Fig 1.
Experimental strategy and subsequent empirical analysis.
The standard accent sample (Stand) is shown in white, the Bavarian accent (Bav) in gray, and the Thuringian accent (Thur) in black. The first stage of the experiment shows the two language informants (LI) who provide two language samples each. In the second stage, we relate economically relevant choices to the assigned treatments and match one of four language samples randomly with experimental participants (EP). In the analysis, we first estimate within-speaker differences to eliminate the effect of individual confounding characteristics (First Differences) and then calculate the difference in those first differences (Second Difference) to account for stochastic discrimination against regional accent. Contrasting the expected choices leaves us with an unbiased discrimination effect δ (cf. following explanations).
Fig 2.
Regional intensity across the language samples.
The figure shows the regional intensity of our language samples relative to codified standard German (d = 0). We find a strong and comparable deviation of both regional language samples from codified standard. At the same time, we find an insignificant difference between these two regional accent samples and between the two standard accent samples.
Table 1.
Procedure of the experiment.
Fig 3.
Color indicates potential gains (green) and losses (orange) compared to piece rate. Piece rate: m = Ʃca * € 0.50; revenue sharing: m = (Ʃca + Ʃcb) / 2 * € 0.50; tournament: m = Ʃca * € 1.40 if Ʃca > Ʃcb otherwise m = Ʃca * € 0.20, with m = payoff; ca = successful completion of task by participant A; cb = successful completion of task by speaker B.
Table 2.
Summary statistics of payment scheme choice pooled over all tasks.
Fig 4.
Predicted margins of Bavarian accent conditional on age and gender controls with the 90% confidence interval.
Standard errors are clustered by subject. This graph shows the predicted margins of the probability of choosing tournament from a multinomial logit model (Table 3) using age and gender as additional control variables. EPs do not chose the tournament more often when perceiving the Thuringian accent, but tournament take up increases strongly when perceiving the Bavarian accent.
Table 3.
Multinomial logit model of scheme choice.