economic growth and institutional variables

8/12/2019 Economic Growth and Institutional Variables

1/22

The Institutional Foundationsof Inequality and Growth

LEWIS DAVIS* & MARK HOPKINS***Union College, Schenectady, New York, USA, **Gettysburg College, Pennsylvania, USA

Final version received June 2010

ABSTRACT After a decade of research, the effect of inequality on long-run economic growthremains unresolved, in part because researchers have treated omitted variable bias as anestimation problem rather than a deeper question of causality. In this article we argue that thekey omitted variable is the quality of economic institutions. Using both cross-country and paneldata specifications, we find no direct effect of inequality on growth in the long-run. Rather, the

protection of property rights simultaneously raises growth rates and reduces income inequality.We interpret these findings as evidence that insecure property rights disproportionatelydisadvantage the poor.

1. Introduction

After more than a decade of empirical work, there is little consensus regarding the

impact of income inequality on economic growth. Early investigations by Alesina

and Rodrik (1994) and Persson and Tabellini (1994) using cross-country growth

regressions concluded that initial income inequality is associated negatively with

future growth. Later researchers argued that these estimates may be subject to bias

from omitted variables. Li and Zou (1998) and Forbes (2000) use panel datatechniques to control for omitted variables, and find that income inequality is good

for growth.

While it effectively controls for omitted variable bias, the second wave of empirical

work on inequality and growth is subject to two criticisms. First, because they

typically use variables measured over five-year periods, panel techniques generate

estimates that capture the relatively high-frequency co-movements of growth and

inequality. In contrast, the theoretical literature on growth and inequality focuses on

mechanisms that may not fully manifest themselves over such short time horizons.

Correspondence Address: Lewis Davis, Department of Economics, 807 Union Street, Schenectady,

Journal of Development Studies,

Vol. 47, No. 7, 977997, July 2011


2/22

For example, Galor and Zeira (1993), Piketty (1997) and Aghion and Bolton (1997)

argue that, in the presence of capital market imperfections, income and wealth

inequality increases the share of potentially profitable investments that are

unrealised due to credit constraints. The impact of credit constraints is likely to be

particularly strong with respect to investments in human capital, since human capital

cannot serve as collateral. But human capital investments also have a long gestation

period and may generate significant returns that are not fully captured within a given

five-year period.

A similarly long time horizon may apply to political economy mechanisms linking

growth and inequality. Alesina and Rodrik (1994) and Persson and Tabellini (1994)

argue a political economy argument in which inequality increases the demand for

redistributional policies that blunt the incentive to invest. But in the presence of

significant rigidities in the political system, the full impact of a change in inequality

on policy outcomes may not be realised in a given five-year period.1 The mismatch

between the time horizons of the theoretical and empirical literature leads to thesuspicion that panel data estimates may not be that informative about the

relationship between inequality and long run economic growth.

Second, the second wave regressions rely on panel estimators that primarily exploit

the time variation in growth and inequality within countries. However, as first

pointed out by Li et al. (1998), most of the variation in income inequality is across

countries rather than time. As a result, panel estimators that rely on the time variation

in inequality are highly inefficient. However, the use of more efficient random effects

(GLS) panel estimators is typically found to be misspecified under the strong

assumption that the country effects are uncorrelated with the included variables.This article presents a unified view of the relationship between growth and

inequality that addresses these weaknesses and resolves the apparent puzzle in the

existing empirical literature. We begin by noting that measures of growth and

inequality are derived from moments of the same, evolving distribution of individual

incomes, as pointed out by Lundberg and Squire (2003), and that as a result it makes

sense to think of growth and inequality as being generated by a common set of

processes. From this perspective, the question that has motivated much of the

research in this area Persson and Tabellinis (1994: 600) query Is inequality

harmful to growth? is itself misleading. We believe a more revealing question isthat pursued in this article: What are the determinants of inequality and how are

they related to economic growth?

Following the theoretical contribution of Davis (2007), we argue that economic

institutions related to the protection of property rights are a fundamental

determinant of income inequality. Indeed, in the most influential work on

institutions and growth, Acemoglu et al.s (2001, 2002) and Engerman and

Sokoloffs (1997, 2002), the authors clearly indicate that their understanding of

institutional quality is fundamentally about the distribution,rather than the average

level, of property rights protection. While the indices used to measure institutional

quality empirically are not well designed to make this distinction, in many cases what

has been referred to as institutional quality may to a large degree reflect institutional

978 L. Davis & M. Hopkins


3/22

Given a large body of evidence that institutions affect economic growth, our

finding that institutions also affect inequality suggests that the omission of

institutional variables may be responsible for the negative relationship between

growth and inequality reported in cross-country regressions.2 In Section 3 we re-

investigate the cross-country relationship between inequality and growth. We find

that when both institutions and inequality are included in a growth regression,

institutions have a statistically significant impact on growth, while inequality does

not. We extend this analysis to address the panel data evidence on inequality and

growth in Section 4, showing that institutional quality explains much of the country

specific intercepts generated by a canonical fixed effects regression. Using this

information, we construct an efficient GLS estimator that allows us to identify

separately the long run, between-country and short run, within-country effects of

inequality on growth. Consistent with earlier evidence from fixed effects models, we

find that short run changes in inequality over time are positively associated with

growth within countries. The variation in long run inequality levels across countriesis not a significant determinant of growth, however, providing the most direct

refutation yet of the claims that inequality directly slows economic growth.

Our research is closely related to the third wave empirical literature that challenges

the conclusions of the second wave panel data estimates on methodological grounds.

Like us, Banerjee and Duflo (2003) motivate their work by noting the conflict

between the cross sectional and fixed effects estimates of the inequality on economic

growth. Barro (2000) uses a random effects estimator to consider the effect of

inequality on growth. However, unlike us, Barro does not conduct a Hausman

specification test, leaving open the possibility that his coefficient estimates are biased.The argument that both growth and inequality should be treated as endogenous

motivates the empirical specification employed by Lundberg and Squire (2003). Our

research is distinguished from this work in two ways. First, none of these papers

considers a fundamental role for institutions in determining income inequality, and

second, none addresses the long run relationship between growth and inequality that

is our focus here.

The debate over the relationship between inequality and growth touches on the

compatibility of two fundamental development goals, increasing economic growth

and achieving a more equitable distribution of income. While the first wave empiricalliterature finds these goals are compatible, the second wave suggests instead that

advances on one margin will inevitably require sacrifices on the other. The results

presented here suggest a third possibility, that whether or not a trade-off between

growth and equality exists may depend on the particular policy instrument in

question. In particular, our results suggest that reforms that improve the security of

property rights for the poor will tend to advance both development goals, increasing

economic growth while reducing income inequality.

2. The Institutional Foundations of Inequality

Over the last decade, a number of authors have presented empirical evidence that a

Inequality and Growth 979


4/22

protections. This is explicit with measures of political institutions, such as democracy,

that intentionally measure the equality of political power and participation. But

measures of the quality of economic institutions also contain a distributional

dimension, as the protection of property rights implicitly depends on the equality of

individuals before the law. Moreover, where private property rights are not well

protected through public institutions, agents will use private resources to protect their

property. In this situation, there will be unequal protection of property reflecting the

inequality of wealth and political power which agents have at their disposal.

Though largely overlooked, our emphasis on the distributional dimension of

institutional quality, and our claim that low quality institutions are associated with

the inequality of economic and political rights, finds support from some of the most

influential work on institutions and development. For example, Acemoglu et al.

(2001) argue that in colonies where European mortality rates were high, settlers

adopted extractive institutions that tended to retard development. In Acemoglu

et al. (2002: 1235), they clearly identify extractive institutions with the unequalprotection of property rights, defining extractive institutions as those that

concentrate power in the hands of a small elite and create a high risk of

expropriation for the majority of the population, [and thus] are likely to discourage

investment and economic development.

Engerman and Sokoloffs (1997, 2000, 2002) influential work on comparative

American development also highlights the distributional dimension of weak

institutions. For example, Sokoloff and Engerman (2000: 221) clearly identify Latin

American underdevelopment with institutions that protected the privileges of the

elites and restricted opportunities for the broad mass of the population to participatefully in the commercial economy. From this perspective, high levels of inequality are

not simply an unintended consequence of weak institutions and poorly protected

property rights. Rather, they reflect a deliberate attempt by colonial elites to

maintain high disparities of economic and political power.

Finally, we note that our emphasis on the institutional foundations of inequality is

consistent with existing empirical on the determinants of income inequality. Li et al.

(1998) and Lundberg and Squire (2003) find that the cross-country pattern of income

inequality is explained by levels of financial development, education, land inequality

and civil liberties, a result they interpret as support for theories that emphasise capitalmarket imperfections and political economy mechanisms. However, as noted by

Engerman and Sokoloff (1997), while these variables may be proximate determinants

of income inequality, they may also be viewed as the expression of deeper institutional

structures that manifest themselves through their influence on land tenure and

settlement, the provision of public education, the regulation of financial organisa-

tions, and restrictions on political participation. That is, they are all reflections of

institutions designed to restrict the economic and political power of the poor.

3. Data

While there are numerous empirical measures of institutional quality, none that we



5/22

and development, choosing one variable each as measures of the quality of economic

and political institutions. Our measure of democracy is the polity 2 variable from the

Polity IV database described in Marshall and Jaggers (2006). Our measure of the quality

of economic institutions is the Freedom from Expropriation variable constructed by

the International Country Risk Guide (ICRG). This variable has been used as a

measure of property rights protection in a number of studies, including Knack and

Keefer (1995) and Acemoglu et al. (2001). As these scores are only available for the

second half of our sample period, we use a single observation for each country

representing the average expropriation risk in the country over all years.

We measure income inequality using a cross-country panel of Gini coefficients

compiled by Milanovic (2006). To facilitate transparency, we use those observations

suggested by Milanovic (2006) and then adjust Gini coefficients for remaining

differences in survey sources and methodologies using a hedonic regression on survey

type, resulting in an unbalanced panel of observations over 6 five-year periods, from

1961 through 2000.3 For consistency, our cross sectional regressions employ theaverage Gini coefficient for each country, as reported in the panel. As noted by Li

et al. (1998), inequality levels are highly stable over time, so that the use of an

average, rather than an initial or final level of inequality, is not important to our

results.

In measuring our controls we depart slightly from the data sources and definitions

used by Li et al. (1998). Following Sokoloff and Engerman (2000), we use primary

rather than secondary school enrollment rates as a measure of access to human

capital. For land inequality, we use a more recent set of estimates compiled by

Frankema (2005). Similarly, we use the ratio of private credit to GDP from Becket al. (2000) to measure financial development rather than the ratio of M2 to GDP.

The differences in variable definitions are not essential to our results.

Results

Column (1) of Table 1 shows that our baseline regression by itself explains over 40

per cent of the cross-country variation in inequality. The variables all have the

expected sign, and two of the four, land inequality and political freedom, are

statistically significant at the 5 per cent level. White heteroskedasticity-correctedstandard errors are reported and used for inference throughout. As seen in column 2,

including our proxy for the quality of economic institutions significantly increases

the fit of the regression, and the coefficient on freedom from expropriation is large

and statistically significant at the 1 per cent level. The partial correlation, shown

in Figure 1, implies that a one-standard deviation increase in freedom from

expropriation (1.8 on a 10 point scale) corresponds to a 7 point decrease in the

average level of the Gini coefficient, or roughly of a standard deviation.

Columns (3) through (6) pair freedom from expropriation risk with each variable

individually in none of these regressions are the results qualitatively different from

those reported in column (2). While economic institutions clearly have a strong

association with inequality, with the exception of land inequality, none of the



6/22

Table1.Institutio

nsandinequality(dependen

tvariableisaverageincome

Gini,19602000)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ment,

70.0729

(0.061)

70.0471

(0.052)

0.0289

(0.042)

redit

76.832

(4.17)

5.579

(3.95)

1.705

(3.98)

ni

17.17**

(8.33)

18.70***

(4.47)

10

.86*

(5

.63)

11.73*

(6.03)

12.22**

(6.03)

5.728(4.66)

8.442*

(4.96)

cy,

70.395**

(0.16)

70.125

(0.15)

70.0906

(0.14)

ation

73.844***

(0.75)

73.801***

(0.37)

74.378***

(0.64)

73

.456***

(0

.42)

73.674***

(0.49)

73.095***

(0.77)

72.585***

(0.76)

71.623

***

(0.53)

71.516**

(0.74)

al

2.854

(27.1)

al d

70.220

(1.60)

sector

0.134

(0.089)

5.896

***

(1.92)

from

r

70.106

**

(0.043

)

ed

2.291(2.17)

N

N

N

N

N

N

N

N

N

Y

43.69***

(6.40)

63.22***

(6.99)

69.68***

(4.26)

76.08***

(3.84)

62

.97***

(6

.24)

72.08***

(3.61)

51.45

(114)

51.31***

(9.70)

51.47*

**

(5.68)

44.46***

(7.05)

ons

59

53

70

58

70

74

7

0

56

69

70

d

0.43

0.67

0.48

0.60

0

.48

0.50

0.48

0.54

0.64

0.71

obuststandarderrorsinparenthese

s.***p5

0.01,**p5

0.05,*p5

0.1.

nclude

EastAsiaPacific,SouthAs

ia,LatinAmericaandCarib

bean,Sub-SaharanAfrica,a

ndMiddleEastandNorthAfrica.



7/22

clearest, most direct impact on income inequality, we retain only land inequality in

further robustness checks of the role of economic institutions.

Because of model uncertainty regarding the proper specification of the inequality

regression, we include additional control variables in columns (7) through (10). To

control for the presence of a Kuznets-curve type pattern, found in cross sectional

data by a number of authors including Li et al. (1998) and Barro (2000), we include

the log of per capita income in 1970 and its square in column (7). Neither income

variable is statistically significant, while expropriation risk remains strongly so. In

column (8) we introduce a measure of the informal sectors share of output,constructed by Friedman et al. (2000), motivated by several papers that have

suggested a link between the share of the informal sector of the economy and income

inequality (for example, Rosser et al., 2000; Davis, 2007).4 The inclusion of

informality results in a lower estimate for the magnitude of the coefficient on

freedom from expropriation, ostensibly because part of the reason expropriation risk

increases inequality is because it discourages participation in the formal sector. The

coefficient on expropriation risk retains its statistical significance, however,

suggesting that it may play a fundamental role in shaping the income distribution

beyond simply creating a barrier to formality.

Omitted geographic and regional characteristics may also bias our estimates.

Geography has been linked to institutional quality by Hall and Jones (1999), and is

Figure 1. The partial relationship of expropriation risk and inequality.



8/22

geographic and regional variables reduces the magnitude of the coefficient on

freedom from expropriation by nearly one half, but it remains statistically significant

at the 5 per cent level or better.5 These results suggest that freedom from

expropriation is not simply a proxy for omitted regional or geographic variables.

Overall, the results presented in Table 1 are consistent with the hypothesis that the

quality of economic institutions plays an important key role in determining the level

of income inequality across countries.

Addressing the Endogeneity of Institutional Quality

Conclusions based on the regression results reported above are subject to three

criticisms related to the treatment of institutional measures as exogenous regressors.

The first, articulated well by Acemoglu et al. (2001), is that the institutional variables

are derived from expert opinion and survey data, and are thus potentially subject to

systematic measurement error. This will occur, for example, if experts tend to seebetter institutions in countries that experience higher growth rates or less income

inequality. The second is reverse causation: a number of papers argue that it is

income inequality that reduces property rights protection, not vice versa (Keefer and

Knack, 2002; Glaeser et al., 2003). Finally, as stated earlier, these regressions are

intentionally parsimonious and the omission of variables that are simultaneously

correlated with institutions and inequality could bias our coefficient estimates.

We address these issues by instrumenting for our key institutional variable, freedom

from expropriation, to avoid bias introduced by the potential correlation between it and

the unobserved determinants of income inequality. In doing so, we rely on instrumentsfor contemporary institutions that have been previously used in the empirical growth

literature. This literature relies heavily on the argument that colonisation resulted in

generating an exogenous shock that influenced the path of institutional development.

The instruments we use are language and geography variables from Hall and Jones

(1999), legal heritage from Beck et al. (2000) and setter mortality rates from Acemoglu

et al. (2001). Many readers will be familiar with these instruments and the arguments

made for why they are likely to be uncorrelated with current income. Our arguments for

why these instruments for the first moment of the income distribution can also plausibly

be considered uncorrelated with higher moments are sufficiently similar that we omitthem here and refer interested readers to the Online Appendix.

Table 2 presents two stage least squares estimates that confirm our earlier finding

that the protection of property rights has a strong, statistically significant and robust

effect on income inequality. For each specification, the second stage regression of

average inequality on predicted institutional protections is reported in the first

column (A), with results from the first stage displayed in the second column (B).

Using a variety of instruments and controls in columns (1) through (5), we

consistently find that freedom from expropriation is significant at the 1 per cent level

with a coefficient of similar magnitude to that reported in Table 1.

Below each specification we report the p-value from Hansens J statistic and in

each case a test of overidentifying restrictions (OIR) fails to reject the joint null



9/22

Tab

le2.Institutionsandinequalityusingtwo-stageleastsqu

ares

(1)

(2)

(3)

(4)

(5)

A

B

A

B

A

B

A

B

A

B

t

avgini

exprop

avgini

exprop

avgini

exprop

av

gini

exprop

avgin

i

exprop

tionrisk

73.695***

(0.41)

74.584***

(0.65)

73.616***

(0.42)

72.3

24***

(0.4

8)

73.171***

(0.75)

14.53***

(5.10)

70.269

(1.28)

15.07**

(6.20)

1.245

(0.80)

15.90***

(4.78)

70.145

(1.17)

10.4

9*

(5.3

2)

0.879

(1.23)

17.11***

(6.01)

71.507

(1.13)

language

0.862**

(0.33)

0.290

(0.29)

0.885**

(0.34)

1.346***

(0.37)

0.906**

(0.41)

rom

70.0123

(0.033)

0.0160

(0.027)

70.00186

(0.029)

70.0509

(0.031)

70.0983*

(0.051)

rom

quared

0.00129

**

(0.00054

)

0.000704

(0.00042)

0.00115**

(0.00050)

0.00155***

(0.00049)

0.00245**

(0.0012)

alorigin

70.988**

*

(0.28)

70.618**

(0.24)

70.971***

(0.28)

70.618**

(0.26)

70.711**

(0.27)

dit

5.296

(3.70)

1.843***

(0.44)

0.0242

(0.023)

0.00381

(0.0063)

anAfrica

8.9

23***

(3.3

5)

71.158***

(0.41)

erica&

n

6.2

26***

(2.1

0)

71.809***

(0.42)

morta

lity

70.608***

(0.15)

ons

64

64

55

55

64

64

64

64

44

44

0.56

0.66

0.64

0.79

0.57

0.66

0.6

8

0.75

0.35

0.64

HansensJ

0.269

0.615

0.391

0.8

91

0.550

naldF-

stat

26.34

16.03

26.08

17.5

6

12.94

luefor

ximal

ias

10.27

10.27

10.27

10.2

7

10.83

buststa

ndarderrorsinparentheses.***p5

0.01,**p5

0.05,*p5

0.1.

B-columnsreportfirst-stageregressionresults.



10/22

the reported Cragg-Donald F-statistic and the critical values calculated by Stock and

Yogo (2005), which vary by number of instruments. In every regression we are able

to reject the null hypothesis that the bias of the IV estimator exceeds the bias of the

OLS estimator by more than 10 per cent. These diagnostics provide us some

confidence that our instruments are neither invalid nor weak.

A potential criticism of the model estimated in column (1) is that our instruments

might not properly be excluded from the second stage regression, influencing

inequality instead through alternate channels. Beck et al. (2004) have argued, for

example, that legal heritage influences inequality through financial sector regulation

and development. A similar criticism is that the share of a population that speaks a

European language might be directly related to income distribution through

channels related to international trade and capital flows. To control for these effects

we include the ratio of private credit to GDP to proxy for financial development in

column (2) and a measure of openness in column (3). In both cases the coefficient on

freedom from expropriation remains significant while neither private credit nor tradeis significant as a determinant of either inequality or expropriation risk.

To control for other potential omitted factors, dummy variables are introduced in

the specification in column (4) indicating whether a country is in either Latin

America and the Caribbean or in Sub-Saharan Africa, both of which have a

statistically significant association with both inequality and expropriation risk.

Adding these controls reduces the economic significance of freedom from

expropriation on inequality, but the coefficient remains strongly significant. These

results suggest that the correlation between inequality and expropriation risk is

somewhat stronger across than within regions, but omitted regional variables cannotfully account for the influence of institutions on inequality.

In the fifth column, we include European settler mortality rate as an additional

instrument. This lowers sample size significantly, but Acemoglu et al. (2001) present

strong arguments for the relevance and exogeneity of this variable as an instrument for

expropriation risk, so we consider it useful for assessing the validity of our other

instruments. With the expanded instrument set, the coefficient on freedom from

expropriation is slightly reduced in magnitude but it remains significant at the 1 per cent

level and the tests statistics for instrument validity and weakness both remain above

their critical values.7

Thus we believe our choice of excluded instruments is reasonableand, more importantly, that our conclusions regarding the importance of economic

institutions for determining inequality are not sensitive to a specific set of instruments.

4. Re-evaluating the Cross Country Evidence on Growth and Inequality

The fact that stronger property rights are associated with lower inequality has

important implications for the empirical literature on inequality and growth. Since

expropriation risk has been shown in previous literature to be an important

determinant of economic growth, regressions that include inequality but not

expropriation risk will generate negatively biased estimates of the influence of

inequality on growth.



11/22

the growth and inequality literature (for example, Alesina and Rodrik, 1994; Persson

and Tabellini, 1994). The dependent variable is the average growth rate of real per

capita income (PPP) from 1970 to 1995, a period that maximised sample size. This is

regressed on average income inequality and several control variables, including

initial income, to capture conditional convergence effects, and the primary

enrollment rate in 1970, a measure of human capital investment used by Alesina

and Rodrik (1994). To avoid potential endogeneity problems between investment

and growth, we do not include the average investment rate directly as a growth

regressor but use instead two variables that influence investment decisions, the

domestic price of investment goods, averaged from 1970 to 1990, and private credit

relative to national income, a measure of financial development.

The results from our benchmark regression, reported in column (1) of Table 3,

confirm the central finding of the first wave of empirical work on growth and

inequality: the coefficient on inequality is large, negative and highly significant, in

this case at the 1 per cent level. With the exception of the price of investment goods,all of our control variables have the expected sign and are significant as well. In

column (2), we introduce our political and economic institutional variables to the

regression. A common hypothesis in the literature is that democracy has a nonlinear

effect on growth, positive at low levels and negative at high levels, so we allow for a

nonlinearity by including a quadratic term for political rights as well. Overall,

political freedoms appear to be negatively associated with growth, although this

effect is not statistically significant in most of the specifications we consider.

Freedom from expropriation has a positive effect on growth that is both

statistically and economically significant. Specifically, in column (2), an increase infreedom from expropriation of one standard deviation (1.83 out of 10 points, or

roughly the difference between Panama at 5.66 and Chile at 7.5) is associated with a

1.43 percentage point increase in the average annual growth rate. In addition,

introducing expropriation risk dramatically reduces the measured effect of inequal-

ity, which is no longer statistically significant. These results are consistent with the

argument above, that the omission of institutional variables introduced bias into

earlier cross-country regression estimates and led to mistaken inference regarding the

role of income inequality on growth.

Institutions, development and inequality all vary systematically with geographyand regional location, as countries in the tropics tend to suffer both from low levels

of property rights protection and from geographic attributes that negatively

influence economic growth through their effect on the disease burden, the fertility

of the land and average temperatures. To consider for this, column (3) adds two

geographic control variables, a dummy variable indicating landlocked countries and

the absolute distance from the equator, and country (4) includes regional dummies.

The inclusion of these geographic variables reduces the coefficient on freedom from

expropriation somewhat, but does not qualitatively alter our conclusions: good

economic institutions have a robust positive relationship with economic growth. In

contrast, we cannot reject the hypothesis that a countrys average level of inequality

does not affect its long run rate of growth.



12/22

Table3.The

roleofinstitutionsandinequalityincross-countrygrowthregressions

nt

Averageper-capitaincomegrowth,19707

1995

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

al

71.322***

(0.38)

71.695***

(0.31)

71.859***

(0.33)

71.835***

(0.36)

71.172***

(0.33)

71.691***

(0.31)

71.908***

(0.32)

71.809***

(0.36)

redit

2.669***

(0.72)

1.259**

(0.58)

1.405**

(0.58)

1.002

(0.66)

2.606***

(0.82)

1.029*

(0.61)

1.331*

(0.66)

0.938

(0.64)

ment

70.0149

(0.0090)

70.00186

(0.0071)

70.00244

(0.0072)

70.00423

(0.0056)

70.00644

(0.0095)

0.000650

(0.0078)

70.000979

(0.0079)

70.00160

(0.0050)

ment

0.0406***

(0.0093)

0.0398***

(0.0095)

0.0406***

(0.0099)

0.0220*

(0.012)

0.0501***

(0.013)

0.0439***

(0.013)

0.0413***

(0.013)

0.0220*

(0.013)

Gini

70.0737***

(0.024)

70.0394

(0.024)

70.0326

(0.026)

0.00784

(0.031)

ni

72.508

(1.51)

71.088

(1.17)

70.252

(1.36)

70.802

(1.21)

ation

0.784***

(0.19)

0.689***

(0.22)

0.786***

(0.28)

0.902***

(0.18)

0.796***

(0.22)

0.832***

(0.28)

cy

70.0706*

(0.038)

70.0658

(0.039)

70.0290

(0.040)

70.0664*

(0.035)

70.0595

(0.037)

70.0446

(0.039)

cy d

70.0121*

(0.0069)

70.0114

(0.0069)

70.00784

(0.0064)

70.0108

(0.0071)

70.0102

(0.0071)

70.00852

(0.0064)

hy

Yes

Yes

Yes

Yes

ons

63

55

55

55

59

53

53

53

d

0.50

0.70

0.71

0.79

0.42

0.68

0.70

0.80

red

0.45

0.65

0.65

0.72

0.37

0.63

0.63

0.73

obuststandarderrorsappearinparenthesesbelowcoefficientes

timates.Significancenoteda

s***p5

0.01,**p5

0.05,*

p5

0.1.

(3)and(7)includetwogeography

controls,adummyvariableforbeinglandlockedanddistancefromtheequator.Colu

mns(4)and(8)

egionaldummyvariablesascontrols.



13/22


14/22

T

able4.Cross-sectionalsimultaneousequationsregression

s

(1)

(2)

(3)

(4)

(5)

t:

growth

inequality

growth

inequality

growth

inequality

gro

wth

inequality

growth

inequality

71.656***

(0.34)

7

1.848***

(0.26)

72

.123***

(0

.35)

71.8

50***

(0.3

5)

72.252***

(0.42)

970

0.0443***

(0.0083)

70.0732

(0.051)

0.0410***

(0.0076)

70.0471

(0.046)

0

.0380***

(0

.0083)

70.0345

(0.055)

0.0

251**

(0.0

11)

0.0454

(0.056)

0.0411***

(0.012)

70.0438

(0.075)

ent

70.00843

(0.0074)

7

0.00412

(0.0070)

70

.00333

(0

.0081)

70.0

0494

(0.0

096)

70.00463

(0.0099)

2.490***

(0.71)

76.833*

(3.58)

1.757***

(0.64)

5.579

(3.42)

1

.063

(0

.86)

6.759

(4.34)

2.0

28**

(0.8

3)

7.812**

(3.67)

1.439

(2.05)

18.16*

(9.65)

ni

70.101**

(0.043)

7

0.0590

(0.054)

70

.0276

(0

.061)

70.1

15

(0.0

92)

70.0000875

(0.072)

17.21**

(7.20)

18.70***

(6.09)

18.31***

(6.80)

12.19*

(7.00)

24.57**

(10.2)

y

70.394***

(0.14)

7

0.0374

(0.024)

70.125

(0.13)

70

.0693*

(0

.038)

0.175

(0.20)

70.0

383

(0.0

34)

0.145

(0.18)

70.199**

(0.078)

0.327

(0.40)

tion

0.543**

(0.27)

73.844***

(0.74)

1

.040**

(0

.48)

74.912***

(1.27)

0.3

75

(0.3

7)

74.257***

(1.36)

1.713**

(0.79)

77.120**

(2.82)

71.1

91

(1.3

9)

10.85***

(3.51)

erica

0.5

90

(1.2

6)

4.666

(3.26)

15.62***

(4.01)

43.68***

(4.86)

11.68***

(4.20)

63.22***

(5.49)

9

.525*

(4

.89)

68.88***

(7.79)

16.7

8***

(6.1

6)

57.18***

(8.40)

4.662

(6.69)

76.75***

(15.7)

ns

59

59

53

53

52

52

52

52

35

35

0.55

0.43

0.66

0.67

0

.63

0.62

0.6

4

0.70

0.41

0.20

ents

None

None

Polity1960,French,

EuropeanLanguage,

distequat,dist2

Polity1960,French,

Eu

ropeanLanguage,

distequat,dist2

Polity1960,French,

mortalit

y,distequat,dist2

us

s

Growthandinequality

Growthandinequality

Growth,inequality,

polity1970,exprop

G

rowth,inequality,

polity1970,exprop

Grow

th,inequality,

polit

y1970,exprop

ndarderrorsinparentheses.***p5

0.0

1,**p5

0.05,*p5

0.1.



15/22

additional controls, however, one key finding is not. In column four, we include

regional dummy variables for Sub-Saharan Africa and Latin America and the

Caribbean, and find that freedom from expropriation is no longer significant in the

growth regression.8 This suggests that the coefficient estimate in column (3) reflects

interregional variations in institutions more than intra-regional variations. Overall,

however, these regressions confirm our hypothesis that economic institutions matter

for both growth and inequality, but inequality itself does not have an independent

effect on economic growth.

5. Identifying the Role of Institutions in Panel Data Fixed Effects

Clearly, we are not the first to claim that the estimates from cross-country

regressions might suffer from omitted variable bias. The standard approach to this

problem has been to alter the type of estimation being done: exploiting the time

variation in panel data to net out unobservable country specific, time invariantcharacteristics. An alternative method of controlling for intrinsic country specific

heterogeneity is through estimation of random effects, or country specific residuals.

The problem for Forbes (2000) and others is that a Hausman specification test rejects

the maintained assumption that the omitted time invariant variables are

uncorrelated with the included variables, suggesting that coefficients estimated using

random effects might not be consistent. Rather than opting for a less efficient

estimator, we address this dilemma by identifying the time invariant omitted

variables that cause the random effects estimator to be rejected in a Hausman

specification test.This approach has a number of advantages. First, parameter estimation using a

random effects estimator is efficient. The estimates generated by the random effects

estimator are, by construction, a weighted average of the fixed effects estimates

derived from period-to-period variations within countries and a between estimator

using cross-country averages. Second, the use of a random effects estimator allows us

to estimate the effect of relatively time invariant variables including many

institutional measures that cannot be identified within a fixed effects specification.

Third, the use of fixed effects estimator changes the question from the effect of

inequality on long run growth in aggregate supply to a much shorter run correlationbetween inequality and growth, typically over a series of five-year periods. By

employing a random effects estimator we are able to produce separate estimates for

the short run and long run effects of inequality on growth.

In particular, we replace the standard specification for panel data regressions,

Yit ai bXit gZi eit 1

with the more flexible specification,

Yit ai bW Xit Xi b

BXi gZi eit 2



16/22

data set. This approach has two advantages. First, the short run and long run effects

of inequality on growth can be estimated at the same time, within the same modelling

framework, using a random effects (GLS) estimator. Perhaps more importantly, it

allows us to test empirically whetherbW bB. That is, our approach allows us to testa key assumption that underlies much of the existing literature on inequality and

growth, that the short run relationship between inequality and growth provides

insight into the longer run relationship.

Random Effects Estimation

The evidence on inequality, institutions and growth from Sections 3 and 4 suggests

that institutions may be a primary source of the omitted variable bias generating

conflicting results between the first and second waves of growth and inequality

regressions. To test whether this is in fact the case we employ a series of panel

regressions reported in Table 5. Column (1) reports the Forbes specification with fixedeffects while column (2) reports the same estimates using a random effects estimator

and the specification in Equation (1). If the unobserved country specific effects are

uncorrelated with the regressors, both estimates will be consistent estimators of the

same quantities but the random effects estimator the GLS estimator will be more

efficient. A Hausman specification test strongly rejects this maintained hypothesis,

however, suggesting that the random effects estimator is inconsistent. This is the

conclusion reached by Forbes (2000) as well as other researchers, which has led to

popularity of less efficient fixed effects estimators in panel growth regressions.

In column (3), we adopt the specification in Equation (2), which allows us toidentify separately the short run and long run effects of inequality on economic

growth. The results in column (3) confirm that, as expected, the use of random effects

estimator in column (2) conflates very different short run and long run effects.

Specifically, the coefficient on inequality reported in column (2) appears not to have

been significant because it is constructed from a weighted average of a short run

positive relationship within countries and a long run negative relationship across

countries. As expected, the coefficient estimate on within country inequality in

column (3) is quite similar to that reported using fixed effects in column (1), while the

between country estimate is similar to that reported in column (1) of Table 3.In columns (4) and (5) we control for the quality of economic and political

institutions. Due to data limitations we continue to use a single time invariant

average score for expropriation risk, but we exploit the long, balanced time series in

the Polity IV database to include a time varying measure of political rights. Because

the uncertainty associated with changes in political institutions in any direction may

have a negative impact on growth, we control for this separately with a dummy

indicating periods involving a significant political transition (during which the Polity

IV database provides no scores). The results from columns (4) and (5) confirm our

earlier conclusions: the coefficient on freedom from expropriation is large, positive,

and statistically significant at the 1 per cent level. In contrast, the coefficient on the

quality of political institutions is not statistically significant, though political



17/22

Table5.Institutions,inequalityandgrow

thusingaGLSrandom-effe

ctsestimator

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

income

73.718***

(0.50)

70.489*

(0.28)

70.751***

(0.28)

71.402***

(0.33)

71.399***

(0.34)

71.874**

*

(0.34)

72.267***

(0.36)

71.275***

(0.31)

71.508***

(0.33)

econdary

g

0.976***

(0.28)

0.274

(0.22)

0.330

(0.21)

0.264

(0.22)

0.263

(0.22)

0.408*

(0.21)

0.515**

(0.21)

0.417*

(0.22)

0.416*

(0.22)

vestment

70.00751*

(0.0040)

70.0131

***

(0.0039

)

70.0113***

(0.0038)

70.0251***

(0.0054)

70.0246***

(0.0054)

70.0251*

**

(0.0053)

70.0216***

(0.0053)

70.0103**

(0.0044)

70.00951**

(0.0038)

ni

0.0857***

(0.029)

70.0285

(0.019)

ni

0.0904***

(0.030)

0.0663**

(0.031)

0.0670**

(0.030)

0.0633*

*

(0.030)

0.0690**

(0.030)

0.0788***

(0.030)

0.0863***

(0.029)

ni

)

70.0917***

(0.022)

70.0273

(0.028)

70.0307

(0.028)

70.0253

(0.028)

0.0188

(0.035)

70.0947***

(0.024)

70.0351

(0.032)

tionrisk

0.990***

(0.19)

0.969***

(0.19)

0.938**

*

(0.19)

1.202***

(0.21)

y

70.00133

(0.024)

ansition

71.612**

(0.81)

d

71.454**

(0.62)

71.474**

(0.60)

alisation

70.0245*

**

(0.0074)

70.0216***

(0.0078)

mmies

Included

F

43.4

Included

F

33.4

ffects

Fixed

Rando

m

Random

Random

Random

Random

Random

Random

Random

ons

450

450

450

402

398

395

402

427

450

roups

88

88

88

72

72

71

72

83

88

(within)

0.183

0.029

0.095

0.176

0.184

0.196

0.201

0.130

0.150

(betwe

en)

0.076

0.043

0.071

0.209

0.209

0.327

0.428

0.107

0.191

(overa

ll)

0.011

0.042

0.089

0.199

0.209

0.270

0.356

0.149

0.206

chi-squ

are

92.12

63.1

19.48

22.4

4.02

0.85

50.09

40.53

testp-v

alue

0.000

0.000

0.002

.0004

0.55

0.973

0.000

0.000

pendentvariableisper-capitaincomegrowthoverfollowingfive-yearp

eriod.



18/22

remaining columns, however, that expropriation risk is the key to eliminating the

omitted variable bias.

In columns (6) and (7) we add two additional sets of controls. Column (6) includes

the landlocked dummy and the ethnic fractionalisation score, while column (7)

includes six regional dummy variables. When economic institutions are used in

conjunction with either of these additional sets of controls, the consistency of the

random effects estimator cannot be rejected. As seen in columns (8) and (9), the

inclusion of expropriation risk is the key to eliminating omitted variable bias. In

these columns, we include the new controls but omit freedom from expropriation,

and in both of these cases the Hausman test rejects the random effects specification.

These results support our contention that, more than other potential excluded

controls, the omission of economic institutions led both to biased estimates in the

first wave and inefficient estimates in the second wave of the growth and inequality

literature.

The most noteworthy aspect of Table 5 is that it nests and ultimately reconciles theresults from the first and second wave of the growth and inequality regressions within

a single unified framework. In column (3), we see that the long run level of inequality

in a country has a negative and significant effect on growth rates, while short run

variations in equality from its average level are positively associated with growth. In

columns (4) through (9) we see clearly that the apparent long run association is, in

fact, spurious: it is an artifact of the joint association of growth and inequality with

omitted property rights. This leads us to conclude that the final answer to the long

pondered question of whether inequality affects long run growth is no.

Our results also suggest caution regarding how we interpret the positiverelationship reported by Li and Zou (1998) and Forbes (2000). Although we too

find evidence of a positive short run relationship within countries, the random effects

model does not support the position that the short run and long run relationships are

the same. Over five-year periods, the mechanism generating this correlation could

well be more closely associated with the business cycles than with long run economic

growth if, for instance, inequality is positively correlated with the unemployment

rate. A full exploration of the short run dynamics of inequality and growth within

countries lies beyond the scope of this paper.

6. Conclusion

This article has argued that a countrys institutions are a critical determinant of the

level of income inequality. While our results do not support the commonly held view

that democratic political institutions are an important determinant of income

inequality, we find a strong negative relationship between the protection of property

rights and income inequality. Moreover, this relationship is robust to the addition of

controls and the use of instruments to control for the endogeneity of our

institutional variables. The evidence presented suggests that the risk of state

expropriation creates an institutional climate more costly to the poor and

disenfranchised than to economic and social elites. We believe this evidence supports



19/22


20/22


21/22

Engerman, S.L. and Sokoloff, K.L. (1997) Factor endowments, institutions, and differential paths of

growth among new world economies: a view from economic historians of the United States, in: S.

Haber (ed.) How Latin America Fell Behind: Essays on the Economic Histories of Brazil and Mexico,

18001914(Stanford: Stanford University Press), pp. 260304.

Forbes, K.J. (2000) A reassessment of the relationship between inequality and growth. American Economic

Review, 90(4), pp. 869887.

Frankema, E. (2005) The colonial origins of inequality: a global investigation of land distribution.

Manuscript. Groningen Growth and Development Centre, University of Groningen, The

Netherlands.

Friedman, E., Simon, J., Kaufmann, D. and Zoido-Lobato n, P. (2000) Dodging the grabbing hand: the

determinants of unofficial activity in 69 countries. Journal of Public Economics, 76(3), pp. 459493.

Galor, O. and Zeira, J. (1993) Income distribution and macroeconomics. Review of Economic Studies,

64(1), pp. 3542.

Glaeser, E., Scheinkman, J. and Shleifer, A. (2003) The injustice of inequality. Journal of Monetary

Economics, 50(1), pp. 199222.

Hall, R.E. and Jones, C.I. (1999) Why do some countries produce so much more output per worker than

others? Quarterly Journal of Economics, 114(1), pp. 83116.

Keefer, P. and Knack, S. (2002) Polarization, politics and property rights: links between inequality and

growth. Public Choice, 111(12), pp. 127154.

Knack, S. and Keefer, P. (1995) Institutions and economic performance: cross-country tests using

alternative institutional measures. Economics and Politics, 7(3), pp. 207227.

Li, H., Squire, L. and Zou, H. (1998) Explaining international and intertemporal variations in income

inequality.Economic Journal, 108(446), pp. 2643.

Li, H. and Zou, H. (1998) Income inequality is not harmful for growth: theory and evidence. Review of

Development Economics, 2(3), pp. 318334.

Lundberg, M. and Squire, L. (2003) The simultaneous evolution of growth and inequality. Economic

Journal, 113(487), pp. 326344.

Marshall, M.G. and Jaggers, K. (2006) Political regime characteristics and transitions, 18002006. Polity

IV Project. Center for International Development and Conflict Management (CIDCM), University ofMaryland, College Park, MD.

Mauro, P. (1995) Corruption and growth. Quarterly Journal of Economics, 110(3), pp. 681712.

Milanovic, B. (2006) Description of all the Ginis database. World Bank, accessed at: http://

go.worldbank.org/9VCQW66LA0.

Persson, T. and Tabellini, G. (1994) Is inequality harmful for growth? American Economic Review, 84(3),

pp. 600621.

Piketty, T. (1997) The dynamics of the wealth distribution and the interest rate with credit rationing.

Review of Economic Studies, 64(2), pp. 173189.

Rodrik, D. (2000) Participatory politics, social cooperation, and economic stability. American Economic

Review, 90(2), pp. 140144.

Rosser, J.B., Jr., Rosser, M.V. and Ahmed, E. (2000) Income inequality and the informal economy intransition economies. Journal of Comparative Economics, 28(1), pp. 156171.

Sokoloff, K.L. and Engerman, S.L. (2000) Institutions, factor endowments, and paths of development in

the new world. Journal of Economic Perspectives, 14(3), pp. 217232.

Stock, J.H. and Yogo, M. (2005) Testing for weak instruments in Linear IV Regression, in: D.W.K.

Andrews and J.H. Stock (eds) Identification and Inference for Econometric Models: Essays in Honor of

Thomas Rothenberg (New York: Cambridge University Press), pp. 80108.

http://go.worldbank.org/9VCQW66LA0http://go.worldbank.org/9VCQW66LA0http://go.worldbank.org/9VCQW66LA0http://go.worldbank.org/9VCQW66LA0


22/22

Copyright of Journal of Development Studies is the property of Routledge and its content may not be copied or

emailed to multiple sites or posted to a listserv without the copyright holder's express written permission.

However, users may print, download, or email articles for individual use.

economic growth and institutional variables

Documents