Abstract
This paper examines the effect of trade openness on the productivity of skilled and unskilled labor in a group of 36 developing countries using panel data and fixed effect approach. We have developed and utilized an empirical model that readily lends itself to testing the hypothesis posed. Our results support the hypothesis that trade openness has a positive and significant impact on labor productivity for both skilled and unskilled labor in the sample countries. We also observe that the beneficial effect of trade openness is relatively stronger for the skilled labor than the unskilled labor. We conclude that contrary to the claim made by Mayda and Rodrik (Citation2001), skilled workers in developing countries may oppose protectionism. When adjusting for the purchasing power parity, the impact of trade openness on labor productivity, although positive and significant, is not as pronounced as it is for other definitions of openness.
Notes
See, Held and McGrew (Citation2000) and Jones (Citation2000) regarding these debates.
See, Weiss (Citation1999) for an excellent discussion on this debate.
See Appendix for the complete list of countries used in our sample (Table A1), detailed definition of variables used and their sources of data.
This form of CES production function is based on the theoretical model presented in Layard and Walters (Citation1978). According to these authors, when it comes to this type of production function, ‘there is no difficulty in introducing a third factor, by, say, distinguishing between skilled labor S and unskilled labor N, (Layard & Walters, Citation1978: 275).
Recent studies by Andersson (Citation2001) and Dar and Amirkhalkhali (Citation2003), among others, have used this definition.
Abizadeh and Grant (Citation1999) are among those who have used this definition in their analysis.
Chi-Square test statistics are 193.84, 195.03 and 202.74, respectively for the regressions reported in , and 84.31, 86.57 and 129.28, respectively for the regressions reported in .
For this we used the Levin-Lin-Chu panel unit root test as explained in Levin et al. (Citation2002). Baltagi (Citation2001: 236 – 238) also explains an earlier version of this test. The test was conducted in STATA using the command ‘levinlin’. This test requires a balanced panel. Hence the panel sample was reduced to 21 countries and the period from 1991 to 1998. To give an example, the calculated adjusted t-statistics for the two dependent variables (skilled labor productivity and unskilled labor productivity) were −151.53 and −13.52, respectively, which suggests rejection of the Null Hypothesis of non-stationarity in these two series.
We found that heteroskedasticity is a problem in all three regressions in . We used robust standard errors in those regressions.