Training, Not Treatment: Why the RCT Model Is a Poor Fit for Biofeedback and Neurofeedback
Clinicians who deliver biofeedback and neurofeedback keep hearing that only randomized controlled trials count as evidence. That standard was built for agricultural plots and pills, not for a skill someone learns with a coach. Here is why the mismatch matters, and what to do about it.
Two kinds of evidence
The randomized controlled trial (RCT) is at best a marginally useful tool for studying the effectiveness of biofeedback and neurofeedback as clinicians actually practice them. It was built to test agricultural variables and then drugs, that is, static things that an experimenter controls (Bothwell & Podolsky, 2016; Meldrum, 2000). Biofeedback and neurofeedback are not drugs, however. They are training: a person learns to change their own physiology, guided by a clinician who adjusts the training parameters as the person changes.
Clinicians and researchers collect different kinds of data. Clinical data follow one person over many sessions: their baseline, learning curve, symptoms, and response when the protocol changes. Research data, in the RCT sense, take a few measurements from many people, average them into groups, and ask whether the group averages differ.
Both have value. But when RCT results are treated as the only legitimate evidence, and clinical data are dismissed as anecdotal, the field judges a training method with a tool designed for something else. This post looks at why, what the research on placebos and trial design actually shows, and which methods fit training interventions better.
A design built for pills
The RCT entered clinical medicine with the 1948 streptomycin trial for tuberculosis and became the gold standard for testing therapeutic agents (Concato et al., 2000). A drug is close to the ideal subject for this design. The dose is fixed, the pill looks the same in every bottle, and a sugar pill can look identical. The design rests on a handful of assumptions.
Everyone in the trial has the same condition, so you can pool their outcomes. Every participant in the active arm receives the same thing, in the same amount, on the same schedule. The control arm receives a believable placebo, something indistinguishable from the real thing, so neither participant nor rater knows who got what. The drug effect is whatever remains after you subtract the placebo response from the active response. Individual differences in response are random error, which averages out with a larger sample.
When those assumptions hold, the RCT is a powerful tool. Strictly, they are design choices rather than logical requirements. Pragmatic trials let clinicians adapt treatment under a prespecified rule (Zwarenstein et al., 2008), and the format can stretch to individualized, session-adaptive neurofeedback (Wang et al., 2026). The question for our field is whether the classic sham-controlled trial, as it has actually been run, fits a learned skill. As the next sections show, its assumptions don't fully hold even for drugs.
The blind is weaker than it looks, even for drugs
When the placebo is made more convincing, the gap between drug and placebo shrinks. That is the central finding of the Cochrane review by Moncrieff et al. (2004), first published in 2004 and reprinted in 2012. It is worth walking through, because it shows that the problem is not unique to our field.
The problem they addressed is simple. Antidepressants have side effects such as dry mouth, and an inert sugar pill does not. In a so-called double-blind RCT, blinding keeps both the participants and those who administer the drug unaware of whether a pill is a drug or placebo. Even so, participants and raters can often tell who is on the real drug, and that knowledge can shape both how people feel and how they are rated. An active placebo contains a substance, atropine in these trials, that mimics the side effects without being expected to treat depression.
The review pooled nine trials with 751 participants that compared tricyclic antidepressants against active placebos. Across all nine trials, the drug beat the active placebo by 0.39 standard deviations, a modest effect size. Most of that came from one outlying trial; without it, the difference fell to 0.17 standard deviations, with a confidence interval running from 0.00 to 0.34. In practical terms, the average person on the drug did better than 57% to 65% of people on the active placebo, with 50% meaning no difference.
In one trial, raters judged the patients they guessed were on the drug as more improved. In another, side effects were more prominent on the drug, a sign the blind was still leaking. The authors concluded that "unblinding effects may inflate the efficacy of antidepressants in trials using inert placebos" (Moncrieff et al., 2004). The trials were old and small, and the authors call their conclusions tentative.
But the lesson is clear. What an RCT reports as the specific effect of a treatment depends heavily on what the control group receives and on whether the blind actually holds. When the control looks and feels more like the real thing, the measured difference shrinks.
A second line of evidence makes the same point from the brain side. Tétreault et al. (2016) studied people with knee osteoarthritis pain across three clinical trials. Resting-state brain connectivity measured before treatment predicted who would respond to placebo. They also found that the active drug, duloxetine, appeared to enhance the placebo response in some patients and diminish it in others.
In other words, placebo response appears to be a measurable property of the individual nervous system, and drug and placebo effects may not simply add together. The sample was small, and the authors called for validation, so this is a hint rather than a verdict. It is the kind of hint, though, that should make the RCT's subtraction logic less comfortable.
Neurofeedback has the same problem in a sharper form. A sham neurofeedback condition is never inert. The participant still sits with a clinician, attends to a screen, is rewarded for something, and spends weeks in a structured routine. In one of the largest double-blind neurofeedback trials for ADHD, researchers randomized 144 children and analyzed 142 (Neurofeedback Collaborative Group, 2021). Both groups improved significantly, and the groups did not differ.
Both groups received guidance on sleep and nutrition, and the control condition presented prerecorded EEG from other children with each participant's own artifacts superimposed (Neurofeedback Collaborative Group, 2021). Schummer and Sguigna (2024) point out that the control group also received real-time muscle-tension feedback.
Children in both arms were therefore rewarded for sitting still and attending to an engaging task. The trial authors themselves wrote that what the shared improvement is due to "requires further research" (Neurofeedback Collaborative Group, 2021, as quoted by Schummer & Sguigna, 2024). That is an active placebo, and the Cochrane review tells us what to expect when one is used: the measured difference shrinks.
It helps to be precise about what such a comparison measures. It estimates the added effect of the targeted feedback over everything both groups share. A null result says that deliberate theta/beta ratio training added little over that shared package in this trial. It does not say the package did nothing, and it does not test neurofeedback as clinicians practice it.
The group-level evidence has since hardened, and it cuts both ways. A 2025 meta-analysis of 38 RCTs with 2,472 participants found no meaningful benefit of neurofeedback on probably blinded ratings of ADHD symptoms (Westwood et al., 2025). Across the 20 trials with such ratings, the standardized mean difference was 0.04, with a confidence interval from −0.10 to 0.18. Trials that used standard protocols showed a small, significant effect of 0.21.
A reanalysis of the 2021 trial found that most children in both arms learned to regulate the theta/beta ratio, suggesting the sham was largely inert (Kuznetsova et al., 2026). Individual learning, however, did not reliably predict symptom improvement.
A 2026 double-blind trial of individualized upper-alpha neurofeedback likewise produced clear EEG learning in eight sessions without beating sham on parent-rated symptoms (Wang et al., 2026). Those findings weigh against a specific added benefit of these protocols as delivered, and they sharpen the task. Documenting learning is necessary, but not sufficient; the field must also show that learning is what helps, which is the last of the questions posed below.
Trullinger et al. (2019) described the reasoning errors that can follow from sham comparisons. Researchers treat a sham as inert when it isn't, treat no difference from sham as proof of no effect, and invoke placebo after the fact to explain away positive results. Trullinger and colleagues filed these under Type III errors, a statistician's name for a correct answer to the wrong question.
Where RCT assumptions break down for training
Each RCT assumption fails in a specific, predictable way when the intervention is neurofeedback or peripheral biofeedback. Table 1 sets the mismatches side by side. The sections that follow take up each in turn.
The trial assumes that everyone with the diagnosis has the same condition, but a diagnosis is a label for a cluster of reported symptoms, not a confirmed mechanism. Different physiological mechanisms can produce the same symptoms. It assumes that participants respond the same way, but each person brings unique physiology, genetics, and developmental history, and learns at a different pace. It assumes an identical intervention for everyone, but good training changes as the client changes, with sites, frequencies, thresholds, and session length adjusted from session to session.
It assumes an interchangeable provider, but the coaching relationship shapes how learning happens. It assumes that an experimenter-delivered substance is the active ingredient, but learning is the active ingredient, and feedback that doesn't produce learning isn't the intervention, real or sham. And it assumes that the control group receives nothing active, but sham sessions still deliver attention, structure, reward, and coaching.
Table 1. Six assumptions behind the RCT and what happens in biofeedback and neurofeedback.
The diagnosis problem
An RCT starts by selecting participants with a specific, restricted diagnosis and pooling everyone who shares it. That works only if the diagnosis is correct and the people who share it are alike enough for an average to mean something. The data on symptom profiles give reason to doubt the second part.
Among 3,703 depressed outpatients in the STAR*D study, Fried and Nesse (2015) found 1,030 distinct symptom profiles. Nearly half of those profiles (501) belonged to a single person, and the most common profile described only 1.8% of patients. Heterogeneity on that scale does not make the category useless, but it means a group average can hide who benefits and who does not.
Under DSM-5 rules, Galatzer-Levy and Bryant (2013) calculated 636,120 possible symptom combinations that qualify as posttraumatic stress disorder (PTSD). A quantitative EEG (qEEG) on two people with the same diagnosis often looks very different. The EEG example below shows a child diagnosed with attention deficit hyperactivity disorder (ADHD) whose recording shows a high-amplitude, 3 per second pattern suggestive of absence seizures. Absence seizures can pass for inattention and can coexist with ADHD, which is exactly why clinicians assess before they train.
Figure 1. A 19-channel EEG recording in the longitudinal bipolar montage with a 350 µV scale showing 3 per second activity.
The diagnosis problem also raises questions about the aims of training, not only the aims of treatment. Is the objective of training to remove a diagnosis, a negative goal? Or is it to produce something positive, such as better sustained attention, longer stretches of on-task classroom behavior, and EEG patterns closer to normative values where that target is tied to function?
Is a superordinate aim to improve resilience and flexibility in psychological and behavioral functioning within the individual context where a person lives (Hayes et al., 2012)? If the latter is the aim, then the most effective training may be highly individualized, because clinicians see wide varieties of interactions between person and context. That, in turn, as described below, points toward single-case experimental designs, which have great value for scientifically validating neurofeedback.
The adaptive nervous system
The central nervous system, like most bodily systems, keeps adapting to the input it receives, within limits. A protocol that suits a client in week one may be wrong in week six because the client is no longer the same. A drug trial can hold the dose constant because the drug doesn't change. A training protocol that is held constant regardless of the client's response isn't good practice. It is a protocol frozen to suit the research design.
The relationship is part of the method
RCTs typically minimize the clinician's influence so that only the specific ingredient remains, and they rarely measure it. In behavioral change, that sets aside a major part of the process, and trials could preserve and measure provider effects if they chose to. The APA task force on evidence-based therapy relationships reviewed 16 meta-analyses (Norcross & Lambert, 2018). It concluded that the relationship is associated with outcomes substantially and consistently, regardless of treatment type. Ute Strehl, writing on learning theory in neurofeedback, describes the therapist as the person who shapes learning, sustains motivation, and helps clients through frustration and boredom (Strehl, 2014).
Automatic thresholds and operant conditioning
Neurofeedback is operant conditioning of brain activity (Sherlin et al., 2011). Operant conditioning requires that reward follow the target behavior. Shaping means raising the criterion as the learner improves.
Many studies instead use automatic thresholds that reset themselves to guarantee reward on a fixed share of trials, often 60% or 70% (Strehl, 2014). When the threshold resets, a person can be rewarded for not changing their EEG, or effectively punished for improving. Strehl (2014) notes that this can reward worse performance than the previous day's. Pigott et al. (2021) reviewed six sham-controlled ADHD trials that used this approach. None showed that participants learned to change their EEG, and most participants (71% to 75%) guessed they had received sham feedback.
They concluded that these trials compared two forms of false feedback. Where there is no evidence of learning, they argue, operant conditioning of the EEG did not occur. A trial that never shows learning can still tell us whether offering that protocol, as delivered, helped. What it has not tested is neurofeedback as a learning intervention. It has tested sitting in front of a screen.
Training is not treatment
The deepest mismatch is categorical. The RCT asks whether something done to a person produces an effect. Biofeedback and neurofeedback are done by the person through self-interaction via the feedback display. A clinician serves as a guide, interpreter, and resource. The clinician sets up conditions for learning, and the client does the learning.
We don't evaluate other kinds of training this way. Nobody runs a placebo-controlled, double-blind trial of piano lessons in which half the students practice on a keyboard that plays random notes and the teacher is told not to correct them. We instead ask whether the student learned to play, how quickly, and how well the skill holds up. We ask whether it carries over to a real piano in a real room, possibly in front of a real audience.
Those are the right questions for neurofeedback too. Did learning occur, meaning that the trained signal changed in the intended direction, within and across sessions? Did it transfer, so that the client can produce the change without feedback, outside the clinic, under the real-world conditions that may have elicited unwanted behaviors in the past? Did function change, meaning that the symptoms, behaviors, or performance that brought the client in improved by a meaningful amount? Did it last at follow-up, and is the change linked to the learning, so that the clients who learned most also improved most?
Strehl (2014) concludes plainly that neurofeedback is a method of behavior therapy, and argues that it should be designed and evaluated with learning principles in mind. An evaluation model for a training intervention has to put learning at the center. Many RCTs of neurofeedback never report whether it happened, and the ones that do are now finding that learning alone does not guarantee a clinical result (Kuznetsova et al., 2026).
Better ways to measure what we do
Rigorous alternatives exist, and some have appeared in the neurotherapy literature since the field's first journal. They treat each client as an experiment, and they treat variability as something to explain rather than average away. Table 2 summarizes six approaches, and the sections that follow describe each one.
Single-case experimental designs take repeated measures on one person across baseline and training phases, with replication, which mirrors how neurofeedback is already delivered: many sessions, many measures, one learner. Structured clinical case studies turn everyday practice into publishable, comparable evidence. N-of-1 randomized trials, in which one person alternates between conditions in random order, rank alongside systematic reviews of RCTs at the top of the Oxford evidence levels for treatment decisions.
Well-designed observational studies track outcomes in real clinical populations under individualized care, and they match RCT results on average with less scatter between studies. Measures of reliable and clinically significant change judge outcomes person by person rather than by group means. Learning-first trial designs show that the trained signal changed from its pre-training baseline before testing whether symptoms changed.
Table 2. Six research approaches that fit training interventions, what each does, and why it fits.
Single-case designs
The case for single-case research in our field was made in the Journal of Neurotherapy in 1996. Blampied et al. (1996) argued that the group-comparison tradition stifles innovation and precludes scientific investigation of the exceptional or novel case. They point out that averaging different learning curves (fast then slow, slow then fast, step-like, flat) can produce a smooth group curve that represents no actual person. They also note that no two people with the same diagnosis are alike in their neurobiology or life history.
The core unit is a baseline phase followed by a training phase, displayed graphically and judged on level, trend, and variability (Cooper et al., 2020). Causal inference comes from replication. Several designs fit neurofeedback especially well.
A changing-criterion design shows that performance tracks the reward criterion as it is raised step by step. This is shaping, and it is how neurofeedback with manually set thresholds already works. A reversal (withdrawal) design withdraws a component of training and then reinstates it.
An alternating-treatments design compares two protocols within the same person. A multiple-baseline design staggers the start of training across different signals, for example, sensorimotor rhythm (SMR) and skin conductance, or across clients who start training at staggered times. These designs now have formal standards. The What Works Clearinghouse requires at least three demonstrations of effect at three time points, with at least three to five data points per phase (Kratochwill et al., 2010). The 26-item SCRIBE guideline sets out how to report them (Tate et al., 2016).
Clinical case studies done well
In a companion paper in the same issue, Barabasz et al. (1996) lamented practitioners' lack of research output as a social and professional tragedy. They listed what a useful neurotherapy case report needs: a literature review, a diagnosis, neurometric data, the client's history, enough procedural detail to replicate, and follow-up data, including unsuccessful cases. They recommend reporting raw data rather than percentages, so that other clinicians can compare their own clients and researchers can pool cases later. They suggested a target of three to five representative cases, including failures.
N-of-1 trials
The 2011 Oxford Centre for Evidence-Based Medicine levels place N-of-1 randomized trials at Level 1 for the question of whether an intervention helps (OCEBM Levels of Evidence Working Group, 2011). That puts them alongside systematic reviews of RCTs. For a decision about an individual client, evidence from that client counts as top-tier evidence.
One caution applies to training. N-of-1 trials work best when an effect can be switched off and on, and a learned skill cannot be switched off. For neurofeedback, the format therefore suits comparisons of reversible components, such as two feedback parameters or two session formats, rather than training versus no training.
Observational and practice-based evidence
The usual objection to clinical data is that, without randomization, they will overstate effects. Concato et al. (2000) tested that claim across 99 reports and 1,871,681 participants on five medical topics. Well-designed cohort studies and case-control studies produced average results "remarkably similar" to RCTs on the same questions (Concato et al., 2000). RCTs varied more, and only RCTs produced results pointing in the opposite direction from the pooled estimate.
The authors also cite a review of 302 meta-analyses of psychological, educational, and behavioral treatments (Concato et al., 2000). In that review, observational designs differed from randomized ones by a trivial 0.05 in effect size. Their explanation is directly relevant here: observational studies include the full range of patients, with treatment tailored to the individual patient, while an RCT's protocol may not represent clinical practice. None of those topics was neurofeedback, so the finding does not make practice data a substitute for randomized comparisons in our field. It does, however, provide evidence that deserves a seat at the table.
Our own field has proposed exactly this kind of infrastructure. Nelson (2003) called for a neurotherapy practitioner research network, an observational network that pools outcome data from clinicians to show how well neurotherapy works as it is actually practiced. That network has yet to be built, and the case for it has only grown.
Person-level outcomes
Jacobson and Truax (1991) introduced the reliable change index and the concept of clinical significance. Instead of asking whether group means differ, they ask whether this person's change is larger than measurement error. They also ask whether it moved them from the clinical range into the normal range. That is the question a client and their family actually care about.
Learning-first research
Within neurofeedback research itself, the CRED-nf checklist is a consensus statement signed by more than 80 neurofeedback researchers (Ros et al., 2020). It sets reporting standards for neurofeedback studies, including how researchers report participants' regulation of the trained signal. Pigott et al. (2021) go further: studies should track within-session learning curves, avoid auto-thresholding, and test whether EEG learning correlates with clinical outcome. Any trial, randomized or not, is more informative when it can show that learning took place.
Making clinical data count
Every clinician already collects most of what a single-case study needs. The difference between a case anecdote and a case experiment is mostly structure. Blampied et al. (1996) argued that, done this way, every client becomes not a case study but a full-fledged experiment.
A practical minimum starts with a real baseline, so there is something stable to compare against. Take several pre-training measures of the physiology (qEEG, and/or peripheral measures such as skin conductance, temperature, or heart rate variability (HRV)) and of the presenting complaints. Next come session-by-session learning data. Record the trained signal, the threshold, and the reward rate every session. Set thresholds deliberately and log each change, so the data form a changing-criterion record rather than an auto-adjusted blur.
Periodic runs without feedback are the transfer check; they show whether the client can produce the change independently. Give standardized outcome measures at intake, at intervals during training, at the end, and at follow-up, scored for reliable and clinically significant change. Use objective tests, such as a continuous performance test, whenever possible.
Put protocol changes on the record, noting when and why a protocol changed. An unplanned pause in training is not a planned reversal, but if you keep measuring through it, the record shows whether gains held without training. Plot the data by phase, so that level, trend, and variability are visible at a glance. Include failures, because clients who didn't respond are as informative as those who did, and leaving them out is what makes clinical data look biased.
Collected this way, with planned phases, repeated measures, and replicated effects, a single client's record can support a defensible causal claim. An ordinary chart note cannot. A series of such records, pooled across clinicians, is the practice-based evidence that RCTs cannot supply. It shows how training works for real, varied clients under individualized care.
The right questions
None of this means randomized trials have nothing to offer, or that every positive clinical report should be taken at face value. Critics such as Thibault and Raz (2017) argue that expectation and motivation may account for much of neurofeedback's benefit, and that is a fair question to put to the field. But it has to be answered with designs that can detect learning.
Consider a trial that standardizes away individualized protocols, auto-adjusts thresholds against the principles of operant conditioning, minimizes the coaching relationship without measuring it, and gives the control group an active placebo. It is set up to find little difference. Its null result says little about how neurofeedback works in clinical practice.
Examining these studies and their findings also raises ethical questions. If a sham condition and neurofeedback training produce similar, significant benefits, can a provider simply deliver a placebo to achieve those effects? Two cautions come first. A trial that fails to show superiority has not shown equivalence. And a research sham is not a clinical placebo: participants in a sham arm consent to the possibility of sham, and a clinic client does not.
The question also assumes that the active feedback was correctly delivered, that changes in the training variables were tracked, and that participants who showed no actual learned behavior were reported separately. Should neurofeedback training be withheld if it is no more effective than a placebo? If the ethical principles of autonomy and beneficence are taken seriously, a clinician cannot knowingly deliver a sham as treatment, because that is deception and it reduces autonomy. What the clinician owes the client is true neurofeedback whose active ingredient, learning, is documented in that client's own record, and the honesty to say when it is not helping.
Even in drug research, a more convincing placebo shrinks the measured drug effect. Placebo response is a measurable trait of individual brains, and well-designed observational studies match RCT results. The case for treating the RCT as the only real evidence is weaker than its reputation.
For a training intervention, the right questions are whether the person learned, whether the learning transferred, whether their life changed, and whether the change lasted. Single-case designs, structured case reports, N-of-1 trials, practice-based outcome data, and person-level measures of change can answer those questions rigorously. Randomized trials answer a different one: whether offering a protocol beats a comparator on average, and the field needs both kinds of answers, with the client's life as the final test. Clinics already collect the data needed to answer the training questions every day.
Key Takeaways
1. The RCT was built to test fixed-dose drugs on pooled, interchangeable participants. Biofeedback and neurofeedback are learned skills, and the design's assumptions fail them in predictable ways.
2. The blind leaks even in drug trials. A more convincing placebo shrinks the measured drug effect, and sham neurofeedback is an active placebo. A null result from a sham-controlled trial measures the added effect of targeted feedback over a shared package; it says little about neurofeedback as clinicians practice it.
3. A trial that never shows that participants learned to change their EEG has not tested neurofeedback as a learning intervention. Automatic thresholds that hold the reward rate fixed can reward no change or punish improvement, and recent trials show that learning alone does not guarantee a clinical result.
4. Single-case experimental designs, structured case reports, N-of-1 trials, well-designed observational studies, and reliable-change indices are rigorous methods that fit training and already carry formal standards.
5. Clinicians can turn routine records into evidence by taking stable baselines, logging every threshold change, running transfer checks, scoring standardized outcomes for reliable change, graphing by phase, and including failures.
Glossary
absence seizures: brief, generalized seizures marked by lapses of awareness and a 3 per second spike-and-wave EEG pattern.
active placebo: a control substance or condition that mimics the side effects or incidental features of the active treatment without its presumed specific action.
alternating-treatments design: a single-case design that compares two or more conditions by rapidly alternating them within the same person.
automatic threshold: a reward criterion that resets itself to hold the reward rate fixed, regardless of whether the trained signal is improving.
biofeedback: training in which a person learns to change a measured physiological signal using real-time information about that signal.
blinding: keeping participants, clinicians, or raters unaware of which condition a participant received.
case-control study: an observational study that compares people with an outcome against people without it to identify prior exposures or treatments.
changing-criterion design: a single-case design in which the reward criterion is raised in steps and the behavior is expected to track each step.
clinical significance: change that moves a person from the clinical range into the range of the functional population.
cohort study: an observational study that follows groups defined by exposure or treatment forward in time and compares their outcomes.
confidence interval: the range of values within which the true effect is estimated to lie at a stated level of confidence, usually 95%.
continuous performance test: a computerized attention task that scores omissions, commissions, and reaction time.
CRED-nf checklist: a consensus checklist for the reporting and experimental design of clinical and cognitive-behavioral neurofeedback studies.
effect size: a standardized measure of the magnitude of a difference or relationship; 0.2 is conventionally small, 0.5 medium, and 0.8 large.
heart rate variability (HRV): the beat-to-beat variation in heart rate, used as an index of autonomic modulation.
longitudinal bipolar montage: an EEG display in which each channel shows the voltage difference between two adjacent electrodes arranged front to back.
meta-analysis: a statistical synthesis that pools the results of multiple studies on the same question.
multiple-baseline design: a single-case design that staggers the start of training across signals, behaviors, settings, or people.
N-of-1 randomized trial: a trial in which one person alternates between conditions in random order and outcomes are compared within that person.
neurofeedback: biofeedback of brain activity, usually EEG, in which a person learns to change a trained signal through operant conditioning.
observational study: a study that measures outcomes in people who receive treatment in the course of care, without assigning treatment by chance.
operant conditioning: learning in which a behavior becomes more or less frequent depending on the consequences that follow it.
placebo: an inert treatment or condition designed to be indistinguishable from the active treatment.
practice-based evidence: outcome data collected from routine clinical practice and pooled across clinicians.
practitioner research network: a group of clinicians who collect outcome data in a shared format so that results can be pooled.
pragmatic trial: a randomized trial run under conditions close to routine care, in which clinicians may adapt treatment under prespecified rules and outcomes are judged as delivered.
quantitative EEG (qEEG): statistical analysis of the digitized EEG, typically compared against a normative database.
randomized controlled trial (RCT): an experiment that assigns participants to treatment or control by chance and compares group outcomes.
reliable change index: a statistic that tests whether an individual's change on a measure exceeds the change expected from measurement error alone.
resting-state brain connectivity: the correlated activity of brain regions measured while a person rests in a brain scanner.
reversal (withdrawal) design: a single-case design in which a treatment component is introduced, withdrawn, and reintroduced to show that behavior tracks it.
SCRIBE: the Single-Case Reporting Guideline In BEhavioural Interventions, a 26-item checklist for reporting single-case studies.
sensorimotor rhythm (SMR): a 12 to 15 Hz EEG rhythm recorded over the sensorimotor strip and often trained in neurofeedback.
sham neurofeedback: a control condition in which the feedback is not linked to the participant's own trained signal.
shaping: reinforcing successive approximations of a target behavior by raising the criterion as performance improves.
single-case experimental design: a design that measures one person repeatedly across baseline and intervention phases and infers effects from replicated changes.
skin conductance: the electrical conductance of the skin, which rises with sweat gland activity driven by sympathetic arousal.
specific effect: the part of a treatment's effect attributable to its presumed active ingredient rather than to context, expectation, or attention.
standardized mean difference: an effect size that expresses the difference between two group means in standard deviation units.
theta/beta ratio: the ratio of EEG theta (4 to 8 Hz) power to beta (13 to 21 Hz) power, often elevated in ADHD and the target of theta/beta neurofeedback.
transfer: the ability to produce a trained change without feedback, in settings outside the clinic.
Type III error: a correct statistical answer to the wrong question, such as treating no difference from an active sham as no effect.
References
Barabasz, M., Barabasz, A., & Blampied, N. M. (1996). A primer on clinical study research in neurotherapy. Journal of Neurotherapy, 1(4), 12–14. https://doi.org/10.1300/J184v01n04_02
Blampied, N. M., Barabasz, A., & Barabasz, M. (1996). Single-case research designs for the science and practice of neurotherapy. Journal of Neurotherapy, 1(4), 15–26. https://doi.org/10.1300/J184v01n04_03
Bothwell, L. E., & Podolsky, S. H. (2016). The emergence of the randomized, controlled trial. New England Journal of Medicine, 375(6), 501–504. https://doi.org/10.1056/NEJMp1604635
Concato, J., Shah, N., & Horwitz, R. I. (2000). Randomized, controlled trials, observational studies, and the hierarchy of research designs. New England Journal of Medicine, 342(25), 1887–1892. https://doi.org/10.1056/NEJM200006223422507
Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.). Pearson.
Fried, E. I., & Nesse, R. M. (2015). Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders, 172, 96–102. https://doi.org/10.1016/j.jad.2014.10.010
Galatzer-Levy, I. R., & Bryant, R. A. (2013). 636,120 ways to have posttraumatic stress disorder. Perspectives on Psychological Science, 8(6), 651–662. https://doi.org/10.1177/1745691613504115
Hayes, S. C., Strosahl, K. D., & Wilson, K. G. (2012). Acceptance and commitment therapy: The process and practice of mindful change (2nd ed.). Guilford Press.
Jacobson, N. S., & Truax, P. (1991). Clinical significance: A statistical approach to defining meaningful change in psychotherapy research. Journal of Consulting and Clinical Psychology, 59(1), 12–19. https://doi.org/10.1037/0022-006X.59.1.12
Kratochwill, T. R., Hitchcock, J., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M., & Shadish, W. R. (2010). Single-case design technical documentation (Version 1.0, pilot). What Works Clearinghouse.
Kuznetsova, E., de Beus, R., Arnold, L. E., Rudakov, E., & Cowley, B. U. (2026). Learning in neurofeedback is heterogenous and does not guarantee ADHD symptom improvement. Applied Psychophysiology and Biofeedback. Advance online publication. https://doi.org/10.1007/s10484-026-09774-w
Meldrum, M. L. (2000). A brief history of the randomized controlled trial: From oranges and lemons to the gold standard. Hematology/Oncology Clinics of North America, 14(4), 745–760. https://doi.org/10.1016/S0889-8588(05)70309-9
Moncrieff, J., Wessely, S., & Hardy, R. (2004). Active placebos versus antidepressants for depression. Cochrane Database of Systematic Reviews, 2004(1), Article CD003012. https://doi.org/10.1002/14651858.CD003012.pub2
Nelson, L. A. (2003). Neurotherapy and the challenge of empirical support: A call for a neurotherapy practice research network. Journal of Neurotherapy, 7(2), 53–67. https://doi.org/10.1300/J184v07n02_04
Neurofeedback Collaborative Group. (2021). Double-blind placebo-controlled randomized clinical trial of neurofeedback for attention-deficit/hyperactivity disorder with 13-month follow-up. Journal of the American Academy of Child & Adolescent Psychiatry, 60(7), 841–855. https://doi.org/10.1016/j.jaac.2020.07.906
Norcross, J. C., & Lambert, M. J. (2018). Psychotherapy relationships that work III. Psychotherapy, 55(4), 303–315. https://doi.org/10.1037/pst0000193
OCEBM Levels of Evidence Working Group. (2011). The Oxford 2011 levels of evidence. Oxford Centre for Evidence-Based Medicine. https://www.cebm.ox.ac.uk/resources/levels-of-evidence/ocebm-levels-of-evidence
Pigott, H. E., Cannon, R., & Trullinger, M. (2021). The fallacy of sham-controlled neurofeedback trials: A reply to Thibault and colleagues (2018). Journal of Attention Disorders, 25(3), 448–457. https://doi.org/10.1177/1087054718790802
Ros, T., Enriquez-Geppert, S., Zotev, V., Young, K. D., Wood, G., Whitfield-Gabrieli, S., Wan, F., Vuilleumier, P., Vialatte, F., Van De Ville, D., Todder, D., Surmeli, T., Sulzer, J. S., Strehl, U., Sterman, M. B., Steiner, N. J., Sorger, B., Soekadar, S. R., Sitaram, R., . . . Thibault, R. T. (2020). Consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies (CRED-nf checklist). Brain, 143(6), 1674–1685. https://doi.org/10.1093/brain/awaa009
Schummer, G. J., & Sguigna, T. (2024). A critical review of: Double-blind placebo-controlled randomized clinical trial of neurofeedback for attention-deficit/hyperactivity disorder with 13-month follow-up. NeuroRegulation, 11(1), 92–110. https://doi.org/10.15540/nr.11.1.92
Sherlin, L. H., Arns, M., Lubar, J., Heinrich, H., Kerson, C., Strehl, U., & Sterman, M. B. (2011). Neurofeedback and basic learning theory: Implications for research and practice. Journal of Neurotherapy, 15(4), 292–304. https://doi.org/10.1080/10874208.2011.623089
Strehl, U. (2014). What learning theories can teach us in designing neurofeedback treatments. Frontiers in Human Neuroscience, 8, Article 894. https://doi.org/10.3389/fnhum.2014.00894
Tate, R. L., Perdices, M., Rosenkoetter, U., Shadish, W., Vohra, S., Barlow, D. H., Horner, R., Kazdin, A., Kratochwill, T., McDonald, S., Sampson, M., Shamseer, L., Togher, L., Albin, R., Backman, C., Douglas, J., Evans, J. J., Gast, D., Manolov, R., Mitchell, G., Nickels, L., Nikles, J., Ownsworth, T., Rose, M., Schmid, C. H., & Wilson, B. (2016). The Single-Case Reporting Guideline In BEhavioural Interventions (SCRIBE) 2016 statement. American Journal of Occupational Therapy, 70(4), 7004320010p1–7004320010p11. https://doi.org/10.5014/ajot.2016.704002
Tétreault, P., Mansour, A., Vachon-Presseau, E., Schnitzer, T. J., Apkarian, A. V., & Baliki, M. N. (2016). Brain connectivity predicts placebo response across chronic pain clinical trials. PLOS Biology, 14(10), Article e1002570. https://doi.org/10.1371/journal.pbio.1002570
Thibault, R. T., & Raz, A. (2017). The psychology of neurofeedback: Clinical intervention even if applied placebo. American Psychologist, 72(7), 679–688. https://doi.org/10.1037/amp0000118
Trullinger, M., Novian, A., Russell-Chapin, L., & Pradhan, D. (2019). Perspectives on Type III statistical errors: Exaggerating the effects of placebo in neurofeedback. NeuroRegulation, 6(1), 38–41. https://doi.org/10.15540/nr.6.1.38
Wang, Y., Hui, K. Y., Lu, S. F., Lui, Q. C. H., Ma, L., Asharaf, S. T., Lak, D., Coghill, D., Ip, P., Lee, T. M. C., & Tso, W. W. Y. (2026). A double-blind randomized controlled trial of personalized upper-alpha neurofeedback in children with ADHD. Journal of Child Psychology and Psychiatry. Advance online publication. https://doi.org/10.1111/jcpp.70188
Westwood, S. J., Aggensteiner, P.-M., Kaiser, A., Nagy, P., Donno, F., Merkl, D., Balia, C., Goujon, A., Bousquet, E., Capodiferro, A. M., Derks, L., Purper-Ouakil, D., Carucci, S., Holtmann, M., Brandeis, D., Cortese, S., Sonuga-Barke, E. J. S., & European ADHD Guidelines Group. (2025). Neurofeedback for attention-deficit/hyperactivity disorder: A systematic review and meta-analysis. JAMA Psychiatry, 82(2), 118–129. https://doi.org/10.1001/jamapsychiatry.2024.3702
Zwarenstein, M., Treweek, S., Gagnier, J. J., Altman, D. G., Tunis, S., Haynes, B., Oxman, A. D., & Moher, D. (2008). Improving the reporting of pragmatic trials: An extension of the CONSORT statement. BMJ, 337, Article a2390. https://doi.org/10.1136/bmj.a2390
About the Author
John S. Anderson, MA, LADC, BCB, BCN, QEEGD, is a veteran neurofeedback practitioner and educator with over five decades of experience in biofeedback and neurofeedback, beginning his work in 1974. He holds a master's degree in psychology and is certified by the Biofeedback Certification International Alliance (BCIA) and the International QEEG Certification Board. As the founder of the Minnesota Neuro-Training Institute, Anderson provides clinical services, mentorship, and professional training in neurotherapy. His clientele includes individuals with ADHD, learning disorders, chronic pain, and addiction. He is also a recognized instructor, offering BCIA-approved courses and QEEG certification programs, and contributes to educational initiatives such as Biosource Software's "Seminars Without Borders." Anderson integrates holistic healing practices with contemporary neurophysiological research to develop effective neurofeedback protocols.
Support Our Friends
Comments