Parents rated children before and after the training on a German standardized DSM IV- questionnaire for ADHD (FBB-HKS) 31, on the Conners' Parent Rating Scale (CPRS) 32 and on the Behavior Rating Inventory for Executive Function (BRIEF) 34. To screen for comorbid clinical conditions, the Child Behavior Checklist (CBCL) 35 was completed only once by parents, before the training. At the end of the training, parents participated in a semi-standardized interview, conducted by a psychologist who had not been involved in the therapy. Teachers rated children before and after the training on the Conners' Teacher Rating Scale (CTRS) 33 and on the teacher's version of the BRIEF 34.

A short form of the German WISC III 36 was used to estimate IQ, with Block Design, Vocabulary, Picture Arrangement and Arithmetics. Before and after the training, all children were tested using a comprehensive battery of neuropsychological tests:

- "Alertness", a subtest from the computerized standardized test battery "Test for Attentional Performance" (TAP 1.7) 37. In this task, participants respond to a visual stimulus by pressing a response button as fast as possible. A condition with auditory warning signal, where subjects are supposed to react faster, is contrasted with a condition without warning tone. Standard deviations of response time from both conditions are taken in this study as measures of response time variability.

- "Go/nogo", a subtest from TAP 37. In this inhibitory control task, subjects respond as fast as possible to a go-stimulus, an "X" presented on the computer screen, by pressing a response button, and inhibit responding when the no/go-stimulus, a "+", appears.

- "D2" 38. In this paper-pencil – test of focussed and selective attention subjects cross out critical letters on a working sheet, working line by line. They dispose of 20 seconds for finishing each line. The concentration performance is calculated as the number of correctly cancelled minus the total number of incorrectly cancelled items.

- "Score!", a subtest from the Test of Everyday Attention for children (Tea-ch) tapping sustained attention 39. In this task children have to count the number of auditory signals which they hear on a CD and which are presented at irregular intervals.

- "Code transmission" (Tea-ch). In this sustained attention task, children listen to a series of numbers from a CD. Whenever the critical number "five" appears twice, subjects have to name the number that immediately came before the fives. Test duration is twelve minutes.

- Trail Making Test (TMT) for children. This paper pencil task is a simpler version of the well-known TMT procedure 40. In part A, numbers spread randomly on a sheet have to be connected in ascending order as quickly as possible. In part B, numbers and letters of the alphabet have to be connected in ascending order, switching continuously between letter and number. Time differences between part A and part B are considered as indicative of switching costs.

In addition children participated pre- and post-training in an EEG and ERP recording. These results will be reported elsewhere 41.

Children were trained with the neurofeedback system "Goefi" which had been developed for the training of children with ADHD (see 7). The training was presented like a computer game for children with well-known cartoon figures from a TV show as protagonists. The child was instructed by different colours to activate or deactivate cortical excitability, the colour "blue" indicating decrease (positivity) and the colour "red" indicating increase of cortical activation (negativity). Successful activation or deactivation was fed back by an action of the animated cartoon figure (i.e. a pole vaulting mouse) and rewarded by points.

A feedback-trial consisted of a baseline period of 2 seconds, where the child was instructed to activate (negativity trials) or to deactivate (positivity trials), and a feedback period of 6 seconds, which started with an acoustic signal. During the feedback period, the feedback signal was updated every 100 ms. The reinforcement described above was provided at the end of the feedback period. The interval between trials was fixed to 5 ± 1 seconds. Negativity (50%) and positivity trials (50%) were presented randomly. Feedback was measured from Cz, with reference on mastoids, a bandwidth from .01–30 Hz, and a sampling rate of 250 Hz. Artefacts due to vertical eye movements were measured by two electrodes above and below the left eye and were corrected online. Increase or decrease of activation exceeding a threshold previously defined (200 μV for the eye channel, 100 μV for the EEG-channel; see 7), were considered as artefacts and were not fed back to participants.

In addition to feedback-trials, children also performed transfer trials with instructions given on the computer screen and with EEG recording, but no feedback was provided on the screen (= EEG transfer training). The aim of the EEG transfer trials was to prepare for generalization of learned activation or deactivation into daily-life situations (see 8). In a second form of transfer training children trained with red and blue transfer cards (= transfer training with cards) and without EEG recording. They were encouraged to carry these cards with them and to use them in everyday settings, e.g. during homework or during school exams.

Each training course consisted of 30 sessions. Two training sessions of 45 minutes each were administered on the same day, usually as double sessions, according to the following protocol: Children started with a block of 40 feedback-trials, in randomized order for positivity and negativity, followed by a block of 30 EEG transfer trials and by a block of 40 transfer trials with training cards. After a break of 15 to 20 minutes another block of 40 feedback-trials started, followed by a block of 30 EEG transfer-trials. The training was conceived according to cognitive behavioural principles, with therapeutic interaction between trainer and child being fundamental. The trainer encouraged the child to develop an appropriate strategy in order to induce activation or deactivation, to work out a plan how and where to use the strategy in daily life, discussed the problems encountered with transfer and introduced a training diary. Information of the parents represented another important part of the training. Parents were invited to participate at training sessions and to supervise transfer training with cards at home.

The training started as a two weeks-vacation-course with daily training sessions. This intense training phase was followed by a break of 5 weeks. Children and parents had been instructed previously to train regularly during this time with transfer cards and to employ the learned strategies in everyday-life situations. During these weeks, the trainer held contact with children and families by phone and e-mails. The second feedback-training period consisted of 5 double sessions, conducted once or twice a week during the school semester over a time of three weeks. Parents' involvement (low vs. high) based upon the contact held during the transfer period was rated by the trainer. Pilot studies had indicated that parents had to assume a very active role in the neurofeedback program, e.g. to remind the child during the transfer period without EEG training to exercise with transfer cards or to encourage the use of transfer cards in everyday situations. Parents were explicitly instructed to reward training efforts with "Goefi"-points which could be exchanged against small presents. Effective parental support (high vs. low) was thus assessed in a semistructured interview by asking parents whether they had reminded the child to exercise, had actively supported the use of transfer cards and had used the reward – system.

The group training program for children with ADHD was based on principles of cognitive behavioural therapy and conducted by two experienced child psychologists. It comprised components of social skill training, self-management, metacognitive skill training and enhancement of self-awareness. A major focus was on how to give and to benefit from feedback. Some modules were taken from or inspired by published programs (e.g. 4243), others had been newly developed. A group of five to six children participated in each group therapy program which consisted of 14 to 15 sessions of ninety minutes, hold with a frequency of one or two sessions per week. A summary of the training design of both treatments is shown in Figure 1.

Parents were invited to participate at the last 15 minutes of each session and received information on program issues on additional parent meetings. Parental involvement (high vs. low) was assessed based on the frequency of participations at treatments. In order to stabilize learning effects of the group sessions, children were assigned with "homework" from one week to the next, such as to measure the duration of all kinds of everyday activities with a stop watch in order to improve time estimation. Parents received information on the objective of these tasks without being explicitly asked to supervise them. In group therapy positive behaviour was also rewarded by points, but only during therapy sessions. Therefore the importance of the commitment shown by the parents in both groups was not directly comparable, and effective parental support at home could not be assessed.

Although therapeutic setting, method, and the distribution of sessions over time are different, the two treatments present a number of similarities: Both treatments are behavioural therapies aiming at the enhancement of self-awareness and self-control. Both types of treatments were roughly comparable in duration and in the amount of time spent in therapy sessions. In both treatments parents regularly received information on ongoing treatment issues and were invited to collaborate.

Composite scores of three behavioural scales rated by parents were entered into a repeated measures MANOVA (group by time by six composite scores – FBB-HKS: Hyperactivity, Impulsivity, Inattention; CPRS: Global Score; BRIEF: Behavioural Index, Metacognition Index). A significant effect for time (p = .011, partial Eta squared = .484) and an interaction of group by time (p = .004, partial Eta squared = .507) but no significant group differences were found (Table 2). The group by time interaction remained significant for the Inattention score of the DSM IV Rating Scale (FBB-HKS) and the Metacognition Index of the BRIEF after Bonferroni correction (Table 2). Post hoc t-tests revealed that children of the neurofeedback group improved significantly on the Inattention score (t = 3.957, p = .001) and on the Metacognition Index (t = 3.368, p = .004) whereas no changes were observed in the group therapy children on either subscale (Inattention score t = -.557, p = .588, Metacognition Index t = -1.035, p = .321). Therefore, according to parents' ratings, children benefited more from neurofeedback training than from group therapy on two subscales related to attention and cognitive performance. Similar advantages for neurofeedback over group training were found on the Impulsivity scale and for the Global Index of the CPRS, but these did not survive the conservative Bonferroni correction. Among the significant changes before correction, only those on the Hyperactivity subscale did not differentiate between the two groups. In addition, when effects sizes of changes on the CPRS Global Score were calculated separately for both groups, it could be shown that changes within the neurofeedback group were responsible for the improvement on the CPRS Gobal Index (neurofeedback: partial Eta squared = .557; group therapy: partial Eta squared = .155). One might thus question whether any behavioural improvement after group therapy can be found at all.

Composite scores of the behavioural scales CTRS and BRIEF rated by teachers before and after training were entered into another repeated measures MANOVA (group by time by three scores). The group effect and the interaction of group by time were found to be significant (group p = .036, partial Eta squared = .275; group by time p = .036, partial Eta squared = .380) (Table 2), while no general improvement over time was observed. In post hoc analyses the group effect was significant for the Behavioral Index of the BRIEF. On this index teachers had rated children of the neurofeedback group as more impaired than the children of the therapy group at both assessments (before training t = 2.344, p = .004, after training t = 2.924, p = .007). A significant group by time interaction was observed for the Metacognition Index of the BRIEF: Teachers rated changes related to cognitive performance as more pronounced after neurofeedback than after group therapy.

Neuropsychological test performances were entered into a repeated measures MANOVA (group by time by 7 performance parameters) (Table 3). Both groups showed a general improvement over time (time p = .000, partial Eta squared = .751), which was significant in post hoc tests (ANOVAS) for the number of commission errors in the Go/nogo test, the concentration performance in the cancellation task (D2), and the switching speed in the Trail Making Test. Interaction of time by group was not significant, indicating that with regard to neuropsychological performance both groups improved similarly over time. Group differences were significant (p = .013, partial Eta squared = .520), but there was only one significant performance difference between groups, the number of commission errors in the Go/nogo task, where children of the neurofeedback group made significantly more errors. These group differences were highly significant before training (t = 3.182, p = .004) and showed a strong trend (t = 2.012, p = .054) after the training. Effect sizes of significant changes over time were calculated separately for both groups. Effects were moderate to large in the neurofeedback group (TAP Go/nogo: partial Eta squared = .616; D2 concentration performance: partial Eta squared = .555; TMT: partial Eta squared = .339) and small to moderate after group therapy (TAP Go/nogo: partial Eta squared = .389; D2 concentration performance: partial Eta squared = .301; TMT: partial Eta squared = .245).

The cortical activation during the sessions at the beginning of the training (double sessions 2 and 3) was compared to sessions at the end of the training (double sessions 13 and 14) in order to evaluate changes in cortical regulation. To this end the mean amplitudes of positivity and negativity trials were calculated separately for feedback and transfer and entered into a repeated measures MANOVA (time by 4 parameters). In order to reduce effects of fatigue and diminishing motivation towards the end of the session, only the first two trials of each category per double session were considered. For two children who showed a high number of artefacts during the first sessions, data from invalid initial transfer trials were replaced using subsequent valid trials at day 2 and day 3. In another case transfer data of positivity trials could not been reliably obtained at time 1 because of problems with artefacts, and the group mean was entered into the analysis instead. The main effect of time was significant (F = 5.035, p = .011). Post hoc test showed that the mean amplitudes during negativity trials, during feedback (F = 4.772, p = .044) as well as during transfer (F = 7.222, p = .016), showed a significant change in the desired direction (Figure 2). Therefore, it can be concluded that in the course of the training children learned to increase slow central negativity. In contrast, no significant changes of the mean amplitudes during positivity trials were observed (feedback F = .193, p = .666; transfer F = .133, p = .720). Figure 2 illustrates that from the beginning of the training a majority of children spontaneously induced positive amplitudes during both types of trials. Obviously children did not need to learn to further increase positive amplitudes throughout the training but rather to differentiate successfully between negativity and positivity trials. To test this hypothesis, the differentiations between negativity and positivity trials at day 2/3 and 13/14 were compared in an additional analysis: A significant increase was found for the feedback condition (mean difference between amplitudes of positivity and negativity trials, time 1 = .41 μV, time 2 = 4.7 μV, t = -2.59, p = .020), whereas the differentiation in transfer conditions did not significantly increase over time (mean difference time 1 = 1.12 μV, time 2 = 3.149 μV, t = 1.440, p = .169).

In additional analyses the ability to differentiate between states of cortical activation in the transfer condition (which is more demanding than under direct feedback) was used as the relevant index of training success. This ability was considered more representative of improved cortical control in real life situations than a performance that is confined to a laboratory situation. To counterbalance for individual variations from one session to the next, which were considerable throughout the training, and for problems with artefacts at the beginning, the transfer trial data from double sessions 7 to 14 were entered into the analysis. Over all these sessions the group produced a mean difference between amplitudes of positivity and negativity trials of 2.711 μV (SD= 3.630). On closer inspection of the data, it became evident that only some of the children had learned well to differentiate between positivity and negativity in transfer condition. When the children were divided into two subgroups according to the median split of difference amplitudes, a group of good performers (N = 8) with a mean differentiation of 5.72 μV (SD 2.5) could be distinguished from a subgroup of poor performers (N = 9) with a mean differentiation of .034 μV (SD 1.9). The group of poor performers did not learn to differentiate between different states of cortical activation over the training and occasionally continued to produce the opposite of the desired state, which is indicated in Figure 3 by negative mean differences.

To summarize, although an overall group effect was found for an increase in the regulation of negativity during feedback and transfer throughout the training phase, the training did not result in an overall improvement in the differentiation between negative and positive activation in the full group. Only a subgroup of good performers improved. When demographic data of good and poor performers were compared, no salient difference emerged (see Table 4), though, as a tendency, a higher proportion of girls and of children with ADD was present among the poor performers. Otherwise, both subgroups seemed more or less comparable in regard to IQ, parental support, medication, age or initial severity of ADHD – symptoms.

To evaluate the relation between training performance and clinical improvement, training parameters were correlated with improvement on behavioural scales and neuropsychological tests. Improvement was conceptualized as difference between time 1 and time 2 (score 1 minus score 2) and correlated with the mean difference of negativity and positivity trials amplitudes in EEG transfer condition of double session 7 to 14 (Table 5). None of these correlation provided significant results, indicating that behavioural improvement cannot simply be related to training performance in EEG transfer condition. However, when correlations were computed for the subgroups of good and poor performers separately, significant correlations emerged for the good performers between the ability to differentiate between negativity and positivity and improvements on the Hyperactivity and Impulsivity subscales (FBB-HKS, parents' ratings) (Table 5, see Figure 4). Other moderate correlations (with CPRS Global Score, BRIEF Behavioral Score) resulted in trends for this subgroup. In the group of poor performers, changes on the behavioural scales were unrelated to the magnitude of differentiation. Improvements according to teachers' ratings did not result in any significant correlation, even when subgroups were considered separately.

Changes in neuropsychological test before and after the training were mostly unrelated to the ability to differentiate between negativity and positivity (Table 5). The only significant – but negative – correlation was found between the difference in the standard deviation of the Alertness task and the mean differentiation between positivity and negativity, which was more pronounced for the subgroup of good performers. This correlation can be explained by the fact that the children of the neurofeedback group did not improve on this task but rather showed a small but not significant increase of reaction time variability. Among the poor performers, improvement on the Go/nogo-task was inversely related to training performance. With exception of the TMT parameters, the remaining correlations, though not significant in this very small group, turned out to be positive for the good performers, which intuitively is in accordance with the expected direction of change associated with good achievement. One has to take into account, however, that difference scores may provide a biased result, as initially severely impaired participants have a better chance to improve.