Neurophysiological markers of emotion regulation predict efficacy of entrepreneurship education

Abstract

Recent evidence shows that programs targeting the socio-emotional dimensions of entrepreneurship-e.g., resilience, personal initiative, and empathy-are more highly correlated with success along with key business metrics, such as sales and survival, than programs with a narrow, technical bent-e.g., accounting and finance. We argue that programs designed to foster socio-emotional skills are effective in improving entrepreneurship outcomes because they improve the students' ability to regulate their emotions. They enhance the individuals' disposition to make more measured, rational decisions. We test this hypothesis studying a randomized controlled trial (RCT, RCT ID: AEARCTR-0000916) of an entrepreneurship program in Chile. We combine administrative data, surveys, and neuro-psychological data from lab-in-the-field measurements. A key methodological contribution of this study is the use of the electroencephalogram (EEG) to quantify the impact of emotional responses. We find that the program has a positive and significant impact on educational outcomes and, in line with the findings of other studies in the literature, we find no impact on self-reported measures of socio-emotional skills (e.g., grit and locus of control) and creativity. Our novel insight comes from the finding that the program has a significant impact on neurophysiological markers, decreasing arousal (a proxy of alertness), valence (a proxy for withdrawal from or approachability to an event or stimuli), and neuro-psychological changes to negative stimuli.

Figure 1

Experiment timeline and experiment paradigm. Four stages are included which are randomized selection, baseline field experiment, entrepreneurship program, and follow-up experiment, respectively. Entire experiment starts at February 2015 and ends in August 2015.

Figure 2

Impact on Emotional Responsiveness. The result of δ_j for different emotion-related variables j are shown. Bars represent the model estimation and 95% CI. Except Valence (positive), all other three variables have δ significantly (p < 0.05) smaller than 0 and the impact on responsiveness to negative stimuli Valence (negative) is the strongest.

Figure 3

EEG biomarker changes in 2-D Valence-Arousal (VA) space for positive stimuli. EEG biomarker changes are between baseline (t = 0, N₀ = 68) and follow-up (t = 1, N₁ = 68) experiment (see Eqs. (4) to (6)). The complete scatter plot, one point per student, with its original scale is shown in the upper right corner. The scatter plot in the center is a zoomed-in version with [− 10, 10] as x/y-axis limit (x-axis: $\mathrm{\Delta }Valance$; y-axis: $\mathrm{\Delta }Arousal$). Blue dots represent students who participated in the entrepreneurship program (students with program, g = 1) and red dots represent students who do not participant the program (students without program, g = 0). Colored vectors show the average of each group in the VA space (e.g., the blue vector is determined by the value of both $\mu \left(\mathrm{\Delta }{V}_{s=1,g=1}^{\mathrm{\Delta }t}\right)$ and $\mu \left(\mathrm{\Delta }{A}_{s=1,g=1}^{\mathrm{\Delta }t}\right)$ . For students with program, the mean and standard error of changes in valence with respect to positive stimuli (i.e., $\mathrm{\Delta }{V}_{s=1,g=1}^{\mathrm{\Delta }t}$) is 3.51 ± 3.46 with a t-test showing that $\mathrm{\Delta }{V}_{s=1,g=1}^{\mathrm{\Delta }t}$ is significantly different from $\mathrm{\Delta }Valance=0$ with a 95% confidence interval ($\mathrm{\Delta }{V}_{s=1,g=1}^{\mathrm{\Delta }t}>0$, p = 0.0495). However, for students without program, the mean and standard error of changes in valence with respect to positive stimuli (i.e., $\mathrm{\Delta }{V}_{s=1,g=0}^{\mathrm{\Delta }t}$) is 0.72 ± 1.64 and the t-test shows that $\mathrm{\Delta }{V}_{s=1,g=0}^{\mathrm{\Delta }t}$ is not significantly different from $\mathrm{\Delta }Valance=0$ within a 95% confidence interval. For changes in arousal domain ($\mathrm{\Delta }Arousal$), there is no significant difference between $\mathrm{\Delta }{A}_{s=1,g=0}^{\mathrm{\Delta }t}$ (0.47 ± 2.52) and $\mathrm{\Delta }{A}_{s=1,g=1}^{\mathrm{\Delta }t}$ (0.36 ± 3.10) and neither are significantly different from $\mathrm{\Delta }Arousal=0$.

Figure 4

EEG biomarker changes in 2D Valence-Arousal (VA) space for negative stimuli. EEG biomarker changes are between baseline (t = 0, N₀ = 68) and follow-up (t = 1, N₁ = 68) experiment (see Eqs. (4) to (6)). Plot is structured the same as above, except that the comparison is for negative stimuli. For students with program (g = 1), the mean and standard error of changes in valence with respect to negative stimuli (i.e., $\mathrm{\Delta }{V}_{s=2,g=1}^{\mathrm{\Delta }t}$) is $2.08\pm 2.58$ with a t-test showing that $\mathrm{\Delta }{V}_{s=2,g=1}^{\mathrm{\Delta }t}$ is not significantly different from $\mathrm{\Delta }Valance$ equals to 0 within a 95% confidence interval. However, for student without program (g = 0), the mean and standard error of changes in valence respect with respect to positive stimuli (i.e., $\mathrm{\Delta }{V}_{s=2,g=0}^{\mathrm{\Delta }t}$) is $-1.44\pm 1.37$ and the t-test shows that $\mathrm{\Delta }{V}_{s=2,g=0}^{\mathrm{\Delta }t}$ is significantly different from $\mathrm{\Delta }Valance$ equals to 0 within 95% confidence interval ($\mathrm{\Delta }{V}_{s=2,g=0}^{\mathrm{\Delta }t}<0$, p = 0.0414). With the t-test between two groups, results shows that there is significant difference between them. For changes in arousal domain ($\mathrm{\Delta }Arousal$), there is no significant difference between $\mathrm{\Delta }{\mathrm{A}}_{s=2,g=0}^{\mathrm{\Delta }t}$ ($0.58\pm 2.22$) and $\mathrm{\Delta }{\mathrm{A}}_{s=2,g=1}^{\mathrm{\Delta }t}$ ($0.18\pm 3.10$) and neither of them is significantly different from $\mathrm{\Delta }Arousal=0$.

Figure 5

Impact on Creativity and Socio-emotional Skills. The result of ${\delta }_j$ for different self-reported test j are shown. Bars represent the model estimation and 95% CI. δ of all four tests are not significantly (p < 0.05) different from 0.