Regional Heterogeneity, Conditional Convergence and Regional Inequality

Abstract

Beenstock M. and Felsenstein D. Regional heterogeneity, conditional convergence and regional inequality, Regional Studies. The paper stresses the importance of accounting for regional heterogeneity in the dynamic analysis of regional economic disparities. Studies of regional growth mainly presume that regions are homogeneous in their socio-demographic composition. It is argued that the analysis of regional convergence needs to be tested conditionally, i.e. conditional upon the socio-demographic structure of the workers in the various regions. To this end, various measures of conditional regional earnings inequality are estimated using Israeli regional data for the period 1991–2002. The results show that about half of regional earnings inequality may be accounted for by the conditioning variables. Conditioning also makes a large difference to estimates of Gini and beta-convergence. Conditional beta and Gini mobility are about half their unconditional counterparts.

Beenstock M. et Felsenstein D. L'hétérogénéité régionale, la convergence sous condition et les écarts régionaux, Regional Studies. Cet article met l'accent sur l'importance de tenir compte de l'hétérogénéité régionale dans l'analyse dynamique des écarts régionaux économiques. En règle générale, les études sur la croissance régionale laissent supposer que la structure socio-démographique des régions est homogène. On cherche ici à affirmer que l'analyse de la convergence régionale doit être tester sous condition, telle la structure socio-démographique des travailleurs dans les diverses régions. Dans ce but, on estime plusieurs mesures de l'écart des salaires régionaux sous condition à partir des données israéliennes pour la période de 1991 à 2002. Les résultats laissent voir que la moitié de l'écart des salaires régionaux environ s'explique par les conditions posées. Poser des conditions influe sensiblement sur les estimations de convergence Gini et béta. Les mobilités béta et Gini sous condition se chiffrent à environ la moitié de leurs équivalents sans condition.

Hétérogénéité régionale Ecart Convergence sous condition Mobilité

Beenstock M. und Felsenstein D. Regionale Heterogenität, konditionale Konvergenz und regionale Unterschiede, Regional Studies. In diesem Beitrag wird die Bedeutung einer Berücksichtigung der regionalen Heterogenität bei einer dynamischen Analyse der regionalen wirtschaftlichen Disparitäten betont. In den meisten Studien zum Regionalwachstum wird davon ausgegangen, dass Regionen hinsichtlich ihrer soziodemografischen Zusammensetzung homogen sind. Wir argumentieren, dass die Analyse der regionalen Konvergenz konditional überprüft werden muss, d.h. in Abhängigkeit von der soziodemografischen Struktur der Arbeitnehmer in den verschiedenen Regionen. Zu diesem Zweck schätzen wir die verschiedenen Maßstäbe für die konditionalen regionalen Einkommensunterschiede mit Hilfe von Regionaldaten aus Israel für den Zeitraum von 1991 bis 2002. Aus unseren Ergebnissen geht hervor, dass sich etwa die Hälfte der regionalen Einkommensunterschiede auf die konditionierenden Variablen zurückführen lässt. Die Konditionierung macht auch bei den Schätzungen des Gini-Koeffizienten und der Beta-Konvergenz einen erheblichen Unterschied. Die konditionale Beta- und Gini-Mobilität beträgt etwa die Hälfte ihrer nicht konditionalen Vergleichswerte.

Regionale Homogenität Ungleichheit Konditionale Konvergenz Mobilität

Beenstock M. y Felsenstein D. Heterogeneidad regional, convergencia condicional y desigualdades regionales, Regional Studies. En este ensayo recalcamos la importancia de tener en cuenta la heterogeneidad regional en el análisis dinámico de las desigualdades económicas a nivel regional. En los estudios de crecimiento regional se supone principalmente que las regiones son homogéneas en su composición sociodemográfica. Sostenemos que debe comprobarse condicionalmente el análisis de la convergencia regional, es decir, condicional en la estructura sociodemográfica de los trabajadores en las diferentes regiones. Con este fin calculamos las diferentes medidas de las desigualdades salariales condicionales en las regiones usando datos regionales de Israel durante el periodo de 1991 a 2002. Nuestros resultados indican que aproximadamente la mitad de las desigualdades salariales regionales se deben a variables condicionales. Este acondicionamiento también muestra grandes diferencias en los cálculos del índice Gini y la convergencia beta. La movilidad condicional de beta y Gini representan casi la mitad de su equivalentes incondicionales.

Homogeneidad regional Desigualdad Convergencia condicional Movilidad

Keywords:

• Regional homogeneity
• Inequality
• Conditional convergence
• Mobility

JEL classifications:

• D63
• O18
• R12

Acknowledgements

The authors thank Olga Kazanina for excellent research assistance and the Eshkol Institute for Economic and Social Research, at the Hebrew University of Jerusalem, for support. The comments of three referees are gratefully acknowledged.

Notes

1. Duranton and Monastiriotis (2002) and Henley (2005) for the UK; Azzoni and Servo (2002) for Brazil; and Combes et al. (2007) for France.

2. Dickie and Gerking (1987, 1998) concluded that US regional wage differentials in 1976 disappeared after conditioning, and provincial wage differentials in Canada in 1988–89 were reduced by 40%. Maier and Weiss (1986) showed that regional wage differentials in Austria persisted even after conditioning on regional demographic characteristics.

3. Several studies use regional deflators but do not condition earnings (e.g. Dumond et al., 1999; Johnston et al., 1996; Deller et al., 1996; Eberts and Schweitzer, 1994).

4. Arabs from East Jerusalem are excluded from the Household Income Survey.

5. Although there are only nine regions, 12 time periods contribute to the effective number of degrees of freedom. Analogous to the panel DW statistic for autocorrelation, a panel Moran's I statistic has been estimated. The panel DW statistic is meaningful when the number of time periods is nine and the number of cross-section units is 12. In the present case, the number of cross sections is nine and the number of time periods is 12. Unfortunately, there is as yet no formal definition of the panel Moran's I statistic.

6. Using macrodata, as in Henley (2005), induces bias in non-linear models, e.g. the mean of lnX_i does not equal the log of the mean.

7. Duranton and Monastiriotis (2002) allow the β’s to vary by region. The present paper tests the restriction that β _i = β.

8. Duranton and Monastiriotis (2002, p. 226) suggested occupation as such an instrument. However, occupation will tend to affect earnings.

9. Restrictions of space prevent the present paper from reporting the full set of estimates of equation (1) for each year. However, a complete set of Mincer equations for 1991–2002 is available from the authors on request.

10. As in note 5, a panel Moran's I statistic is also calculated for average Mincer residuals for each year of estimation (1991–2002). In only four of these years is the Moran's I statistic significant at p < 0.05.

11. In Duranton and Monastiriotis (2002) the same phenomenon occurs but even more acutely.

12. Since robust standard errors are larger than regular standard errors, some regional dummies become marginally insignificant.

13. Using Stata's Treatreg procedure.

14. A simple method for proxying regional COLs is applied in Beenstock and Felsenstein (2007).