Auditory perceptual task.

Sinusoidal tones of 75 milliseconds (ms) duration (including 10 ms rise and fall times) with an intensity of 45 dB sensation level (hearing threshold measured individually for each participant with a staircase procedure using the experimental sounds) were presented according to the auditory streaming paradigm (a cyclically repeating”ABA-”pattern; Fig 1, top panel). The frequency difference was 4 semitones with the”A” tone frequency set at 400 Hz and”B” tone at 504 Hz. The stimulus onset asynchrony (SOA, onset to onset time interval) was 150 ms. Participants were presented with eleven four-minute-long sequences with an additional ca. 40 second catch-trial segment (see the Test procedure section) appended without a break to the end of each block. Tones were delivered through Sennheiser HD600 headphones by an IBM PC computer using the Cogent 2000 toolbox [56] under MATLAB [57].

Listeners were instructed to mark their perception continuously in terms of four possible categories: a) “integrated” (ABA-; Fig 1, second panel; response: depressing one of the two response keys), b) “segregated” (A-A-/B—; Fig 1, 3rd panel; depressing the other response key), c) “combined” (-AB-/—A or -BA-/A; Fig 1, bottom panel; simultaneously depressing both response keys), and d) “none” (no repeating pattern perceived; releasing both response keys). Participants received both instructions and training to help them identify the different perceptual categories. The description of the integrated percept emphasized hearing all tones as part of a single repeating pattern with a galloping rhythm. The description of the segregated percept emphasized hearing two isochronous sound streams in parallel, one in the foreground, the other in the background, each with a uniform (one slower, the other faster) delivery rate. The description of the combined percept emphasized the perception of two parallel streams of sound, at least one of which included a repeating pattern composed of both high and low tones. Finally, the “none” choice allowed listeners to indicate that they did not hear any repeating pattern or could not decide between the patterns previously described to them. The left and right arrow keys of a standard Hungarian IBM PC keyboard were used as response keys with their roles counterbalanced across participants. Participants were instructed to keep one or both keys depressed for as long as they continued hearing the corresponding pattern but to switch to another combination as soon as their perception changed. They were asked to employ a neutral listening mind-set (termed “neutral instructions”; for a detailed discussion of the instructions, see [13]). The state of the response keys was continuously recorded at a nominal sampling rate of 10Hz.

Participants were trained to indicate their perception in terms of the above listed four categories without hesitation. Training started with the participant listening to six one-minute long demonstration sequences, each promoting the perception of one of the alternatives shown in Fig 1. The integrated percept was introduced by using a smaller frequency difference between the “A” and “B” tones (1ST; 400 and 426 Hz, respectively) than the parameters chosen for the experiment, while the segregated percept was initially demonstrated by using a larger frequency difference (890 Hz). Subsequently, the two segregated and the three combined percepts in Fig 1. were demonstrated by emphasizing the corresponding repeating tone pattern within an auditory streaming sequence with the parameters used in the experiment. Emphasis was created by attenuating by 18 dB those tones which were not part of the intended foreground stream and their timbre was also manipulated by including eight harmonics with equal weights (thus promoting perception of the normal-intensity puretones as a single coherent sequence). After the response key assignment and the “none” choice were explained to the participant, training continued in blocks of six sequences separated by short silent intervals. The first sequence was one-minute-long and its parameters were identical to those used in the experiment. This was followed by five sequences of 6–9 s, each, one sequence for each of the categories the participant was required to identify. The order of the five short sequences (small-frequency-difference sample and the 4 attenuated-intensity tones examples), which served as catch trials, was randomized separately for each training block. The training was completed when the participant made the intended response for at least 80% of the presentation time during the catch trials or when 20 training blocks had been delivered. During the training blocks, the experimenter gave feedback and further help as needed. No participant was rejected on the basis of the accuracy of their responses within the training blocks.

A ca. 40-second long catch-trial period, consisting of the five 6–9 s long segments of the example patterns used in the demonstration blocks, was appended to the end of each four-minute stimulus block in the experiment, with the purpose of checking correct response assignment for the non-ambiguous patterns throughout the experiment. The order of the catch-trial patterns was randomized across participants and blocks with each pattern appearing only once within a catch-trial period. If the average of the catch-trial matching performance across perceptual patterns and blocks during the experiment was below 60%, the participant was excluded.

Executive functions.

Inhibition, updating, and shifting [31] were assessed as follows. Inhibition was measured using a computerized version of the Stroop task [58,59] using E-Prime 2.0 [60]. The colours red, green, and blue were used, and participants were instructed to press the arrow key assigned to the colour appearing on a 15.6” screen with a resolution of 1366 x 768 pixels as quickly as possible. Participants were approximately 60cm away from the screen and stimuli had a width of 12cm and height of 3cm, giving them a vertical visual angle of 2.9° and horizontal visual angle of 11.4°. The arrow keys were used as response keys: “up” was paired with red, “left” was paired with blue, and “right” was paired with green. The task consisted of four conditions, each measured by 60 trials and delivered in three stimulus blocks. The first block was the neutral-word condition. In this condition, the names of the three colours appeared on the screen written in black. In the neutral-colour condition (2nd block) four “X” letters appeared in one of the three colours. The number of the “X” letters was decided by the average number of letters (4) in the Hungarian names of the colours used in the experiment. The third and final block mixed together the congruent and the incongruent condition in equal proportion. In congruent trials, the colour names appeared in their respective colours, whereas in the incongruent trials, the colour names appeared in one of the other two colours. The participants’ task was to press the correct response key as quickly and accurately as they could. The stimuli were shown on the screen until one of the response keys was depressed. This was followed by a blank screen for 250 ms. The Stroop interference effect was measured in the following way: first, the median reaction times of the correct responses measured in the colour-neutral and word-neutral conditions were averaged to get a neutral reaction time; second, the neutral reaction time was subtracted from the median reaction time of the correct responses in the incongruent conditions. Thus, a smaller reaction time difference indicates stronger inhibitory control of a prepotent response. The task was run using E-prime 2.0.

The executive function of updating was measured using a computerized version of the N-back task [61] using PsychoPy2 [62]. The same screen and viewing distance was used as in the Stroop task (see above). Stimuli had a width and height of 5cm each, giving visual angles of 4.8°. The task consisted of three blocks, each with 50 trials, of which 20 trials contained targets. 14 different capital letters were used as stimuli. Each letter was presented for 500 ms followed by a blank screen for 250 ms. In the first stimulus block, the participant’s task was to press a key when the current stimulus matched the preceding one (1-back condition). In the second and the third stimulus block, the letters to be matched were separated by one (2-back condition) and two letters (3-back condition), respectively. Responses were recorded during the 500 ms intervals of letter presentation but not during the 250-ms blank intervals. A response that was made within 500 ms from the onset of a target letter was scored as a hit, whereas a response within that time frame for a non-target was coded as a false alarm. Corrected Recognition Rate (CRR) was calculated for each condition using the following formula: (Hits / Targets)–(False alarms / Non-targets). The 1-back condition showed a ceiling effect, thus only the CRRs from the 2- and 3- back conditions were taken into account for further analysis.

Shifting was measured by semantic fluency [63], where participants were asked to name as many animals as they could. Unknown to them, they had one minute to complete the task. Responses were written down by the experimenter. During the spontaneous production of words, participants often name animals from various subcategories, such as African animals or pets. Shifts between these subcategories or semantic clusters can be used to assess the shifting executive function. Identification and analysis of the clusters was based on the protocol of Troyer and colleagues [63,64], adapted to Hungarian by Mészáros, Kónya, and Csépe [65] and was scored by an independent rater. Cluster-size was given as the length of the cluster minus one, such that a single word from a category was regarded as a cluster of the size zero, while for example, a cluster with five consecutive words belonging to the same category had the size four. The average cluster size and the number of subcategory changes (number of clusters minus one) were used as the output measures. All participants had at least 18 correct responses.

Personality questionnaires.

Participants completed four personality questionnaires on a computer using E-prime 2.0 software [60]. Computer-based administration helped to ensure that all participants responded to all questions, thus, there were no missing values in the final dataset.

First, they completed the ER89 questionnaire [66], which measures ego-resiliency (ER). The ER89 is a 14-item questionnaire, where participants have to indicate to what extent the items apply to them on a four-level Likert scale from “Does not apply to me at all” to “Applies to me very much”. In the Hungarian version [48], ER is measured using eleven items of the questionnaire (Cronbach’s α = .639 for the current sample).

Participants then filled out the Big Five Inventory [67] (Hungarian adaptation by personal communication from Z. Vass, Károli Gáspár University of the Reformed Church in Hungary), which measures the big five traits: Extraversion (8 items, α = .768 for the current sample), Agreeableness (nine items, α = .740), Conscientiousness (9 items, α = .833), Emotional stability (8 items, α = .855), and Openness (10 items, α = .806). Participants were instructed to rate the items on a 5-level Likert scale ranging from”I strongly disagree” to “I strongly agree”.

The next inventory was the UPPS Impulsive Behavior Inventory (Hungarian adaptation by personal communication from G. Orosz, Hungarian Academy of Sciences [68]). This inventory consists of 45 items that assess the four different aspects of impulsivity, labelled Urgency (12 items, α = .850 for the current sample), (lack of) Premeditation (11 items, α = .881), (lack of) Perseverance (10 items, α = .862), and sensation seeking (12 items, α = .881). Items on the questionnaire are scored on a 4-level Likert scale from ‘‘I strongly agree” to ‘‘I strongly disagree”.

Next, the Internal/External Encoding Style Questionnaire (NISROE [49]) was completed (the Hungarian version was adapted by the authors following the protocol of [69]). NISROE measures the internal and external encoding styles on 21 items using a 6-level Likert scale ranging from”I strongly disagree” to “I strongly agree”. As suggested by Lewiczki [49], only items 5, 8, 11, 15, 18, and 21 are used to calculate the final score on the questionnaire (α = .708 for the current sample).

Finally, participants filled out the Balanced Inventory of Desirable Responding (BIDR; [70,71] Hungarian adaptation by personal communication from G. Orosz, Hungarian Academy of Sciences). This questionnaire uses 20 items to measure two dimensions, self-deceptive positivity (SDP, α = .790) and impression management (IM, α = .562), to assess individual sensitivity to social expectations. Respondents indicate their agreement with items on a seven-level Likert scale ranging from “Not true” to “Very true”. The scores of this questionnaire can be used to identify respondents whose responses are likely to be affected by implicit expectations present in the social context. We have reported in a separate paper that the social desirability bias as measured by the BIDR did not have a significant effect on perception [72] in the present dataset. Therefore, the data obtained with this instrument will not be reported here.

Creativity.

Creativity was measured through two divergent thinking tasks, because this aspect of creativity can be best assessed in laboratory settings [73]. The first task was the Use of Objects Task [74], in which participants were instructed to produce as many novel uses as they could for three common objects (brick, paperclip, and newspaper). They had one minute for each word. Their voice was recorded and later transcribed. The second task was caption generation [73] in which participants were instructed to write as many captions as they could for three “New Yorker Magazine” cartoons, which were provided to the authors by R. E. Jung (University of New Mexico) through personal communication. Participants had two minutes for each of the cartoons and recorded the captions using paper and pen. All participants produced at least one response for each task.

Three independent raters (2 females) with MA in psychology (age: 25–28 years; M = 26.33, SD = 1.53) assessed the responses using the consensual assessment technique [75–77]. They rated 283 captions and 1106 uses of common objects in a randomized order on a 6-level Likert scale ranging from “Not creative at all” to “Very creative”. Raters were instructed to attempt to achieve a normal distribution of the assigned levels and creativity was not defined to them. Inter-rater reliability was assessed as adjusted inter-class correlations with the Reliability Calculator 1.5 (Mangold International GmBH). Inter-rater reliability was .811 for the Use of Objects and .706 for the caption generation task, both being acceptable values. Item ratings were averaged and linearly transformed to the 0–1 range, separately for the two tasks. The average of the two task-ratings were taken into account as the Composite Creativity Index (CCI, inter-rater reliability = .797) of each participant.

Pre-processing perceptual data from the auditory streaming paradigm.

Perceptual phases shorter than 300ms were discarded from the data recorded in the auditory streaming paradigm, because these were assumed to result from inaccurate switching between key combinations (see [79]). The data removed by this pre-processing step amounted to 0.5% of the total record duration.

Transition matrices were constructed from the perceptual reports using the method described in [4,23]. Each transition matrix had 4 rows and 4 columns (each corresponding to one of the 4 alternative perceptual classes, described previously) with elements representing the conditional probability of the percept changing from the one assigned to the column (starting percept) to the one assigned to the row (next percept; including the probability of no percept change). The conditional probabilities were estimated for each listener and block separately (block transition matrix), but also for each listener (by pooling data from all blocks for each listener: listener transition matrix), and for the whole experiment (by pooling all data from all neutral blocks for all participants: global transition matrix). Denham et al. [23] showed that the global transition matrix can be used to estimate missing data for individual listeners (i.e., transitions that were not observed for a given listener) and listener transition matrices can be used to estimate missing data for the each individual's block transition matrices. Therefore, this procedure was employed to provide a principled method for assigning default values in the event of missing data. Because the switching patterns obtained in the very first experimental block substantially differed from those obtained in the subsequent blocks [72], data from the first block were excluded. Four participants did not experience the combined percept. In the whole sample 37.67% of the sixteen possible transitions were missing. Most of the missing transitions were related to the ‘none’ percept, whose overall proportion was less than 0.8% in the data. Since we did not analyse the ‘none’ responses, the effective proportion of missing transitions was 13.19%.

Perceptual variables.

Most previous studies investigating individual differences in perceptual multi-stability (e.g., [21]) measured perceptual variables, such as the proportions and average phase durations of the alternative perceptual reports as well as the number of switches between the alternative perceptual reports. In order to allow comparisons between the current and previous results, the mean number of switches, and the proportions and mean phase durations of the integrated, segregated, and combined perceptual reports were estimated from the listener transition matrices. Variables for the “none” perceptual report were not analysed, because the overall proportion of this percept was less than 0.8%. The remaining variables characterizing the perception of the tone sequences were used in the analyses to be reported (see below).

Motivated by the idea that some listeners might discover the perceptual alternatives faster than others, we also analysed the time to discover all three perceptual alternatives, which is a potential source of individual differences in the data. This new measure provides information about the viability of the possible perceptions. Short discovery times suggest that all alternatives are relatively easy to perceive for the given participant, whereas long discovery times suggest that some perceptual alternatives are less viable; i.e. more difficult to discover and/or less able to win the competition for perceptual dominance. The measure was calculated by means of simulation: the listener transition matrices were used to simulate switching behaviour. Each simulation was run until all patterns were discovered. For each listener, the median duration of 1000 simulation runs was used as the “Time to discover all patterns” value. Lower values indicate quicker discovery of all perceptual patterns. The descriptive statistics of all variables are included in the S1 Table.

Testing the presence of idiosyncratic switching patterns.

In order to determine whether a given listener showed an idiosyncratic switching pattern, the intra-individual and inter-individual similarities were calculated for each participant and compared using Wilcoxon’s Signed Rank test. Intra-individual similarity was assessed by calculating the K-L distance values between each pair of the 4 (2nd to 5th) block transition matrices of the same listener. Inter-individual similarity was assessed by calculating the K-L distance values between transition matrix pairs for each combination having one of the 4 blocks of one listener and one of the 4 blocks of every other listener. In the Wilcoxon’s Signed Rank test, one-tailed assessment of significance was employed, because intra-individual similarity was expected to be lower than inter-individual similarity if the listener showed an idiosyncratic switching pattern. In previous studies (e.g. [3]), inter-individual variability was investigated using the number of switches measure. Therefore, we repeated the above analysis using the number of switches. The possibility of a different number of listeners being identified as having an idiosyncratic switching pattern by the two methods was tested with McNemar’s test [80].

Multidimensional Scaling of the transition matrices.

Based on the K-L dFistances between the listener transition matrices, Multi-Dimensional Scaling (MDS [81]) was used to find the main dimensions characterising listeners’ perceptual switching behaviour in the auditory streaming paradigm. MDS is similar to factor analysis, but instead of using the covariance matrix of the investigated variables, it uses a distance measure between the points in order to extract the underlying dimensions. The scree test [82], which evaluates the stress values of the dimension configurations was used to decide on the number of dimensions to be used. The stress value assesses how well the observed distance matrix is reproduced by an MDS configuration. A linear stress criterion was used as an index of the goodness-of-fit, which is the stress normalized by the sum of squares of the inter-response distances. The scree test indicated that three dimensions were sufficient for describing the transition matrix space of the listeners' switching patterns (MDS stress = .0408).

Interpretation of the MDS dimensions was attempted by correlating them with the perceptual variables. Because some of the variables did not have a normal distribution, and a normality transformation could have distorted their magnitude, Spearman’s rank-order correlations were used for this analysis. The significance levels of the correlations were determined by two randomization (permutation) methods. In the first approach, the distribution of the correlation coefficients under the null hypothesis was estimated for each perceptual variable separately, by permuting the values of the given perceptual variable and correlating them with the given MDS dimension 10,000 times. The p-value of the observed correlation was established as the proportion of random correlations higher than or equal to the observed value (more precisely, the absolute values of the correlation coefficients were compared to obtain a two-tailed test). The second method controlled for the family-wise error rate of the above procedure by registering the highest (absolute) correlation between the given MDS dimension and all perceptual variables in each permutation run, and using the distribution of these maximal coefficients to compute the p-value of the observed correlations [83,84].

Correlations between perceptual and other variables.

Spearman’s correlations were used to test the relationship between the MDS dimensions and personality, executive functions, and creativity variables. Correlations between these variables and the perceptual variables were also tested. Family-wise error correction was done as described in the section titled ‘Multidimensional Scaling of the transition matrices’, separately for the four groups of variables described at the beginning of the Data Analysis section. For correlations with significant p-values after family-wise error correction (marked as p_fwe), both types of p-values are reported. The correlation matrix between the perceptual variables and the executive functions, personality traits, and creativity is included in S2 Table with p-values without family-wise error correction.