Efficiency and productivity comparisons between outsourcers and non-outsourcers: Evidence from a metafrontier production function with endogenous switching

Abstract

This paper aims to compare productivity and technical efficiency (TE) between outsourcers and non-outsourcers with an empirical methodology that accounts for both heterogeneous production technologies and non-random sample separation. Using plant-level data on six two-digit manufacturing industries in Taiwan over the period 2002–2005, the endogenous switching regression and the stochastic metafrontier methodology enable us to generate TE scores that are comparable across production units that operate under different technologies without the standard assumption in the literature that the outsourcing status is out of the control of the plant and that outsourcers and non-outsourcers use the same technology. We find that outsourcers are, on average, more technically efficient and technologically advanced than non-outsourcers. Productivity differences account for the lion's share of the outsourcer–non-outsourcer output gap.

Keywords:

• outsourcing
• productivity
• technical efficiency
• Oaxaca–Ransom decomposition
• metafrontier
• technology gap ratios

JEL Classifications:

• C24
• D24
• F14
• L6
• O14
• O31
• O33

Acknowledgements

We would like to thank Tim Coelli for making the computer program Frontier version 4.1 publicly available and two anonymous referees for their insightful comments that greatly improved this paper. All remaining errors are our own.

Notes

Note: Outsourcing intensity is defined as the ratio of the expenditure paid to subcontractors to wage bill.

Note: ***significant at the 1% level.

Note: Standard errors (in parentheses) are constructed using the method in Greene (2000), 251–3.

Table 3 Oaxaca–Ransom decomposition of the output gap without selectivity bias correction.

Note: Standard errors (in parentheses) are constructed using the method in Greene (2000), 251–3.

***significant at the 1% level.

**significant at the 5% level.

***significant at the 1% level.

**significant at the 1% level.

Note: The critical value at the 1% level with 14 degrees of freedom is 29.141.

1. In this paper, we identify outsourcers as plants that paid positive amounts to external subcontractors that carry out a part of the production process for those plants.

2. As a firm transfers parts of its production activities to external specialist providers, it is able to shift its internal production towards its core competences.

3. The benefits of outsourcing are measured by better incentives for external providers and enhanced technology transfer and information transmission efficiency from outsourcers to production workers hired by the external provider via managers of the external provider firms.

4. Since offshoring, which is the act of relocating production activities to any foreign country without distinguishing whether the provider is external or affiliated with the firm, could potentially destruct a large number of jobs, much of the research effort has been on estimating the effects of offshoring on domestic labor market outcomes.

5. Due to the use of the SFA, which requires the maximum likelihood, we do not deal with the simultaneity of input choices, as noted by Olley and Pakes (1996) and Levinsohn and Petrin (2003). Yang and Chen (2009) also used the endogenous switching regression and the metafrontier approach. However, our paper analyzes a completely distinct question from the one considered there. Specifically, Yang and Chen applied these methods to compare the TE between small and large firms, whereas our current paper compares TE scores between outsourcers and non-outsourcers. Moreover, Yang and Chen studied the effect of being a subcontractor on plant performance, whereas this current paper investigates the performance differences between outsourcers and non-outsourcers.

6. See, for example, Antràs and Helpman (2004) and Antràs and Rossi-Hansberg (2009).

7. See Maddala and Nelson (1975) for an excellent review of the endogenous switching regression approach.

8. One can alternatively rely on the maximum-likelihood technique to estimate the simultaneous equations of (1)–(3). Due to the highly nonlinear nature of those equations, getting maximum-likelihood estimators is computationally cumbersome. In addition, a non-negative error term representing production inefficiency of a firm will be appended to (2) and (3) later, which makes the likelihood function more difficult to be derived. To keep the estimation problem tractable, we turn to the two-stage method of Lee (1978) and Heckman (1979).

9. The selection variable derived from a probit model is also known as the inverse Mills ratio. In what follows, we use them interchangeably.

10. It is worth mentioning that the coefficient estimates are still consistent even though ν_1it and ν_2it are heteroscedastic, but their estimated variances tend to be biased. Recall that our main interest is utilizing the coefficient estimates to assess the productivity and efficiency differences between outsourcing and non-outsourcing firms, instead of making statistical inferences. The use of consistent parameter estimates is enough to fulfill our purpose here.

11. We refer interested readers to the appendix of Battese and Coelli (1992).

12. The other two popular inefficiency estimates are the conditional mean, E(U_it |ɛ _i ), and the mode, Mode(U_it |ɛ _i ). See, for example, Kumbhakar and Lovell (2000).

13. In the study of labor market discrimination, usually represents the competitive wage structure purged off discrimination. Oaxaca and Ransom (1994) discuss four alternatives in constructing Here, can be referred to as the estimated common vector of technology parameters in the absence of discriminating outsourcers from non-outsourcers. See Oaxaca and Ransom (1988, 1994), Christofides and Pashardes (2002), and Neuman and Oaxaca (2004) for applications of the Oaxaca–Ransom decomposition.

14. Measure TE ^M _jit can be alternatively expressed as TE ^M _jit = TE _jit × TGR _jit . This shows that the TE relative to the metafrontier function, TE ^M _jit , is equal to the product of the TE score relative to the group frontier, TE _jit , and the TGR. It can be seen that TE ^M _jit must range from zero to 1 and is less than or equal to TE _jit , as both TE _jit and the TGR lie between zero and 1.

15. In an older version of this paper, we also analyzed data covering the period 1997–2000 and found very similar results. To save space, the results are unreported here, but are available upon request.

16. Görg, Hanley, and Strobl (2008) and Jabbour (2010) also adopted this way of scaling outsourcing expenditure.

17. We are grateful to an anonymous referee for pointing this out.

18. For brevity, the parameter estimates of the six probit models (equations [1] and [4]), the 12 group production frontiers (equations [10] and [11]), the six pooled production frontiers, i.e. production frontiers estimated using the pooled data of outsourcers and non-outsourcers, and the six metafrontier production functions (equation [17]) obtained by quadratic programming are not reported, but are available upon request.

19. All of the lagged explanatory variables are shifted by one period to alleviate the potential simultaneity problem.

20. This follows from equations (8) (9), which shows that the coefficient on λ₁ (λ₂) is equal to ().

21. A total of 18 additional production functions without correcting for self-selectivity are estimated, but are not shown to save space, yet are available upon request.

22. See Battese, Rao, and O’Donnell (2004) for the derivation of the LR statistic.

23. To name a few, according to the coefficient reported in Table 3, column 6 in Görg, Hanley, and Strobl (2008), a one percentage point increase in service (material) outsourcing intensity is predicted to increase the total factor productivity by 0.076% (0.003%). Table 4, column 2, of Jabbour (2010) predicts that a one percentage point increase in the intensity of outsourcing from developed (developing) countries is associated with a 0.00004 (0.00012) increase in total factor productivity.

24. See Coase (1937), Antràs and Helpman (2004), Antràs, Garicano, and Rossi-Hansberg (2008), and Antràs and Rossi-Hansberg (2009).

25. Most existing studies on the performance impact of an outsourcing deal with manufacturing firms.