Policy Research Working Paper 8961 What Determines the Size of Public Employment? An Empirical Investigation Santiago Herrera Ercio Muñoz Macroeconomics, Trade and Investment Global Practice August 2019 Policy Research Working Paper 8961 Abstract This paper explores the determinants of public employment the Arab Republic of Egypt and the Islamic Republic of across the world and finds that it is negatively associated Iran. East Asian and Pacific countries’ public employment with country size (by population) and positively associ- is significantly below the predicted levels, particularly in ated with the income level. The findings show that a Hong Kong SAR, China; Japan; the Republic of Korea; country’s openness to trade is positively associated with and Mongolia. Countries in Europe and Central Asia show public employment in low- and middle-income coun- higher than predicted public employment, mostly in Roma- tries, but inversely related in high-income countries. The nia, Denmark, Sweden, Armenia, and Belorussia. Public estimated models are used to predict the expected public employment in Sub-Saharan Africa appears to be below the employment for a country given its income, population, predicted levels, with the notable exceptions of Botswana and openness to trade, and to compare the actual levels and South Africa. The deviations from predicted levels are with the predicted ones. In general, public employment positively correlated with the union density rate, which is in Latin American countries is below the predicted levels, negatively associated with private employment rates. Finally, except for Argentina, Brazil, Ecuador, Mexico, Suriname, the study finds no statistical association between public Trinidad and Tobago, and the República Bolivariana de and private employment, suggesting the absence of crowd- Venezuela. Public employment in the Middle East and ing-out in the employment levels. North Africa is above the predicted levels, particularly in This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at sherrera@worldbank.org and emunozsaavedra@gradcenter.cuny.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team What Determines the Size of Public Employment? An Empirical Investigation Santiago Herrera and Ercio Muñoz 1 Keywords: Public employment, government size, openness, country size, Wagner’s law JEL codes: E62, E24, F16, F41, J45 1 The authors thank Elena Ianchovichina, Guillermo Vuletin, Eduardo Olaberria, and Julio Velasco, for useful comments and suggestions to an earlier draft. Table of Contents 1. Introduction .......................................................................................................................................... 3 2. Data, Stylized facts and Methodology ................................................................................................. 3 i. Data................................................................................................................................................... 3 ii. Expenditure-based measures and public employment: the wage bill ........................................... 4 iii. The wage bill: price versus quantity ............................................................................................ 6 iv. Public employment across countries and over time ....................................................................... 7 v. Determinants of public employment ............................................................................................ 11 vi. Actual public employment compared with its predicted level..................................................... 19 vii. Does public employment crowd out private employment? ..................................................... 22 3. Concluding Remarks ........................................................................................................................... 25 References .................................................................................................................................................. 27 Appendix..................................................................................................................................................... 30 2 1. Introduction While extensive theoretical and empirical literature has studied the size and composition of government expenditure (see for example Shelton, 2007 for a review of the literature), less is known about the determinants of public employment. Two notable exceptions are Rodrik (2000) and Alesina, et al. (2000) who developed theoretical models to explain the behavior of public employment. The first one shows that public employment could play a welfare-enhancing social insurance role in an economy buffeted by external shocks, and presents evidence of a positive association between the exposure to external risk and the share of public employment across countries. The second one motivates public employment as a redistributive tool to circumvent opposition to explicit tax-transfer schemes, predicting a positive relationship between the size of public employment and inequality or fractionalization, and provide empirical evidence of cities in the United States. Public employment and its determinants is an important topic because it affects not only the size of expenditure, but it also affects its composition due to the rigidity of the wage bill (Vegh, et al., 2017). Recent empirical literature that uses expenditure-based measures of government size explores the relationship between trade openness and size, but its counterpart with employment measures has received less attention. The robustness in a panel data context over a period that goes beyond the 1990s has not been tested, in part because data on public employment is much more scarse than data on government expenditure. In the same vein, there is not much evidence about how alternative hypotheses to explain the size of the government can help to explain the size of public employment. This paper attempts to fill this gap in the empirical literature. 2. Data, Stylized Facts, and Methodology i. Data The paper uses three measures of public employment from the International Labor Organization Statistics (ILO Stats): total public-sector employment, general government employment, and central government employment. The coverage of each employment aggregate is the standard one. 2 Data on private employment and the total labor force come from the same source. 2 Total public-sector employment covers all employment of general government sector as defined in the System of National Accounts 1993, plus employment of publicly owned enterprises and companies, resident and operating at central, state (or regional) and local levels of government. It covers all persons employed directly by those institutions, without regard for the type of employment contract. The general government sector employment is the total employment of all resident institutional units operating at central, state (or regional) and local levels of 3 Altogether there are 145 countries 3 with at least one observation of public employment in any of the three aggregates, with 60 of the countries classified as high income. On average, countries have 13 years of data when considering the general government aggregate. The sample period starts in 1980, although the number of countries with data changes over time, making it an unbalanced panel. It is important to have in mind the level of agregation (public sector, general government, and central government) when comparing with wage bill data (IMF, 2016). Total government expenditure as a share of GDP and government expense in compensation to employees as share of GDP are obtained from the IMF’s World Economic Outlook and Government Finance Statistics (October 2018). ii. Expenditure-based measures and public employment: The wage bill Most of the literature that explores the size of government uses expenditure-based measures, either of the wage bill or other aggregates. The relationship between the size of the government, measured as expenditure as percentage of GDP, and public employment as percentage of the labor force can differ considerably (Figure 1). On the other hand, the wage bill, although highly correlated with total government expenditure (Figure 2), shows significant heterogeneity across countries, with some having a wage bill that is twice the size of others that have similar government expense levels. Figure 1: Public employment (% labor force) and general government expenditure (% GDP) (each dot is a country-year obs.) government; i.e. all government units, social security funds and non-market nonprofit institutions (NPIs) that are controlled and mainly financed by public authority (Hammouya, 1999). Finally, the central government aggregate is composed of departments or ministries, of autonomous agencies carrying out special functions, and of all NPIs which are controlled and mainly financed by public authority. Their fiscal, legislative and executive authority extends over the entire territory of the country. 3 The number is reduced to 122 when we consider employment as percentage of the labor force. 4 4.5 4.5 log of General Gov. expenditure log of General Gov. expenditure 4 4 3.5 3.5 3 3 2.5 2.5 2 2 -8 -6 -4 -2 0 -10 -8 -6 -4 -2 Log of General Gov. employment Log of Central Gov. employment 4.5 log of General Gov. expenditure 2.5 3 2 3.5 4 -6 -4 -2 0 Log of Public Sector employment Source: ILO Stats and IMF Figure 2: The wage bill and total government expense 20 30 CE (% GDP) 10 0 0 20 40 60 80 General Government Expense (% GDP) Source: ILO Stats and IMF 5 iii. The wage bill: Price versus quantity Changes in the public wage bill can be due to variations in wage and/or changes in the public employment level. We observe a positive correlation between the three measures of public employment (as share of the labor force) and size of the wage bill (as percentage of GDP), with the General Government and Public Sector aggregations showing the clearest association (Figure 3). Figure 3: Public employment (% labor force) and compensation to employees (% GDP) 30 30 20 20 CE (% GDP) CE (% GDP) 10 10 0 0 0 .05 .1 .15 .2 .25 0 .2 .4 .6 General Government Central Government 30 20 CE (% GDP) 10 0 0 .2 .4 .6 .8 Public Sector Source: ILO Stats and IMF A variance decomposition of the wage bill into the variance of its components (employment and wages) may be a useful starting point. 4 The decomposition is done using country-year level data (80 countries) 4 The variance of the wage bill is equal to the variance of employment, plus the variance of wages (proxied by total wage bill divided by public employment) and twice the covariance between both components: (ln( )) = (ln( )) + (ln( )) + 2 ∗ (ln( ), ln( )) Where corresponds to the wage bill in constant 2011 international dollars, is general government employment (in thousands), and is a proxy of wage level constructed by dividing by . 6 (Figure 4, left panel) or using 5-year averages to decrease the potential role of measurement error in employment levels (Figure 4, right panel). Both results suggest that, across countries, changes in employment explain a large share of the variance of the wage bill. This result is only suggestive and must be taken with a grain of salt, given the quality of the information and the assumptions to construct the average wage. Figure 4: Variance decomposition (annual frequency on the left and 5-year averages on the right) 4 4 3 3 2 2 1 1 0 0 1 1 Var(Wage) Var(Employment) 2*Covariance Var(Wage) Var(Employment) 2*Covariance Source: calculations based on ILO Stats and IMF iv. Public employment across countries and over time The different measures of public employment are highly correlated (Table 1), although when using a cross section, like year 2005 in the table, the correlation may be stronger. The data set is an unbalanced panel which increases its coverage in the mid-1990s reaching almost 60 countries in some years in the mid- 2000s. However, as Figure 5 shows, the coverage varies over time and across the three different measures. Table 1: Correlation of different measures of public employment as a share of the labor force a) Panel data Central Government General Government Public Sector Central Government 1.00 7 General Government 0.37 1.00 (0.000) Public Sector 0.20 0.71 1.00 (0.000) (0.000) b) Cross section in 2005 Central Government General Government Public Sector Central Government 1.00 General Government 0.41 1.00 (0.02) Public Sector 0.40 0.81 1.00 (0.05) (0.00) c) Cross section of country average values Central Government General Government Public Sector Central Government 1.00 General Government 0.50 1.00 (0.00) Public Sector 0.048 0.50 1.00 (0.74) (0.00) Source: calculations based on ILO Stats. Figure 5: Number of countries with data per year (max in 2006, 2005, and 2008) 8 60 40 20 0 1980 1990 2000 2010 2020 year General Gov. Public Sector Central Gov. Source: ILO Stats The size of public employment, as percentage of the labor force, varies widely across regions. Middle East and North Africa (MENA) shows the highest levels of public employment in the three measures (Figure 6). Europe and Central Asia (ECA) and Sub-Saharan Africa (SSA) also show high levels of public employment compared to other regions. South Asia (SA) and East Asia Pacific (EAP) show lower levels of public employment at all the levels. The evolution of absolute employment levels over time is volatile and noisy, so we plot the average over each decade to better visualize the trends by region (Figure 7). Most of them have a declining trend, except LAC and MENA, which show the opposite in central and general government levels since the 1990s. These regional trends are only suggestive because the country composition changes over time due to data availability. 5 Figure 6: Public employment (% of Labor Force), by region 5 Figures 13, 14, and 15 in the Appendix show the three time series of employment by country. 9 Central Gov. General Gov. Public Sector EAP 6.2 EAP 7.4 EAP 9.7 ECA 7.3 ECA 17.0 ECA 22.8 LAC 3.7 LAC 12.5 LAC 14.0 MENA 15.5 MENA 21.2 MENA 23.4 NA 2.2 NA 14.3 NA 17.9 SA SA 5.7 SA 12.3 SSA 7.4 SSA 11.2 SSA 13.6 0 5 10 15 0 5 10 15 20 0 5 10 15 20 25 Mean Mean Mean Notes: Simple averages for each region. The Appendix has weighted averages by the size of the labor force. The regions in the case of general government employment are constructed in the following way 6: EAP includes Fiji; Hong Kong SAR, China; Indonesia; Japan; Republic of Korea; Macau; New Zealand; Philippines; Singapore; Thailand; and Timor-Leste. ECA includes Albania, Armenia, Austria, Azerbaijan, Belarus, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Georgia, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Macedonia, Moldova, Netherlands, Norway, Poland, Portugal, Russian Federation, San Marino, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Turkey, Ukraine, and United Kingdom. LAC includes Argentina, Aruba, Belize, Bolivia, Brazil, Costa Rica, Cuba, Dominican Republic, Guatemala, Mexico, Nicaragua, Panama, Paraguay, Trinidad and Tobago, and Uruguay. MENA includes Arab Republico f Egypt, Israel, Malta, Oman, Qatar, and United Arab Emirates. NA includes Canada and the United States. SA includes Afghanistan and Sri Lanka. SSA includes Botswana, Cabo Verde, Ethiopia, Guinea, Madagascar, Mauritius, Senegal, Seychelles, South Africa, Tanzania, and Zimbabwe. Source: ILO Stats Figure 7: Employment (%LF) over time, by region 6 Our total sample consists of 88 countries, but central government employment data are reported for only for 56 countries, while public sector employment data are reported for 113 countries. 10 Central Government General Government Public Sector Private Sector 100 35 35 35 30 30 30 25 25 25 80 20 20 20 15 15 15 60 10 10 10 5 5 5 40 0 0 0 1980 1990 2000 2010 1980 1990 2000 2010 1980 1990 2000 2010 1980 1990 2000 2010 year year year year EAP ECA EAP ECA EAP ECA EAP ECA LAC MENA LAC MENA LAC MENA LAC MENA NA SA NA SA NA SA NA SA SSA SSA SSA SSA Source: calculations based on ILO Stats v. Determinants of public employment This section explores the potential factors that explain these differences across regions and over time, and has two parts. The first one focuses on Rodrik’s (2000) hypothesis of public employment as a tool to mitigate the country exposure to undiversifiable external risk. The second one expands the list of potential determinants of public employment. We initially replicate Rodrik’s baseline econometric specification, but with more recent data and additional countries. The summary statistics and cross correlations of the main variables (Tables 2 and 3) are presented to facilitate comparisons with the original work. The central variable in Rodrik’s model is exposure to external risk, which is calculated as the product of the volume of trade with the unanticipated component of variability in the external terms of trade, which he argues, it is the theoretically appropriate measure of external risk, as it yields the unpredictable variation in the streams of incomes associated with foreign trade. Hence, let x, m, and y stand for the volumes of exports, imports, and GDP, respectively, and T represent the terms of trade. The measure of exposure to external risk corresponds to: 11 + = � � . . (()) Table 2: Descriptive Statistics N mean sd min max Central gov. employment (% LF) 6.99 4.39 0.01 22.33 General gov. employment (% LF) 15.49 6.42 1.58 31.78 Public sector employment (% LF) 19.93 7.60 2.33 54.09 GDP per capita, PPP (constant 2011 29340.27 16493.10 599.95 98184.64 international $) Urbanization rate (%) 72.34 15.12 14.55 100.00 Trade openness (% GDP) 101.00 72.13 15.16 455.37 External risk 3.50 2.81 0.43 27.59 Population (Millions of persons) 35.24 59.68 0.03 325.44 Observations 948 Table 3: Matrix of Cross Correlations Unbalanced panel data: 948 observations lcg lgg lps lGDPPC lurb open R Log of central gov. employment (% LF) 1 Log of general gov. employment (% LF) 0.639 1 0 Log of public sector employment (% LF) 0.361 0.803 1 0 0 Log of GDP 0.135 0.253 0.094 1 0.001 0 0.011 Log of urbanization rate -0.008 0.147 -0.096 0.676 1 0.85 0 0.01 0 Openness (% GDP) 0.374 0 0.005 0.222 0.183 1 0 0.996 0.884 0 0 External risk 0.182 -0.053 0.053 -0.322 -0.28 0.263 1 0 0.104 0.157 0 0 0 Note: p-value reported below the correlation coefficient. The baseline regression model (omitting the subscripts i or it) has government employment (as percentage of the labor force) as dependent variable, and the per capita income, urbanization rate and exposure to risk as explanatory variables using a cross section of data: log() = 0 + 1 log � � + 2 log() + 3 R + 12 Our cross-section results are not as robust as Rodrik’s, as we find that trade openness and external risk are positively associated with government employment in only two of the six specifications, while the GDP per capita is the only significant variable, in addition to regional dummies (Table 4). Table 4: Cross section for 2005: Rodrik’s framework (1) (2) (3) (4) (5) (6) VARIABLES Central General Public Central General Public Gov. Gov. sector Gov. Gov. sector log GDP per capita 0.268 0.260** 0.298*** 0.622** 0.258** 0.370*** (0.220) (0.101) (0.095) (0.277) (0.109) (0.101) log urbanization -1.225 -0.150 -0.118 -1.191 -0.066 -0.135 (0.878) (0.338) (0.249) (0.997) (0.346) (0.242) Trade openness 0.006*** -0.000 0.001 (0.002) (0.001) (0.001) External risk 0.078 0.009 0.037* (0.054) (0.021) (0.019) East Asia & Pacific 0.831 -0.656*** -0.764*** 0.656 -0.812*** -0.629*** (0.646) (0.201) (0.190) (0.728) (0.240) (0.196) Sub-Saharan Africa -0.701 -0.673** -0.569* -0.465 -0.256 0.133 (0.686) (0.283) (0.328) (0.812) (0.310) (0.421) Constant -0.906 -3.846*** -4.236*** -4.358 -4.217** -4.958*** (3.942) (1.159) (0.762) (4.832) (1.569) (0.842) Observations 31 52 57 30 48 52 R-squared 0.407 0.423 0.430 0.260 0.311 0.389 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 When the regression is estimated with pooled country data over time, trade openness and external risk are not robust and are only statistically significant for central government employment (Table 5). Table 5: Pooled OLS: Rodrik’s framework (1) (2) (3) (4) (5) (6) 13 VARIABLES Central Gov. General Gov. Public sector Central Gov. General Gov. Public sector log GDP per capita 0.302 0.204** 0.181** 0.551** 0.228** 0.182** (0.225) (0.102) (0.082) (0.241) (0.107) (0.080) log urbanization -0.786 0.133 -0.195 -0.780 0.145 -0.253 (0.709) (0.229) (0.209) (0.727) (0.238) (0.228) Trade openness 0.319** 0.007 0.016 (0.125) (0.051) (0.055) External risk 6.878** 1.065 -0.461 (2.647) (1.283) (2.171) EAP 0.781 -0.614*** -0.700*** 0.818 -0.630*** -0.669*** (0.567) (0.137) (0.171) (0.605) (0.135) (0.153) SSA 0.731 0.002 -0.351 1.178* 0.044 -0.322 (0.615) (0.259) (0.278) (0.612) (0.252) (0.292) LAC 0.701* 0.108 -0.249** 0.875** 0.103 -0.281** (0.375) (0.174) (0.118) (0.362) (0.169) (0.131) MENA 1.899*** 0.566*** 0.401** 2.264*** 0.572*** 0.447** (0.257) (0.181) (0.162) (0.224) (0.172) (0.170) SA -0.289 -0.457 -0.189 -0.519* (0.331) (0.296) (0.376) (0.297) ECA 0.911*** 0.266** 0.279*** 1.028*** 0.263** 0.289*** (0.189) (0.121) (0.088) (0.153) (0.109) (0.082) Constant -3.674 -4.615*** -2.505*** -6.238* -4.919*** -2.281*** (2.972) (0.966) (0.692) (3.288) (1.086) (0.723) Observations 634 953 949 623 926 908 R-squared 0.289 0.327 0.513 0.285 0.315 0.519 Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 Finally, estimating the same regression but using a panel with country fixed effects, shows that external risk has a statistically significant positive sign in two of the three specifications (for central and general government), though the sign switches for the public sector aggregation; trade openness is not statistically significant (Table 6). Hence, controlling for unobserved time-invariant heterogeneity seems to be relevant to confirm the positive relationship between openness and public employment. 14 Moving from Rodrik’s framework to a more encompassing one, and following Shelton (2007), we include additional potential explanotory variables, including those proposed by Alesina et al. (2000) ( Table 7 ). Table 6: Determinants of the government expenditure size Variable Expected sign Main references: Openness + Rodrik, 1998; Rodrik, 2000; Benarroch & Pandey, 2008; Garen & Trask, 2005; Ram, 2009; Vianna & Mollick, 2018 Country size - Alesina & Wacziag, 1998; Jetter & Parmeter, 2015 Fragmentation +- Easterly & Levine, 1997; Alesina, et al., 1999; Alesina, et al., 2003; Alesina, et al., 2001; Alesina & Wacziag, 1998; Alesina, et al., 2000 Income + Henrekson, 1993; Oxley, 1994; Ram, 1987; Stein, et al., 1998; Easterly & Rebelo, 1993 Income inequality + Meltzer & Richard, 1981; Meltzer & Richard, 1983; Alesina, et al., 2000 Political rights + Lott & Kenny, 1999; Husted & Kenny, 1997; Mulligan, et al., 2004; Mulligan, et al., 2002; Easterly & Rebelo, 1993 Institutions of +- Milesi-Ferretti, et al., 2002; Persson, et al., 1998; government Austen-Smith, 2000; Persson & Tabellini, 1999 Table 7: Panel data regression: Rodrik’s framework (1) (2) (3) (4) (5) (6) VARIABLES Central Gov. General Gov. Public Sector Central Gov. General Gov. Public Sector 15 log GDP per capita 0.140 0.136** -0.097*** 0.217 0.179** -0.104*** (0.151) (0.069) (0.035) (0.157) (0.072) (0.036) log urbanization 0.800* 0.604*** 0.739*** 1.289*** 0.608*** 0.734*** (0.454) (0.206) (0.126) (0.471) (0.225) (0.132) Trade openness 0.001 -0.000 -0.000 (0.001) (0.001) (0.000) External risk 0.055** 0.023* -0.011* (0.026) (0.012) (0.006) EAP 0.319 -0.652* -0.991*** 0.115 -0.668* -1.013*** (0.617) (0.341) (0.353) (0.616) (0.346) (0.358) SSA 0.402 -0.209 -0.732* 0.938 -0.171 -0.703* (0.717) (0.378) (0.376) (0.742) (0.398) (0.387) LAC 0.718 -0.083 -0.633* 0.624 -0.102 -0.668* (0.630) (0.345) (0.345) (0.614) (0.352) (0.348) MENA 1.079 0.365 -0.082 0.911 0.404 -0.074 (0.705) (0.373) (0.377) (0.768) (0.407) (0.402) SA -0.021 -0.078 0.158 -0.081 (0.660) (0.426) (0.680) (0.429) ECA 0.642 0.083 0.084 0.521 0.062 0.087 (0.553) (0.319) (0.336) (0.539) (0.325) (0.337) log of Population -0.168** -0.071** -0.076*** -0.248*** -0.069** -0.092*** (0.067) (0.032) (0.025) (0.072) (0.035) (0.028) Constant -5.536** -4.631*** -2.365*** -6.976*** -5.142*** -1.978** (2.310) (1.068) (0.872) (2.444) (1.225) (0.924) Observations 632 948 944 623 925 907 Number of countries 55 82 100 49 73 90 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Using 5-year non-overlapping averages, we estimate a panel with random effects model 7: log( ) = 0 + 1 log( ) + 2 log( ) + 3 +4 +5 + 7 As in Shelton (2007), we use a random effects model over a fixed effects (FE) one, considering the trade-off between measurement error bias and omitted variable bias, and that an FE model would exacerbate the first one. 16 where corresponds to the measures of public employment over labor force, log of population, stands for GDP per capita, for openness (imports plus exports as share of GDP), for ethnic fractionalization and for other controls. Table 8 reports a set of descriptive statistics of the main variables used in the analysis, where each observation corresponds to a 5-year average. Table 8: Descriptive statistics N mean sd min max Central gov. employment (% LF) 6.97 4.55 0.22 21.82 General gov. employment (% LF) 14.85 6.61 1.58 31.25 Public sector employment (% LF) 19.72 8.03 2.33 54.09 Population (Millions of persons) 33.32 56.38 0.03 321.02 GDP per capita, PPP (constant 2011 25993.43 17031.91 601.82 91798.26 international $) Openness (% GDP) 99.50 73.30 15.16 422.39 Ethnic fractionalization 0.32 0.22 0.00 0.88 High income 0.66 0.47 0.00 1.00 Age dependency ratio, young 34.32 16.03 15.09 94.00 Age dependency ratio, old 18.02 7.75 0.90 42.47 Gini 36.27 9.48 17.50 69.70 Observations 286 The extended model shows a negative and statistically significant association of public employment (as a % of labor force) with country size by population and a positive association with income (Table 9). The size of the income coefficient is similar to Rodrik’s’ and the impact of a change in income is similar to recent IDB estimates according to which a 25% increase in GDP per capita in LAC is associated with a 1 percentage point increase in public employment. 8 However, openness is negatively associated with public employment (see column 1 in Table 9), contradicting Rodrik’s model prediction. 9 Table 9: Determinants of general government employment; extended model. Dependent variable: Log of general government employment as share of labor force. Variables (1) (2) (3) (4) (5) (6) 8 The impact of a 25 percent increase in GDP per capita evaluated at the average general government employment level (14.85%) is 0.25x0.223x0.1485=0.008, or 0.8%. 9 This result also holds when external risk is defined as in Rodrick’s paper (see Table 17 in the appendix). 17 Log of Population -0.147*** -0.130*** -0.137*** -0.138*** -0.143*** -0.138*** (0.029) (0.029) (0.032) (0.035) (0.035) (0.036) Log of GDP per capita 0.239*** 0.223** 0.222 0.275** 0.226 0.250 (0.076) (0.093) (0.139) (0.108) (0.189) (0.195) Openness -0.190*** 0.217 0.195 0.179 -0.077 0.155 (0.072) (0.224) (0.220) (0.275) (0.293) (0.251) Ethnic fractionalization -0.739*** -0.607** -0.617** -0.465 -0.462 (0.269) (0.267) (0.280) (0.330) (0.321) High income 0.490* 0.472* 0.433 0.333 0.430 (0.285) (0.280) (0.367) (0.379) (0.349) High income * Openness -0.448** -0.413* -0.463* -0.233 -0.437* (0.211) (0.213) (0.262) (0.280) (0.249) Age dependency ratio, young -0.001 -0.007 -0.002 (0.005) (0.007) (0.008) Age dependency ratio, old -0.003 -0.003 -0.001 (0.005) (0.007) (0.007) Gini 0.002 0.003 0.003 (0.008) (0.010) (0.011) Constant 0.904 0.587 0.718 0.058 0.819 0.376 (0.743) (0.795) (1.439) (0.801) (1.616) (1.751) Observations 286 286 284 223 224 223 Number of countries 80 80 79 72 73 72 Country FE No No No No No No R-2 0.347 0.355 0.358 0.352 0.372 0.351 Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 This result is driven by the assumption of homogeneity of the coefficients across all countries. When that assumption is relaxed and heterogeneity across groups of countries (by income group) is allowed, results change. When a dummy variable for high-income countries is included and interacted with openness (Table 9, Column 2), a robust negative coefficient associated with the interaction term is present in most specifications. Hence, openness is negatively associated with public employment in high-income countries, but positively associated in the rest. 10 In the case of ethnic fractionalization, we find a negative association, implying that more fragmented countries have lower public sector employment, while the age dependency ratio and income inequality are not statistically significant determinants of public 10 Openness appears imprecisely estimated, but when the dependent variable is changed to public employment over population instead of labor force (to increase the sample size by 15%), results are similar, with the openness coefficient being positive and statistically significant (Table 13 in the Appendix). 18 employment. The hypothesis of public employment being used as a redistribution tool is not supported by these findings, while the role of public employment as an insurance to mitigate exposure to undiversifiable external risk finds better support, though differently across groups of countries. It is possible that the use of public employment varies along the business cycle, expanding during booms, but being rigid during recessions. We tested the hypothesis of asymmetric or ratchet effects in the response of public employment to changes in GDP per capita, with a dummy variable equal to 1 when GDP per capita growth is positive, included by itself and interacted with GDP per capita. We found no significative difference in the response of public employment to changes in GDP along the cycle and hence reject the ratchet effects hypothesis. 11 vi. Actual public employment compared with its predicted level Based on the preferred specification in terms of goodness of fit and statistical significance (column 2 in Table 9), we predict the public employment levels to compare with actual public employment. 12 The comparisons, grouped by geographic region and averaging over all the years available, 13 show clear differences across regions (Figure 8). EAP shows public employment lower than the predicted by the model, while MENA shows the opposite. In Latin America and the Caribbean, most countries have lower public employment than the predicted levels, except for Argentina, Brazil, Mexico, Suriname, Trinidad and Tobago, and the República Bolivariana de Venezuela, supporting the IDB’s conclusion that the wage bill in LAC is driven by large public sector wage premiums (IDB, 2018). ECA has higher employment than predicted by the model, while AFR has mixed results, with Botswana and South Africa showing significantly higher public employment levels. We examine the potential role of political ideology driving these deviations, under the hypothesis that left-leaning governments would have larger public employment levels. Hence, we correlate the deviations with a variable that captures the political ideology of the government in office (left, center of right) obtained from the Database of Political Institutions 2017 from the Inter-American Development Bank. The deviations from the predicted level do not appear to be driven by political ideology, as suggested by 11 The model is estimated at an annual frequency (not five-year averages) to examine the public employment along the business cycle. Results are not presented to save space but are available from the authors upon request. 12 To increase the number of countries, in cases where general government employment data are not available, we substitute it with public sector employment. 13 Figure 16 shows the comparison using only the last available 5-year window. 19 the null correlation (Figure 17 in the Appendix). 14 In contrast, we find that these deviations are positively correlated with union density rate, which is obtained from the Database on Institutional Characteristics of Trade Unions, Wage Setting, State Intervention and Social Pacts and is available for 51 countries (see Figure 9). Figure 8: Actual minus predicted (general government employment as share of labor force), average over the available years by country a) East Asia Pacific b) Latin America and the Caribbean THA VEN URY SGP TTO SUR PHL SLV PRY NZL PER PAN NIC MYS MEX HND MNG GTM ECU KOR DOM DMA JPN CRI COL IDN CHL BRB HKG BRA BOL FJI BLZ BHS AUS ARG -15 -10 -5 0 5 -10 0 10 20 30 40 c) Middle East and North Africa d) North America OMN USA MLT ISR IRN EGY BHR ARE CAN -5 0 5 10 15 0 2 4 6 8 e) South Asia f) Sub-Saharan Africa 14 Using the political ideology variable in the model would imply losing a significant part of the sample due to data availability. Hence, we do the correlation analysis between the deviations estimated with the entire sample and the available data on political ideology. 20 LKA ZWE ZAF UGA TZA SYC IND SEN MUS MDG KEN BGD GIN ETH CPV BWA AFG BFA -10 -5 0 5 -10 -5 0 5 10 g) Europe and Central Asia 1 h) Europe and Central Asia 2 UKR ITA IRL TUR HUN SWE HRV SVN GRC SVK GEO GBR SMR FRA RUS FIN ROU EST ESP PRT DNK POL DEU NOR CZE NLD CYP CHE MKD BLR MDA BGR LVA BEL AZE LUX AUT LTU ARM KGZ ALB -10 0 10 20 -10 0 10 20 Notes: The prediction model is log(E/LF) = -1.38 - .11*log(Population) + .10*log(GDPPC) + .11*Openness - .56*Ethnic + .47*HighIncome - .27*HighIncome*Openness Figure 9: Union density rate and difference between public employment and predicted level 40 Public employment minus predicted 0 10 -10 20 30 0 20 40 60 80 100 Union density 21 vii. Does public employment crowd- out private employment? So far, we have analyzed the determinants of public employment without any reference to the private sector employment. However, it is possible that both variables are related and we have omitted the relationship from the analysis. There is mixed evidence of crowding out in the literature. On the one hand, Behar & Mok (2013) find that public employment fully crowds out private employment using a cross section of developing and advanced countries. Similarly, Malley & Moutos (1996) argue that increases in government employment can have a negative effect on private employment and support this hypothesis with Swedish data. On the other hand, Faggio & Overman (2014) use data on local labor markets in England to show that the impact of public sector employment has no identifiable effect on total private sector employment. First, we examine some stylized facts of our data set. The scatter plot between private and public employment shows a negative correlation between public sector and private sector employment (Figure 10).15 The plot of public employment and unemployment rates shows no clear relationship in the data (Figure 11). The measure of public sector employment seems to be more closely related with private employment or the unemployment rate, while the other two measures show more dispersion and a flatter relationship. Figure 10: Public and private employment 1 1 .8 .8 Private Sector Private Sector .6 .6 .4 .4 .2 .2 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Central Gov. General Gov. 15 Each observation is a country in a specific year. 22 1 .8 Private Sector .4 .2 0 .6 0 .2 .4 .6 .8 1 Public Sector Source: ILO Stats Figure 11: Public employment and unemployment rate .4 .6 .5 .3 Unemployment rate Unemployment rate .4 .2 .2 .3 .1 .1 0 0 0 .1 .2 .3 .4 0 .2 .4 .6 Central Gov. General Gov. .8 .7 .6 Unemployment rate .3 .4 .2 .1 0 .5 0 .1 .2 .3 .4 .5 .6 .7 .8 Public Sector Source: ILO Stats 23 To examine more carefully the relationship between private and public employment, we follow Behar and Mol (2013) and estimate the following two regressions: = + 1 + + + = + 1 + + + where corresponds to private employment over labor force, public employment over labor force, unemployment rate, and Z a vector of controls that includes log of GDP per capita, urbanization rate, trade openness, and union density rate. The data come from sources already described and we use 5- year non-overlapped averages. We do not find evidence of a crowding-out of private employment or an effect on the unemployment rate (see Tables 10 and 11). This result holds using the three measures of public employment. In addition, we find that the union density rate decreases the size of private employment (see column 2 in Table 11) and the unemployment rate (see column 2 and 4 in Table 10 and 11). This is consistent with our finding that union density increases the size of the public sector, while the effect on the unemployment rate suggests a stronger positive effect in public employment than the negative impact on private employment, however further research is needed to make that statement categorically. Table 10: Dependent Variable: Unemployment rate (1) (2) (3) (4) (5) (6) General gov. (% LF) -0.158 -0.172 (0.112) (0.150) Central gov. (% LF) 0.021 0.281 (0.175) (0.252) Public sector (% LF) -0.019 -0.050 (0.056) (0.105) Urbanization rate 11.022 -5.910 11.579 -7.137 8.058 -6.857 (8.240) (11.021) (10.738) (13.361) (8.623) (13.844) Openness -0.001 -0.001 -0.002 0.006 0.002 -0.006 (0.009) (0.015) (0.014) (0.018) (0.009) (0.015) log of GDP per capita -10.248*** -14.034*** -13.275*** -17.466*** -10.991*** -15.544*** (1.234) (2.180) (1.963) (2.641) (1.311) (2.607) Union density rate -0.073** -0.100* -0.077 (0.031) (0.051) (0.050) Constant 103.412*** 158.969*** 131.728*** 191.178*** 109.165*** 171.264*** (14.398) (24.664) (21.897) (29.856) (15.151) (29.993) Observations 270 174 183 141 287 152 24 R-squared 0.314 0.319 0.348 0.390 0.315 0.337 Number of countries 70 43 51 38 84 41 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 11: Dependent Variable: Private employment (1) (2) (3) (4) (5) (6) General gov. (% LF) -0.106 -0.196 (0.598) (0.627) Central gov. (% LF) -0.547 2.181 (0.815) (2.178) Public sector (% LF) -0.150 -0.065 (0.472) (0.660) Urbanization rate -10.657 102.695 80.421 3.972 6.315 118.322 (70.828) (90.142) (110.580) (85.469) (64.484) (99.628) Openness -0.047 -0.006 -0.082 0.006 -0.029 0.016 (0.041) (0.058) (0.050) (0.049) (0.039) (0.053) log of GDP per capita 24.658*** 32.782*** 37.814*** 27.785*** 18.886*** 27.231*** (7.318) (8.574) (9.307) (8.809) (6.571) (8.950) Union density rate -0.640** -0.201 -0.228 (0.261) (0.348) (0.201) Constant -153.845 -302.752** -351.076** -221.599* -104.241 -269.917* (93.478) (121.838) (140.327) (120.366) (82.424) (132.392) Observations 116 61 87 59 131 62 R-squared 0.255 0.600 0.374 0.567 0.203 0.450 Number of countries 55 31 39 29 64 30 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 3. Concluding Remarks We examined the determinants of public employment based on the ILO data set, which is the most complete source for analysis of public employment across countries, We found a negative association between the size of public employment and country size by population, and a positive association with 25 income. Openness was found to be positively associated in low- and middle-income countries, but negatively associated in high-income countries. Based on the estimated models, we predicted the public employment levels for each country, given its population size, income level, and level of openness to trade. The deviations between the actual and predicted levels show clear regional differences. EAP has, in general, lower public employment than that predicted by the model, while MENA has the opposite. In Latin America and the Caribbean region, Argentina, Suriname, Trinidad and Tobago, and the República Bolivariana de Venezuela show the largest positive deviation with respect to the prediction. These differences do not appear to be driven by political ideology of the government holding office, but they are positively correlated with union density rates. 26 References Alesina, A., Baqir, R. & Easterly, W., 1999. Public goods and ethnic divisions. Quarterly Journal of Economics, 114(4), pp. 1214-1284. Alesina, A., Baqir, R. & Easterly, W., 2000. Redistributive Public Employment. Journal of Urban Economics, Volume 48, pp. 219-241. Alesina, A. et al., 2003. Fractionalization. Journal of Economic Growth, Volume 8, pp. 155-194. Alesina, A., Glaser, E. & Sacerdote, B., 2001. Why doesn't the US have a european style welfare state?. Harvard Inst. Econ. Research Disc. Paper, Volume 1933. Alesina, A. & Wacziag, R., 1998. Openness, country size and government. Journal of Public Economics , 69(3), pp. 305-321. Austen-Smith, D., 2000. Redistributing income under proportional representation. The Journal of Political Economy, 108(6), pp. 1235-1269. Behar, A. & Mok, J., 2013. Does public-sector employment fully crowd out private-sector employment?. IMF Working Paper, 13(146). Benarroch, M. & Pandey, M., 2008. Trade openness and government size. Economics Letters, Volume 101, pp. 157-159. Benarroch, M. & Pandey, M., 2012. The relationship between trade openness and government size: Does disaggregating government expenditure matter?. Journal of Macroeconomics, Volume 34, pp. 239-252. Easterly, W. & Levine, R., 1997. Africa's growth tragedy: policies and ethnic divisions. Quarterly Journal of Economics, 111(4), pp. 1203-1250. Easterly, W. & Rebelo, S., 1993. Fiscal policy and economic growth: an empirical investigation. Journal of Monetary Economics, Volume 32, pp. 417-458. Faggio, G. & Overman, H., 2014. The effect of public sector employment on local labour markets. Journal of Urban Economics, Volume 79, pp. 91-107. Garen, J. & Trask, K., 2005. Do more open economies have bigger governments? Another look. Journal of Development Economics, Volume 77, pp. 533-551. Hammouya, M., 1999. Statistics on public sector employment: Methodology, structures and trends. Henrekson, M., 1993. Wagner's law - a spurious relationship. Public Finance, 2(406-415), p. 48. Husted, T. & Kenny, L., 1997. The effect of the expansion of the voting franchise of the size of government. The Journal of Political Economy, 105(1), pp. 54-82. IMF, 2016. Managing government compensation and employment - Institutions, policies, and reform challenges. IMF Policy Papers. Inter-American Development Bank, 2018. The (in)efficiency of public spending. Alejandro Izquierdo, Carola Pessino, and Guillermo Vuletin ed. Washington: Inter-American Development Bank. 27 Jetter, M. & Parmeter, C., 2015. Trade openness and bigger governments: The role of country size revisited. European Journal of Political Economy, Volume 37, pp. 49-63. Lott, J. & Kenny, L., 1999. Did women's suffrage change the size and scope of government?. The Journal of Political Economy, 107(6), pp. 1163-1198. Malley, J. & Moutos, T., 1996. Does government employment "crowrd-out" private employment? Evidence from Sweden. The Scandinavian Journal of Economics, 98(2), pp. 289-302. Meltzer, A. & Richard, S., 1981. A rational theory of the size of government. The Journal of Political Economy, 89(5), pp. 914-927. Meltzer, A. & Richard, S., 1983. Tests of a rational theory of the size of government. Public Choice, 41(3), pp. 403-418. Milesi-Ferretti, G.-M., Perotti, R. & Rostagno, M., 2002. Electoral systems and public spending. Quarterly Journal of Economics, 117(2), pp. 609-657. Mulligan, C., Gil, R. & Sala-i-Martin, X., 2002. Social security and democracy. NBER working paper, Issue 8958. Mulligan, C., Gil, R. & Sala-i-Martin, X., 2004. Do democracies have different public policies than non- democracies. Journal of Economic Perspectives, 18(1), pp. 51-74. Oxley, L., 1994. Cointegration, causality, and Wagner's law: a test for Britain 1870-1913. Scottish Journal of Political Economy, 41(3), pp. 286-298. Persson, T., Roland, G. & Tabelini, G., 1998. Towards micropolitical foundations of public finance. European Economic Review, Volume 42, pp. 685-694. Persson, T. & Tabellini, G., 1999. The size and scope of government: comparative government with rational politicians. European Economic Review, Volume 43, pp. 699-735. Ram, R., 1987. Wagner's hypothesis in time-series and cross-section perspectives: evidence from 'real' data for 115 countries. Review of Economics and Statistics, pp. 194-204. Ram, R., 2009. Openness, country size, and government size: Additional evidence from a large cross- country panel. Journal of Public Economics, Volume 93, pp. 213-218. Rodrik, D., 1998. Why do more open economies have bigger governments?. The Journal of Political Economy, pp. 997-1032. Rodrik, D., 2000. What drives public employment in developing countries?. Review of Development Economics, pp. 229-243. Shelton, C., 2007. The size and composition of government expenditure. Journal of Public Economics, Volume 91, pp. 2230-2260. Stein, E., Talvi, E. & Grisanti, A., 1998. Institutional arrangements and fiscal performance: the latin american experience. Working paper. Office of the Chief Economist. Inter-American Development Bank, Issue 367. 28 Vegh, C., Lederman, A. & Bennett, F., 2017. Leaning against the wind: Fiscal policy in Latin America and the Caribbean in a historical perspective, Washington DC: World Bank Group. Vianna, A. & Mollick, A., 2018. Government size and openness: Evidence from the commodity boom in Latin America. Resources Policy, Volume 59, pp. 318-328. 29 Appendix Table 12: Descriptive statistics Regio stats Central General Gov. Public Sector Private Unemploym n Gov. employment employment employment ent employm ent EAP mea 0.062 0.074 0.097 0.859 0.047 n sd 0.041 0.020 0.039 0.119 0.025 min 0.005 0.030 0.004 0.502 0.007 max 0.118 0.136 0.230 1.022 0.118 ECA mea 0.073 0.170 0.228 0.655 0.092 n sd 0.037 0.060 0.071 0.128 0.060 min 0.000 0.000 0.120 0.112 0.002 max 0.158 0.318 0.751 0.941 0.384 LAC mea 0.037 0.125 0.140 0.861 0.092 n sd 0.029 0.056 0.092 0.102 0.045 min 0.012 0.048 0.025 0.146 0.012 max 0.139 0.283 0.831 0.979 0.276 MENA mea 0.155 0.212 0.234 0.666 0.095 n sd 0.045 0.045 0.063 0.102 0.035 min 0.085 0.146 0.079 0.307 0.034 max 0.223 0.306 0.361 0.902 0.201 NA mea 0.022 0.143 0.179 0.785 0.073 n sd 0.003 0.034 0.019 0.056 0.019 min 0.019 0.069 0.123 0.687 0.040 max 0.031 0.189 0.218 0.902 0.120 SA mea 0.057 0.123 0.891 0.056 n 30 sd 0.022 0.079 0.078 0.024 min 0.041 0.043 0.594 0.023 max 0.072 0.323 0.963 0.159 SSA mea 0.074 0.112 0.136 0.806 0.126 n sd 0.062 0.063 0.087 0.168 0.086 min 0.002 0.016 0.010 0.105 0.017 max 0.169 0.255 0.370 1.081 0.278 Total mea 0.070 0.153 0.189 0.764 0.085 n sd 0.044 0.065 0.088 0.152 0.054 min 0.000 0.000 0.004 0.105 0.002 max 0.223 0.318 0.831 1.081 0.384 Figure 12: Public employment (% labor force) by region (weighted by labor force size) Central Gov. General Gov. Public Sector EAP 1.7 EAP 6.4 EAP 9.3 ECA 5.6 ECA 15.6 ECA 21.0 LAC 2.0 LAC 11.1 LAC 11.3 MENA 13.5 MENA 21.4 MENA 24.1 NA 2.1 NA 15.1 NA 17.1 SA SA 5.5 SA 5.6 SSA 2.9 SSA 6.2 SSA 5.2 0 5 10 15 0 5 10 15 20 0 5 10 15 20 25 Mean Mean Mean Source: ILO Stats Figure 13: General Government Employment by country 31 East Asia & Pacific Europe & Central Asia General gov. employment (% LF) .2 .4 General gov. employment (% LF) .2 .3 .1 0 .1 1980 1990 2000 2010 2020 year 0 1980 1990 2000 2010 2020 iso = ALB/iso = FRA/iso = POL iso = ARM/iso = GBR/iso = PRT iso = AUT/iso = GEO/iso = RUS year iso = AZE/iso = GRC/iso = SMR iso = BEL/iso = HRV/iso = SVK iso = BGR/iso = HUN/iso = SVN iso = FJI iso = HKG iso = IDN iso = JPN iso = BLR/iso = IRL/iso = SWE iso = CHE/iso = ITA/iso = TUR iso = CYP/iso = LTU/iso = UKR iso = KOR iso = MAC iso = NZL iso = PHL iso = CZE/iso = LUX iso = DEU/iso = LVA iso = DNK/iso = MDA iso = SGP iso = THA iso = TLS iso = ESP/iso = MKD iso = EST/iso = NLD iso = FIN/iso = NOR Latin America & Caribbean Middle East & North Africa .3 .4 General gov. employment (% LF) General gov. employment (% LF) .3 .2 .2 .1 .1 0 1980 1990 2000 2010 2020 year 0 iso = ABW iso = ARG iso = BLZ iso = BOL 1995 2000 2005 2010 2015 iso = BRA iso = CRI iso = CUB iso = DOM year iso = GTM iso = MEX iso = NIC iso = PAN iso = ARE iso = EGY iso = ISR iso = MLT iso = PRY iso = TTO iso = URY iso = OMN iso = QAT North America Sub-Saharan Africa .4 .3 General gov. employment (% LF) General gov. employment (% LF) .3 .2 .2 .1 .1 0 1980 1990 2000 2010 2020 year 0 1980 1990 2000 2010 2020 iso = BWA iso = CPV iso = ETH iso = GIN year iso = MDG iso = MUS iso = SEN iso = SYC iso = CAN iso = USA iso = TZA iso = ZAF iso = ZWE Figure 14: Public sector employment by country 16 16 Estonia is the country with high employment that falls rapidly. Bolivia and Ecuador in LAC shows a rapid fall. 32 East Asia & Pacific Europe & Central Asia Public sector employment (% LF) .8 .3 Public sector employment (% LF) .6 .2 .4 .1 0 .2 1980 1990 2000 2010 2020 0 1980 1990 2000 2010 2020 year year iso = ALB/iso = GBR/iso = NLD iso = ARM/iso = GEO/iso = NOR iso = AUT/iso = GRC/iso = POL iso = AUS iso = FJI iso = HKG iso = JPN iso = AZE/iso = HRV/iso = PRT iso = BGR/iso = HUN/iso = ROU iso = BLR/iso = IMN/iso = RUS iso = KOR iso = MAC iso = MNG iso = MYS iso = CHE/iso = IRL/iso = SMR iso = CYP/iso = ITA/iso = SVK iso = CZE/iso = KGZ/iso = SVN iso = NCL iso = NZL iso = PHL iso = THA iso = DEU/iso = LTU/iso = SWE iso = DNK/iso = LUX/iso = TUR iso = ESP/iso = LVA/iso = UKR iso = VNM iso = EST/iso = MCO iso = FIN/iso = MDA iso = FRA/iso = MKD Latin America & Caribbean Middle East & North Africa Public sector employment (% LF) .4 .4 Public sector employment (% LF) .3 .3 .2 .2 .1 .1 0 1980 1990 2000 2010 2020 year 0 iso = ABW/iso = MEX iso = ARG/iso = NIC iso = BHS/iso = PAN 1995 2000 2005 2010 2015 year iso = BLZ/iso = PER iso = BOL/iso = PRI iso = BRA/iso = PRY iso = BRB/iso = SLV iso = CHL/iso = SUR iso = COL/iso = TTO iso = ARE iso = BHR iso = EGY iso = IRN iso = CRI/iso = URY iso = DMA/iso = VEN iso = DOM iso = MLT iso = OMN iso = PSE iso = QAT iso = ECU iso = GTM iso = HND iso = SYR North America South Asia .4 .4 Public sector employment (% LF) Public sector employment (% LF) .3 .3 .2 .2 .1 .1 0 0 1980 1990 2000 2010 2020 1980 1990 2000 2010 2020 year year iso = BMU iso = CAN iso = USA iso = BGD iso = IND iso = LKA iso = MDV 33 Public sector employment (% LF) Sub-Saharan Africa .1 0 .2 .3 1980 1990 2000 2010 2020 year iso = BFA iso = BWA iso = CPV iso = KEN iso = MDG iso = MUS iso = SEN iso = SYC iso = TZA iso = UGA iso = ZAF iso = ZWE Figure 15: Central government employment by country 17 East Asia & Pacific Europe & Central Asia Central gov. employment (% LF) .15 .2 Central gov. employment (% LF) .15 .1 .05 .1 .05 0 1980 1990 2000 2010 2020 year 0 iso = ALB/iso = GRC/iso = SWE iso = AZE/iso = HUN/iso = TUR iso = BEL/iso = IRL/iso = UKR 1980 1990 2000 2010 2020 iso = BGR/iso = ITA iso = CHE/iso = LTU iso = CYP/iso = LUX year iso = CZE/iso = LVA iso = DEU/iso = NLD iso = DNK/iso = NOR iso = FJI iso = IDN iso = JPN iso = KOR iso = ESP/iso = POL iso = EST/iso = PRT iso = FIN/iso = RUS iso = MAC iso = NZL iso = FRA/iso = SMR iso = GBR/iso = SVK iso = GEO/iso = SVN Latin America & Caribbean Middle East & North Africa .15 .8 Central gov. employment (% LF) Central gov. employment (% LF) .6 .1 .4 .05 .2 0 1990 2000 2010 2020 0 year 1995 2000 2005 2010 2015 year iso = ARG iso = BOL iso = BRA iso = CRI iso = CUB iso = DOM iso = MEX iso = ARE iso = ISR iso = MLT 17 Estonia shows a rapid fall from the beginning. Russia goes to almost zero at the end of the sample. Hungary increases rapidly since 2010. Norway increase rapidly around 2000. 34 North America Sub-Saharan Africa .1 .2 Central gov. employment (% LF) Central gov. employment (% LF) .08 .15 .06 .1 .04 .05 .02 0 1980 1990 2000 2010 2020 0 1980 1990 2000 2010 2020 year year iso = BWA iso = CPV iso = ETH iso = ZAF iso = CAN iso = USA iso = ZWE Table 13: Determinants of public employment. Dependent variable: Log of general government employment as share of population Variables (1) (2) (3) (4) (5) (6) Population -0.132*** -0.138*** -0.164*** -0.121*** -0.117*** -0.123*** (0.029) (0.027) (0.034) (0.030) (0.029) (0.029) GDP per capita 0.404*** 0.462*** 0.285*** 0.304*** 0.136 0.160 (0.064) (0.090) (0.088) (0.083) (0.148) (0.149) Openness -0.137 0.213*** 0.104 0.128*** 0.031 0.079 (0.110) (0.054) (0.078) (0.042) (0.083) (0.049) Ethnic fractionalization -0.785*** -0.789*** -0.473** -0.734** -0.570** (0.225) (0.227) (0.229) (0.295) (0.275) High income 0.255 0.205 0.417** 0.542*** 0.460** (0.199) (0.226) (0.184) (0.196) (0.197) High income * Openness -0.490*** -0.404*** -0.383*** -0.305** -0.331*** (0.091) (0.094) (0.092) (0.120) (0.104) Age dependency ratio, young -0.013*** -0.016** -0.011* (0.005) (0.006) (0.006) Age dependency ratio, old -0.000 -0.008 -0.007 (0.008) (0.008) (0.008) Gini -0.003 -0.000 -0.001 (0.006) (0.007) (0.007) Constant -11.277*** -11.898*** -9.282*** -10.634*** -8.775*** -8.847*** (0.644) (0.723) (1.176) (0.886) (1.563) (1.575) Observations 355 355 353 263 264 263 Number of countries 92 92 91 80 81 80 Country FE No No No No No No Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 35 Table 14: Basic specification with central government employment Variables (1) (2) (3) (4) (5) (6) Log of Population -0.259*** -0.233** -0.195* -0.337*** -0.318*** -0.283*** (0.088) (0.101) (0.109) (0.105) (0.102) (0.107) Log of GDP per capita 0.093 0.006 0.136 0.104 0.259 0.166 (0.200) (0.244) (0.318) (0.274) (0.401) (0.415) Openness -0.084 0.330 0.532 1.037 1.200 1.266 (0.186) (0.647) (0.558) (0.937) (0.784) (0.779) Ethnic fractionalization -1.202** -0.972* -1.283** -1.272* -1.529** (0.526) (0.537) (0.525) (0.705) (0.749) High Income 0.673 0.873 1.177 1.610* 1.454* (0.616) (0.539) (0.928) (0.866) (0.828) High Income * Openness -0.415 -0.478 -1.306* -1.452** -1.341** (0.598) (0.508) (0.769) (0.636) (0.603) Age dependency ratio, young 0.003 -0.002 -0.002 (0.010) (0.016) (0.016) Age dependency ratio, old -0.031*** -0.039*** -0.043*** (0.007) (0.013) (0.013) Gini 0.027 0.020 0.026 (0.022) (0.023) (0.025) Constant 0.670 0.515 -1.096 0.027 -1.489 -0.809 (2.270) (2.003) (3.027) (2.296) (3.399) (3.421) Observations 192 192 190 155 155 155 Number of countries 54 54 53 48 48 48 Country FE No No No No No No Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 15: Basic specification with public sector employment Variables (1) (2) (3) (4) (5) (6) Log of Population -0.118*** -0.086** -0.098** -0.050 -0.076 -0.061 (0.037) (0.037) (0.043) (0.059) (0.054) (0.057) Log of GDP per capita -0.181** -0.262*** -0.263** -0.245** -0.276** -0.292** (0.088) (0.091) (0.109) (0.105) (0.136) (0.134) Openness -0.062 -0.176 -0.169 -0.131 -0.144 -0.160 (0.081) (0.195) (0.202) (0.224) (0.224) (0.225) Ethnic fractionalization -1.328*** -0.762* -0.752* -0.872* -0.858* (0.379) (0.399) (0.409) (0.513) (0.506) High Income 0.608** 0.601** 0.747** 0.873*** 0.723** (0.242) (0.254) (0.314) (0.315) (0.311) High Income * Openness 0.158 0.139 0.082 0.092 0.126 36 (0.200) (0.208) (0.284) (0.285) (0.290) Age dependency ratio, young -0.001 -0.004 -0.004 (0.004) (0.006) (0.006) Age dependency ratio, old -0.001 -0.003 -0.004 (0.005) (0.008) (0.009) Gini 0.006 0.007 0.007 (0.007) (0.007) (0.007) Constant 2.191** 2.005** 2.276* 0.992 1.527 1.823 (0.940) (0.864) (1.206) (1.100) (1.337) (1.385) Observations 305 305 301 239 240 239 Number of countries 97 97 95 82 83 82 Country FE No No No No No No Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 16: Yearly estimates of the basic specification Variables (1) (2) (3) (4) (5) (6) Log of Population -0.138*** -0.124*** -0.127*** -0.068 -0.044 -0.059 (0.030) (0.029) (0.034) (0.050) (0.054) (0.051) Log GDP per capita 0.150 0.149 0.108 -0.027 0.065 0.091 (0.094) (0.110) (0.192) (0.095) (0.082) (0.079) Trade openness -0.117 0.221* 0.187** 0.064 0.047 0.114 (0.072) (0.124) (0.091) (0.077) (0.115) (0.077) Ethnic fragmentation -0.852*** -0.600** -0.589** -0.965*** -1.073*** (0.295) (0.268) (0.275) (0.365) (0.402) High income 0.534** 0.516** 0.398** 0.625*** 0.418** (0.239) (0.222) (0.195) (0.214) (0.196) High income * openness -0.405*** -0.352*** -0.165* -0.179 -0.229** (0.109) (0.095) (0.092) (0.129) (0.095) Age dependency ratio, young -0.005 0.007 0.010 (0.009) (0.006) (0.007) Age dependency ratio, old -0.006 0.005 0.005 (0.007) (0.007) (0.007) Gini -0.005 -0.006* -0.006* (0.003) (0.003) (0.003) Constant -0.983 -1.648* -0.925 -0.422 -2.443*** -2.124** (0.931) (0.880) (1.936) (1.147) (0.914) (0.979) Observations 936 936 934 505 506 505 Number of countries 80 80 79 61 62 61 Country FE No No No No No No Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 37 Table 17: Determinants of general government employment (basic specification). Dependent variable: Log of general government employment as share of labor force Variables (1) (2) (3) (4) (5) (6) Log of Population -0.125*** -0.127*** -0.126*** -0.120*** -0.125*** -0.121*** (0.029) (0.031) (0.033) (0.036) (0.036) (0.037) Log of GDP per capita 0.146** 0.139 0.133 0.161 0.139 0.113 (0.074) (0.102) (0.152) (0.118) (0.207) (0.208) External risk -0.003 0.020 0.020 0.022 0.023 0.020 (0.013) (0.019) (0.017) (0.023) (0.019) (0.019) Ethnic fractionalization -0.849*** -0.592** -0.590* -0.507 -0.485 (0.299) (0.301) (0.311) (0.340) (0.337) High income 0.428* 0.427** 0.461 0.556 0.463 (0.220) (0.215) (0.337) (0.343) (0.344) High income * External risk -0.071* -0.069* -0.083* -0.100** -0.083* (0.036) (0.039) (0.049) (0.050) (0.050) Age dependency ratio, young -0.001 -0.003 -0.003 (0.005) (0.008) (0.008) Age dependency ratio, old -0.001 0.003 0.002 (0.006) (0.008) (0.008) Gini 0.006 0.007 0.008 (0.010) (0.013) (0.013) Constant 1.627** 1.391 1.487 0.915 0.979 1.379 (0.754) (0.907) (1.545) (0.880) (1.732) (1.755) Observations 275 275 275 220 220 220 Number of countries 73 73 73 69 69 69 Country FE No No No No No No Robust standard errors clustered by country in parentheses *** p<0.01, ** p<0.05, * p<0.1 Figure 16: Actual versus predicted, last 5-year window a) EAP b) LAC 38 ARG HKG BLZ JPN BOL CHL KOR COL CRI MNG ECU PHL GTM MEX THA PRY -.1 -.05 0 -.05 0 .05 .1 .15 .2 Actual-predicted Actual-predicted c) MENA d) NA EGY CAN ISR MLT USA OMN 0 .05 .1 -.04 -.02 0 .02 .04 Actual-predicted Actual-predicted e) SA f) SSA CPV LKA TZA UGA 0 .005 .01 .015 -.03 -.02 -.01 0 .01 .02 Actual-predicted Actual-predicted g) ECA 39 ARM AUT AZE BEL CHE CYP CZE DEU DNK ESP EST FIN FRA GBR GRC HRV HUN IRL ITA KGZ LTU LVA MDA NLD NOR POL PRT RUS SMR SVK SVN SWE TUR -.1 -.05 0 .05 .1 Actual-predicted Notes: The prediction model is log(E/LF) = -1.38 - .11*log(Population) + .10*log(GDPPC) + .11*Openness - .56*Ethnic + .47*HighIncome - .27*HighIncome*Openness Figure 17: Political ideology and public employment in excess of the predicted level 40 Public employment minus predicted 0 10 -10 20 30 1 1.5 2 2.5 3 1=right; 2=center; 3=left Note: Correlation coefficient is 0.06 with p-value=0.3 40