Median-voter, size of the government and budget spillover: evidence for US states

Abstract

The article tests an extended version of Meltzer and Richard's model where the redistributive policy of the government is made through in kind transfer. The median-voter of one region decides the size of the government of her region taking into account the expenditures made by neighbours' location. The theoretical model predicts that the poorer the median-voter, the higher the size of government must be. Also, the effect of neighbour's region expenditures on own location's decision of government size is undetermined. Estimations for the American states for 1999–2000 year data suggest that there is no spatial interaction among states' expenditures. However, there is evidence in favour of the positive relation between the size of the government and inequality which supports Meltzer and Richard's theoretical model.

Acknowledgements

I want to thank Luc Anselin for useful comments. This article was prepared at the University of Illinois while I was in the PhD programme at that University.

Notes

¹ The standard way used on this literature to measure the size of government is to consider the per capita expenditures.

² See Mueller (1989) for a survey on literature.

³ See Person and Tabellini (2000).

⁴ Suppose that one state's (A) increase its expenditures in police equipments; this may imply that some of the criminals from that particular state will flee to the closest state in order to commit crime, with lower chance of being punished. The A's neighbour has incentive to increase their expenditure also to avoid the criminal's migration. Another example that can be used to illustrate such spatial interaction is the construction of roads. The construction of roads in one state may or may not benefit the residents of neighbour's states.

⁵ See Keleijan and Robinson (1992) for evidence of this spatial interaction.

⁶ See Bera and Anselin (1998) and Brueckner (2003) for an overview. See Besley and Case (1995) for the yardstick competition model.

⁷ See Anselin (1988).

⁸ This variable is calculated as in Alesina et al. (1997). It measures how a location is socially fragmented. There are seven different ‘racial’ structures: white (not latino), African–American, Asian, American–Indian, Hawaian or Pacific Islander, Latino and with two or more races. Even though Latino cannot be considered a different race, it captures a difference in tastes that may drive the government expenditures. If the society is uniformly socially distributed this variable would be 0.875.

⁹ The choices of these variables are uncontroversial and they are typically used in similar cases, see for instance Case et al. (1993).

¹⁰ See for instance Boarnet and Glazer (2002).

¹¹ All the data comes from www.census.gov in the sub site State and Local Government Finances by Level of Government and State, 2002.

¹² Since it does not consider the results of the regression, it just point to a possible spatial correlation among the variables.

¹³ I also estimated a basic equation for per capita government expenditures on: police, environmental, highway, public welfare, administrative expenditures, hospital and education and the results are available upon request.

¹⁴ An inclusion of income square would have a negative coefficient but would imply a loss in degrees of freedom without any improvment in the likelihood of the model.

¹⁵ See Anselin and Florax (1995).

¹⁶ Every likelihood ratio test could not reject the null of no-lag and no error dependence. The p-values for LR test against from columns (2)–(7) are respectively: 0.31, 0.62, 0.625, 0.93, 0.46, 0.24.

¹⁷ The classical LM tests tend to over-reject the null under misspecification, so it is not necessary to implement the specification robust LM. Other alternative models are tested and the results confirm the conclusions stated above.