Time allocation by the self-employed: the determinants of the number of working hours in start-ups

Abstract

This article examines determinants of working hours by self-employed, explicitly discriminating between preference and productivity effects. A simple model of working hours is derived, not requiring expected profit data. The model is estimated using data from a Dutch survey of 1350 start-ups. Outsourcing and number of employees appear related to both preference and productivity. The model has out-of-sample predictive power for the number of working hours.

Acknowledgements

We are grateful to EIM Business and Policy Research for having been able to use the Starterspanel data set. Ingrid Verheul acknowledges financial support of the Fund Schiedam Vlaardingen e.o. and the Trust Fund Rotterdam. We also thank Gerard Pfann for his comments.

Notes

¹ Thornton (1998, p. 90) finds the Cobb–Douglas specification to be sufficiently flexible against more general functional forms to model the (self-employed) physicians utility function. An alternative would be to choose a CES-type of specification for the utility function, following Wales (1973). An analytic expression for the optimal number of hours then requires the assumption of a linear relation between expected profit and hours worked (γ _i  = 1) having β _i left for measuring individual differences in productivity.

² Thornton (1998, p. 89) finds an estimate of the hours elasticity of patient visits to (self-employed) physicians equal to 0.746, significantly less than unity.

³ Thurston (2002) uses a similar substitution explaining working hours by young physicians.

⁴ We fix a ₀ 0.5 for reasons of identification. Choosing different values for this parameter had limited effects on the estimation results.

⁵ The seven categories are (1) < 10 000 guilders; (2) 10–25 000 guilders; (3) 25–50 000 guilders; (4) 50–100 000 guilders; (5) 100–250 000 guilders; (6) 250–500 000 guilders; (7) > 500 000 guilders. One Dutch guilder is equal to 0.45 Euros.

⁶ The estimated residuals closely follow a normal distribution: the Jarque–Bera statistic is 0.41 (p-value is 0.82).

⁷ The distribution of the preference parameter across the self-employed is approximately normal: the Jarque–Bera statistic is 8.29 (p-value is 0.02).

⁸ It is obvious that the use of a follow-up survey may introduce selection effects. Hence, the results should be interpreted with some care.