Corruption, Productivity and Transition *

CENTRE FOR ECONOMIC REFORM AND TRANSFORMATION School of Management and Languages, Heriot-Watt University, Edinburgh, EH14 4AS Tel: 0131 451 8143/3485 Fax: 0131 451 3498 email: ecocert@hw.ac.uk World-Wide Web: http://www.som.hw.ac.uk/cert Corruption, Productivity and Transition * Geoffrey Wyatt ** Heriot-Watt University May 2002 Discussion Paper 2002/05 Abstract The level of productivity is correlated across countries with measures of (lack of) corruption, but this appears to be due to a common association of these variables with measures of social infrastructure, here measured by a combination of governance indexes labelled rule of law and government effectiveness. Econometric use of instruments establishes that causation runs from social infrastructure to productivity. Socialism has had a positive direct effect on productivity but one that is dominated by a negative indirect effect via social infrastructure. Keywords: corruption; governance; productivity; social infrastructure; transition. JEL Classification: H10, O11, P20 * Paper prepared for PHARE-ACE project P98-1125-R. Presented at the Project Workshop in Sofia, Bulgaria, 17-18 May 2002. This research was undertaken with support from the European Community s Phare ACE Programme 1998. The content of the publication is solely the responsibility of the author and it in no way represents the view of the Commission or its services. ** Department of Economics, School of Management and Languages, Heriot-Watt University, Edinburgh EH14 4AS. email: G.J.Wyatt@hw.ac.uk. 1

1. Introduction Many empirical studies of economic growth have reported a connection between growth and diversion, meaning unproductive economic activity pursued for gain via legal or illegal transfers of resources. This is called corruption where it involves transfers to government sector employees arising from side transactions between the private and public sectors. Since much of this type of activity is illegal, the transactions are for the most part unrecorded, and data measuring the extent of corruption has to rely on perceptions of corruption as recorded in surveys of businessmen, country experts and the like. We make use of such data in the empirical analysis reported below, where we find a clear positive correlation across countries between measures of corruption and the level of output per worker (productivity), see Figures 1 and 2. The main question addressed in this paper concerns the interpretation of this correlation: is it cause or effect, or perhaps even spurious? We follow Hall and Jones (1999) in arguing that corruption is a reflection, or indeed an expression, of a wider concept called social infrastructure. We confirm that social infrastructure appears to be causally related to the level of productivity. The principal contribution of this paper is to consider the manner in which corruption features in this relationship and, further, how the former socialist countries, including the transition countries, fit into the picture. The plan of this paper is as follows. In Section 2 we consider the recent literature on the topic. Sections 3 discusses the available data. The empirical econometric analysis is reported in Section 4 and the results obtained are discussed in the concluding Section 5. The data used in the empirical analysis is listed in an Appendix. 1

3.5 Productivity and Control of Corruption USA LUX log(productivity) 3 2.5 2 1.5 1 0.5 0 ZAR NER TKM TJK BEL JPN ITA ARE QAT LBY BHR KWT TWN SAU OMN GRC CZEMLT BHS KOR ARG VEN URY HUN GAB SVK MUS BWA RUS MEX ZAF PAN COL CRI MYS EST IRN POL TTO LTU BLR BRA HRV ROM LBN TURLVA NAM THA DZA KAZ ECU BGR GTM TUN UKRDOM MKD SUR GEO PER PRY FJI SWZ JOR SYR PHL JAM MAR BOL EGY IDN SLV AZE KGZ MDA ALB ZWE HND GUY UZBPNG CMR NIC ARM CHN AGO GIN COG LKA CIV SEN GHA PAK TGO SDN LSO HTI GMB BEN VNM IND KEN MNG UGA ZMB TCD BFA GNB MDG NGA BGD YEM MMR MLI MOZ MWI SVN CHL NOR CHE ISL SGPCANDNK FRA HKG AUT AUS DEU SWE GBR NLD FIN IRL ISR NZL ESP PRT CYP -0.5 TZA ETH SLE -1-2 -1.5-1 -0.5 0 0.5 1 1.5 2 2.5 Control of corruption Figure 1. Productivity and Control of Corruption 2.5 Measures of "Governance" 2 1.5 1 SVN ISR DNK SWE CAN FIN NZL CHE NLD SGP CYP ISL LUXNOR GBR DEU IRL AUS AUT USA FRA HKG PRT ESP CHL Control of corruption 0.5 0-0.5-1 FJI GRC ITA JPN BEL EST KWT HUNMYS TWN CRI QAT BWA BHS TTOPOL MLT OMN URY NAM CZE ZAF MUS LSO MAR JOR KOR BRA SVKLTU SLE SUR SWZ TUN GUY GMB ARE CIV JAMLKA GNB MNG MWI THA PER TGO SEN BHR BGD CHN EGY PHL GHA MEX LVA IND ARG ZWE SLV VNM TUR MDA BFA BOL ETH LBN MDGCOL ROM PAN MLI UGA HRV HTI MOZ MKD COG TCD BGR SAU KEN BLRRUS ZMB VEN PAK GEO SYR IDN KGZ ARM BEN DOM GTM ECU AGOLBY DZA YEMKAZ NIC PNGGIN IRN UKR NGA PRY HND TZA UZB SDN GAB ALBAZE MMR CMR -1.5 ZAR TJK NER TKM -2-4 -3-2 -1 0 1 2 3 4 5 Government Effectiveness + Rule of Law Figure 2. Control of Corruption and Government Effectiveness + Rule of Law 2

2. Brief review of literature Study of the causes and consequences of corruption has increased in recent years with greater availability of the output of international surveys of corruption, e.g. Kaufmann et al. (1999) and Hellman et al, (2000). The recent empirical literature relating corruption, among other determinants, to growth begins with Knack and Keefer (1995) and Mauro (1995). The Knack and Keefer study is based on Barro's (1991) cross country regression formulation. They demonstrate that an index of the quality of institutions, which included a (lack of) corruption component, was highly significant as an explanatory variable for economic growth, indeed as significant as the education variables which proxied for human capital. This study used two survey sources for the institutional data: ICRG and BERI. Mauro, using BI data showed that indexes for bureaucratic effectiveness and for corruption were significant explanatory variables in Barro regressions for the share of investment, but less significant for economic growth 1. This study instrumented the explanatory variables by an index of linguistic fractionalization in order to treat for their potential endogeneity. It concludes that bureaucratic efficacy and corruption affect growth through the investment ratio. In a later paper (Mauro, 1999) it is shown that a composite index of corruption is significantly associated with the share in gdp of government spending on education. This would seem to provide another channel from corruption to economic growth as human capital is usually positively associated with growth. In Barro and Sala-i-Martin (1995) growth regressions involving corruption and rule of law variables are reported; it is found that the corruption variable was insignificant when included together with a "Rule of Law" variable, which was highly significant. In an influential paper, Hall and Jones (1999) argue that "social infrastructure" explains a large part the cross-country variation in the level of per capita output, and does so through physical and human capital formation and productivity. The measure of social infrastructure they use is a combination of ICRG indexes and openness to trade. The potential endogeneity of social infrastructure is treated by the use of geographical and linguistic characteristics of the countries in the world-wide dataset. Kaufmann et al (2000) build on the Hall- 1 ICRG is International Country Risk Guide, BERI is Business Environment Risk Intelligence and BI is Business International, part of the Economist Intelligence Unit. 3

Jones analysis by applying their own compilation of measures of governance ("voice", "political violence", "government effectiveness", "regulatory burden", "rule of law" and "graft") separately, using a language instrument - the fraction of the population speaking a major European language. However they encountered a problem with the graft (i.e. corruption) variable, namely that as an explanatory variable in a regression of per capita output it fails a test of the overidentifying restrictions, implying either that it is insufficiently well accounted for by the chosen instruments or that it has a separate direct influence on output per worker. 3. Data 3.1 Governance data A World Bank study (Kaufmann et al, 1999) lists eleven sources of data on corruption. All of these sources conduct questionnaire based surveys, in some cases several surveys, which are very heterogeneous: they have different aims, scope, methodology and sampled population. This study makes use of three of the six indexes synthesised from these disparate sources by the Governance Research Unit at the World Bank, namely those for "Control of Corruption" (CC), "Rule of Law" (RL) and "Government Effectiveness" (GE). Table 1 shows the origin of the data used to compile the indexes further details can be found in the source cited. The World Bank Governance Research Unit created the indices for our three variables by using a statistical methodology based on an unidentified components model. Although there is a substantial range of uncertainty around the estimated indexes, they have the considerable advantages of being comparable across countries and world-wide in coverage. 4

Table 1. Sources of governance data Source coverage: number of countries GE RL CC Standard & Poor's DRI / McGraw-Hill n/a 3 4 1 Economist Intelligence Unit 114 3 2 1 Heritage Foundation / Wall Street Journal 161 0 2 0 Political Risk Services n/a 2 1 1 World Bank / University of Basel 69 9 4 2 Business Environment Risk Intelligence 50 1 1 1 Wall St. Journal Central Europ. Econ. Review 27 0 1 1 Freedom House 28 1 1 1 Gallup International 44 0 0 1 World Economic Forum 83 9 21 4 Political Economic Risk Consultancy 12 - - 1 Institute for Management Development 47 3 5 1 3.2 Productivity The variable we seek to explain is the level of output per worker, not its growth over a number of years. An important reason for this is that the countries of particular interest in this study are the transition countries. Many of these are states that have emerged or re-emerged from the split-up of the Soviet Union. Obviously, for these countries growth data is at best available for a period of about 10 years, which is firstly rather short and secondly covers the exceptional period of transition, which is not yet over. Even the data for those formerly independent socialist states such as Romania and Bulgaria, which have data spanning back several decades, are affected by the distortions and corrections experienced through the transition years. Productivity is GDP at purchasing power parity divided by the labour force, defined as the population aged between 16 and 65. Globally comprehensive figures (excepting Taiwan) for real GDP up to the year 2000 are extracted from the IMF's website. These data extend the Penn World Tables, both in time and in the coverage of countries, to include importantly the transition economies. Labour 5

force data comes from the World Bank s online databank. In order to minimise the effects of any particular year which may be abnormal, the cross-country productivity data to be used in Section 4 are the averages over the 1991 to 2000 period. 3.3 Instruments Our aim is to investigate the connections between corruption, social infrastructure and productivity, and for this we need as instruments for social infrastructure variables that are expected to be independent of the level of output per worker in the 1990s. As Hall and Jones point out, it is not easy to find such instruments. Most economic variables are influenced by the level of productivity, which vitiates their use as instruments. Hence the search for valid instruments led Hall and Jones to consider two types of variable: (i) inherent physical characteristics of countries like their latitude (distance from the equator) or their proximity to other countries, and (ii) features which reflect the distant history of countries, such as their exposure to early European influence, possibly by colonization, as currently reflected in the everyday languages spoken. The languages spoken do of course change over time, partly in response to economic circumstances, implying some feedback. It is a matter of judgement whether there is enough feedback to invalidate these variables as instruments independent of the level of productivity. However, entry to the modern world of technology more or less involves the use of English, so the problem with the English language variable in particular may not be minor. The instruments we use are derived from two sources. One is the World Bank s file on Social Indicators and Fixed Factors. These include geographic variables (latitude, land area and landlocked dummy), oil-exporter and transition dummy variables. The other source is a geographic gazetteer listing the names, populations, administrative status and geographic positions of all cities with population greater than 50,000 plus others that have administrative functions as state or regional capitals (see www.world-gazetteer.com). There are over 10,000 such cities. These data were used to construct instruments as follows. 6

3.3(i) City sizes The size distribution of cities in most countries approximately follows a Pareto distribution (also called Zipf s law ). This is evident from visual plots of log(size) against log(rank). A roughly straight line is the indicator of a twoparameter Pareto distribution, with the crucial shape parameter being the slope. If it is equal to one then it may be said that "Zipf s law" applies. In fact the Pareto "alpha" (i.e. slope) parameters for empirical size distributions of cities vary across countries, mostly between about 0.5 and 2, see Figure 2. The histogram shows the international dispersion of alphas, calculated for each country by ordinary least squares regression of log(population) against log(size) with a dummy variable for the capital city, for all towns with over 50,000 inhabitants. 100 Histogram of Pareto alpha coefficients, all countries 90 80 70 60 50 40 30 20 10 0-4 -3.5-3 -2.5-2 -1.5-1 -0.5 0 Figure 2. The international distribution of Pareto exponents Output per worker in a particular city and the city s population are surely causally connected, both ways. But what about the distribution of city sizes in a country? Recently there have been put forward theories (e.g. Gabaix, 2000) in terms of stochastic processes which lead to the typical Pareto shape. Empirically, 7

while the Pareto curve does move over time (e.g. Parr, 1985), and with it the slope coefficient alpha, that movement, even over many decades, is slight in comparison with the cross-country variations in alpha. For the UK, for example, while it is true that the second city, now Birmingham, used to be Liverpool and a century before that, Glasgow, these swaps can take place without alpha changing by much. Alpha seems to be characteristic of a country s urban make-up over very long periods of time, even centuries. A lower absolute value for alpha implies a greater size equivalence among cities, suggesting perhaps greater autonomy and independence, whereas a higher alpha suggests something of a hierarchical structure among cities, and perhaps less social cohesion. A key question for us is therefore, whether the distribution of city sizes influences social infrastructure and hence national productivity, or whether it has a separate effect of its own on productivity? Accordingly, we shall consider the validity of alpha as an instrument for social infrastructure in a Hall and Jones type regression. 3.3(ii) City density The size distribution of the cities within a country says nothing about their geographical spread. But the spatial distribution of economic activity could also affect social infrastructure and hence national productivity. External economies are often invoked to suggest that agglomerations of economic activity have lower costs than dispersed activity. That would represent a direct effect on productivity. But the spatial distribution of urban centres could have an indirect effect via social infrastructure. The clustering of economic activity may both lower the costs of diversionary activity and increase the opportunities for it, especially where this requires networking among the diverters. On the other hand, concentrated economic activity offers greater opportunities for the observation and control of miscreants. But, again, we shall need to consider whether the spatial dispersion of urban economic activity is exogenous for the level of productivity when there are clear economic incentives to drive clustering. We attempt to construct a simple measure from our city size database which reflects the degree to which cities are clumped together or spread out across 8

the country. It should of course allow meaningful comparisons to be made across countries. Consider two cities, of sizes C 1 and C 2 respectively, and let them be separated by a distance d 12. Let us suppose that these two cities make up the whole country. We wish to formulate a measure that tells how clumped or dispersed it is. This will depend on the three variables we have defined. The measure must be increasing in C 1 and C 2, and decreasing in d 12, though always non-negative. Also C 1 and C 2 should be interchangeable, and if d 12 = 0 (so that the cities are contiguous), it would be desirable for the conurbation to reflect simply the combined size, C 1 + C 2. A simple formula that meets these desiderata is: (C 1 + C 2 ).exp( d 12 ). If there are, say, n cities then the extent to which the others are clustered with respect to the first would then be represented: n (C 1 + C i )e d 12 i =1 and summing this over all n cities we arrive at: n 1 n i =1 j= i +1 (C i + C j )e d ij which is basis of the index we use. Now, bearing in mind the Pareto law for city sizes: C i AR α where R is the city rank within a country, we use ranks instead of population numbers in the formula. And to make them comparable across countries, we use not the within-country ranks, but the ranks of the cities in the world distribution, W. For the choice of α, a benchmark of 0.5 seems acceptable based on the data, so our index becomes: n 1 n D = ( W i + W j )e d ij. i =1 j = i+1 Figure 4 shows how the logarithm of city density is related to the Pareto alpha coefficient across countries. The log-transform was taken to avoid excessive clustering on the vertical axis, and indeed the transformed variable seems better 9

behaved overall, and is used in the regression analysis of Section 4. Although alpha and log(d) are positively correlated, the dispersion of the scatter in Figure 4 confirms that they are basically distinct variables. 0 Alpha and urban density KOR -0.5 EGY TWN JPN -1 BGD PHL GBR DEU NLD NGA IDN MEX BRA IND PAK VEN ISR log(city Density) -1.5-2 -2.5-3 DJI BHS SUR QAT GRCZAF POL COL CHN DZA THA MYS FRA ITA JOR TUR ESP DOM MAR UKR CAN CHL BEL SLV IRNLKA SYR ECU VNM ROM ARG LBN HTI CIV MMR PRY PRT USA ARE TUN KEN ARMGTM NIC SAU GHA RUS CZE HUN BGR YEM AUS SEN BLR CMR UZB ZWE PER HND LBY KWT OMN BOL ZMB JAM GEO SDN AZE BHR MUS PAN MDA BEN ETHNPL COG ZAR MKD NZL KHM LTU BFA AUT SWE ALBTJK TZASVK CHE SLE DNK GIN MWI GAB GMB TGO RWA LVAUGA IRL MOZ FIN AGO BDI KGZ URY GNB HRV NOR KAZ CRI MLI MDG MNG GUY LAO SVN NER PNG CAF EST TCD NAM TKM FJI LSO BWA MRT -3.5 NEA SLU CPV BTN STP TTO GNQ ISL SLB SWZ BLZ CYP LUX MDV -4-2.5-2 -1.5-1 -0.5 0 Alpha Figure 4. Pareto "alpha" and urban density, D 4. Empirical analysis 4.1 Government Effectiveness, Rule of Law and Control of Corruption These three indicators probably capture different aspects of social infrastructure. While Government Effectiveness (GE) and the Rule of Law (RL) relate fairly obviously to institutions of economic policy, Control of Corruption (CC) is an outcome and reflects the behavioural response of government officials and those with whom they do business to the opportunities for diversion. 10

Corruption is an opportunistic reflection of weak social norms and constraints, which are themselves an aspect of social infrastructure. It is a form of behaviour in which the parties to the corrupt exchange lack respect for the social rules. But it can also be seen as a perversion of the profit seeking which is the basis of efficient resource allocation in well functioning markets. With appropriate constraints in place, this form of socially harmful profit seeking could presumably be channelled into more socially useful directions. Whereas CC directly reflects the extent and prevalence of corruption, GE and RL probably reflect restraints to diversionary activity, and thereby the cost of engaging in a corrupt transaction. They are also, presumably, variables which can be influenced by government policies and actions. Since a loosening of the restraints reduces the costs of corruption, the three variables should be related, and in particular it seems likely that GE and RL together cause CC, at least in part. Furthermore, one might suppose that the impact on CC of an increment in GE would be larger the higher the level of RL, and similarly that the impact of an increment in RL on CC would probably be greater the higher the level of GE. This argues for the plausibility of an interaction effect, which suggests that an appropriate regression model might be: CC = a + b.ge + c.rl + d.(ge*rl) + eps (1) Table 2. Regressions of equation (1), dependent variable: CC Full dataset Hall & Jones subset constant -0.122 (0.032) -0.105 (0.033) GE 0.675 (0.061) 0.674 (0.063) RL 0.243 (0.057) 0.227 (0.059) GE*RL 0.154 (0.029) 0.146 (0.028) R 2 0.898 0.906 s.e.e. 0.289 0.289 nobs 142 126 DWH p-value 0.224 0.387 Instruments Note: Standard errors in brackets a, A, a*a, ll, lat, log(d), a*log(d), -- see Section 4.2 EngFrac EurFrac log(frankrom) Latitude This regression model was estimated on the World Bank dataset. The possibility of feedback, i.e. that CC also causes GE and RL, which would render ordinary least squares estimates inconsistent, was examined by a Durbin-Wu- 11

Hausmann (DHW) test. The regression was estimated both for the full dataset, and also for a subset of the data for which the social infrastructure instruments of Hall and Jones were available. The results reported in Table 2 imply that GE and RL are very good predictors of CC, and that the interaction term is highly significant. There is no evidence of inconsistency of ordinary least squares from the DHW tests 2. The lack of evidence of feedback in this regression is taken to be consistent with our interpretation that Government Effectiveness and the Rule of Law act as constraints on corrupt behaviour. Following Hall and Jones, we take these two variables to be indicators of social infrastructure, and use a simple average of the two to represent social infrastructure in the analysis that follows. 4.2 Social infrastructure, transition and the level of productivity Between them, social infrastructure and productivity probably affect almost every economic variable we can define. Nevertheless, as Hall and Jones demonstrate, it is possible to analyse the effect of social infrastructure on productivity by careful use, as instruments, of the very limited set of exogenous variables that determine social infrastructure. In their analysis these variables included geographic position, distance between countries (in the Frankel-Romer gravity model predictor of trade) and linguistic heritage. We follow their analysis but with certain variations which stem partly from our wish to examine how the long experience of socialism has influenced outcomes for the current and former socialist states. In particular, we do not have the Frankel-Romer predictions for states which did not exist prior to the break up of the Soviet Union, and also we have doubts about exogeneity with respect to productivity of the two language variables (proportions of the population speaking English and another Western European language). However, we retain distance from the equator as an instrument. Empirically, it was the most important instrument for Hall and Jones, and their conceptual justification for it is persuasive. The other instruments we use include the slope and intercept of the city log(size):log(rank) regressions and the city density variables describes in Section 3. In addition, inspection of the 12

productivity data reveals the importance of natural resources, especially oil, for certain countries. It is improbable that the impact of oil on output works through social infrastructure, so this is treated as a specifically included exogenous variable, having a direct effect on productivity. Finally, two dummy variables for socialist and former socialist regimes are used to represent those countries that were already socialist before WWII and those that became socialist after WWII respectively. These variables could affect productivity directly, and also indirectly via social infrastructure. Reduced form regressions The exogenous variables in the system are: oil exporting dummy (Oil); early socialist regime (T1); later socialist regime (T2); Pareto exponent from distribution of city sizes (a); Pareto intercept (A); interaction between A and a (A*a); landlocked dummy (ll); latitude (lat); city density (log(d)); and interaction between log(d) and a (log(d)*a). The first three of these variables are exogenous variables assumed to have a direct effect on productivity, and the remainder are assumed to be uncorrelated with the disturbance term in the productivity equation, but correlated with social infrastructure. The reduced form regressions are shown in Table 3. The reduced form regression of GE+RL can be regarded as producing the instrument for the structural equation for productivity to be discussed later, and for that the F ratio in the table relates to the joint significance of the excluded instruments. Recent papers (e.g. Staiger and Stock, 1997) suggest that an F ratio of this size may be adequate to avoid a serious problem of finite-sample bias associated with weak instruments in IV regression. Note that the dummy variables for both the pre-wwii and the post-wwii socialist regimes, T1 and T2 respectively, are negative and highly significant in both equations. Moreover, in the GE+RL regression, the absolute size of the coefficient on T1 exceeds that on T2 by a margin that is statistically significant at the 5% critical level on a one-way test. But in the log(y) regression, the null that the coefficients on these dummies are equal cannot be rejected. 2 Not reported are further regressions with dummy variables for the 12 pre-wwii socialist regimes and for the 13 post-wwii socialist regimes which turn out to be insignificant. 13

Table 3. Reduced form regressions Dep Variable GE+RL log(y) const 22.890 (7.741) 10.509 (4.657) Oil -0.494 (0.285) 0.667 (0.171) T1-2.670 (0.413) -0.915 (0.248) T2-1.927 (0.369) -0.641 (0.222) a 19.910 (7.386) 7.148 (4.443) A -1.332 (0.460) -0.546 (0.277) A*a -1.029 (0.4297) -0.382 (0.258) ll -0.411 (0.271) -0.249 (0.163) lat 0.076 (0.007) 0.044 (0.004) log(d) 2.640 (0.857) 1.193 (0.516) log(d)*a 2.456 (0.848) 0.891 (0.510) F ratio 1 25.487 R-squared 0.612 0.572 s.e.e. 1.111 0.669 nobs 142 142 OLS regressions, standard errors in brackets 1 F ratio for last seven variables (excluded instruments) Structural regression for productivity The reduced form regressions raise the question how the socialist regime variables affect productivity. Is the effect direct or indirect, working through the social infrastructure? The structural equation for productivity is: ln y = α + βs(x ) + γ 1 T1 + γ 2 T2 + δoil + v (2) in which the coefficients can be consistently estimated using the set of variables in X as instruments for S. It is assumed that the Oil variable only affects productivity, not social infrastructure. This regression is reported in Table 4. 14

Table 4. IV estimation of productivity equation Dep Variable log(y) log(y) const 1.393 (0.059) 1.482 (0.054) 0.552 (0.040) 0.513 (0.038) S=GE+RL T1 0.572 (0.191) T2 0.399 (0.174) Oil 0.902 (0.149) 0.885 (0.150) R-squared 0.650 0.633 s.e.e. 0.590 0.601 Nobs 142 142 OverID test, 0.262 0.011 p-value Instruments a, A, A*a, ll, lat, log(d), log(d)*a Standard errors in brackets a, A, A*a, ll, Lat, log(d), log(d)*a, T1, T2 The panel on the left of Table 4 shows that the coefficients of the productivity equation are all significant and the instruments appear to be valid according to the test for overidentifying restrictions 3. The regression reported in the panel on the right assumes that the socialist dummy variables do not affect productivity directly, but only indirectly via social infrastructure, so these variables are in the excluded instrument set for this regression. However, the test for overidentifying restrictions now rejects the expanded set of instruments, so our preferred formulation is the regression reported in the left panel. The socialist dummy variables are significantly positive, and a test for equality of their coefficients does not reject. A significant positive effect from socialism was also found by Hall and Jones, for whom it was a puzzle. As they note, it implies that (former) socialist countries produce substantially more output per worker than otherwise similar capitalist countries 4. But this may not be such a puzzle. The reduced form equations give the total effects of socialism on productivity, including feedback effects between productivity and social infrastructure. These total effects were found from the reduced form regression to be negative, and were 3 Note that the regression reported in the left panel of Table 5 was also estimated with control of corruption CC as the proxy for social infrastructure S instead of GE+RL, with the result that the overidentifying restrictions were decisively rejected (at the 1% critical level). This confirms the earlier findings of Mauro and of Kaufmann that this formulation appears to be a misspecification. 4 They conjectured that the unexpected finding was due to their definition of capitalism, which included a number of sub-saharan African economies. 15

highly significant. With a positive and highly significant coefficient of social infrastructure in the IV regression reported in Table 4, the explanation for the positive signs on T1 and T2 could only be that socialism has a strongly negative effect on social infrastructure, and that the implied indirect negative effects on productivity outweigh the direct positive effects reported in Table 4. But it should not be surprising that, allowing for social infrastructure, socialism enhances productivity. The socialist regimes generally placed a great weight, even if misguidedly, on output and productivity. The failure of socialism for productivity may well be due to the overriding importance of the indirect effects working through social infrastructure, which was not nurtured in those regimes. 5. Concluding discussion The cross-country correlation between "control of corruption" and productivity evident in Figure 1 stems from the fact that both variables are correlated with social infrastructure. The causal relationships between these variables were explored in this paper, which finds that the underlying factor connecting governance indicators and productivity is social infrastructure, as summarized in Figure 5. There, X and ξ represent respectively exogenous variables (the instruments in the regression analysis) and disturbances affecting social infrastructure, while Z and ζ represent respectively an exogenous "oil exporter" dummy variable and other unspecified factore (disturbances) affecting productivity. The (control of) corruption variable reflects social infrastructure but does not directly influence productivity. The transition countries of Eastern Europe and the former Soviet Union are represented by the exogenous "socialism" node in Figure 5, which indicates that their socialist history has depressed social infrastructure and thereby productivity, to such a degree that it offsets the directly positive effect on productivity. Consistent with this, we also find that these effects of socialism are larger the longer the country experienced a socialist regime. 16

ε X, ξ Social Infrastructure + Corruption Socialism + + (+) Z, ζ Productivity Figure 5. Causal structure This interpretation of the empirical results suggests a policy prescription for the former socialist countries currently undergoing transition to a market economy. Namely that, in order to improve the economy i.e. productivity the most effective route could be to repair the social infrastructure, in particular the institutions of governance such as the legal framework (e.g. the law of contract and property rights) and the quality and integrity of the civil service. In addition, perhaps, the State should desist from direct interference in production where possible, thus reducing opportunities for diversion and corruption. 17

References Barro, R. J. (1991) "Economic growth in a cross-section of countries", Quarterly Journal of Economics, 106, 407 444. Barro, R. J. and X. Sala-i-Martin (1995) Economic Growth, McGraw-Hill, New York. Gabaix, X. (1999) "Zipf's Law for cities: an explanation", Quarterly Journal of Economics 114(1) 83 115. Hall, R. E. and C. I. Jones (1999) "Why do some countries produce so much more output per worker than others?" Quarterly Journal of Economics 83-116. Hellman, J. S., G. Jones, D. Kaufmann and M. Schankerman (2000) "Measuring governance, corruption and state capture", Policy Research Working Paper 2312, The World Bank, Washington USA. Kaufmann, D., A. Kraay and P. Zoido-Lobaton (1999) "Aggregating governance indicators ", Policy Research Working Paper 2195, The World Bank, Washington USA. Kaufmann, D., A. Kraay and P. Zoido-Lobaton (2000) "Governance matters", Policy Research Working Paper 2196, The World Bank, Washington USA. Knack, S. and P. Keefer (1995) "Institutions and economic performance: crosscountry tests using alternative institutional measures" Economics and Politics, 7(3) 207-227. Lambsdorff, J. G. (1999) "The Transparency International corruption perceptions index - Framework Document" mimeo. Mauro, P. (1995) "Corruption and growth" Quarterly Journal of Economics 681-712. Mauro, P. (1998) "Corruption and the composition of government expenditure" Journal of Public Economics 69, 263-279. Parr, J. B. (1985) "A note on the size distribution of cities over time", Journal of Urban Economics, 18, 199 212. 18

Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil 1990-1999 Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Albania ALB 2.771-0.653-0.918-0.985 12.844-1.135 0.077 41.31 0 1 0 0 Algeria DZA 5.364-1.087-1.103-0.878 13.971-0.715 0.232 36.72 0 0 0 1 Angola AGO 2.339-1.390-1.225-0.863 13.647-1.253 0.062-8.84 0 0 0 1 Antigua & Barbuda ANB NaN NaN NaN NaN 9.270-1.180 NaN 17.07 0 0 0 0 Argentina ARG 10.033 0.262 0.319-0.275 15.533-1.125 0.153-36.68 0 0 0 0 Armenia ARM 2.556-0.655-0.146-0.803 13.192-1.116 0.133 40.25 1 0 1 0 Australia AUS 20.973 1.459 1.596 1.601 15.788-1.481 0.110-32.22 0 0 0 0 Austria AUT 21.297 1.219 1.812 1.457 13.517-1.192 0.081 48.23 0 0 1 0 Azerbaijan AZE 2.997-0.833-0.563-0.998 13.380-1.047 0.096 40.35 1 0 1 0 Bahamas BHS 11.873 0.474 0.563 0.497 12.184-2.250 0.045 24.70 0 0 0 0 Bahrain BHR 16.072 0.235 0.665-0.215 12.118-0.982 0.092 26.02 0 0 0 1 Bangladesh BGD 1.182-0.565-0.929-0.289 14.866-0.952 0.448 23.88 0 0 0 0 Barbados BRB 11.209 NaN 0.411 NaN 8.005-0.764 NaN 13.18 0 0 0 0 Belarus BLR 6.313-0.659-0.876-0.654 14.186-1.021 0.108 53.00 1 0 1 0 Belgium BEL 22.008 0.883 0.797 0.672 13.314-0.800 0.189 50.84 0 0 0 0 Belize BLZ 5.294 NaN 0.088 NaN 10.605-1.020 0.022 17.84 0 0 0 0 Benin BEN 1.637-0.066-0.422-0.781 13.329-1.041 0.088 6.36 0 0 0 0 Bhutan BTN 1.948 NaN NaN NaN 11.328-1.495 0.029 27.48 0 0 1 0 Bolivia BOL 3.296-0.223-0.355-0.438 14.657-1.477 0.098-15.19 0 0 1 0 Botswana BWA 8.980 0.221 0.502 0.535 12.388-1.015 0.043-21.54 0 0 1 0 Brazil BRA 6.397-0.220-0.222 0.058 16.198-0.878 0.339-19.56 0 0 0 0 Bulgaria BGR 5.274-0.814-0.150-0.557 13.550-0.890 0.117 42.07 0 1 0 0 Burkina Faso BFA 1.292-0.059-0.350-0.368 13.608-1.350 0.077 12.05 0 0 1 0 Burundi BDI 0.991 NaN -0.881 NaN 12.233-1.236 0.062-3.37 0 0 1 0 Cambodia KHM 1.362 NaN -0.235 NaN 13.681-1.535 0.079 12.03 0 0 0 NaN Cameroon CMR 2.517-0.645-1.015-1.105 13.775-0.979 0.111 10.73 0 0 0 0 Canada CAN 22.927 1.717 1.549 2.055 14.846-0.910 0.190 43.73 0 0 0 0 Cape Verde CPV 3.366 NaN 0.088 NaN 11.621-1.547 0.030 15.09 0 0 0 0 Central Africa CAF 1.692-0.747 NaN NaN 12.420-0.914 0.049 4.33 0 0 1 0 Chad TCD 1.353-0.714-0.827-0.587 12.481-1.033 0.046 10.38 0 0 1 0 Chile CHL 11.342 1.166 1.086 1.029 14.313-0.882 0.189-33.55 0 0 0 0 China CHN 2.511 0.016-0.040-0.289 16.489-0.817 0.236 29.56 0 0 0 0 China, Hong Kong HKG 20.959 1.248 1.333 1.313 NaN NaN NaN 22.70 0 0 0 0 Colombia COL 7.010-0.057-0.783-0.490 15.574-1.111 0.254 4.79 0 0 0 0 i

Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil 1990-1999 Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Comoros COM 2.226 NaN NaN NaN 10.629-0.753 NaN -11.67 0 0 0 0 Congo (Dem. Rep.) ZAR 1.449-1.769-2.153-1.556 15.139-1.160 0.084 0.00 0 0 1 0 Congo COG 2.331-0.580-1.435-0.596 13.540-1.676 0.084-3.68 0 0 0 1 Costa Rica CRI 7.284 0.554 0.553 0.577 11.892-0.604 0.059 9.94 0 0 0 0 Cote d'ivoire CIV 2.278-0.180-0.335-0.079 14.040-1.042 0.141 5.50 0 0 0 0 Croatia HRV 5.960 0.150 0.146-0.464 13.257-1.159 0.058 45.17 0 1 0 0 Cyprus CYP 12.090 1.041 0.928 1.811 11.442-0.874 0.034 35.08 0 0 0 0 Czech Republic CZE 12.103 0.595 0.543 0.384 13.508-0.911 0.117 49.75 0 1 1 0 Denmark DNK 23.032 1.721 1.691 2.129 13.025-0.948 0.075 55.72 0 0 0 0 Djibouti DJI 3.389 NaN -0.235 NaN 12.356-2.454 0.082 11.51 0 0 0 0 Dominica DCA 4.415 NaN NaN NaN 9.453-0.932 NaN 15.43 0 0 0 0 Dominican Republic DOM 4.825-0.833 0.380-0.773 14.027-1.210 0.186 18.56 0 0 0 0 Ecuador ECU 5.383-0.562-0.721-0.819 14.490-1.196 0.169-2.06 0 0 0 0 Egypt EGY 3.390-0.138 0.128-0.267 15.242-1.015 0.535 30.00 0 0 0 0 El Salvador SLV 3.127-0.262-0.656-0.354 13.334-1.034 0.174 13.78 0 0 0 0 Equatorial Guinea GNQ 2.432 NaN -1.204 NaN 11.437-1.598 0.022 2.33 0 0 0 0 Estonia EST 7.083 0.258 0.507 0.593 12.567-1.241 0.046 58.69 0 1 0 0 Ethiopia ETH 0.591-0.146 0.269-0.436 13.460-0.926 0.088 9.01 0 0 0 0 Fiji FJI 4.399 0.635-0.495 0.807 12.515-1.793 0.044-17.83 0 0 0 0 Finland FIN 19.242 1.635 1.736 2.085 13.180-0.870 0.064 60.21 0 0 0 0 France FRA 20.704 1.280 1.077 1.282 13.902-0.671 0.229 48.86 0 0 0 0 Gabon GAB 8.658-1.127-0.525-1.015 12.491-1.331 0.066 0.37 0 0 0 1 Gambia GMB 1.765 0.164 0.274-0.019 12.164-1.253 0.068 13.26 0 0 0 0 Georgia GEO 4.474-0.512-0.494-0.744 13.642-1.237 0.096 42.03 1 0 0 0 Germany DEU 20.612 1.409 1.483 1.620 15.008-0.797 0.412 48.16 0 0 0 0 Ghana GHA 2.016-0.287-0.014-0.301 13.832-1.076 0.118 6.69 0 0 0 0 Greece GRC 13.027 0.560 0.496 0.825 14.148-1.081 0.274 38.06 0 0 0 0 Grenada GRE 5.732 NaN NaN NaN 8.566-0.858 NaN 12.12 0 0 0 0 Guatemala GTM 5.096-0.225-1.106-0.819 13.097-1.056 0.131 14.62 0 0 0 0 Guinea GIN 2.341-0.029-0.762-0.848 13.660-1.426 0.067 11.67 0 0 0 0 Guinea Bissau GNB 1.318-0.334-1.615-0.176 11.481-1.272 0.058 12.26 0 0 0 0 Guyana GUY 2.793 0.009-0.140-0.019 12.137-1.761 0.051 5.76 0 0 0 0 Haiti HTI 1.739-1.232-1.495-0.535 13.970-1.562 0.140 18.93 0 0 0 0 Honduras HND 2.721-0.409-0.895-0.938 13.777-1.381 0.103 14.19 0 0 0 0 Hungary HUN 9.327 0.606 0.706 0.614 13.506-0.903 0.121 47.42 0 1 1 0 Iceland ISL 23.866 1.504 1.469 1.831 11.349-1.321 0.035 63.89 0 0 0 0 ii

Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil 1990-1999 Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil India IND 1.663-0.264 0.160-0.306 16.735-0.868 0.347 25.27 0 0 0 0 Indonesia IDN 3.144-0.528-0.918-0.799 15.603-0.886 0.373-6.56 0 0 0 0 Iran IRN 6.756-0.339-0.364-0.848 15.428-0.944 0.173 35.38 0 0 0 1 Ireland IRL 18.173 1.361 1.395 1.567 12.716-1.088 0.068 54.61 0 0 0 0 Israel ISR 16.960 0.685 0.966 1.277 13.408-0.742 0.315 32.08 0 0 0 0 Italy ITA 19.865 0.773 0.861 0.802 14.398-0.742 0.220 45.42 0 0 0 0 Jamaica JAM 3.936-0.484-0.728-0.116 13.238-1.608 0.093 18.06 0 0 0 0 Japan JPN 22.155 0.839 1.422 0.724 15.924-0.816 0.692 35.71 0 0 0 0 Jordan JOR 4.207 0.630 0.708 0.139 13.902-1.268 0.199 31.60 0 0 0 0 Kazakhstan KAZ 5.578-0.824-0.590-0.869 13.968-0.836 0.058 44.31 1 0 1 1 Kenya KEN 1.562-0.899-1.220-0.651 14.396-1.277 0.131-0.51 0 0 0 0 Kiribati KIR 1.519 NaN NaN NaN 10.181-1.238 NaN 1.85 0 0 0 NaN Korea (South) KOR 10.947 0.409 0.943 0.159 16.384-1.255 0.813 37.55 0 0 0 0 Kuwait KWT 16.436-0.063 0.907 0.619 12.033-0.581 0.107 29.33 0 0 0 1 Kyrgyzstan KGZ 3.046-0.575-0.468-0.763 12.865-1.016 0.063 41.00 1 0 1 0 Laos LAO 1.482 NaN -1.204 NaN 13.066-1.318 0.052 16.55 0 0 1 0 Latvia LVA 5.860 0.068 0.155-0.264 12.864-1.178 0.068 56.86 0 1 0 0 Lebanon LBN 5.815 0.174 0.262-0.397 13.920-1.669 0.145 34.11 0 0 0 0 Lesotho LSO 1.829-0.462-0.240 0.188 11.755-1.273 0.043-29.60 0 0 1 0 Libya LBY 17.204-1.322-1.113-0.882 13.865-0.942 0.104 32.61 0 0 0 1 Lithuania LTU 6.642 0.127 0.180 0.034 13.465-1.209 0.081 55.32 0 1 0 0 Luxembourg LUX 30.155 1.674 1.621 1.671 11.021-0.884 0.029 49.78 0 0 1 0 Macedonia MKD 4.821-0.576-0.256-0.517 12.788-1.126 0.084 41.83 0 1 1 0 Madagascar MDG 1.220-0.295-0.825-0.469 13.226-1.054 0.055-18.96 0 0 0 0 Malawi MWI 0.875-0.625-0.409-0.195 13.161-1.386 0.068-15.81 0 0 1 0 Malaysia MYS 7.446 0.714 0.834 0.633 14.284-0.804 0.229 3.27 0 0 0 0 Maldives MDV 4.377 NaN NaN NaN 9.869-0.751 0.027 16.73 0 0 0 0 Mali MLI 0.965-0.052-0.465-0.476 13.229-1.115 0.055 12.51 0 0 1 0 Malta MLT 12.077 0.629 0.864 0.497 10.120-0.390 NaN 35.89 0 0 0 0 Mauritania MRT 2.103 NaN -0.558 NaN 12.683-1.303 0.040 17.93 0 0 0 0 Mauritius MUS 8.629 0.172 1.279 0.336 12.078-0.967 0.092-20.23 0 0 0 0 Mexico MEX 8.136 0.179-0.474-0.277 16.122-0.992 0.355 16.76 0 0 0 0 Moldova MDA 2.982-0.460-0.019-0.387 13.227-1.225 0.092 47.17 1 0 1 0 Mongolia MNG 1.585 0.018 0.039-0.145 12.390-1.013 0.055 47.49 0 0 1 0 Morocco MAR 3.671 0.267 0.678 0.125 15.100-1.173 0.188 33.59 0 0 0 0 Mozambique MOZ 0.887-0.331-1.046-0.535 13.866-0.961 0.066-18.50 0 0 0 0 iii

Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil 1990-1999 Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Myanmar MMR 1.057-1.461-0.839-1.096 14.370-1.020 0.142 17.68 0 0 0 0 Namibia NAM 6.094 0.044 0.954 0.382 11.853-1.006 0.047-17.98 0 0 0 0 Nepal NPL 1.274 NaN -0.558 NaN 13.136-0.856 0.090 27.71 0 0 1 0 Netherlands NLD 20.080 2.030 1.584 2.026 13.455-0.624 0.424 51.87 0 0 0 0 Netherlands Antilles NEA 14.725 NaN NaN NaN 10.847-2.210 0.023 12.19 0 0 0 NaN New Zealand NZL 16.744 1.571 1.824 2.075 14.857-1.642 0.081-36.89 0 0 0 0 Nicaragua NIC 2.507-0.547-0.726-0.836 13.165-1.017 0.123 12.21 0 0 0 0 Niger NER 1.188-1.387-1.144-1.567 13.192-1.176 0.048 13.88 0 0 1 0 Nigeria NGA 1.141-1.321-1.097-0.954 15.789-0.974 0.387 6.54 0 0 0 1 Norway NOR 24.645 1.666 1.833 1.687 13.229-1.113 0.057 59.98 0 0 0 1 Oman OMN 12.882 0.900 1.077 0.484 12.566-0.628 0.103 20.45 0 0 0 1 Pakistan PAK 1.892-0.744-0.760-0.769 15.715-1.043 0.334 31.17 0 0 0 0 Panama PAN 7.415-0.277-0.392-0.458 13.171-1.400 0.089 9.21 0 0 0 0 Papua New Guinea PNG 2.725-0.694-0.307-0.854 12.103-1.161 0.048-6.60 0 0 0 0 Paraguay PRY 4.328-1.100-0.695-0.958 13.981-1.349 0.137-25.58 0 0 1 0 Peru PER 4.651 0.173-0.522-0.200 14.926-1.158 0.107-11.79 0 0 0 0 Philippines PHL 3.830 0.126-0.078-0.228 14.469-0.780 0.462 13.92 0 0 0 0 Poland POL 6.913 0.674 0.538 0.492 14.447-0.794 0.274 50.24 0 1 0 0 Portugal PRT 13.587 1.151 1.083 1.218 13.022-0.800 0.139 38.82 0 0 0 0 Qatar QAT 17.886 0.480 1.269 0.570 12.789-2.092 0.076 25.31 0 0 0 1 Romania ROM 6.020-0.570-0.088-0.457 14.050-0.782 0.160 44.53 0 1 0 0 Russia RUS 7.895-0.595-0.722-0.616 16.248-0.927 0.118 55.68 1 0 0 1 Rwanda RWA 0.968 NaN -1.204 NaN 12.250-1.347 0.067-2.03 0 0 1 0 Samoa SAM 3.939 NaN NaN NaN 9.424-1.305 NaN -13.63 0 0 0 0 Sao Tome STP 1.411 NaN NaN NaN 10.903-1.688 0.023 43.94 0 0 0 NaN Saudi Arabia SAU 12.728-0.349 0.494-0.575 15.393-1.246 0.118 23.07 0 0 0 1 Senegal SEN 2.173 0.047-0.097-0.235 14.074-1.369 0.115 14.77 0 0 0 0 Seychelles SEY 10.488 NaN NaN NaN 9.974-1.632 NaN -4.66 0 0 0 0 Sierra Leone SLE 0.702 0.009-0.906-0.019 13.219-1.462 0.074 8.70 0 0 0 0 Singapore SGP 22.410 2.082 1.939 1.948 15.107 NaN 0.309 1.36 0 0 0 0 Slovakia SVK 8.710-0.032 0.134 0.030 12.733-0.788 0.078 48.67 0 1 1 0 Slovenia SVN 12.342 0.567 0.825 1.023 12.178-1.103 0.053 46.00 0 1 0 0 Solomon Islands SLB 2.903 NaN NaN NaN 10.345-1.344 0.022-9.63 0 0 0 0 South Africa ZAF 8.110-0.010-0.351 0.299 15.212-1.018 0.270-29.13 0 0 0 0 Spain ESP 15.366 1.603 1.032 1.214 14.703-0.803 0.203 37.40 0 0 0 0 Sri Lanka LKA 2.321-0.612-0.361-0.124 13.418-0.902 0.173 6.87 0 0 0 0 iv

Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil 1990-1999 Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Saint Kitts SKN 8.491 NaN NaN NaN 8.776-0.986 NaN 17.33 0 0 0 0 Saint Lucia SLU 6.272 NaN NaN NaN 10.761-2.039 0.024 13.90 0 0 0 0 Saint Vincent SVG 4.175 NaN NaN NaN 9.189-1.485 NaN 13.25 0 0 0 0 Sudan SDN 1.792-1.697-1.346-1.015 14.700-1.138 0.096 14.04 0 0 0 0 Suriname SUR 4.813-0.146-0.730-0.019 11.660-2.181 0.050 5.61 0 0 0 0 Swaziland SWZ 4.168-0.474-0.062 0.007 11.098-1.243 0.027-26.55 0 0 1 0 Sweden SWE 21.117 1.573 1.623 2.085 13.389-0.886 0.080 59.28 0 0 0 0 Switzerland CHE 25.011 1.986 1.996 2.072 12.462-0.709 0.078 47.41 0 0 1 0 Syria SYR 3.905-1.181-0.291-0.789 14.734-1.246 0.163 33.46 0 0 0 0 Taiwan TWN 14.358 1.294 0.928 0.626 14.890-0.946 0.650 23.50 0 0 0 0 Tajikistan TJK 1.756-1.423-1.335-1.316 12.968-1.076 0.076 37.81 1 0 1 0 Tanzania TZA 0.715-0.485 0.161-0.924 13.827-0.859 0.077-2.15 0 0 0 0 Thailand THA 5.817 0.010 0.413-0.165 13.912-0.804 0.224 13.77 0 0 0 0 Togo TGO 1.855-0.374-0.799-0.242 12.598-1.058 0.070 6.19 0 0 0 0 Tonga TGA 4.399 NaN NaN NaN 9.806-1.289 NaN -21.17 0 0 0 0 Trinidad & Tobago TTO 7.009 0.521 0.514 0.511 11.351-1.539 0.023 10.42 0 0 0 1 Tunisia TUN 5.205 0.633 0.648 0.020 13.088-0.722 0.139 36.82 0 0 0 0 Turkey TUR 5.918-0.412-0.010-0.349 15.416-0.945 0.199 41.20 0 0 0 0 Turkmenistan TKM 7.542-1.252-0.971-1.289 13.024-0.960 0.048 40.00 1 0 1 1 Uganda UGA 1.441-0.251-0.013-0.466 13.114-1.116 0.066 0.23 0 0 1 0 Ukraine UKR 4.851-0.893-0.707-0.892 15.286-0.957 0.190 50.28 1 0 0 0 United Arab Emira. ARE 18.612 0.138 0.767-0.027 14.217-1.650 0.137 23.39 0 0 0 1 United Kingdom GBR 19.776 1.966 1.689 1.707 14.410-0.692 0.453 51.51 0 0 0 0 United States USA 28.562 1.366 1.254 1.407 15.544-0.728 0.144 34.36 0 0 0 0 Uruguay URY 9.020 0.618 0.270 0.430 12.909-1.000 0.063-34.82 0 0 0 0 Uzbekistan UZB 2.620-1.305-0.870-0.963 14.135-0.935 0.114 41.27 1 0 1 0 Vanuatu VAN 4.274 NaN NaN NaN 10.250-1.947 NaN -15.23 0 0 0 0 Venezuela VEN 9.585-0.849-0.662-0.725 14.895-0.930 0.330 9.84 0 0 0 1 Vietnam VNM 1.627-0.300-0.437-0.332 14.812-1.094 0.164 10.80 0 0 0 0 Yemen YEM 1.113-0.621-1.008-0.854 14.516-1.509 0.113 15.23 0 0 0 0 Zambia ZMB 1.344-0.399-0.402-0.614 14.092-1.323 0.098-12.94 0 0 1 0 Zimbabwe ZWE 2.834-1.129-0.146-0.319 14.434-1.532 0.108-17.88 0 0 1 0 v