(p.201) Appendix C Chapter Appendices
(p.201) Appendix C Chapter Appendices
Appendix to Chapter 3
The discussion in chapter 3 refers to the results from regression analyses of (1) the determinants of international differences in compensation (labor costs) and (2) the relationship between labor costs, level of development, economic and political characteristics, and social diversity. This appendix provides details about the underlying statistical analyses. Appendix A summarizes the definitions and data sources of all variables.
Analysis of Compensation (Labor Costs)
The discussion of links between compensation and productivity in chapter 3 refers to the regression analyses of international compensation differences for the manufacturing sector and for apparel and footwear—two lowwage manufacturing industries that play a significant role in discussions of globalization. For each industry, the analysis relates total compensation (labor costs) per worker to value added per worker and the average price level of consumption in purchasing power parity (PPP) terms to capture international costofliving differences not accounted for by the exchange rate conversion. All variables are in natural logarithms and are weighted by the size of the labor force in each country observation.
Total compensation “includes direct wages, salaries, and other remuneration paid directly by employers plus all contributions by employers to social (p.202) security programs on behalf of their employees.” As a broad measure of productivity, value added per worker captures the influence of capital, technology, education, training, experience, as well as unobservable influences on worker efficiency. Both the compensation and productivity data are from United Nations surveys of relatively large establishments in the formal sector, and “the data are converted into U.S. dollars using the average exchange rate for each year” (World Bank 2001a, table 2.5). The United Nations warns that the effectiveness in capturing some elements of compensation may vary from country to country. (The fixedeffects estimation discussed below removes constant countryspecific anomalies in the measurement of compensation.)
For total manufacturing, apparel, and footwear industries, the crosscountry analyses are conducted on 1995–99 averages for 58 countries (total manufacturing) or 28 countries (apparel and footwear industries). National differences in unobserved labor regulations, collective bargaining arrangements, or other institutions may be correlated with both compensation and productivity, however, producing biased crosscountry estimates of the links between those variables. Therefore, fixedeffects estimates supplement the crosscountry estimates for total manufacturing and apparel. Panels of 58 countries (manufacturing) and 28 countries (apparel) with 1980–84 and 1995–99 averages of the data provide the raw material for the fixedeffects estimation. (There are insufficient observations to conduct a fixedeffects analysis of compensation in the footwear industry.)
Crosscountry variations in labor productivity and price levels account for more than 90 percent of the international variation in labor compensation for all three industries. Both the productivity and price variables are also economically and statistically significant in the fixedeffect results for manufacturing. Productivity is significant throughout the analyses of all industries (table A3.1). The international price level is significant only for total manufacturing and is somewhat weaker in the fixedeffects estimates.
To summarize, a strong positive correlation between international productivity and compensation differences emerges in both the crosscountry and fixedeffects analyses for total manufacturing. Much the same may be said of the relationship between productivity and compensation in two lowwage industries, apparel and footwear, qualified slightly by the inability to conduct a fixedeffects analysis for footwear.
Analysis of Income Elasticities of Labor Conditions
Chapter 3 provides a discussion of the sensitivity of labor conditions to a country's level of development (real per capita income) and changes in the level development. The discussion is also based on crosscountry and panel estimates of the income elasticities of labor conditions. The following regression equation (with all variables weighted by labor force size) provided crosscountry estimates for 1995:
Table A3.1 Regression Analysis of Compensation (Labor Costs)
Estimation 
Productivity 
Price 
Constant 
R^{2} 

A. Manufacturing 

1. Crosscountry (1995–99) 
0.712 (.063)* 
0.731 (.132)* 
−1.204 
0.97 
2. Fixed effects 
0.909 (.036)* 
0.453 (.070) 
−2.100 
0.91 
B. Apparel 

1. Crosscountry (1995–99) 
0.985 (.111)* 
−0.139 (0.459) 
0.003 
0.93 
2. Fixed effects 
0.942 (.173)* 
−0.04 (0.486) 
0.153 
0.84 
C. Footwear 

1. Crosscountry (1995–99) 
0.721 (0.129) 
0.529 (0.318) 
−0.582 
0.92 
Notes: Dependent Variable: Annual labor costs (compensation) per worker.
All variables in natural logarithms. All observations are weighted by labor force.
Robust standard errors in parentheses for crosscountry estimates.
(*) pvalue < .01
Sources: UNIDO (2001); World Bank (2001a)
The dependent and independent variables are, respectively, the natural logs of a labor condition (LC) and real per capita income (GDPCAP) in country i, and ε_{i} is the countryspecific error term. For each labor condition, table A3.2 reports the estimated income elasticity, a_{1}. Many countries report values of zero for two measures of labor rights—the child labor force participation rate and the number of varieties of forced labor. For these two measures, equation 3.1A is estimated using tobit analysis. For all other measures of labor conditions, the equation is estimated using ordinary least squares. The reported 1995 crosscountry estimates are representative; unreported crosscountry estimates for every 5 or 10 years (as data permitted) from 1970 to 2000 showed little difference in estimates of income elasticities.
For some measures of labor conditions, one may question whether causality runs from per capita GDP to the labor condition or the other way around. (Work hours, an input to production, may present the strongest case for reverse causality.) Taking the view that per capita GDP might influence (p.204)
Table A3.2 Income Elasticities of Labor Conditions
1995 Crosscountry 
Fixed effects,^{a} 1970–2000 


Labor conditions 
Coefficient 
R^{2} 
n 
Coefficient 
Period 
Long hours 
−0.015 (.10) 
0 
37 
−0.005 (.024) 
1990–2000 
Annual hours 
−0.06 (.04) 
0.06 
27 
−0.05* (.006) 
1990–2000 
Weekly hours 
−0.048* (.007) 
0.40 
43 
−.033* (.005) 
1980–2000 
Fatal accidents 
−0.23 (.23) 
0.09 
32 
−0.47* (.07) 
1970–2000 
Life expectancy 
0.10* (.014) 
0.63 
105 
0.14* (.008) 
1970–2000 
Child labor 
−7.62* (.60) 
.15^{b} 
110 
−5.42* (1.74) 
1970–2000 
Civil liberties 
−0.46* (.09) 
0.57 
104 
−0.21* (.04) 
1972–2000 
Freedom of association 
−0.34** (.14) 
0.21 
92 
n.a. 
Late 1990s 
Forced labor 
−29.01* (10.05) 
.08^{b} 
101 
n.a. 
Late 1990s 
Slavery 
−1.47** (.69) 
0.23 
69 
n.a. 
Late 1990s 
Gender differential^{c} 
−0.006 (.031) 
0 
56 
n.a. 
About 1985 
Notes: All variables in natural logarithms. All observations weighted by labor force. Robust standard errors in parentheses for crosscountry estimates.
n = number of observations
(a) Randomeffects estimates reported when a Hausman test finds no significant difference between fixed and random effects. Also, fixed effects not available for tobit estimator.
(b) Pseudo R^{2} from tobit estimation.
(c) 1985 crosssection (see text)
n.a. not available
(*) pvalue < .01
(**) pvalue < .05
Table A3.2 also reports panel data estimates of income elasticities for the late twentieth century obtained from the following regression model:
In this regression, t indexes the year, and λ_{i} represents the fixed effect for country i. The panel consists of country data for intervals during 1970–2000. (With complete data, the panel consists of (1) observations every five years during this period for civil liberties, the fatal job accident rate, life expectancy, and real per capita GDP, (2) observations for 1970, 1980, 1990, 1995, and 2000 for child labor and weekly hours of work in manufacturing, and (3) observations for 1990, 1995, and 2000 for annual work hours and long workweeks.) Data limitations impose unbalanced panels for most measures. Sample sizes range from more than 100 countries (for life expectancy) to fewer than three dozen countries (for some measures of work hours).
Most panel results are fixedeffects estimates. Where a Hausman test rejects the hypothesis of no difference between fixedeffects and randomeffects estimates (weekly work hours and fatal job injuries), randomeffects estimates are reported. For variables with only one annual observation (both measures of forced labor [late 1990s], the measure of discrimination [roughly 1985], and the FACB index [mid1990s]), only crosscountry estimates of income elasticities are available.
Table A3.3 reports income semielasticities, also estimated by the ordinary least squares, tobit, and panel methods used for table A3.2. Each figure describes the absolute change in the measure of labor conditions to a 1 percent change in per capita GDP. For example, the fixedeffect estimates indicate that a 1 percent difference in per capita GDP between countries is associated with a difference of about five fatal onthejob injuries per 100,000 employees, nine years life expectancy, and 1 point on the 7point civil liberties scale. Where results are statistically significant, higher per capita income is associated with superior working conditions and labor rights, as discussed in chapter 3.
Table 3.1 in chapter 3 reports outliers—countries with unusually large positive or negative deviations from the value of a labor condition predicted by the crosssection semielasticity for their level of development in 1995. Each figure reported in table 3.1 is the ratio of the regression residual (actual minus predicted value of the labor conditions) for the country to the mean squared error for the crosscountry regression for the labor condition. All figures are computed from the regressions reported in table A3.3.
Institutions, Social Diversity, and Labor Conditions
Chapter 3 also discusses the role of economic and political institutions and social diversity (measured by ethnic and religious diversity) on labor conditions around the world. To estimate their actual influence, measures of these factors are added as independent variables in equation 3.2A. Indices of the rule of law and risk of expropriation represent economic institutions. Each index is measured on a 10point scale, with higher values indicating (p.206)
Table A3.3 Income SemiElasticities of Labor Conditions, 1995
1995 Crosscountry 
Fixed effects,^{a} 1970 2000 


Labor conditions 
Coefficient 
R^{2} 
n 
Coefficient 
Period 
Long hours 
−0.53 (5.76) 
0 
37 
−2.20* (.84) 
1990–2000 
Annual hours 
−116.16 (73.60) 
0.07 
27 
93.83* (10.40) 
1990–2000 
Weekly hours 
−2.08* (.32) 
0.38 
43 
−1.37* (.20) 
1980–2000 
Fatal accidents 
−2.36** (.90) 
0.16 
32 
−6.01* (1.32) 
1970–2000 
Life expectancy 
6.71* −0.44 
0.69 
105 
8.78* (0.40) 
1970–2000 
Child labor 
−12.67* (1.04) 
0.18^{b} 
110 
1970–2000 

Civil liberties 
−1.36* (.16) 
0.43 
104 
−0.68* (.13) 
1972–2000 
Freedom of association 
−1.26** (.57) 
0.17 
95 
n.a. 
Late 1990s 
Forced labor 
−2.68* (.85) 
.12^{b} 
106 
n.a. 
Late 1990s 
Slavery 
−3034030 (2226022) 
0.14 
69 
n.a. 
Late 1990s 
Gender differential 
−.004 (.017) 
0 
56 
n.a. 
About 1985 
Notes: See notes to table A3.2.
(a) Random effects estimates reported where Hausman test finds no significant differences between fixedeffect and randomeffect estimates.
(b) Pseudo R^{2} from tobit estimation
(*) pvalue < .01
(**) pvalue < .05
Estimates of the effects of institutions and social diversity on labor conditions are developed from country panels of data for 1980, 1985, and 1990—dates imposed by the availability of data on economic institutions. (Measures of social diversity exist only for the early to mid1990s but are unlikely to change markedly over time.) Random effects estimation captures the effects of the latter variables from crosscountry variation, and the effects of development and economic and political institutions from both the crosssection and overtime variation available in the panel. For the FACB index, gender pay differences, and both varieties of forced labor, the absence of panel data permits only a crosssection analysis. The measures of child labor and forced labor are truncated at zero—most developed countries report no child or forced labor. Tobit analysis is used to estimate these regressions, using panel data in the case of child labor and crosscountry data in the case of forced labor.
The results indicate several links between a country's labor conditions and its economic, political, and social characteristics (table A3.4). Correlations between labor conditions and per capita income survive the addition of the new variables to the model. Most measures of labor conditions are significantly correlated with DEMOCRACY. Countries with democratic political institutions tend to have superior labor conditions. Institutions and social diversity variables are selectively influential on various labor conditions. (See discussion of these results in text of chapter 3.)
Appendix to Chapter 4
Chapter 4 discusses (1) how international trade theories imply that trade improves working conditions by increasing the efficiency of resource use, which should raise per capita income, and (2) how liberalized trade may have an additional direct influence on some labor rights. In contrast, the racetothebottom hypothesis predicts that free trade degrades labor conditions. This appendix provides details of the econometric analyses of links between a country's openness to international trade and the labor conditions that are discussed in the chapter. The regression analyses reported below add measures of openness to the crosscountry and panel regression models described and applied in the appendix to chapter 3. Appendix A summarizes the definitions and data sources of all variables.
Table A3.4 Labor Conditions, Institutions, and Social Divisions, 1980–95
Weekly hours^{a} 
Fatal accidents^{a} 
Life expectancy^{a} 
Civil liberties^{a} 
Child labor^{b} 
Forced labor^{c} 
Gender differential^{d} 


Per capita GDP 
−0.076 (.020)* 
−0.138 (−0.152) 
0.082 (.009)* 
−0.07 (.024)* 
−3.219 (.154)* 
−59.988 (21.191)* 
−0.077 (0.048) 
Rule of law 
0.001 (0.009) 
0.058 (0.092) 
0.001 (0.004) 
−0.056 (.013)* 
−0.423 (.108)* 
−13.186 (10.214) 
0.016 (0.031) 
Expropriation 
0.004 (0.005) 
−0.017 (0.049) 
0.014 (.002) 
0.01 (0.008) 
0.132 (.072)*** 
31.236 (10.751)* 
0.057 (.032)*** 
Democracy 
−0.096 (.042)** 
−1.01 (.412)** 
0.019 (0.016) 
−1.546 (.058)* 
−1.473 (.406)* 
−77.458 (32.948)** 
0.128 (0.110) 
Ethnic 
0.042 (0.066) 
1.085 (.503)** 
−0.21 (.040)* 
−0.002 (0.079) 
9.630 (.473)* 
−49.731 (21.174)** 
0.254 (0.169) 
Religion 
−0.004 (0.064) 
−1.078 (.444)** 
−0.013 (0.038) 
−0.121 (.073)*** 
−12.689 (.570)* 
−42.219 (32.764) 
−0.153 (0.144) 
Constant 
4.453 
4.106 
3.427 
2.747 
26.467 
303.171 
−0.081 
R^{2} 
0.38 
0.49 
0.77 
0.91 
— 
0.35 
0.51 
Countries 
51 
38 
97 
97 
96 
93 
55 
Observations 
76 
98 
272 
281 
273 
93 
55 
(Notes: a) Randomeffects estimates.
(b) Randomeffects tobit estimates.
(c) Tobit estimates.
(d) Ordinary least squares estimates.
(*) pvalue < .01
(**) pvalue < .05
(***) pvalue < .10
(p.209) CrossCountry Analysis
The crosscountry analysis tests for the effect of a country's openness to trade on the measures of labor conditions using data for 1995, except in the case of discrimination, where data for 1985 are used to accommodate the fact that the gender wage differential is centered on 1985. Sample sizes vary with data availability for measures of working conditions and labor rights: weekly hours (50 country observations), annual hours of work (27), long work schedules (37), fatal accident rate (33), life expectancy (113), civil liberties index (108), child labor force participation rate (107), forced labor varieties (106), and gender wage difference (57).
The underlying crosscountry regression model is:
As in the analyses for chapter 3, the dependent variable for each regression, LC_{i}, is a measure of a working condition or labor right in country i. The independent variables respectively measure real per capita income, openness to international competition, and j measures of economic and political institutions and social diversity. (The vector of variables, INSTITUTIONS_{j}, includes the measures of rule of law, risk of expropriation, democracy and civil liberties, ethnic diversity, and religious diversity used in chapter 3.) All variables except the INSTITUTIONS vector are in natural logarithms, and all observations are weighted by the country's labor force. The random error term is ε_{i}. The analysis tests for a relationship between each labor condition and OPEN, measured alternately by exports plus imports as a fraction of GDP (TRADE SHARE), and a multihurdle, updated SachsWarner indicator of open trade policies (OPEN POLICY) (Wacziarg and Welch 2003).
The coefficient, a_{2}, provides the crucial test of whether the direct effect of openness is to improve or degrade labor conditions. Note that the total effect of openness on labor conditions includes the direct effect plus the indirect effect that occurs as a country's per capita income is affected. The a_{2} coefficient tests only for the direct effect of openness because the regressions statistically control for the effects of per capita income on labor conditions. That is, the indirect effect of openness is included in the a_{1} coefficient. Since higher per capita income is associated with superior labor conditions (chapter 3), and since, as discussed in chapter 4, there is a consensus that openness raises per capita GDP, the hypothesis that openness degrades labor conditions requires more than a negative direct effect, a_{2}. It requires that a_{2} be sufficiently negative to overwhelm the positive indirect effect of openness. To degrade a country's labor conditions, greater economic openness would have to cause a direct deterioration in labor conditions that overwhelms its positive indirect influence.
(p.210) If labor conditions and some measures of openness are jointly determined (see discussion in main text of chapter 4), ordinary least squares (OLSQ) will produce biased estimates of the relationship between openness and labor conditions. As a result, table A4.1 reports both OLSQ and instrumental variables (IV) estimates of the crosscountry, laborforceweighted relationships. “Gravity” models of international trade flows provide suitable instruments for the trade share of GDP—exogenous variables that are correlated with trade but are unlikely to influence labor conditions except through their influence on trade. The gravity variables used to instrument TRADE SHARE and OPEN POLICY are the labortoland ratio, a dummy variable for small countries, and a dummy for island economies. (See appendix A for details.) The firststage regressions account for 25 to 50 percent of the variance in TRADE SHARE and from 40 to more than 90 percent of the variance in OPEN POLICY, depending on the labor condition under analysis.
Given the scaling of the different measures of labor conditions, a finding that a_{2}>0 will indicate that openness directly improves wages and life expectancy, and a finding that a_{2}<0 will indicate that openness directly improves hours of work, job safety, civil liberties, freedom of association, forced labor, and slavery. The opposite findings would be consistent with the hypothesis that poorer labor conditions accompany openness to international trade.
Table A4.1 reports OLSQ and IV estimates of the a_{2} coefficients (with robust standard errors) for working conditions (pay, hours of work, job risks) from the regression model described in equation 4.1A. The OLSQ and IV estimates of a_{2} often differ substantially in magnitude, supporting concerns that OLSQ may yield biased estimates. None of the IV estimates of TRADE SHARE and OPEN POLICY associations with working conditions are statistically significant. These results imply that openness has only an indirect influence on these working conditions. Larger trade shares or open trade policies improve working conditions by raising per capita GDP (as shown in chapter 3) but have no further direct influence on working conditions. (Recall that the a_{2} estimates come from regressions that already control for per capita GDP.) In two cases, the statistical significance of the estimated openness coefficient changes with the estimation method. The OLSQ implications that countries with open trade policies have lower wages and a larger proportion of the labor force working more than 50 hours per week, ceteris paribus, disappear in the IV estimates. (The small sample of countries reporting annual hours data includes no closed economies, so a test for an association with open policy is not possible.)
In contrast, OLSQ, IV, and, for the measures of child labor and forced labor, tobit estimates indicate that openness has more direct associations with most measures of labor rights (table A4.2). (Tobit estimation follows from the fact that data on child labor force participation rates and the number of varieties of forced labor are censored at zero for many countries.) Countries with open trade policies have significantly more civil liberties (although not (p.211)
Table A4.1 Openness and Working Conditions
Crosssection 
Panel 


Working condition 
OLSQ 
IV 
Fixed effects 
Compensation 

Trade share 
−.101 (.111) 
.007 (.217) 
.003 (.114) 
Open policy 
−.323 (.185)*** 
−.500 (0.356) 
−.062 (0.062) 
Weekly work hours 

Trade share 
−.029 (.023) 
−.051 (.041) 
−.009 (.008) 
Open policy 
−.008 (.046) 
−.040 (.087) 
.004 (.005) 
Annual work hours 

Trade share 
−.072 (.067) 
−.099 (.131) 
−.011 (.009) 
Open policy 
n.a. 
n.a. 
−.001 (.004) 
Long work hours 

Trade share 
−.054 (.143) 
−.103 (.266) 
−.002 (.027) 
Open policy 
.407 (.120)* 
−2.862 (5.049) 
−.005 ( .011) 
Fatal accidents 

Trade share 
.214 (.520) 
−1.483 (1.118) 
−.234 (.169) 
Open policy 
1.357 (1.040) 
−3.807 (7.031) 
−.379 (.156)** 
Life expectancy 

Trade share 
.0001 (.0115) 
.023 (.027) 
.093 (.009)* 
Open policy 
−0.012 (.030) 
−.0068 (.0793) 
.048 (.007)* 
Notes: Regression coefficients and robust standard errors (in parentheses) for OPEN variable in regression model (4.1A) described in the text of this appendix. Each coefficient is from a different regression.
(*) pvalue <.01
(**) pvalue <.01
(***) pvalue <.01
Table A4.2 Openness and Labor Rights
Crosssection 
Panel 


Labor right 
OLSQ 
IV 
Fixed effects 
Civil liberties 

Trade share 
.204 (.147) 
.505 (.360) 
−.036 (.040) 
Open policy 
−.613 (.144)* 
−.834 (.511)*** 
−.115 (.032)* 
FACB index 

Trade share 
−1.286 (.600)** 
−1.024 (1.184) 
n.a 
Open policy 
−.197 (.788) 
.624 (2.697) 
n.a. 
Discrimination 

Trade share 
−.024 (.058) 
.177 (.108) 
n.a. 
Open policy 
.146 (.075)** 
.428 (.127)* 
n.a. 
Child labor^{#} 

Trade share 
−.435 (1.410) 
n.a. 
−1.254 (.437)* 
Open policy 
−4.966 (1.512)* 
n.a. 
−1.352 (.507)* 
Forced labor# 

Trade share 
−25.763 (14.172)*** 
n.a. 
n.a. 
Open policy 
−36.547 (16.393)** 
n.a. 
n.a. 
Slavery 

Trade share 
−2.885 (.778)* 
−1.986 (1.89) 
n.a. 
Open policy 
1.856 (1.53) 
−4.069 (6.66) 
n.a. 
Note: See notes to table A4.1.
(#) Tobit estimates
Panel Analysis
Further biases may exist if unobserved countryspecific influences on labor conditions are also correlated with openness. The (unobserved) domestic regulation of labor conditions by national governments or by labor unions might be correlated with a country's openness to international competition, for example. For all labor conditions except for the FACB index, discrimination, and the forced labor measures, the availability of panel data permits fixedeffects estimation, which uses withincountry changes over time to estimate the coefficients of the regression model. The appendix to chapter 3 describes the country panel data used in the analysis. Tables A4.1 and A4.2 also report fixedeffects estimates of the relationship between openness and working conditions and between openness and labor rights, respectively. These estimates confirm the absence of a statistically significant relationship between openness and pay or any of the measures of hours of work. They also provide a more positive read on the relationship between openness and job safety; countries that change to more open trade policies experience a reduction in fatal job accidents and an increase in life expectancy, ceteris paribus. The fixedeffects estimates also confirm that openness is associated with more civil liberties and less child labor.
Appendix to Chapter 6
Chapter 6 includes a discussion of whether countries with poor labor conditions gain unusually large shares of foreign direct investment (FDI), other things equal. This appendix provides details of crosscountry and panel regression analyses underlying that chapter's discussion of this issue. (Appendix A summarizes the definitions and data sources of all variables.)
The analysis describes why the share of FDI inflows varied among countries in the 1980s and 1990s. The approach is to estimate a baseline model of FDI inflows and then to test for whether the addition of measures of labor conditions and national labor regulations improves our understanding of the variation in FDI among countries. The dependent variable is the natural logarithm of a country's share of world FDI inflows. The baseline analysis (p.214) assumes that decisions about the location of FDI consider the risks to investment, the scope of the market, the availability of complementary inputs, such as land and labor skills, and the openness of alternative host countries. These considerations govern the choice of independent variables. Expropriation, repudiation of contracts, corruption, and other failures of the rule of law all threaten investment returns. In practice, these factors are highly intercorrelated among countries, so an index of the risk of expropriation (EXPROP) serves as a measure of risk in the crosscountry analysis. (EXPROP is scaled so that high values indicate a lower risk of expropriation.) The share of government consumption in GDP (GOVSHARE), often used as a proxy for the scope government intervention in markets, serves as a second measure of potential investment risks. Population (POP) and per capita GDP (GDPCAP) measure the scope of the market. A variable for the AREA of a country (measured in millions of square kilometers) is included to test whether FDI and land are complementary. Chapter 6 reviews the debate over the role of labor force skill in attracting FDI. The years of schooling (educational attainment) of people over 25 years of age (EDUC) serves as proxy for skill and provides an opportunity to test for whether FDI is complementary with high or lowskill labor. This analysis tests for the relationship between FDI shares and two measures of openness: exports and imports as a fraction of GDP (TRADE SHARE) and the updated SachsWarner indicator of open trade policies (OPEN POLICY).
The regression model of FDI shares is estimated on crosscountry data for the early 1990s and panel data for the 1980s and 1990s. The crosscountry sample consists of a maximum of about 80 countries at various stages of economic development. (Limitations in data availability for some labor conditions and labor regulations reduce the sample size for some regressions.) Given the annual volatility in FDI inflows, the dependent variable in the crosssection analysis is the natural logarithm of a country's share of world FDI averaged over 1991–96. To mitigate concerns about causality, the values of independent variables are for 1990, the year preceding the beginning of the period over which the dependent variable is measured.
The randomeffects panel estimates of the model take advantage of withincountry variance over time as well as crosscountry differences. (A Hausman specification test did not reject the hypothesis that the randomeffects and fixedeffects coefficients are the same.) The panel consists of data for a country's share of world FDI inflows averaged over 1980–85, 1986–1991, and 1991–96 with corresponding values of independent variables for 1980, 1985, and 1990.
The regression results strongly support the baseline model of why FDI inflow shares vary among countries (table A6.1). Regressions (1) and (2) report crosscountry, laborforceweighted OLSQ estimates of the FDI model, while regression (3) reports the randomeffects estimates. This discussion and the narrative in chapter 6 focus on these crosscountry results, but notable differences between the two sets of estimates are discussed below.
Table A6.1 Baseline FDI Share Regressions, All Countries
Crosscountry 
Random effects 


(1) 
(2) 
(3) 

EXPROP 
.805 (.233)* 
1.073 (.254)* 
.057 (.051) 
GOVSHARE 
−5.784 (2.367)** 
−7.870 (2.349)* 
−3.882 (1.475)* 
In POP 
.599 (.149)* 
.196 (203) 
.731 (.095)* 
In GDPCAP 
−0.096 (.320) 
−0.884 (.404)** 
1.215 (.194)* 
AREA 
.00025 (.00006)* 
.00026 (.00008)* 
.00020 (.00006)* 
EDUC 
.102 (071) 
.149 (.078)*** 
.002 (0.068) 
In TRADE SHARE 
1.360 (.379)* 
0.916 (0.224)* 

OPEN POLICY 
.727 (.326)** 

Constant 
−18.552 
−5.089 
−22.724 
R^{2} 
.906 
.881 
.850 
Root MSE 
.683 
.776 

Countries 
78 
77 
86 
Notes: Dependent variable is natural logarithm of a country's share of world FDI inflows. All observations weighted by labor force. Robust standard errors in parentheses. Variables defined in text and in Appendix A.
(*) pvalue < .01
(**) pvalue < .05
(***) pvalue < .10
In the crosscountry estimates, both of the investment risk factors—the risk of expropriation and the proxy for government intervention into the economy, significantly influence a country's share of FDI inflows in the predicted direction. (Recall that high values of EXPROP indicate a lower risk of expropriation.) More populous countries receive larger FDI shares. Both the land size and (more weakly) education level of a country appear to complement FDI share. Countries with a relatively large trade sector (regression 1) or open trade policies (regression 2) receive larger FDI shares, other factors equal, but the open trade policy specification is weaker statistically. The overall regression fit is good, with the model accounting for more than ninety percent of the variance in FDI inflow shares among 78 countries. (A similar pattern emerges from unreported, unweighted estimates, although the effect (p.216) of the risk factors is measured much less precisely.) Randomeffects estimates parallel the crosscountry findings, except that the risk of expropriation variable is no longer statistically significant.
As a check on the robustness of the findings, the crosscountry model was also estimated on a sample of nonOECD countries. A similar pattern of findings emerges except that the positive correlation between education and FDI share is measured more precisely for these lower income countries (table A6.2). Both measures of openness remain significant, but the regression with trade volumes continues to have superior statistical properties.
The various measures of working conditions and labor rights were added one at a time to the baseline specification to test for an influence of labor conditions on a country's share of world FDI inflows. The left columns in table A6.3 report the coefficients and robust standard errors from those tests. Each coefficient is from a separate (laborforceweighted) regression. With
Table A6.2 FDI Share Regression, NonOECD Countries
Crosscountry 
Random effects 


(1) 
(2) 
(3) 

EXPROP 
.679 (.236)** 
.739 (.257)* 
0.075 (0.060) 
GOVSHARE 
−8.289 (2.259)* 
−8.351 (3.338)** 
−4.138 (1.618)* 
In POP 
.575 (.175)* 
.372 (.168)** 
0.618 (.121)* 
In GDPOP 
−0.239 (.423) 
−0.556 (.438) 
1.124 (.246)* 
AREA 
.00021 (.00004)* 
.00024 (.00007)* 
0.00028 (.00010)* 
EDUC 
.185 (.093)** 
.241 (.102)** 
0.007 (0.087) 
In TRADE SHARE 
.819 (.227)* 
0.871 (.264)* 

OPEN POLICY 
.806 (.408)*** 

Constant 
−13.999 
−7.302 
−20.95 
R^{2} 
.94 
.94 
.99 
Root MSE 
.566 
.574 

Countries 
56 
56 
63 
Note: See notes to table A6.1.
(*) pvalue < .01
(**) pvalue < .05
(***) pvalue < .10
Table A6.3 Effects of Labor Conditions and Labor Regulations on FDI Shares 1991–1996
Labor condition 
Coefficient 
n 
Labor regulation 
Coefficient 
n 

Work hours 
−0.081 (.056) 
39 
Employment relations 
2.424 (1.271)*** 
58 
Fatal accidents 
−0.04 (.015)** 
31 
Firing cost 
2.283 (1.261)*** 
58 
Child labor 
−0.013 .029 
78 
Hours cost 
.958 (.560)*** 
58 
Civil liberties 
−.029 (.132) 
79 
Collective relations 
−.971 (1.915) 
58 
FACB 
−0.006 (0.058) 
66 
Civil rights 
−1.224 (1.633) 
58 
Forced labor 
−0.091 (.161) 
76 
Social security 
1.556 (.943) 
58 
Slavery 
.016 (.025) 
54 

Gender differential 
−1.621 (.983) 
51 
Note: See notes to table A6.1. Each coefficient (with robust standard error) is from a different regression.
n = number of countries
(*) pvalue < .01
(**) pvalue < .05
(***) pvalue < .10
One of the concerns expressed by globalization skeptics is that governments may compete for FDI by limiting their labor regulations. The Lex Mundi project has developed and published indices of the strength of national labor regulations in several areas (Botero et al. 2004). The next chapter discusses this study in more detail, but the indices are used here to test for significant associations between the strength of national labor regulations and (p.218) a country's ability to attract FDI. There are indices of the strength of national government regulations of the employment relationship (EMPLOYMENT), collective relations (COLLECTIVE), CIVIL RIGHTS, and SOCIAL SECURITY. There are also indices of the cost that employers incur if they dismiss a worker (FIRING COSTS) or lengthen employee work hours (HOURS COSTS). All indices range from zero (weak regulation) to one (strong regulation) and are available for 60 countries. These indices were added one at a time to the baseline FDI specification to test for an influence of national labor regulations on a country's share of world FDI inflows. The right columns in table A6.3 report the coefficients and robust standard errors from those tests in the weighted regressions.
These (laborforceweighted) estimates indicate reasonably precisely measured positive relationships between international differences in FDI shares and differences in national regulation of employment, social security, hours, and dismissals after controlling for the influence of the baseline variables. (The pvalues for the coefficients on the regulation measures range between .062 and .105.) No significant relationship emerges for regulation of collective relationships or civil rights in the crosscountry regressions. (Interestingly, unreported, unweighted regressions reveal no statistically significant relationship between FDI shares and measures of labor regulation.) Two conclusions may be drawn about the relationship between FDI and labor regulations. First, the relationship is strongest in countries with the largest labor forces. The largest countries in the sample are China (641.5 million workers), India (366.5), the United States (123.5), Japan (62.5), Pakistan (60.5), and Bangladesh (52.5). Second, where a significant relationship exists, stronger national regulations are associated with larger FDI shares, ceteris paribus. (At worst, the unweighted estimates reveal no statistically significant relationship.) These estimates reveal no tendency for countries with weak labor regulations to attract larger shares of world FDI inflows, ceteris paribus. In unreported analyses for nonOECD countries, FDI shares were not significantly related to any of the measures of national labor regulations.
Appendix to Chapter 7
Chapter 7 refers to regression analyses of (1) the relationship between openness and the scope of national labor regulations and (2) the effects of national labor regulations on labor conditions. This appendix provides details on these analyses. (Appendix A summarizes the definitions and data sources of all variables.)
Openness and National Labor Regulations
The discussion in chapter 7 notes that the strength of national labor regulations is not significantly associated with measures of openness. That (p.219) conclusion is based on an analysis of the following regression model estimated across i countries for each variety of national labor regulation:
The dependent variables are indices of the strength of national employment, collective relations, civil rights, and social security regulations and the costs of instituting longer work hours or of dismissing workers as of the late 1990s (Botero et al. 2004). (The availability of data for only one date ruled out a panel data analysis of the relationship between national labor regulations and openness.) As discussed in chapter 7, a country's “score” for each index is based on an analysis of relevant national statutes and normalized to fall between zero and 1, with higher scores denoting stronger protection of labor. The independent variables include the per capita GDP (GDPCAP), dummy variables indicating the origin of a country's legal system (LEGAL), a variable for the percentage of years between 1975 and 1995 in which the country's chief executive and largest party in congress have left or center political orientation (LEFT), and the two measures of openness introduced earlier in the book: TRADE SHARE (the share of exports and imports in GDP) and OPEN POLICY (the update of the multihurdle, SachsWarner open policy measure. Data for all variables but the openness measures are from Botero et al. (2004).
As discussed in the chapter, the direction of causality between labor regulations and openness is ambiguous, in principle. To address the direction of causality issue, we supplement OLSQ estimation with IV estimates, in which gravity variables instrument the trade variables. The gravity variables used are AREA, DISTANCE from the capital city of major trading partners, and a dummy variable for whether the country is an ISLAND. These variables are correlated with openness but should not influence the strength of labor regulations except through their effect on the openness measures. In the firststage regressions, the gravity variables explain 27 (38) percent of the variance in TRADE SHARE (OPEN POLICY). Effectively, we ask whether the variation in openness measures attributable to gravity variables influences international differences in labor regulation.
For each type of labor regulation, four regressions were estimated, testing the two measures of openness with two estimation methods. Table A7.1 reports the openness coefficients (and robust standard errors) from these regressions. The two dozen coefficients are easily summarized: Neither the OLSQ nor the IV estimation finds significant relationships between the measures of openness and most varieties of labor regulation. The main exception to this summary is social security regulation. In the OLSQ estimation there is a marginally significant relationship between trade volumes and social security regulation: social security regulation tends to be weaker in countries with relative large trade volumes. In the instrumental variables estimation, (p.220)
Table A7.1 Openness and Employment Regulations
Trade share 
Open policy 


Employment relations 

OLSQ 
.0001 (.0003) 
.0003 (.0628) 
IV 
.0003 (.0005) 
−.1255 (.1980) 
Collective relations 

OLSQ 
−.0001 (.0003) 
.0116 (.0429) 
IV 
.0003 (.0006) 
−.1207 (.1717) 
Civil rights 

OLSQ 
−.0011 (.0003)* 
−.0414 (.0445) 
IV 
−.0006 (.0006) 
−.2093 (.2293) 
Social security 

OLSQ 
−.0007 (.0004)*** 
−.0619 (.0657) 
IV 
−.0012 (.0007)*** 
−.3841 (.2087)*** 
Hours costs 

OLSQ 
.0005 (.0007) 
−.1277 (.1626) 
IV 
.0013 (.0011) 
−1.0024 (.4742)** 
Firing costs 

OLSQ 
.0004 (.0006) 
.1085 (.1033) 
IV 
.0004 (.0012) 
.3995 (.3445) 
Note: Robust standard errors in parentheses.
(*) pvalue<.01
(**) pvalue<.05
(***) pvalue<.10
Labor Regulations and Labor Conditions
Chapter 7 also discusses the effectiveness of national labor regulations in improving domestic labor conditions. Part of that discussion is based on the following crosscountry regression analysis, which tests for a link between the strength of national labor regulations and the measures of working conditions and labor rights. The underlying regression model is:
That is, the measures of labor conditions used throughout this book are regressed on real per capita income (controlling for a country's level of development) and the indices of national labor regulations developed by Botero et al. (2004).
We are interested in whether the implementation of labor regulations alters labor conditions, but we must recognize that some countries may simply legislate regulations that codify existing workplace practice. The latter scenario, which has lower political costs, will introduce bias in ordinary least squares estimates of the relationship between national labor regulations and labor conditions. Therefore, both ordinary least squares and instrumental variables techniques were used to estimate the relationship. Following Botero et al. (2004), each country's legal tradition instruments the level of national labor regulation.
Neither estimation method produced significant estimates for the national labor regulations variables. International differences in labor conditions continued to reflect the influence of differences in the level of development (real per capita GDP) but did not reflect the strength of national labor regulations. (p.222)