Gender Equality and Economic Growth: Is it Equality of Opportunity or Equality of Outcomes?

ABSTRACT

This article explores the impact of gender equality on economic growth. In particular, we focus on the multidimensional nature of gender equality with the object of identifying the relative salience of different aspects of equality. Using exploratory factor analysis on five measures of gender equality, we identify two distinct dimensions: equality of economic opportunities and equality in economic and political outcomes. Regression analysis conducted on an unbalanced panel of 101 countries taken over nonoverlapping five-year periods from 1990 to 2000 reveals that a standard deviation improvement in equality in economic opportunity increases growth by 1.3 percentage points and a corresponding improvement in participatory equality improves growth by an average of about 1.2 percentage points. However, this impact is contingent on a country's stage of development: while developing economies experience significant improvements in growth from greater equality in opportunity, developed societies see significant improvements resulting from greater equality in outcomes.

KEYWORDS:

• Gender
• education
• labor force participation
• economic growth

JEL Codes:

• O43
• J16

NOTES ON CONTRIBUTORS

James T. Bang is Professor of Economics at St. Ambrose University in Davenport, Iowa and teaches courses on microeconomics, econometrics, international economics, and the economic analysis of institutions. His research explores the topics of gender discrimination, international migration, and institutions, with a focus on their impacts on developing countries.

Arnab Biswas is Assistant Professor of Economics at the University of Wisconsin-Stout, where he teaches courses in microeconomics. His research has focused on discrimination in housing markets, as well as topics related to discrimination and economic development.

Aniruddha Mitra is Assistant Professor of Economics at Bard College in Annandale-on-Hudson, New York, where he teaches courses on microeconomics, and migration. His research investigates the phenomena of discrimination, conflict, and international migration.

ACKNOWLEDGMENTS

We would like to thank Nita Mitra for helpful comments on earlier drafts of the paper.

Notes

¹ This should also reduce the precautionary demand for children and the resultant decline in fertility should stimulate growth.

² It should be mentioned that Fernando Ferreira and Joseph Gyourko (2010) fail to obtain any impact of gender on expenditure by American city councils.

³ While Svaleryd (2009) investigates the consequences of female representation in government, John R. Lott, Jr. and Lawrence W. Kenny (1999) consider the impact of women's suffrage in America and identify it as a key determinant of the increase in public expenditure on education observed from 1870 to 1940.

⁴ Annual percentage growth in GDP per capita = (GDP p.c.it - GDP p.c.i,t-1)/GDP p.c.i,t-1.

⁵ Also note that in constructing the time dimension of the panel as five-year intervals, the one-period lagged value of any variable in our specification represents the five-year lagged value of the five-year average.

⁶ A list of countries included in our sample is also available online as a supplemental table at http://dx.doi.org/10.1080/13545701.2014.930163.

⁷ We have not considered the gender wage gap due to the unavailability of reliable wage data for many of the countries in our sample.

⁸ All gender gaps are defined as the ratio of female to male magnitudes of the relevant variables, where a higher value indicates greater equality. To maintain parity with this convention, we take the inverse of adolescent fertility.

⁹ For countries with a bicameral legislature, we take the percentage of women in the lower chamber. The Inter-Parliamentary Union (http://www.ipu.org/wmn-e/classif-arc.htm) does not provide data prior to 1997. We rely on version 3.0 of the Democracy Time Series Data compiled by Pippa Norris (http://www.pippanorris.com) for the missing years.

¹⁰ The most commonly used aggregation procedure is to perform principal component analysis and consider the first component as institutional quality (Knack and Keefer 1995; Alesina and Perotti 1996; Perotti 1996).

¹¹ Highlighting this problem for institutional variables, Laura Langbein and Stephen Knack (2010) undertake a confirmatory factor analysis of the World Governance Indicators and fail to confirm the hypothesis that these measures are causally related to a single variable good governance. Also, unidimensional indices of gender equality such as the GEM and GDI have been criticized on the grounds that they do not reflect gender equality per se (A. Geske Dijkstra 2002; Klasen and Lamanna 2009).

¹² A limitation of dynamic GMM panel estimators is that they assume that the lagged values of the endogenous regressors are strong and only test their validity, using either a Sargan or Hansen statistic. Further, the finite-sample properties of these estimators are not well known (Bazzi and Clemens 2013).

¹³ The unique part of the decomposed variance can be seen as a residual, consisting of a random component and measurement error. The uniqueness factor reported in Table 2 consists of the total variability of each variable minus the sum of its squared factor loadings.

¹⁴ In obtaining the underlying factors, one faces the choice between several extraction methods, including principal component, principal factor, and maximum likelihood. Of these, the principal component extraction method is inappropriate for our purpose since it seeks to explain all of the variance in the observed variables and not the common variance, and hence leads to correlated errors. Maximum likelihood extraction requires the assumption of multivariate normality. One advantage of principal factor extraction is that it requires no distributional assumption on the observed variables. With respect to rotation, one faces the choice between orthogonal and oblique methods. Orthogonal methods, such as orthomax or quartimax, force the assumption of orthogonality onto the factors, which leads to loss of information if the factors are correlated. We have followed the prescription of Anna B. Costello and Jason W. Osborne (2005) in choosing the oblique promax rotation method. We have replicated our analysis using alternative extraction and rotation procedures and obtained virtually identical results, which are available on request.

¹⁵ It should be clarified that we have not restricted the number of factors. Rather, the process determined that two was the appropriate number of factors, based on the proportion of common variance they explain.

¹⁶ Serge Coulombe and Jean-François Tremblay (2006) address this problem by considering results from the International Adult Literacy Survey as the measure of literacy so as to standardize the definition of literacy across countries. But this survey and other cross-national initiatives, such as the Adult Literacy and Lifeskills Survey and the Programme for the International Assessment of Adult Competencies, only cover OECD countries.

¹⁷ As an example of the lack of clarity on the topic, while Barro (1991) finds a negative impact of net government consumption on growth, Xavier Sala-i-Martin (1997) fails to find any robust association between the variables.

¹⁸ Of the 101 countries in our sample, 22 have been members of the OECD over the entire sample period, 76 have been nonmembers for the entire time period, and three – Mexico, Hungary, and South Korea – joined the OECD in 1994, 1996, and 1996, respectively.

¹⁹ Recall that with interactions between a dummy and a continuous variable, the non-interacted coefficient on the latter represents the impact of the continuous factor on the excluded group, here non-OECD countries. The impact of the factor for the included group, here OECD countries, is the sum of the non-interacted and the interacted coefficients and its standard error is calculated as the square root of the sum of the squared standard errors and twice the covariance between them. Note also that we have excluded the non-interacted OECD dummy variable since its effects are almost perfectly correlated with country fixed effects.

²⁰ This is consistent with results obtained by a number of studies on the topic (Hill and King 1995; Dollar and Gatti 1999; Kristin J. Forbes 2000). Given that each of our regressions includes the gender gap in the access to education, we interpret the insignificance of the male secondary completion rate for developing countries as indicating that the importance of human capital as a determinant of growth depends critically on the level of equality allowed in its acquisition (Francesco Caselli, Gerardo Esquivel, and Fernando Lefort 1996; Coulombe and Tremblay 2006).

²¹ The results of these tests are available online as supplemental tables at http://dx.doi.org/10.1080/13545701.2014.930163.

²² Since it is well documented that the volume of trade is correlated with the geographical area and population of a country, we follow Barro (1991) in filtering our measure of openness for the impact of these variables.

²³ Results are available online as a supplemental table at http://dx.doi.org/10.1080/13545701.2014.930163. It is interesting to note that both inflation and trade openness mostly fail to achieve statistical significance. The first result is consistent with Michael Bruno and William Easterly (1998), who also find no impact of inflation on growth from 1960 to 1992, except in extreme episodes of inflation. Also, the ratio of trade to GDP is not the only measure of openness. Despite critiques by Francisco Rodríguez and Dani Rodrik (2001) and others, the literature has predominantly followed Frankel and Romer (1999) in filtering the trade volume for the effects of geographical characteristics. Also, while our difference–GMM methodology implicitly follows the prescription of Caselli, Esquivel, and Lefort (1996) in addressing the endogeneity between trade and growth by estimating the growth model in differences and using lags of the explanatory variables as instruments (Dollar and Kraay 2004), this is not the only way to address the endogeneity problem.

²⁴ We have also reconstructed the predicted factors for gender equality on the basis of the balanced sample. Neither the factor loadings nor the interpretations of the factors change significantly with the change in sample.

²⁵ Results are available online as a supplemental table at http://dx.doi.org/10.1080/13545701.2014.930163.