Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2026 Jul 28;126(2):e70128. doi: 10.1002/jeab.70128

Conditional discrimination of electroencephalogram alpha amplitudes in humans

Toshikazu Kuroda 1,✉, Motoaki Kawanabe 1
PMCID: PMC13416005  PMID: 42522153

Abstract

Conditional discrimination of high and low alpha amplitudes of electroencephalogram by humans was demonstrated in the 1950s. Despite several attempts, the original finding was only replicated once, but discrimination accuracy in the replication was only slightly above chance level and confounds may have contributed to it. Potential confounds included a systematic difference in time to exceed thresholds for high and low amplitudes (i.e., temporal cue), unequal numbers of high‐ and low‐amplitude trials, and short trial durations that allowed for correct choices in a preceding trial to serve as a cue for the next trial due to gradual shifts in alpha amplitude. In the current study, we modified the procedure for eliminating these confounds: Participants observed a bar motion corresponding to changes in alpha amplitude for a constant duration. The direction of the bar motion (up and down) was positively or negatively correlated with the amplitude depending on the trial, and the participant's task was to discriminate the two types of trials. Three of five evaluable participants (60%) showed alpha discrimination without the above confounds, and this success rate was similar to that reported in the original study. Thus, alpha discrimination appears to be a replicable phenomenon.

Keywords: alpha band, conditional discrimination, electroencephalogram, human, private event

In the 1950s, a young researcher, Joseph Kamiya, initiated a new research program that eventually turned into neurofeedback—a variant of operant conditioning that arranges contingencies for neural activities instead of overt behavior (see Kamiya, 2011, for his recollections). According to his recollection, Kamiya was interested in the introspective report of subjective experience, and his research goal was “to bridge the gap between the earlier goals and methods of psychology (e.g., those of Wundt, Titchener, James, and others) with those of more historically recent goals and methods, à la Watson, Tolman, Skinner, and others” (Kamiya, 2011, p. 68). Even those behaviorists had different views on introspection and the world within the skin it was intended to reveal: Watson (1913) rejected introspection as a scientific method, Tolman (1932) considered the role of cognitive processes to account for behavior, and Skinner (1945) proposed an analysis of subjective terms as verbal responses to private stimuli.

Being exposed to both research on electroencephalogram (EEG) measurements and operant conditioning while working at the University of Chicago, Kamiya developed an experimental procedure for studying “the discrimination of private correlates of physiological activity” (Kamiya, 2011, p. 67). It basically was a conditional‐discrimination procedure but with alpha‐band (8–12 Hz) amplitudes of EEG serving as sample stimuli. Kamiya continuously monitored the alpha amplitude measured at an electrode placed on the scalp at the back of the head while participants kept their eyes closed. He then asked them to discriminate whether the amplitude was “high” or “low” (determined with respective thresholds) when a bell rang. Although the sample stimuli were invisible to the participants, many of them discriminated the two states of alpha amplitude, with one participant reaching near 100% accuracy in 4 days of training. Those participants often reported that they were alert at low amplitudes and relaxed at high amplitudes (Kamiya, 1968, 1969; see also Nowlis & Kamiya, 1970, for more details of the verbal report; cf. Orne & Wilson, 1978). Kamiya also noted that participants with high discrimination accuracy could also produce either a high or low state of alpha amplitude upon an experimenter's command, implying that discrimination of alpha amplitude contributes to voluntary control over the amplitude. If neurofeedback is indeed a variant of operant conditioning, voluntary control over a target neural activity (e.g., alpha amplitude) and discrimination of the activity may be prerequisites for the generation of neurofeedback effects, paralleling the role of response discrimination in operant conditioning (cf. Alsop & Davison, 1991, 1992; Jones & Davison, 1998; Killeen, 1978; Kuroda & Lattal, 2014; Lattal, 1975, 1979; Madden & Perone, 1999; Nussear & Lattal, 1983; Pliskoff & Goldiamond, 1966; Tanno et al., 2012).

Despite his pioneering research, Kamiya did not publish much of his alpha‐discrimination data, instead providing anecdotes in his subsequent articles (Kamiya, 1968, 1969). Moreover, alpha discrimination has largely remained uninvestigated since then except for a few reports in which participants generally failed to acquire alpha discrimination (Cott et al., 1981; see also Orne & Wilson, 1978, for a summary of unpublished studies). Frederick (2012) appears to be the first successful replication of Kamiya's original finding on alpha discrimination. In this and his subsequent research (Frederick et al., 2015, 2016, 2019), however, discrimination accuracy was only slightly above chance level, ranging from a mean of 53% to 58% given binary choices (high/low amplitude), but statistically significant. These findings suggest alpha discrimination is a more difficult task for many participants than Kamiya (1968, 1969) originally suggested.

Issues include not only the task difficulty but also the possibility of confounds in the previous alpha‐discrimination studies. First, unequal numbers of high‐ and low‐amplitude trials could have led to discrimination accuracy greater than 50%. Unlike conventional conditional‐discrimination procedures that use exteroceptive stimuli as samples (e.g., colors), experimenters have no control over a participant's alpha amplitude. In an extreme hypothetical case, participants could achieve 100% accuracy by always selecting a choice associated with, for example, high amplitude if the amplitude only exceeds a high‐amplitude threshold but not at all for a low‐amplitude threshold, without discriminating the two states. Thus, the actual chance level varies depending on the ratio of high‐ and low‐amplitude trials. In this respect, Kamiya (1968, 1969) did not describe the number of trials for the two alpha‐amplitude states. In a replication study by Cott et al. (1981), there was evidence of alpha‐discrimination acquisition when the ratio of high‐ and low‐amplitude trials was left uncontrolled, but it disappeared when the ratio was kept near .50. According to their descriptions of procedures, Frederick (2012) and Frederick et al. (2015, 2016, but see 2019) appeared to have equated the number of trials for the two alpha‐amplitude states.

As a second possible confound, temporal cues could have contributed to discrimination accuracy. For example, some participants may take a longer time for the alpha to exceed a high‐amplitude threshold than a low‐amplitude threshold, and this could serve as a temporal cue for the discrimination. Neither Kamiya (1968, 1969) nor Frederick (2012) provided relevant data, but Frederick used a procedure that reasonably controlled the temporal cue. Specifically, he collected alpha amplitudes in the previous 150‐s period, which was continuously updated with every incoming amplitude. From a distribution of the collected alpha amplitudes, he used high (e.g., 70th) and low (e.g., 30th) percentiles of the distribution as thresholds on high‐ and low‐amplitude trials, respectively. In this type of procedure, the time taken for the alpha amplitude to reach a threshold is theoretically equal between the two percentiles, similar to percentile schedule described by Platt (1973).

Third, trial durations could also have affected discrimination accuracy because the amplitude of EEG alpha generally changes slowly in the order of seconds (Frederick, 2012). This allows for a positive correlation in the amplitude across two consecutive trials, thereby introducing an advantage of repeatedly selecting the same choice alternative, especially when trial durations are short. Given that the same trial type (e.g., high‐amplitude trial) could have occurred up to five consecutive times in their procedure, Frederick et al. (2015) examined effects of this variable1 on the participants' choice. Participants with relatively high discrimination accuracy selected the same choice alternative as the preceding trial for approximately 84% and 40% of trials when the durations were short (e.g., 8 s) and long (e.g., 45 s), respectively. In contrast, participants with low accuracy selected the same alternative on approximately 54% and 27% of trials given those respective trial durations. These results suggest trial duration affected not only the choice but also discrimination accuracy (see also Frederick et al., 2016).

Last, muscle movements near the eyes are a common artifact in EEG research and could have affected discrimination accuracy. Opening/closing eyes contaminates EEG signals at electrodes placed both near and farther from the eyes. There are several established methods for removing eye artifacts (e.g., Kobler et al., 2020; Pion‐Tonachini et al., 2019). A more troublesome problem is that closing the eyes itself can increase alpha amplitude—another common finding in EEG research (e.g., van der Meer et al., 2016). Participants in previous alpha‐discrimination studies generally were instructed to keep their eyes closed throughout sessions (Cott et al., 1981; Frederick, 2012; Frederick et al., 2015, 2016, 2019; Kamiya, 1968), but there have been a few exceptions (Nowlis & Kamiya, 1970; see also Kamiya, 1968, addressing a related study). Even when instructed to keep their eyes closed, it was not guaranteed that the participants followed the instruction. Thus, interpreting results of these studies needs to consider whether the participants' eyes were open or closed.

The goal of the present study was to examine whether human participants could discriminate their own alpha amplitudes with a more refined experimental procedure that dealt with those possible confounds. Specifically, we developed a new alpha‐discrimination procedure in which participants observed a bar that reflected their own alpha amplitude. Unlike the previous studies that used high‐ and low‐amplitude trials, we arranged positive and negative trials: The bar moved upward as the amplitude increased (positively correlated) on positive trials or as it decreased (negatively correlated) on negative trials. The participant's task was to discriminate the two types of trials by selecting one of two choice alternatives after observing the bar motion for a constant duration. The bar motion could serve as a reliable cue for discrimination only if alpha amplitude was associated with a participant's subjective experience (e.g., relaxation or attention). Correct choices were made independently of those in the preceding trial by withholding trial onsets until the bar (invisible during intertrial intervals [ITIs]) was near the central point. Eye artifacts resulting from blinks and eyeball motions were computationally removed from EEG signals in real time using a method described by Kobler et al. (2020), and suspicious eye status (e.g., keeping eyes closed) was continuously monitored through a webcam. Thus, the present procedure eliminated the previous confounds by having the following features: an equal number of trials for the two trial types, no unauthorized temporal cues for discrimination, independence of correct choice across two consecutive trials, and minimum eye artifacts.

METHOD

Participants

Six participants were recruited who met the following criteria: (1) ages ranging from 20 to 59 years, (2) no color blindness, and (3) no claustrophobia or nyctophobia for conducting the experiment in a small dark room. We informed participants that they could earn extra bonus money depending on their performance in addition to 8,000 yen for a single 4‐hr laboratory visit. The participants were 23 to 49 years old (mean = 28.5, SD = 9.66, median = 24) and three (50%) were male. All instructions were given in Japanese, and all the participants provided written informed consent prior to their participation. The protocol for this study was approved by the Review Board Ethics Committee at the author's research institute that followed the Declaration of Helsinki.

Apparatus

Experimental tasks were conducted in a sound‐proof shield room with the interior size of 145 cm wide, 185 cm long, and 210 cm high. The room had a table, a comfortable chair (with a cushion if requested), two ceiling lights, and an air conditioner. On the table were an 8BitDo Micro Bluetooth game controller for measuring button presses and a Dell computer screen with the resolution of 1920 by 1080 pixels (60 × 34 cm), with its center being at eye level and approximately 80 cm away when participants sat on the chair. An infrared webcam was attached to the screen for monitoring a participant's face (i.e., eyes) in real time in an adjacent room. The ceiling lights and the air conditioner each were covered with a metal‐mesh box for minimizing noise. A preliminary assessment indicated that EEG signals recorded inside the shield room were quite clean (i.e., little noise) except for the presence of 60‐Hz line noise from a power outlet, which was outside the range of EEG alpha band (i.e., 8–12 Hz).

EEG was recorded with an EEG cap with 32 pin‐type active electrodes with its associated amplifier (ActiveTwo System, BioSemi, Netherlands). Compared with passive electrodes, active electrodes amplified EEG signals before transmitting them to the main amplifier via a cable. As a result, the signals were more resistant to noise that leaked in the cable during signal transmission and were more suitable for real‐time EEG processing. An appropriate size of the cap (either small‐medium or medium‐large size) was selected for each participant. The EEG electrodes were placed on the scalp according to a modified International 10–20 system. Conductive gel (Signa gel; Parker Laboratories, USA) was applied to each electrode for lowering electric impedance. Raw EEG signals were collected at sampling rate of 2048 Hz and then down‐sampled to 512 Hz by the App‐BioSemi software2 while it sent the signals to a main Python program described below.

A custom program written in Python (version 3.10.13), running on a desktop computer (with Windows 10, Intel Core i9‐7900X, NVIDIA GeForce GTX 1080 Ti, and 128‐GB RAM installed), processed EEG signals in real time, executed all but one experimental task, and recorded all experimental events. We used OpenCV (version 4.8.1.78) functions for displaying the experimental task on the computer screen and SciPy (version 1.11.3) functions for computing alpha amplitude. An exception was the experimental task for measuring eye artifacts (see below). The program for this task originally was written by Kobler et al. (2020),3 slightly modified for the present purpose, and run in MATLAB (version R2019b) with Psychtoolbox (version 3.0.19). In the latter task, a laptop computer (with Windows 10, Intel Core i7‐7600U, and 16‐GB RAM installed) running the MATLAB program sent experimental‐event markers to the above Python program via a LAN cable for recording the markers together with EEG signals in the Python program.

EEG Processing and alpha‐amplitude presentation

After receiving EEG signals from the App‐BioSemi software, the main Python program converted them to a visual stimulus (a bar) representing alpha amplitude. Here, we provide a summary of the steps for this conversion, whereas the Appendix shows the technical details for these steps. First, the signal at each electrode was referenced to a mean of all 32 electrode signals (i.e., the mean was subtracted from each of those signals) to keep the mean zero. This was to make the signals more resistant to factors affecting all electrodes (e.g., cable movement). Second, eye artifacts were removed using the method described in Kobler et al. (2020). Third, Welch's (1967) method was applied to the EEG signals for computing amplitudes at each frequency in steps of one from 8 to 12 Hz (corresponding to the alpha band) and the mean across these frequencies was computed for each electrode. The mean at the Pz electrode located on the back of the head (see Figure A1 in the Appendix), generally considered resistant to muscle artifact, served as the target alpha amplitude, which is consistent with previous studies (e.g., Frederick et al., 2012). Fourth, a moving average was applied to the target alpha amplitude for smoothing its change. Last, this moving‐average amplitude went into a queue for computing its percentile in the last 60‐s period, which was updated for every incoming amplitude. This percentile was displayed on the computer screen in a form of a bar (see Figure 1). Thus, the bar position was an index of transient changes rather than sustained levels of alpha amplitude.

FIGURE 1.

FIGURE 1

Open in a new tab

A diagram of the present procedure for examining alpha‐amplitude discrimination. In the sample component, the bar position represents the current percentile of the amplitude relative to the last 60 s. On positive trials, the bar moved above a horizontal line when the percentile was above the median and moved below the line when the percentile was below the median. On negative trials, the bar was an upside‐down image of that on the positive trial. Two circles appeared in the choice component, with the yellow for the positive trial and the blue for the negative trial. Participants chose either color by pressing a corresponding button on the game controller (all buttons on the controller actually were white but with colors just for an illustration purpose). Correct and incorrect choices were followed by a pink and gray surrounding box, respectively, in the outcome component. Tones were presented for indicating stimulus changes. A 4‐s ITI was followed by a presample component that lasted until alpha amplitude was in the range between 40th and 60th percentiles.

Procedure

After checking with the BioSemi ActiView902‐Win software that offset voltages of all electrodes were in an acceptable range between 20 and − 20 mV, each participant went through experimental tasks in the following order: (1) resting state, (2) eye‐artifact measurement, (3) alpha‐discrimination practice, and (4) alpha discrimination. The first two tasks were preparatory for the latter two. All the tasks took place with room lights off except for the computer screen.

Resting state

Resting state was for measuring EEG when participants were not performing any specific task. They were instructed not to ponder anything specific without falling asleep for 4 min. EEG was recorded with eyes closed first, followed by another 4‐min recording with eyes open (blinks were allowed). In the latter recording, the participants were instructed to keep staring blankly at a fixation cross (a “+” symbol) presented in the center of the screen.

Eye‐artifact measurement

This task was for subsequent real‐time removal of eye artifacts from EEG signals. Details of the task were described elsewhere (Kobler et al., 2020). Briefly, the task consisted of a single session of 27 trials. On each trial, participants were instructed to move their eyes according to one of four different motions of a small blue ring (vertical, horizontal, blink, or no blink) for 10 s. The ring oscillated up and down in vertical‐motion trials and moved left and right in horizontal‐motion trials, and participants were instructed to keep tracking the ring motion by moving their eyeballs while holding their face rigid. In blink trials, the ring shrank briefly at 1‐s intervals (like heartbeats) and participants were instructed to blink their eyes accordingly. In no‐blink trials, the ring remained at the center of the monitor and participants were instructed to keep their eyes open without blinks. After the task, EEG signals were processed with software provided by Kobler et al. (2020) to generate a 32 × 32 matrix (given 32 EEG electrodes) containing parameters for removing eye artifacts. This matrix was used in subsequent experimental tasks. Figure A2 in the Appendix shows a sample segment of EEG signals before and after the removal of eye artifacts, demonstrating the validity of this method.

Alpha‐discrimination practice

This practice was for participants to get accustomed to the alpha‐discrimination task. Prior to the practice, the diagram in Figure 1 was presented to participants with the following written instructions:

You will see a bar moving up and down in the upcoming task. This bar represents an aspect of your brain state. There are two types of trials: positive and negative. On positive trials, the bar moves upward when you are in a high state and downward when in a low state. On negative trials, the bar moves downward when you are in a high state and upward when in a low state. The horizontal line at the center of computer screen represents an average of the state in the last 60‐s bar observation.

After a 15‐s observation of the bar, you will hear a tone and see a yellow circle on top and a blue circle on bottom. Choose the yellow by pressing the top button on the game controller you have on your right hand if you think it was a positive trial or the blue by pressing the bottom button for a negative trial. You will see a pink box and hear a high‐pitch tone if the choice is correct; otherwise, you will see a gray box and a low‐pitch tone. Please keep your eyes open (blinks are allowed) during sessions: You will not be able see the bar motion and, thus, cannot tell which of the trials is in effect if you close your eyes.

To give you a hint, on positive trials, the bar typically moves upward when you are relaxed and downward when you pay attention to or ponder something. On negative trials, the bar typically moves downward when relaxed and upward when paying attention to or pondering something. If these tips do not apply to you, try to find your own tips. For other advice, there is a delay of a few seconds in bar motion relative to the actual change in the brain state. Thus, a good strategy is to decide which of the states (relaxation or attention) you want to be in and then wait for a while until the bar motion changes accordingly. Note that the next trial does not begin until the bar (invisible but still moving in the background) is near the horizontal line. If the next trial does not start after waiting for a while, try to relax or pay attention to something. And finally, as a general rule of thumb, keep your body still except when pressing a button; otherwise, noise will contaminate your EEG signals, making the task more difficult than it is designed.

Participants had 10 practice trials for alpha discrimination. As shown in Figure 1, each trial consisted of presample (described in detail below), sample, choice, and outcome components, followed by an ITI. In the sample component, which always lasted for 15 s, a bar moved up and down, representing the percentile of current alpha amplitude relative to the last 60 s as described above. This 60‐s period only included time in the presample and sample components in the current and previous trials. This period was not available on the first practice trial such that the last 60‐s segment of resting‐state EEG with eyes open was used as an alternative, which subsequently was updated with every incoming EEG signal. For this reason, participants were informed that the bar motion on the first few practice trials was inaccurate and, thus, just practice button pressing.

A horizontal line at the center of the screen represented the median (50th percentile) of alpha amplitudes in the last 60 s. The bar position at a given moment depended on the trial. On positive trial types, the bar was below the horizontal line for the 1st to 49th percentiles and above the line for the 51st to 100th percentiles. Thus, the bar position was positively correlated with the percentile. On negative trial types, the bar was an upside‐down image of that in the positive trial. Thus, the bar position was negatively correlated with the percentile. Unlike previous studies (e.g., Kamiya, 1968) that arranged high‐ and low‐amplitude trials, we arranged positive and negative trials. This arrangement removed the necessity of setting thresholds for determining whether alpha amplitude was high or low, which, in turn, allowed for keeping the sample‐component duration constant, thereby removing temporal cues for discrimination. The sequence of positive and negative trials was quasirandom, with the following restrictions: (1) there was an equal number of each trial type in a session and (2) the same trial type of trial occurred no more than four times consecutively.

The participant's task was to discriminate the two trial types based on the observation of bar motions. The bar motion on its own would not serve as a reliable cue, however. This was because the bar, representing the current percentile of alpha amplitude, would theoretically be above the horizontal line (median) for approximately 50% of time and below the line for the remaining time. The maintenance of the 50:50 ratio was further promoted by updating the last 60‐s of data with every incoming EEG signal, which accommodated gradual shifts in the baseline alpha amplitude that often occur over the course of measurement (see Orne & Wilson, 1978). For empirical evidence, Table A1 in the Appendix shows a summary of time spent below and above the horizontal line when simulating the bar position with resting‐state data, indicating the maintenance of an approximately 50:50 ratio for each participant. Thus, the bar motion would likely serve as a reliable cue for the discrimination of positive and negative trials only if alpha amplitude, presented as the bar position, was associated with a participant's subjective experience (e.g., relaxation and attention).

After a 15‐s presentation of bar motions in the sample component, the bar disappeared and two circles appeared with a 0.4‐s 700‐Hz tone in the choice component. A yellow circle, always presented on top, was associated with the positive trial, whereas a blue circle, always presented at bottom, was associated with the negative trial.4 Participants chose either color by pressing a button on the game controller. Making a correct choice led to a 0.4‐s 2000‐Hz tone with a pink box surrounding the selected button in the outcome component, whereas making an incorrect choice led to 0.4‐s 400‐Hz tone with a gray surrounding box. The circles and the surrounding box disappeared after 2 s, followed by a 4‐s ITI during which only the horizontal line was presented at the center of the screen.

After the ITI, a presample component started during which the horizontal line remained on the screen. The presample component lasted until alpha amplitude, computed in the background without the bar presentation, fell in the range between 40th and 60th percentiles. This restriction was included to eliminate the possibility that bar positions presented immediately after the onset of sample component would serve a cue for discrimination in case those locations were somehow correlated with the type of trial. This could occur, for example, when alpha amplitude had remained high since a preceding trial (see Frederick et al., 2015), resulting in the bar being close to the higher edge of the computer monitor, indicative of a positive trial, or being close to the lower edge, indicative of a negative trial. At the end of the 10th practice trial, a text message appeared on the computer screen indicating how many correct choices the participant had made, which did not affect the payment.

Participants were informed about the relation between the bar position and private correlates of EEG (e.g., relaxation and attention), unlike Kamiya's (1968, 1969) original study but consistent with Frederick (2012) and Frederick et al. (2015). This was to maximize the possibility of acquiring alpha discrimination in a limited number of sessions while attempting to make the strategy similar among participants.

Alpha discrimination

This was the main task of the present experiment and was the same as alpha‐discrimination practice except for the following points. First, each session consisted of 30 trials. Second, the last 60‐s segment of EEG in the practice session (or in the last alpha‐discrimination session for subsequent sessions) was used for computing the percentile of alpha amplitude in the first trial. Third, participants could earn bonus money if their percentage correct was greater than 50% (i.e., more than 15 correct choices given 30 trials) in a session, after which each correct choice resulted in an additional 100‐yen bonus. Thus, they could earn a maximum of 1,500 yen in a session. This instruction was given to participants before the discrimination task. This payment scheme was part of an establishing operation to encourage achieving a percentage correct higher than at chance level of 50%. Fourth, at the end of each session, the amount of bonus earned and percentage correct were displayed on the screen. Participants took a short break between sessions, usually 2–3 min but extended when requested. The alpha‐discrimination session was repeated until the total of bonuses earned was near or exceeded a maximum of 5,000 yen or until 3.5 hr had passed since the onset of the laboratory visit. Participants generally took five or six sessions to complete a study.

After the last discrimination session, participants were instructed to fill out a postexperiment questionnaire about general feelings during sessions (e.g., back pain) as well as an open‐ended question about strategies they took, if any, for making correct choices. Then payment was made at the end of laboratory visit after washing hair.

RESULTS

Data were analyzed individually for each participant. Figure 2 shows accuracy as the percentage of correct responses across sessions of the alpha‐discrimination task (excluding practice sessions). For Participant 01 M, accuracy increased to 100% by the fourth session and then slightly decreased in the last session. Accuracy by Participant 02F maintained around 80% throughout the sessions. Accuracy varied across sessions while remaining above chance (50%) for Participants 03 M and 06 M. In contrast, accuracy was near chance for Participants 04F and 05F. Thus, four of the six participants successfully discriminated their own alpha amplitudes. When examining percentage of correct responses separately for each trial type, Participant 02F showed a slight bias toward the choice associated with negative trials (see Figure A4 in the Appendix).

FIGURE 2.

FIGURE 2

Open in a new tab

Percentage of correct responses across sessions of the alpha‐discrimination task for each participant. Dashed lines represent chance level (50%) given equal numbers of the two trial types (i.e., positive and negative trials). “M” and “F” in the participant ID indicate males and females, respectively.

We examined potential cues for discrimination. Figure 3 shows mean position of the bar relative to the horizontal line in the entire 15‐s period of the sample component across discrimination sessions. Positive and negative values on the y‐axis indicate that the bar was above or below the median, respectively. This measure systematically differed between positive and negative trials for all six participants: The bar tended to be in the negative range during positive trials and be in the positive range during negative trials, although the magnitudes of mean difference varied across the participants. The mean difference was large, especially for Participants 01 M and 02F. These two participants also showed higher discrimination accuracy than the others (see Figure 2), suggesting that bar positions during the sample component likely served as a cue for discrimination. This appears to be the case for other participants, but this cue was not effective for Participants 04F and 05F. Moreover, the bar being in the negative range during positive trials suggests that participants generally took the “attention” strategy rather than the “relaxation” strategy for controlling the bar motion.

FIGURE 3.

FIGURE 3

Open in a new tab

Mean position of the bar relative to the horizontal line during the entire 15‐s period of the sample component across alpha‐discrimination sessions for each participant. Positive and negative values on the y‐axis indicate that the bar was above or below the line, respectively. Filled and unfilled circles represent positive and negative trials, respectively. Error bars represent ±1 standard deviation. “M” and “F” in the participant ID indicate males and females, respectively.

We next examined other cues for discrimination, especially undesirable ones that would otherwise invalidate the experimental procedure if present. One cue could be the bar position at the onset of the sample component. If there was a systematic difference in this measure between positive and negative trials, there would have been high discrimination accuracy based on the bar position per se, not the bar position associated with alpha amplitude. Accordingly, Figure 4 shows the mean position of the bar during the first 1‐s period of the sample component across discrimination sessions. This measure generally was near zero for both positive and negative trials. There was a systematic difference in the mean bar position between the two trial types for Participants 04F and 05F, but the mean differences appeared to be marginal, thereby unlikely contributing to discrimination accuracy. These results indicate that withholding the onset of the sample component until alpha amplitude fell in the range between 40th and 60th percentiles was effective in eliminating this potential confound.

FIGURE 4.

FIGURE 4

Open in a new tab

Mean position of the bar relative to the horizontal line during the first 1‐s period of the sample component across alpha‐discrimination sessions for each participant. Positive and negative values on the y‐axis indicate that the bar was above or below the line, respectively. Filled and unfilled circles represent positive and negative trials, respectively. Error bars represent ±1 standard deviation. “M” and “F” in the participant ID indicate males and females, respectively.

Conversely, withholding the sample‐component onset allowed for variations in the time prior to the onset of sample component (i.e., presample component). This could be another potential cue for discrimination if it was somehow correlated with trial type. Accordingly, Figure 5 shows the time between onsets of the 4‐s ITI and sample component across discrimination sessions, separately for positive and negative trials. Those times generally overlapped. One possible exception was for Participant 01 M, but the substantial variability indicated by the standard deviations suggests that this variable unlikely served as a cue for discrimination.

FIGURE 5.

FIGURE 5

Open in a new tab

Time between onsets of 4‐s ITI and sample component across alpha‐discrimination sessions for each participant. Filled and unfilled circles represent positive and negative trials, respectively. Error bars represent ±1 standard deviation. “M” and “F” in the participant ID indicate males and females, respectively.

Figure 6 shows raw alpha amplitude across time. The x‐axis shows the time across all sample components in the last alpha‐discrimination session (for a total of 450 s given 15 s per trial for 30 trials). For resting‐state sessions, alpha amplitudes in an entire 240‐s (4‐min) period are presented. The amplitude generally remained lower in the discrimination session than the resting‐state session with eyes closed, suggesting that participants' eyes were generally open during the discrimination session (see Figure A5 in the Appendix for more details).

FIGURE 6.

FIGURE 6

Open in a new tab

Raw alpha amplitude across time for each participant. For resting state with eyes closed (red) and with eyes open (green), amplitudes for an entire 4‐min (240‐s) session are displayed. For the last discrimination session (blue), amplitudes across all thirty 15‐s (450‐s) sample components are displayed. The scale on the y‐axis differs across participants for accommodating individual differences in this measure. “M” and “F” in the participant ID indicate males and females, respectively.

Figure 7 shows power spectral density across 0.5–30‐Hz frequencies. This is a common measure of EEG expressed in the unit of decibel (a logarithmic scale) and provides a summary of the amplitudes, shown in Figure 6. There is a direct relation between amplitude and power spectral density (Stoica & Moses, 2005) such that having a high peak in the 8–12 Hz range indicates high alpha amplitude. There was a clear peak in that frequency range during resting state with eyes closed for all six participants, although the specific frequency for the highest peak differed across participants and also across tasks for some participants. The peak was below 10 Hz for Participants 01 M, 02F, and 05F and was above 10 Hz for the other three participants.5 The peaks were similar across the three tasks for Participants 01 M and 02F. The peaks in the resting state with eyes open and the last discrimination session were lower than those in the resting state with eyes closed but still were present for Participants 04F, 05F, and 06 M; this pattern was less clear for Participant 03 M. Although correlational, these individual differences in alpha peaks might partially account for the difference in discrimination accuracy.

FIGURE 7.

FIGURE 7

Open in a new tab

Power spectral density in a frequency range between 0.5 and 30 Hz for each participant. Red, green, and blue lines respectively represent resting state with eyes closed, resting state with eyes open, and the last discrimination session. “M” and “F” in the participant ID indicate males and females, respectively. Note the presence of peaks in the alpha band of frequency (8–12 Hz).

The participants' faces were continuously monitored through an infrared webcam during alpha‐discrimination sessions to see whether they were closing their eyes to control the bar motion. All but Participant 06 M kept their eyes open throughout the task except for blinking. This latter participant appeared to have been closing his eyes for several seconds from time to time toward the end of the experiment. This was confirmed with an informal postsession question that revealed he was not attempting to control the bar motion by closing his eyes. Rather, he was trying to relax. For this reason, we decided to withhold a conclusion as to whether he acquired the alpha discrimination.

An open‐ended question in the postexperiment questionnaire about strategies participants took indicated that Participants 01 M, 02F, and 03 M used the attention strategy. This was generally consistent with the above interpretations of bar positions in the sample component (see Figure 3). Participant 04F mentioned that one third of the bar motion matched her perceived state (i.e., relaxation/attention). Participant 05F mentioned that she used the relaxation strategy but could not keep making correct choices. In contrast, Participant 06 M mentioned that he started with the attention strategy but without much success so that he switched to the relaxation strategy, which was consistent with the observation that he sometimes closed his eyes toward the end of the experiment.

Finally, to complement the visual assessment of discrimination accuracy in Figure 2, we conducted a binomial test to examine how likely the observed accuracy would have occurred by chance for each participant. Using the Bonferroni correction with the five evaluable participants (excluding Participant 06 M), alpha level was set at .01. Given the number of correct choices being 28, 26, 23, 17, and 14 out of 30 trials in the last discrimination session for Participants 01 M–05F, the binomial test revealed significant results for the first three participants with p values of <.001, <.001, .003, .585, and .856 for each participant, respectively. In addition, conclusions of these binomial tests remained the same (i.e., p < .01 for the first three participants, whereas p ≥ .01 for the other two) when considering the last n sessions where n varied from two to five (or six for those who had six sessions). Thus, the statistical test supported the visual assessment of the data.

DISCUSSION

The present study investigated discrimination of EEG alpha amplitude in humans. Participants in Kamiya's (1968, 1969) original study showed high discrimination accuracy, but those in subsequent studies by other researchers either did not acquire the discrimination (Cott et al., 1981; Orne & Wilson, 1978) or showed accuracy slightly higher than at chance level (Frederick 2012; Frederick et al., 2015, 2016, 2019). Of the six participants in the present study, discrimination accuracy was considerably high for four (01 M, 02F, 03 M, and 06 M). Two participants (04F and 05F) did not acquire the alpha discrimination in five or six sessions. Continuous observation through an infrared webcam, however, revealed that Participant 06 M sometimes closed his eyes for several seconds during a sample component toward the end of the experiment. This behavior by itself could increase alpha amplitude (e.g., van der Meer et al., 2016) and may have contributed to high accuracy; thus, we decided to withhold a conclusion about whether 06 M acquired the alpha discrimination. Thus, excluding this participant, three of five participants (60%) showed alpha discrimination in a single 4‐hr laboratory visit. This success rate was similar to that in Kamiya (1968, 1969), where 9 of 12 participants (75%) acquired the discrimination within seven 1‐hr sessions.

Several researchers have pointed out the possible presence of confounds that could account for Kamiya's (1968, 1969) original finding. These included unequal numbers of high‐ and low‐amplitude trials (Cott et al., 1981), temporal cues arising from unequal durations of a sample component (Frederick, 2012), and correct choice in a previous trial serving as a cue for that in the ongoing trial due to the slow change in alpha amplitude (Frederick et al., 2015, 2016). We removed these possible confounds in the present procedure and still demonstrated alpha discrimination in three of five participants.

The present procedure differed from previous ones (e.g., Cott et al., 1981; Kamiya, 1968; Frederick 2012) mainly in how alpha discrimination was assessed. In the previous studies, participants typically kept eyes closed throughout sessions. When alpha amplitude reached either a high or low level determined with some thresholds, participants were instructed to report whether the amplitude was high or low. In the present study, in contrast, participants were instructed to keep their eyes open and a bar representing alpha amplitude was displayed. They observed the bar for a constant duration and then reported whether its motion was positively or negatively correlated with alpha amplitude.

This procedural modification brought a couple of advantages. First, it was unnecessary to set thresholds for determining whether alpha amplitude was high or low. Using these thresholds was the source of a potential confound contributing to discrimination accuracy in the previous studies. Specifically, time required for alpha amplitude to reach the threshold could be systematically different between high‐ and low‐amplitude trials, thereby introducing a temporal cue for discrimination (Frederick, 2012). This confound was completely removed by keeping sample‐component durations constant in the present study.

Another advantage of presenting bar motions, from both practical and theoretical perspectives, was the continuous exposure to exteroceptive stimuli associated with alpha amplitude. From a practical perspective, the continuous exposure offers more learning opportunities. In previous studies, participants were exposed to such stimuli (a bell, tone, etc.) only when their alpha amplitude exceeded a threshold. For example, Frederick et al. (2016) indicated that alpha amplitude exceeded thresholds approximately three times per minute in their study. Fewer opportunities for learning a relation between a participant's subjective experience and the discrete stimulus presentation, compared with the continuous exposure, seems to make alpha discrimination difficult, which may partially account for low discrimination accuracy in previous studies (e.g., Frederick 2012). In contrast, continuous exposure to an alpha‐associated exteroceptive stimulus as in the present study offers more learning opportunities. Nonetheless, evidence for higher discrimination accuracy with the continuous than discrete exposure is needed (see Cott et al., 1981; Frederick et al., 2019); thus, future research should examine the difference in discrimination accuracy between continuous and discrete exposures to exteroceptive stimuli associated with alpha amplitude.

From a theoretical perspective, continuous exposure to the alpha‐associated stimulus offers a common ground for examining the relation between discrimination and neurofeedback effectiveness. Kamiya (1968, 1969) speculated that discrimination of alpha amplitude contributes to a participant's voluntary control over it, thereby making the amplitude subject to an operant contingency (i.e., neurofeedback). However, his speculation had not been supported by empirical studies showing that participants generally did not acquire alpha discrimination in conditional‐discrimination procedures but still showed increases in a target alpha amplitude in neurofeedback procedures (Cott et al., 1981; see also Kotchoubey et al., 2002; Lacroix & Gowen, 1981). Regarding this issue, Frederick et al. (2016) pointed out that continuous exposure to alpha‐associated stimuli is common in neurofeedback studies, whereas discrete exposure is common in alpha‐discrimination studies. They argued this procedural difference may account for the higher success rate of learning in the neurofeedback setting compared with the conditional‐discrimination setting. The present conditional‐discrimination procedure, characterized by continuous exposure to alpha‐associated stimuli, offers a fairer common ground for examining the relation between discrimination and neurofeedback effectiveness, which parallels the role of response discrimination in operant conditioning (e.g., Kuroda & Lattal, 2014; Lattal, 1975, 1979; Madden & Perone, 1999). Moreover, this type of procedure may help to clarify the factor—namely, a lack of response discrimination—underlying common problems in the field of brain–computer interface (BCI), so‐called BCI illiteracy, in which approximately 30%–40% of participants do not acquire a new neural response in BCI/neurofeedback settings (e.g., Alkoby et al., 2018; Vidaurre & Blankertz, 2010).

An interesting finding in the present study was that mean bar positions during the entire 15‐s sample component systematically differed depending on whether positive or negative trials were in effect, especially for those participants with high discrimination accuracy. For example, for Participant 01 M accuracy was around 65% in the first session and mean bar positions were around −5 and + 10 on positive and negative trials, respectively (see Figures 2 and 3). Then accuracy increased in subsequent sessions, and so did the difference between mean bar positions. Likewise, discrimination accuracy of Participant 02F was stable across all five sessions, ranging approximately between 80% and 90%, and the difference in mean bar position between positive and negative trials was relatively large, remaining stable across those sessions. One could argue that such a large difference in mean bar position depending on the trial type could serve as a cue for discrimination, thereby being a possible confound. Yet we think this rather was an indication of participants' voluntary control over their own alpha amplitude. Specifically, the bar represented the percentile of alpha amplitude in the last 60‐s period and, thus, the bar would theoretically be above the horizontal line for approximately 50% of time and below the line for the remaining time. This indeed was the case in the simulation of bar position with resting‐state data (see Table A1 in the Appendix). The ratio of time spent below and above the horizontal line could deviate from the theoretical ratio of 50:50 only if participants had voluntary control over the bar position. In this respect, the present finding was relevant to Kamiya's (1968, 1969) original speculation that discrimination of alpha amplitude contributes to a participant's voluntary control over it. We likewise speculate that discrimination of one's own activity (overt/covert behavior) correlated with the event that follows it (e.g., bar motion) is almost synonymous with having voluntary control over the latter.

Another interesting finding in the present study was that some participants successfully discriminated alpha amplitudes while others did not. Similar individual differences have been reported in previous studies on alpha discrimination (e.g., Frederick, 2012; Kamiya, 1968). Although correlational, Participants 01 M and 02F had high discrimination accuracy and showed high alpha peaks during resting state regardless of whether their eyes were closed or open (see Figure 7). This relation suggests that those who naturally show a high alpha peak for whatever reason (genes, age, learning history, etc.) tend to show high discrimination accuracy. For example, higher alpha peaks might have been associated with higher signal‐to‐noise ratios, making the discriminative stimuli in the sample component clearer. Similarly, Nowlis and Kamiya (1970) showed that those with very high alpha peaks during resting state with eyes open tended to show greater voluntary control over the change in alpha amplitude than others without such a peak. Neurophysiologically, alpha amplitude increases when a greater number of neurons, being close to each other spatially, synchronically fire around 10 Hz (Klimesch, 2012). Identification of factors contributing to the synchronic firing of neurons (e.g., meditation as discussed by Kamiya, 1968) might lead to higher alpha amplitude, which, in turn, might increase alpha‐discrimination accuracy.

Limitations of the present procedure and their implications should also be addressed. First, it was a task of discrimination between positive and negative correlations between alpha amplitude and bar positions, so it remains unclear how discrimination accuracy in such a task is related to discrimination of high and low amplitudes as examined in previous studies (e.g., Frederick 2012; Kamiya, 1968, 1969). The bar position was an index of transient changes in alpha amplitude, not its absolute level, in the present procedure such that these perceptual events may be different from each other. However, the discrimination of the transient change may have stronger practical implications, such as relevance to neurofeedback. In standard neurofeedback, it has been a common practice to set a threshold on the absolute level of a target neural activity, serving as a criterion for reinforcer deliveries (e.g., Asai et al., 2022). Such a practice might not be ideal for training if the neural activity fluctuates not only on a short timescale but also on a long timescale, like the air temperature fluctuating not only across days but also across seasons. Specifically, the difficulty of producing reinforcers by target neural activities likely changes depending on the baseline (operant levels) when using an absolute activity level as a threshold. Thus, neurofeedback targeting transient changes in a target neural activity rather than its absolute level might better help participants acquire the activity. Likewise, the present discrimination task might serve as a better training module for standard neurofeedback, although different from Kamiya's original discrimination task.

A second limitation was that the present procedure required participants to keep their eyes open, which left unauthorized behavior (closing eyes) uncontrolled as exemplified by the case of Participant 06 M. Nonetheless, such behavior can be detected. We continuously monitored participants' eyes with an infrared webcam during sessions, and this can be automated with modern computer‐vision technology. When unauthorized behavior is detected, a contingency can be arranged for promoting participants to keep their eyes open (presenting a reminder, a punisher, etc.) or data during the occurrence of such behavior can simply be discarded. Alternatively, some stimulus modalities other than visual (e.g., auditory) can be used for stimuli representing alpha amplitude, although it remains unclear how those stimuli per se affect the amplitude. The issue would be more complicated when unauthorized behavior is less salient to a third person. This indeed was a controversy in the early days of neurofeedback research (Black, 1982), but it might not be as critical as it was once thought from the standpoint of operant conditioning. Specifically, different participants show different forms of behavior for producing reinforcers in an identical contingency only because the contingency allows for high degrees of freedom in how to produce the reinforcer. For example, rats can produce reinforcers by pressing a lever with their left paw, right paw, head, and so on because a given contingency does not specify a topography of behavior. An emphasis on the function of behavior rather than its structure/topography is a major characteristic of operant conditioning, whereas the topography matters only when a contingency specifies it. Thus, we argue that the possibility of alternative overt/covert behavior for producing reinforcers, different from the one intended by a researcher, should be ruled out only when those are considered undesirable in terms of the research question (see Black, 1982, who made a similar argument regarding neurofeedback).

A third limitation was that we provided participants detailed instructions regarding the relation between bar positions and possible subjective experience (relaxation/attention). This was our attempt to obtain as high discrimination accuracy as possible in a small number of sessions. It has been common practice in human operant research to keep instructions minimal (e.g., Madden & Perone, 1999; Perone & Baron, 1980) because instructions can affect behavior in a different way from control by a given contingency (Baron & Galizio, 1983). However, that does not necessarily mean all instructions are undesirable—it depends on the research question—and researchers have to carefully assess pros and cons in each case. In the present experiment, there was no apparent reason not to provide explicit instructions about possibly effective strategies (relaxation/attention); thus, we took the above practical approach. Nonetheless, it would certainly be ideal if participants still show high accuracy with minimal instructions, perhaps with longer exposure to the contingency.

Kamiya (1968, 1969) started his research on alpha discrimination for examining introspective reports of subjective experience. Skinner also appeared to have had a similar research goal as implied in his writings on subjective terms and verbal behavior (Skinner, 1945, 1957, 1974). Indeed, Kamiya (2011) offered an anecdote on Skinner's visit to his laboratory in the early 1960s for observing how to administer alpha‐discrimination training. From the beginning, Kamiya collected verbal reports on subjective experience when alpha amplitude was high and low in conditional‐discrimination studies as well as in neurofeedback studies (see Nowlis & Kamiya, 1970, for details). This parallels Skinner's work on verbal behavior and his discussion of private events that take place when a contingency is arranged for overt behavior (Skinner, 1957, 1974). Specifically, these researchers both viewed changes in private events as the result of a contingency. Thus, it is of particular interest in future research to examine how participants will describe their private events when discriminating their own alpha amplitude in the present procedure when given no instruction about the relation of the two. Although further validations are required in future research, the present procedure may be viewed as a method for the empirical study of private events that Skinner (1974) pursued as exemplified by the following quote:

A science of behavior must consider the place of private stimuli as physical things, and in doing so it provides an alternative account of mental life. The question, then is this: What is inside the skin, and how do we know about it? The answer is, I believe, the heart of radical behaviorism. (p. 232)

AUTHOR CONTRIBUTIONS

Toshikazu Kuroda was responsible for conceptualization, writing the original draft, reviewing, editing, software, data curation, formal analysis, whereas Motoaki Kawanabe was responsible for reviewing, editing, and supervision.

CONFLICT OF INTEREST STATEMENT

We have no conflict of interest.

ETHICS APPROVAL

The study protocol for this project was approved by the Review Board Ethics Committee at the author's research institute that followed the Declaration of Helsinki.

ACKNOWLEDGMENTS

We thank the Department of Computational Brain Imaging at ATR for sharing their EEG amplifier with us and members in the Department of Dynamic Brain Imaging for their assistance: Takeshi Ogawa and Reinmar Kobler for their advice on EEG processing, Miki Kitagawa for preparing materials for the experiment, and Yoko Matsumoto for recruiting participants. We also thank Chris Podlesnik at the University of Florida for his comments on an earlier version of this manuscript.

APPENDIX A.

A.1.

This section provides technical details and supplementary figures not described in the main text.

A.1. EEG processing and alpha‐amplitude presentation

Figure A1 shows a diagram of steps for EEG preprocessing. After the Python program received EEG signals from the App‐BioSemi software at sampling rate of 512 Hz, it further down‐sampled the signals to 64 Hz for allowing the execution of subsequent computational processes without delays. This 64‐Hz EEG was saved for reproducing alpha amplitudes in post hoc analyses. A real‐time band‐pass filter was applied to EEG signals using a second‐order Butterworth filter in a state‐space form with cutoff frequency set between 0.5 and 30 Hz (see algorithm in the butter.m file of the Signal Processing Toolbox in MATLAB) for keeping the signals in that frequency range intact while reducing the signal in other ranges. This filter was mainly for removing 0‐Hz direct current from alternating current, which was considerably different in magnitude across electrodes. Next, for every incoming EEG signal, the signal at each electrode was referenced to a mean of all 32‐electrode signals. That is, the mean was subtracted from each of those signals to keep their mean zero to make the signals more resistant to factors affecting all electrodes (e.g., cable movement). Then eye artifacts were removed in real time by multiplying a 32 × 32 matrix (given 32 EEG electrodes), created for each participant following the method described in Kobler et al. (2020), by the incoming EEG signal (see Figure A2 for the outcome). The resultant EEG will hereafter be referred to as preprocessed EEG.

FIGURE A1.

FIGURE A1

Open in a new tab

A diagram of EEG preprocessing. A diagram of steps for EEG preprocessing (left panel) and a main function or an outcome of each step (texts on the right). The spatial map (top‐left panel) with a nose on top shows locations of all 32 electrodes according to a modified International 10–20 system. The Pz electrode, located on the back of the head and indicated with a red square, served as the target in the present experiment.

FIGURE A2.

FIGURE A2

Open in a new tab

A sample outcome of Kobler et al.'s (2020) method for removing eye artifacts, using participant 02F's EEG data. Blue and red lines represent EEG signals before and after the artifact removal. Yellow arrows indicate the point at which the participant blinked. Note that raw signals at electrodes near eyes (e.g., Fp1 and Fp2) were strongly affected by the blinks but also at some other electrodes as well. These artifacts were removed/attenuated while other signals not affected by the blinks generally remained intact. See the main text for details. EEG electrodes: Fp1 to Cz. Auxiliary channels used in the computation: HEOG to artifact classes.

Figure A3 shows steps for computing alpha amplitude. First, the preprocessed EEG in the last 4‐s period at a given moment was stored in a queue, serving as a 4‐s sliding window that was updated with every incoming EEG signal. Second, Welch's (1967) method was applied to the data stored in the queue for computing the squared magnitude spectrum. Third, the squared magnitude spectrum was multiplied by 2 and then square‐rooted for obtaining amplitudes at each frequency in steps of 1 from 1 to 30 Hz. Fourth, amplitudes between 8 and 12 Hz (corresponding to the alpha band) were extracted and their mean computed for each electrode. Finally, the mean amplitude at the Pz electrode located on the back of the head (see Figure A1), generally resistant to muscle artifacts, was selected as the target alpha amplitude in the present study. The alpha amplitude computed as such fluctuated rapidly, looking quite “busy” and, thus, was not suitable for a sample stimulus in a conditional‐discrimination procedure. To smooth the fluctuation, a 2‐s moving average of alpha amplitudes was computed. As a result, the moving‐averaged amplitude at a given moment consisted of preprocessed EEG data in the last 6‐s period. This moving‐averaged amplitude went into another queue for computing its percentile in the last 60‐s period. This percentile was displayed on a computer screen in a form of bar.

FIGURE A3.

FIGURE A3

Open in a new tab

A diagram of steps for computing alpha amplitude and how it was presented on a computer screen. Welch's (1967) method was applied to preprocessed EEG in the last 4‐s period (top‐left panel) for obtaining squared magnitude spectrum (bottom‐left panel). This was multiplied by 2 and then was squared‐rooted to obtain amplitude (bottom‐right panel). Then a 2‐s moving average was applied to the mean amplitude in the frequency range of alpha band (8–12 Hz) and its percentile in the last 60‐s period was presented in a form of bar with the median being at the center of the screen (top‐right panel).

FIGURE A4.

FIGURE A4

Open in a new tab

Percent correct separately for each type of trial across sessions of the alpha‐discrimination task for each participant. Left panel: percent correct separately for each type of trial across sessions of the alpha‐discrimination task for each participant. Filled and unfilled circles represent positive‐ and negative‐correlation trials, respectively. Right panel: percent choice associated with the positive‐correlation trial across sessions of the alpha‐discrimination task for each participant. “M” and “F” in the participant ID indicate males and females, respectively.

FIGURE A5.

FIGURE A5

Open in a new tab

Raw alpha amplitude across time, separately for positive‐ and negative‐correlation trials, for each participant in the last discrimination session. Yellow and blue solid lines represent positive‐ and negative‐correlation trials, respectively. Vertical dashed lines represent the end of trials, with yellow and blue indicating a participant's choice associated with the positive‐ and negative‐correlation trials, respectively. Match and mismatch between a solid line and a dashed line in a given trial indicate correct and incorrect choices, respectively. The scale on the y‐axis differs across participants for accommodating individual differences in this measure. “M” and “F” in the participant ID indicate males and females, respectively.

TABLE A1.

Results of a simulation with resting‐state data.

Time spent below/above the horizontal line Mean bar position
eyes closed eyes open eyes closed eyes open
ID below above below above mean (SD) mean (SD)
01 M 45.6% 54.4% 49.5% 50.5% 2.3 (34.7) 1.5 (31.6)
02F 49.7% 50.3% 45.6% 54.4% −1.0 (31.9) 2.9 (29.3)
03 M 45.4% 54.6% 50.9% 49.1% 3.7 (29.2) −1.0 (32.3)
04F 46.1% 53.9% 53.3% 46.7% 2.2 (31.2) −2.4 (28.9)
05F 43.0% 57.0% 51.6% 48.3% 5.9 (29.0) −1.5 (30.8)
06 M 49.4% 50.5% 52.1% 47.9% −0.3 (32.8) −0.6 (28.7)
Open in a new tab

Note: (Left) Percentage of time spent below and above the horizontal line and mean bar position for each participant, separately for when using resting‐state data with eyes closed and those with eyes open. The last 60‐s period of the other resting‐state session data (i.e., eyes‐open data for eyes‐closed data and vice versa) was used for computing the initial bar position, which was updated with every incoming EEG signal. The bar position was computed as if a positive trial was in effect. (Right) A value of 0.0 corresponds to the horizontal line (50th percentile), and 1.0 corresponds to the 51st percentile. Note that these were post hoc analyses using the saved resting‐state EEG data; thus, the bar actually was never presented to participants.

Kuroda, T. , & Kawanabe, M. (2026). Conditional discrimination of electroencephalogram alpha amplitudes in humans. Journal of the Experimental Analysis of Behavior, 126(2), e70128. 10.1002/jeab.70128

Editor‐in‐Chief: Suzanne H. Mitchell

Academic Editor: Karen Lionello‐DeNolf

Footnotes

1

These researchers appeared to have misused the term “intertrial interval” when referring to the time between onset of two consecutive trials.

4

Locations of yellow and blue circles were fixed given that some participants in a pilot study found the task too difficult if their locations were randomized across trials. This randomization may have introduced a process that affected alpha amplitude, unnecessarily making the task more difficult. It was considered that fixing circle locations would not adversely affect interpretations of data; thus, the randomization was removed in the present experiment.

5

It could be argued that using the amplitude at a peak alpha frequency instead of a mean amplitude across 8–12 Hz might have led to higher discrimination accuracy. We did not take this approach for a couple of reasons. First, Frederick (2012) showed higher accuracy with a mean amplitude than with a highest peak frequency. Second, peak alpha frequency slightly differed across tasks within a participant, leaving the “true” peak frequency unclear.

6

Conventional band‐pass filters generally take a few seconds for processing signals. In contrast, state‐space band‐pass filters only take a moment, being more suitable for real‐time processing. In both cases, the frequency can go up to a half of the sampling rate (64 Hz), which would be 32 Hz in the present case. Thus, the upper limit of 30 Hz in the present experiment was within an acceptable range.

7

The 6‐s period may sound quite long from the perspective of operant conditioning given that a few‐second unsignaled delay is often enough for reducing reinforcement effects (e.g., Sizemore & Lattal, 1978). Yet this was consistent with Kamiya's point that it takes 2–6 s for a distinct alpha state to appear according to Frederick's (2012) personal communication with him. Frederick himself showed higher accuracy of alpha discrimination with a longer period (4 s) than with shorter ones (1 and 2 s).

DATA AVAILABILITY STATEMENT

The present dataset can be requested at our website (https://bicr.atr.jp/dbi/download/conditional‐discrimination‐of‐electroencephalogram‐alpha‐amplitudes‐in‐humans/), after which a link to the data repository will be supplied.

REFERENCES

  1. Alkoby, O. , Abu‐Rmileh, A. , Shriki, O. , & Todder, D. (2018). Can we predict who will respond to neurofeedback? A review of the inefficacy problem and existing predictors for successful EEG neurofeedback learning. Neuroscience, 378, 155–164. 10.1016/j.neuroscience.2016.12.050 [DOI] [PubMed] [Google Scholar]
  2. Alsop, B. , & Davison, M. (1991). Effects of varying stimulus disparity and the reinforcer ratio in concurrent‐schedule and signal‐detection procedures. Journal of the Experimental Analysis of Behavior, 56(1), 67–80. 10.1901/jeab.1991.56-67 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Alsop, B. , & Davison, M. (1992). Discriminability between alternatives in a switching‐key concurrent schedule. Journal of the Experimental Analysis of Behavior, 57(1), 51–65. 10.1901/jeab.1992.57-51 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Asai, T. , Hamamoto, T. , Kashihara, S. , & Imamizu, H. (2022). Real‐time detection and feedback of canonical electroencephalogram microstates: Validating a neurofeedback system as a function of delay. Frontiers in Systems Neuroscience, 16, Article 786200. 10.3389/fnsys.2022.786200 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Baron, A. , & Galizio, M. (1983). Instructional control of human operant behavior. The Psychological Record, 33(4), 495–520. [Google Scholar]
  6. Black, A. H. (1982). The operant conditioning of central nervous system electrical activity. In Bower G. H. (Ed.), Psychology of learning and motivation (Vol. 6, pp. 47–95). Academic Press. [Google Scholar]
  7. Cott, A. , Pavloski, R. P. , & Black, A. H. (1981). Operant conditioning and discrimination of alpha: Some methodological limitations inherent in response‐discrimination experiments. Journal of Experimental Psychology. General, 110(3), 398–414. 10.1037//0096-3445.110.3.398 [DOI] [PubMed] [Google Scholar]
  8. Frederick, J. A. (2012). Psychophysics of EEG alpha state discrimination. Consciousness and Cognition, 21(3), 1345–1354. 10.1016/j.concog.2012.06.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Frederick, J. A. , Dunn, K. N. , Collura, T. F. (2015). Interactions between discrimination and control of EEG alpha. NeuroRegulation, 2(3), 126–136. 10.15540/nr.2.3.126 [DOI] [Google Scholar]
  10. Frederick, J. A. , Heim, A. S. , Dunn, K. N. , Powers, C. D. , & Klein, C. J. (2016). Generalization of skills between operant control and discrimination of EEG alpha. Consciousness and Cognition, 45, 226–234. 10.1016/j.concog.2016.09.009 [DOI] [PubMed] [Google Scholar]
  11. Frederick, J. A. , Heim, A. S. , & Dunn, K. N. (2019). Self‐prompted discrimination and operant control of EEG alpha. NeuroRegulation, 6(2), 81–92. 10.15540/nr.6.2.81 [DOI] [Google Scholar]
  12. Jones, B. M. , & Davison, M. (1998). Reporting contingencies of reinforcement in concurrent schedules. Journal of the Experimental Analysis of Behavior, 69(2), 161–183. 10.1901/jeab.1998.69-161 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Kamiya, J. (1968). Conscious control of brain waves. Psychology Today, 1(11), 56–61. 10.1037/e400092009-006 [DOI] [Google Scholar]
  14. Kamiya, J. (1969). Operant control of the EEG alpha rhythm and some of its reported effects on consciousness. In Tart C. (Ed.), Altered states of consciousness (pp. 489–501). John Wiley & Sons. [Google Scholar]
  15. Kamiya, J. (2011) The first communications about operant conditioning of the EEG. Journal of Neurotherapy: Investigations in Neuromodulation, Neurofeedback and Applied Neuroscience, 15(1), 65–73. 10.1080/10874208.2011.545764 [DOI] [Google Scholar]
  16. Killeen, P. R. (1978). Superstition: A matter of bias, not detectability. Science, 199(4324), 88–90. 10.1126/science.199.4324.88 [DOI] [PubMed] [Google Scholar]
  17. Klimesch, W. (2012). Alpha‐band oscillations, attention, and controlled access to stored information. Trends in Cognitive Sciences, 16(12), 606–617. 10.1016/j.tics.2012.10.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kobler, R. J. , Sburlea, A. I. , Lopes‐Dias, C. , Schwarz, A. , Hirata, M. , & Müller‐Putz, G. R. (2020). Corneo‐retinal‐dipole and eyelid‐related eye artifacts can be corrected offline and online in electroencephalographic and magnetoencephalographic signals. NeuroImage, 218, Article 117000. 10.1016/j.neuroimage.2020.117000 [DOI] [PubMed] [Google Scholar]
  19. Kotchoubey, B. , Kübler, A. , Strehl, U. , Flor, H. , & Birbaumer, N. (2002). Can humans perceive their brain states? Consciousness and Cognition, 11(1), 98–113. 10.1006/ccog.2001.0535 [DOI] [PubMed] [Google Scholar]
  20. Kuroda, T. , & Lattal, K. A. (2014). Signal functions in delayed discriminative stimulus control by reinforcement sources. Journal of the Experimental Analysis of Behavior, 101(3), 355–372. 10.1002/jeab.85 [DOI] [PubMed] [Google Scholar]
  21. Lacroix, J. M. , & Gowen, A. H. (1981). The acquisition of autonomic control through biofeedback: Some tests of discrimination theory. Psychophysiology, 18(5), 559–572. 10.1111/j.1469-8986.1981.tb01826.x [DOI] [PubMed] [Google Scholar]
  22. Lattal, K. A. (1975). Reinforcement contingencies as discriminative stimuli. Journal of the Experimental Analysis of Behavior, 23(2), 241–246. 10.1901/jeab.1975.23-241 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Lattal, K. A. (1979). Reinforcement contingencies as discriminative stimuli: II. Effects of changes in stimulus probability. Journal of the Experimental Analysis of Behavior, 31(1), 15–22. 10.1901/jeab.1979.31-15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Madden, G. , & Perone, M. (1999). Human sensitivity to concurrent schedules of reinforcement: Effects of observing schedule‐correlated stimuli. Journal of the Experimental Analysis of Behavior, 71(3), 303–318. 10.1901/jeab.1999.71-303 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Nowlis, D. P. , & Kamiya, J. (1970). The control of electroencephalographic alpha rhythms through auditory feedback and the associated mental activity. Psychophysiology, 6(4), 476–484. 10.1111/j.1469-8986.1970.tb01756.x [DOI] [PubMed] [Google Scholar]
  26. Nussear, V. P. , & Lattal, K. A. (1983). Stimulus control of responding by response‐reinforcer temporal contiguity. Learning and Motivation, 14(4), 472–486. 10.1016/0023-9690(83)90028-0 [DOI] [Google Scholar]
  27. Orne, M. T. , & Wilson, S. K. (1978). On the nature of alpha feedback training. In Schwartz G. E. & Shapiro D. (Eds.), Consciousness and self‐regulation: Advances in research and theory (Vol. 2). Plenum Press; [Google Scholar]
  28. Pion‐Tonachini, L. , Kreutz‐Delgado, K. , & Makeig, S. (2019). ICLabel: An automated electroencephalographic independent component classifier, dataset, and website. NeuroImage, 198, 181–197. 10.1016/j.neuroimage.2019.05.026 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Perone, M. , & Baron, A. (1980). Reinforcement of human observing behavior by a stimulus correlated with extinction or increased effort. Journal of the Experimental Analysis of Behavior, 34(3), 239–261. 10.1901/jeab.1980.34-239 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Platt, J. R. (1973). Percentile reinforcement: Paradigms for experimental analysis of response shaping. In Bower G. H. (Ed.), The psychology of learning and motivation: Vol. 7. Advances in theory and research (pp. 271–296). Academic Press. [Google Scholar]
  31. Pliskoff, S. S. , & Goldiamond, I. (1966). Some discriminative properties of fixed ratio performance in the pigeon. Journal of the Experimental Analysis of Behavior, 9(1), 1–9. 10.1901/jeab.1966.9-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Sizemore, O. J. , & Lattal, K. A. (1978). Unsignalled delay of reinforcement in variable‐interval schedules. Journal of the Experimental Analysis of Behavior, 30(2), 169–175. 10.1901/jeab.1978.30-169 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Skinner, B. F. (1945). The operational analysis of psychological terms. Psychological Review, 52(5), 270–277. 10.1037/h0062535 [DOI] [Google Scholar]
  34. Skinner, B. F. (1957). Verbal behavior. Appleton‐Century‐Crofts. 10.1037/11256-000 [DOI] [Google Scholar]
  35. Skinner, B. F. (1974). About behaviorism. Alfred A. Knopf. [Google Scholar]
  36. Stoica, P. , & Moses, R. L. (2005). Spectral analysis of signals. Prentice Hall. [Google Scholar]
  37. Tanno, T. , Silberberg, A. , & Sakagami, T. (2012). Discrimination of variable schedules is controlled by interresponse times proximal to reinforcement. Journal of the Experimental Analysis of Behavior, 98(3), 341–354. 10.1901/jeab.2012.98-341 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Tolman, E. C. (1932). Purposive behavior in animals and men. Century/Random House UK. [Google Scholar]
  39. van der Meer, J. N. , Pampel, A. , Van Someren, E. J. W. , Ramautar, J. R. , van der Werf, Y. D. , Gomez‐Herrero, G. , Lepsien, J. , Hellrung, L. , Hinrichs, H. , Möller, H. E. , & Walter, M. (2016). Carbon‐wire loop based artifact correction outperforms post‐processing EEG/fMRI corrections—A validation of a real‐time simultaneous EEG/fMRI correction method. NeuroImage, 125, 880–894. 10.1016/j.neuroimage.2015.10.064 [DOI] [PubMed] [Google Scholar]
  40. Vidaurre, C. , & Blankertz, B. (2010). Towards a cure for BCI illiteracy. Brain Topography, 23(2), 194–198. 10.1007/s10548-009-0121-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological Review, 20(2), 158–177. 10.1037/h0074428 [DOI] [Google Scholar]
  42. Welch, P. (1967). The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Transactions on Audio and Electroacoustics, 15(2), 70–73. 10.1109/TAU.1967.1161901 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The present dataset can be requested at our website (https://bicr.atr.jp/dbi/download/conditional‐discrimination‐of‐electroencephalogram‐alpha‐amplitudes‐in‐humans/), after which a link to the data repository will be supplied.

Articles from Journal of the Experimental Analysis of Behavior are provided here courtesy of Wiley

RESOURCES