know the truth.
There is also a question of public policy. This new study by Jaswal and colleagues is likely to be cited in a future lawsuit. Parents in Virginia and Pennsylvania have attempted to get their local school boards to include RPM in their children’s educational plans and have it used by teachers at public expense (Autistic Self Advocacy Network 2016.; Vyse 2020). In many cases, this might also involve placing the children in much higher grade-level classes. Most of these efforts have failed due to the lack of evidence in support of RPM, but studies like Jaswal et al. (2020) may someday tip the balance, further institutionalizing an unfounded and potentially harmful communication method.
Finally, the stakes are understandably very high for the parents of children who use these techniques. All of us who are parents struggle with the challenge of raising our children as best we can, and from time to time, we make mistakes and get distracted from the important goals. Raising kids is a difficult job for anyone, and parents of children who have profound language deficits face enormous challenges. If they’ve found FC or RPM and are now carrying on conversations with their children and transcribing their poetry and college papers, it is easy to understand the reason they might be reluctant to probe their beliefs about their children. RPM came on the scene after the FC debacle, and that history may also be part of the reason that, for almost two decades, supporters of RPM have avoided the kinds of tests that would prove authorship. Failure would shatter cherished ideas about some of the people they care most about in the world.
And so, we are at a standoff. Until someone has the courage to step forward and test RPM using simple methods that have been around for over a hundred years (Pfungst 1911), this technique will remain controversial, in doubt, and discouraged by large segments of the scientific and professional communities. My hope is that, in the spirit of collaborative science, someday someone will take that step and move us toward ending this logjam.
- This technique goes by several names. Jaswal et al. (2020) did not call it by any of the common names; they simply described their participants as “letterboard users” (1). Because it is the most commonly used name, I have adopted RPM as a generic label to refer to all these techniques.
- American Speech-Language-Hearing Association. 2018. Rapid prompting method [Position Statement]. Available online at www.asha.org/policy/; doi:10.1044/policy.PS2018-00351.
- Autistic Self Advocacy Network. 2016. ASAN files ADA complaint on communication access in schools. Autistic Self Advocacy Network (May 31). Available online at https://autisticadvocacy.org/2016/03/asan-files-ada-complaint-on-communication-access-in-schools/.
- Behavior Analysis Association of Michigan. N.d. Resolutions and statements by scientific, professional, medical, governmental, and support organizations against the use of facilitated communication and rapid prompting. Available online at http://www.baam.emich.edu/baam-fc-resolutions-compilation.html.
- Gardner, R.A., and B.T. Gardner. 1984. A vocabulary test for chimpanzees (Pan troglodytes). Journal of Comparative Psychology 98(4): 381–404. Available online at https://doi.org/10.1037/0735-7036.98.4.381.
- Hemsley, Bronwyn, Lucy Bryant, Ralf W. Schlosser, et al. 2018. Systematic review of facilitated communication 2014–2018 finds no new evidence that messages delivered using facilitated communication are authored by the person with disability. Autism & Developmental Language Impairments 3: 2396941518821570.
- Jaswal, Vikram K., Allison Wayne, and Hudson Golino. 2020. Eye-tracking reveals agency in assisted autistic communication. Nature Scientific Reports 10(7882): 1–10. Available online at https://doi.org/https://doi.org/10.1038/s41598-020-64553-9.
- Lilienfeld, Scott O. 2007. Psychological treatments that cause harm. Psychological Science 2(1): 53–70.
- Lyn, Heidi. 2012. Apes and the evolution of language: Taking stock of 40 years of research. In J. Vonk and T.K. Shackelford (eds.), Oxford Library of Psychology: The Oxford Handbook of Comparative Evolutionary Psychology Oxford University Press, 356–377.
- Moriarty, Thomas. 2018. Ex-Rutgers prof admits it was a crime to have sex with disabled man. NJ.com (March 19). Available online at https://www.nj.com/essex/2018/03/ex-rutgers-newark_prof_admits_criminal_sexual_cont.html.
- Pfungst, O. 1911. Clever Hans (The Horse of Mr. von Osten): A Contribution to Experimental Animal and Human Psychology (Trans. C.L. Rahn). New York: Henry Holt. (Originally published in German, 1907).
- Schlosser, R., Balandin, S., Hemsley, B., et al. 2014. Facilitated communication and authorship: A systematic review. Augmentative and Alternative Communication 30: 359–368.
- Vyse, S. 2020. A small victory for science in suburban Philadelphia. Skeptical Inquirer 44(2): 5.