Research design
We assembled a unique dataset by counting the number of consumers (out of 80,471 consumers, 49% males) who donated to 72 street musicians in the city center of Cologne. The focus on one city is critical, as it enabled us to observe some musicians on different days.
To prevent our presence from affecting the buskers and consumers, we conducted observations from angles where the consumers and musicians would not notice us. The observations occurred from mid-December 2016 to March 2017, and all observations were made by a coauthor and an additional observer between 10:00 a.m. and 9:00 p.m.
Sampling strategy
To generate a sample of consumers and buskers, we first identified public busking locations in the city center, such as shopping streets, side streets, and public squares. We then randomly chose one of these locations at different times on different days.
Strategy for determining busker success
To determine how successful a busker was, we counted the number of consumers who donated to the busker in 10-min windows (e.g., from 10:20 a.m. to 10:30 a.m.)—a timeframe sufficient to approximate the musician’s success. We verified this sufficiency with a random sample of 15 musicians who we observed for two to three consecutive 10-min timeslots at the same location. A t-test did not indicate a significant absolute deviance of more than two donations across the timeslots (p = 0.437; SD = 1.49).
Underlying assumptions
Our identification strategy rests on the assumption that our observations are allocated randomly across musicians and consumers. Specifically, we do not assume that only successful musicians self-select by choosing to play only under conditions that increase their earnings (e.g., playing at specific spots). Conducting our study in Cologne helps, as buskers in Cologne must change locations every 30 min by law, which creates the opportunity for a quasi-randomized sample. Similarly, we do not assume that potential drivers (e.g., bad weather) change the average share of generous consumers in the population of pedestrians. Note: We control for the exact number of passersby. We are also aware that our assumptions may be violated under certain conditions and discuss these in the limitations section of this paper.
Data collection and operationalization
Data collection required two observers. The first observer counted the number of consumers who donated within 10-min intervals (our dependent variable) and noted observable personal characteristics of the donating consumers (e.g., gender). The second observer used a mechanical frequency counter to tally the number of male and female pedestrians who walked by during the 10-min interval.
The observers also noted environmental factors (e.g., weather) and conducted brief interviews with the buskers after each observation period to determine musician-specific factors (e.g., age).
After the 10-min interval, one of the observers took a sample video of the musician and secured permission to use the video anonymously in our study. In a later step, three professional musicians watched the video of each musician and evaluated the music quality on a 7-point scale. The answers of the three evaluators were averaged with a sufficient inter-coder reliability (Cronbach’s α = 0.794; Fleiss-Kappa: p < 0.01). Table 1 provides a more detailed overview of the measurement of all variables.
Summary statistics
This study covers a total of 80,471 consumers and 72 buskers observed in 122 10-min intervals, yielding an average observation time of 17 min. Forty buskers were observed only once, and the remaining 32 were observed in up to five time intervals. Buskers who were observed more than once were observed again on average after 10 days. The unit of analysis is the number of female and male donors per observation period. This yields 244 unique observations. Appendix 2 provides summary statistics for potential success factors. Appendix 3 gives an overview of the observed buskers over time and a histogram of the number of donating consumers. Appendix 4 provides correlations of all variables.
Model specification
Equation 1 gives the econometric specification of our regression model. The dependent variable “musician success” measures the absolute number of male and female consumers (g) who donate to the street musician (s) during the observation period (o). We measured the dependent variable separately for men and women to identify the effect of consumer gender. Consequently, we use 244 unique observations to estimate the model. The independent variables comprise musician characteristics (m), consumer characteristics (c), environmental characteristics (e), and controls (k).
$$\begin{aligned}{\text{MusicianSuccess}}_{\text{gso}} &=\alpha_{\text{s }}+ {\textstyle\sum_{\mathrm{m}=1}^{\mathrm{M}}}{\beta }_{1,m} {\text{MusicianCharacteristics}}_{\text{s}}\\&+ {\textstyle\sum_{\mathrm{e}=1}^{\mathrm{E}}}{\beta }_{2,e} {\text{EnvironmentCharacteristics}}_{\text{go}}\\&+ {\textstyle\sum_{\mathrm{c}=1}^{\mathrm{C}}}{\beta }_{3,c} {\text{ConsumerCharacteristics}}_{\text{gso}}\\&+ {\textstyle\sum_{\mathrm{k}=1}^{\mathrm{K}}}{\beta }_{4,k} {\mathrm{Controls}}_{\mathrm{gso}}+ {\varepsilon }_{\mathrm{gso}}\end{aligned}$$
(1)
with \({ \alpha }_{\mathrm{s }}= {\overline{\alpha }} +{\updelta }_{\mathrm{s}}, {\updelta }_{\mathrm{m}}\mathrm{ i}.\mathrm{i}.\mathrm{d}.\mathrm{ N}\left(0, {\upsigma }_{\text{gso}}^{2}\right)\) and ε kbmi i.i.d. N(0,σ2gso)
The parameter \(\alpha_{\text{s}}\) denotes the intercept, and ε and \(\updelta\) denote the error terms, which are assumed to be normally distributed. \({\upbeta }_{1,m}, {\upbeta }_{2,e},\) and \({\upbeta }_{3,c}\) are the estimated parameters of interest. \({\beta }_{4,k}\) designates the estimated control parameters.
By specifying a musician-specific constant \(\alpha_{s}\), we control for musician-specific characteristics that are unobservable to us. Specifically, we capture their joint influence in the unobserved term \({\updelta }_{\mathrm{s}}\), which is assumed to be normally distributed with zero mean and variance \({\upsigma }^{2}\).
Regression results
Table 2 reports our estimates in determining the characteristics that are associated with the number of donations to street musicians. We estimate an additional modelFootnote 1 to investigate the potential impact of consumers being accompanied and consumers’ ages (as control). The results of both models are highly consistent. Further, both models are highly significant, show good model fit, and do not yield multicollinearity issues (all VIFs < 10).
Table.2 Results of field study: consumer responses to buskers on the street Musician, environment, and consumer characteristics
As expected, we find empirical evidence for the importance of high-quality execution. Street musicians are associated with more donations if the music quality is high (β = 0.745, p < 0.01). Child musicians receive significantly more donations than adult musicians (β = 3.86 = , p < 0.01), and a large audience is associated with the number of consumers who donate (β = 2.258, p < 0.01).
While results on weather conditions (e.g., rainy) are not significant (β = -0.999, p > 0.05), we do find an effect from temperature: the colder the weather, the more likely people seem to donate (β = -0.263, p < 0.01). Day of the week also has an effect; Sunday is associated with the highest number of donations (p < 0.01).
We also find support for the relevance of consumer-specific variables. More women donate than men (β = 0.978, p < 0.01), and more consumers in company donate than consumers who are alone (β = 0.245, p < 0.01).
Control variables
For the sake of brevity, we only highlight significant control variables but fully report all parameter estimates in Table 2. First, we find some evidence that consumers are most generous when classical music is played (compared to rock: β = -2.814, p < 0.10). Second, the location of the musician on the street has a significant impact; standing against a wall (β = -2.28, p < 0.01) or in the middle of the street (β = -2.480, p < 0.01) is associated with fewer donations than standing in a square. Third, more consumers seem to donate in March than in the reference category February (β = 2.612, p < 0.05). Fourth, background noise is negatively associated with donations (β = -1.447, p < 0.05). Fifth, the analysis reveals that the more people who walk by, the more donations the musicians receive (β = 0.007, p < 0.01). Finally, more consumers between 30 and 65 years of age donate than in any other age category (e.g., consumers from 18 to 30, β = -0.316, p < 0.01).
Robustness checks
We conducted several further tests to ensure robustness. First, following prior research (Stäbler & Fischer, 2020), we used an inductive approach to test for potential interaction effects among our focal variables (Appendix 5). None of the interactions were significant or met the statistical requirements. Second, instead of counting the number of consumers who donated while controlling for the number of pedestrians, we operationalized the dependent variable as the donation share (equal to the number of donating consumers divided by the number of pedestrians). Appendix 6 shows that the results are fully consistent with our main model. Third, we tested for omitted variable bias and included further control variables. We included a measurement of consumers’ familiarity with the busker as well as the time of the day. The variables did not add value to the model according to the likelihood ratio test. Fourth, we removed (blocks of) controls, and the results do not indicate any other conclusions (Appendix 7).