The weight of the median voter ageing on public debt

Abstract

This paper studies the relationship between an ageing society and voter preference for financing public expenditures with debt versus income taxation. We focus on advanced economies and OECD members where ageing is relatively severe, the political process is more democratic, and economic institutions are subject to electorate scrutiny. Since there are no international data sources available for the median voter age, one of the contributions of this paper is the construction of a median voter series for the countries included in the study. The sample period covers 30 years, from 1980 to 2010. The panel estimates confirm the Brennan hypothesis (Const Polit Econ 23(3):182–198, 2012). A 1-year ageing of the median voter increases the public debt to GDP ratio by on the order of 4.5 percentage points.

Appendices

Appendix 1: The median voter age variable

In this appendix a relatively accurate estimate of median voter age is constructed drawing on population data available in the OECD statistical database. Data on population densities by age group are collected from the population OECD database. The population data in the OECD demographic dataset are grouped by subsets of age categories. Every age-group is 5 years long except for the very elderly of 85 years or older. The age-groups are the following: 0–4, 5–9, 10–14, 15–19, 20–24 and so forth until 85+.

We construct the age profile of the electorate by excluding the four youngest groups. One problem with this procedure is that many voters of age 18 and 19 have voting rights. Nevertheless, this problem should not be substantial, because the elderly usually tend to have higher turnout rates than younger voters, as suggested for example by previous research. [See IDEA (1999) for an appraisal of the tendency of older generations in the OECD countries to be more participative in the electoral process.]

The proposed measure of the median age of the electorate counts the median position of persons older than age 20. We have the number of people in each age group and can use this information to order the electorate as a function of age and subsequently take the person positioned in the median of this ordered list to interpolate the median age of the electorate. We call the person positioned in the median spot as the Median Voter Position Count for the population aged 20–85+. The interval age-groups make it necessary to interpolate the median voter age from the group in question. After ordering the number of people in a given country by age and taking the median position count we apply the following procedure: if the median voter belongs to the bracket of 40–44 age-group the formula of the approximation for the median voter ratio (MVR) is the following:

$$\begin{aligned} MVR & = \\ & {{\left[ {Median\;Voter\;Position\;Count\;\left( {20 - 85 + } \right) - \sum {population\;20 - 39} } \right]} \mathord{\left/ {\vphantom {{\left[ {Median\;Voter\;Position\;Count\;\left( {20 - 85 + } \right) - \sum {population\;20 - 39} } \right]} {\left( {\sum {population \, 40 - 44} } \right)}}} \right. \kern-0pt} {\left( {\sum {population \, 40 - 44} } \right)}} \\ \end{aligned}$$

where: 0 < MVR < 1

This ratio gives the fraction of the 40–44 population which resides below the position of the median voter. The median voter’s age is determined by the below rules:

• If 0.0 < MVR < 0.2 the median voter is allocated the age of 40

• If 0.2 < MVR < 0.4 the median voter is allocated the age of 41

• If 0.4 < MVR < 0.6 the median voter is allocated the age of 42

• If 0.6 < MVR < 0.8 the median voter is allocated the age of 43

• If 0.8 < MVR < 1.0 the median voter is allocated the age of 44

The same reasoning is applied to every country and to other age groups in each country. The linear approximation is not ideal, but it is a substantial improvement over average and median population age for the purposes of this paper.

Appendix 2: Random versus fixed effects

See Tables 4 and 5.

Table 4 Hausman test fixed versus random effects

Table 5 Fixed effect versus random effect regressions

Appendix 3: Robustness of empirical findings

See Tables 6, 7 and 8.

Table 6 First differences estimates, dependent variable debt/GDP

Table 7 Lagged variable estimation, dependent variable debt/GDP

Table 8 Robustness checks, dependent variable debt/GDP