Bounded rationality in music decision making

2.2 Bounded rationality in music decision making

To further develop the BEM within this thesis, four empirical studies explored whether bounded rationality and related insights from behavioural economics may prove valuable for music research (RQ3). The four studies showed that when making musical judgments and decisions, listeners are limited by their mental capacity (e.g., memory constraints), time, and information available (e.g., song titles, post-event information, or descriptions about the performer). Consequently, listeners rely on cognitive biases and heuristics that do not depend on the music stimuli themselves.

2.2.1 The repeated recording illusion (S3)

This study investigated the extent to which listeners are limited by memory constraints and the context when evaluating music performance (see Appendix C for the full text). To do so, a novel experimental paradigm was developed, namely, the repeated recording illusion.

In this paradigm, participants (N= 72) were told to listen to three “different” musical performances of an original piece. However, unbeknownst to them, they were exposed to the same repeated recording three times in succession. Each time, the recording was accompanied by a text suggesting a low, medium, or high prestige of the performer.

Participants evaluated the music using several rating scales (i.e., liking, timing, tone quality, pitch accuracy, emotional quality, and overall quality). The procedure was repeated using a piece of highly familiar rock music and a piece of unfamiliar classical music. Potentially related extrinsic factors (i.e., explicit information and repeated exposure) and individual differences were investigated.

Results showed that most participants (75%) believed that they had heard different musical performances while, in fact, they were identical. In the two music conditions, participants evaluated the same recording significantly more positively when it was presented with a high-prestige text compared to low and medium texts. However, the position of the recording only had a significant impact on the familiar music condition. To capture higher-order interactions between extrinsic and individual difference factors, a regression tree model based on permutation tests was computed. The dependent variable was a one-factor solution indicated by a Principal Component Analysis, where a negative score indicated an overall negative evaluation and a positive score an overall positive evaluation. The predictor variables were prestige effect, the position of the recording, music conditions (classical-unfamiliar vs. rock-familiar), and seven individual difference variables, including age, personality, and musicality. Figure2.2 shows the structure of the regression tree the model, which shows that only 3 of the predictor variables had a significant impact on performance evaluation, i.e., explicit information, repeated exposure, and the music condition. Note that none of the individual differences were significant predictors.

Overall, these findings highlight the fallibility of music evaluation and support the notion of bounded rationality in musical behaviour, showing that musical judgments are limited by

16 2.2 Bounded rationality in music decision making

memory constraints, cognitive biases, and the context. The influence of explicit information and the partial effect of repeated exposure are discussed in terms of the anchoring heuristic (Tversky & Kahneman, 1974) and processing fluency (Reber, Schwarz, & Winkielman, 2004).

Fig. 2.2 The influence of non-musical factors on music performance evaluation.

The regression tree model is useful in identifying the most predictive variables influencing music performance evaluation, as well as specific conditions that lead to particularly high (left node, 3) and low (right node, 9) ratings. The tree model can be interpreted by starting at the top and following each branch down, to arrive at a

terminal node. A path to a terminal node describes the interaction of experimental conditions that lead to a particular subset of ratings.

17 2.2 Bounded rationality in music decision making

2.2.2 False memories in music listening (S4)

When people listen to music or experience music in a live performance, they are normally exposed to related information at some point after the event. This study examined for the first time whether post-event misinformation can induce false memories in music (see Appendix D for the full text). Though misinformation effects have been demonstrated extensively within visual tasks, they have not yet been explored in the realm of non-visual auditory stimuli. Besides, the study explored individual difference factors potentially associated with false memory susceptibility in music, including age, suggestibility, personality, and musical training. In two music recognition tasks, participants (N = 151) listened to an initial music track, which unbeknownst to them was missing an instrument.

They were then presented with post-event information which either did or did not suggest the presence of the missing instrument. The presence of misinformation resulted in significantly poorer performance on the music recognition tasks (d = .43), suggesting the existence of false musical memories. A random forest analysis indicated that music expertise was not significantly associated with misinformation susceptibility. These findings support previous research on the fallibility of human memory and demonstrate, to some extent, the generality of the misinformation effect to a non-visual auditory domain. In the context of the BEM, this is important to further support the notion of bounded rationality in music decision making, in particular demonstrating the fallibility of memory-based judgments of music.

2.2.3 Names and titles matter: Linguistic fluency and the affect heuris- tic (S5)

This study manipulated the song titles and artist names presented with music to examine the influence of two well- known heuristic principles on aesthetic and value judgments of music:

processing fluency (Experiment 1) (Reber et al., 2004) and the affect heuristic (Experiment 2) (Slovic, Finucane, Peters, & MacGregor, 2002) (see Appendix E for the full text). In Experiment 1, the same music excerpts were presented with easy-to-pronounce (fluent) and difficult-to-pronounce (disfluent) names. The names consisted of a list of Turkish names that were shown in a previous study to be fluent or disfluent to English speakers (Shah &

Oppenheimer, 2007). Native English-speaking participants (N= 48) listened to the music stimuli and provided evaluations on different scales measuring aesthetic properties (e.g., like, emotional expressivity, quality) and subjective value of the music (e.g., likelihood to attend a concert of the artists or to recommend the song to a friend). Results indicated a main significant effect of fluency. In particular, participants evaluated the same music excerpts significantly more positively when presented with fluent names than when presented with disfluent names.

In Experiment 2 (N= 100), the same procedure was used, but instead manipulating the emotional content of the titles. Thus, the music excerpts were presented with positive (e.g.,

18 2.2 Bounded rationality in music decision making

Kiss), negative (e.g., Suicide), and neutral (e.g., Window) titles. This time, at the end of the experiment, participants also performed an unexpected free recall task (i.e., write down the songs they remembered). In both aesthetic and subjective value evaluations, presenting the music with negative titles resulted in the lowest judgments. When looking at the effects of emotionality on memory, results showed that music excerpts presented with neutral and negative titles were remembered significantly more often than positive titles.

Overall, these findings suggest that like any other human judgments, evaluations of music also rely on heuristic principles that do not necessarily depend on the aesthetic stimuli themselves. These heuristics operate even when the information processed is minimal, such as changing the linguistic properties of titles presented with music.

2.2.4 The effect of name recognition on listener choices (S6)

When searching for and choosing music in playlists, individuals may rely on judgment heuristics to make fast (in terms of computing time) and frugal (in the use of information) decisions. This study addressed this issue by investigating for the first time the role of the recognition heuristic on musical choices when listeners search for music in playlists (see Appendix F for the full text). The recognition heuristic states that when people are faced with recognised and unrecognised options, they infer that the recognized one has the higher value with respect to the criterion being judged and, therefore, they tend to choose it (Goldstein & Gigerenzer, 2002). In particular, the study extended the paradigm used in Oeusoonthornwattana and Shanks, 2010 to a listening task with 10 alternative choices, simulating a common listening playlist. Before the main experimental task, participants (German and English speakers) had to learn a list of Spanish names. This manipulation made it possible to create playlists using novel music paired with Spanish names that had been previously learned (i.e., recognisable) or were completely novel. In the main choosing task, participants were presented with ten songs in a playlist format and had to choose their favourite five. To study the role of recognition-based heuristics in the presence and absence of music information, participants searched for and selected music in two playlist conditions:

a titles-only condition (where they could only choose music based on visual cues – i.e., the Spanish names) and a titles-and-music condition (where they could choose music based on both visual and auditory cues – i.e., they could also listen to the music).

19 2.2 Bounded rationality in music decision making

Figure 2.3 depicts the mean choice proportion of music clips paired with learned and novel names in the two choosing conditions. Results confirmed that there was a significant effect of name recognition in the two choosing conditions, but this effect was larger when participants chose music based on visual information only (titles condition). Moreover, participants’ preferences for the selected music were also influenced by recognition - i.e., the same music clips were significantly more liked when paired with learned names than when paired with novel ones. These results show that listeners rely on the recognition heuristic when both deciding which songs to choose in a playlist and developing music preferences.

Fig. 2.3 Mean choice proportion of music clips when paired with learned and novel names in both playlist conditions (Error bars represent 95% CI).

1.00

0.75

0.50

0.25

0.00

Titles Only Titles and Music

Name recognition Novel Learned

Listeners rely on recognition cues when searching for and choosing music in the two playlist conditions.

However, the effect of name recognition was larger when listeners chose music only based on visual cues (title only) than they chose music based on both visual and music cues.

Mean choice proportion of music clips

20