• Keine Ergebnisse gefunden

Voting in the UN General Assembly

3. 1 Introduction

Government ideology has been shown to influence foreign policy choices. In particular, right-wing governments tend to be pro-military and in favor of national defence, taking more hawk-ish foreign policy stances. Left-wing governments tend to be pro-peace and more concerned with domestic issues, advocating dovish positions in international relations (see Klingemann et al. 1994; Whitten and Williams 2011). Empirical studies provide evidence that right-wing governments are more likely to engage in interstate conflicts, whereas left-wing governments are more likely to face escalations of disputes (Palmer et al. 2004; Arena and Palmer 2009).

Right-wing governments also engage in longer disputes (Koch 2009). Clare (2010) shows that even an ideological outlier party in a coalition government can push the government towards more belligerent or conciliatory actions, depending on whether the outlier party is further to the right or the left of the core of coalition.

Consistent with their dovish foreign policy attitudes, one would expect that left-wing governments are also more supportive of disarmament than right-wing governments are. By redirecting money otherwise spent on armament towards meeting human needs and creating jobs in socially useful sectors, left-wing governments can fulfill their unemployment, welfare, and health care policy promises, thus gratifying their constituents.

This paper empirically investigates to what extent government ideology predicts attitudes towards disarmament. I use United Nations General Assembly (UNGA) vote data on two issue areas: arms control/disarmament and nuclear weapons/material. The hypothesis tested is that left-wing governments vote more in favor of disarmament and reducing nuclear material

87

than right-wing governments do. Data on 133 countries over the period 1975-2012 shows that switching from a right-wing to a left-wing government increases the share of votes in favor of disarmament by about 3.5 percentage points (or 1/5 of a standard deviation) and the share of votes in favor of reducing nuclear material by 5 percentage points (or 1/4 of a standard deviation).

Comparing government ideologies across countries with different political regimes might not be appropriate in this context. In particular, left-wing and right-wing autocratic govern-ments, even though at the opposite ends of the political spectrum, might use military force to the same extent, even against their own constituencies. Therefore, I consider the sub-sample of OECD countries in which governments are easier to place on a left-right ideological spec-trum. Inferences do not change when focusing on the OECD countries: left-wing governments are more in favor of disarmament and reducing nuclear weapons than right-wing governments are.

The question addressed in this paper is related to the literature on government ideology and military expenditure. Partisan theories suggest that left-wing governments pursue more expansionary policies than right-wing governments, which implies a larger share of govern-ment (Hibbs 1977; Alesina 1987). Thus, left-wing governgovern-ments might also increase military expenditure in order to stimulate the economy. Left-wing governments take however more dovish positions in foreign relations, whereas right-wing governments take more hawkish po-sitions (Whitten and Williams 2011). Military expenditure can therefore be lower under left-wing governments than under right-wing governments.

Empirical research on the link between ideology and military expenditure provides mixed evidence. Country studies show that while in some countries, such as the Netherlands and Turkey, military expenditures are higher under right-wing administrations (Van Dalen and Swank 1996; Karagol and Turhan 2008), in other countries, such as Greece and Sweden, military expenditures are somewhat higher under left-wing administrations (Kollias and Pa-leologou 2003; Eichenberg and Stoll 2003). For Germany, Kauder and Potrafke (2016) do not find any effect of government ideology on the growth rate of military expenditures. Using data of 22 OECD countries from 1988 to 2009, Bove et al. (2016) find that military expenditures are higher during right-wing administrations. T¨ong¨ur et al. (2015) use a political regime dataset that classifies governments into different types of democracies and autocracies. In particular, they distinguish between social democracies, conservative democracies, one-party democracies, dictatorships, military dictatorships, civil wars, and communist regimes. Using data on 130 countries for the period 1963-2002, they find that social democracies have a lower military burden than all other regime types.

My paper is also closely related to the literature on UN General Assembly voting. Voting patterns in UNGA have been shown to be influenced by international relations between states and other international institutions. Dreher and Sturm (2012), for example, provide evidence

88

that countries receiving adjustment projects and larger non-concessional loans from the World Bank and countries receiving non-concessional IMF loans are more likely to vote in line with the G7 countries. When focusing on voting in line with the United States, the authors find that World Bank non-concessional loans have a significant impact, but IMF loans do not. Using data on UNGA voting on repeated resolutions, Brazys and Panke (2016) show that states with limited financial capacity are more likely to engage in one-time vote shifting, whereas aid-recipient states are more likely to engage in serial shifting. Potrafke (2009a) finds that left-wing governments are less likely to vote in line with the United States, especially when the U.S. President was a Republican. Boockmann and Dreher (2011) focus on UN human rights resolutions and show that peer groups, as defined by countries within the same world region, are an important predictor of voting behavior. Countries in regions with poor human rights are more likely to vote against a human rights resolution, whereas countries in regions with better human rights records are more likely to vote in favor of a human rights resolution. This finding suggests that voting at the UN General Assembly is rather expressive than instrumental.1

When using UNGA voting as a setting to investigate whether government ideology influ-ences attitudes towards disarmament one, thus, needs to acknowledge that voting cohesion or vote buying in the UN General Assembly may occur. In that case the results obtained would be downward biased; left-wing governments, for example, might vote against disarma-ment and reducing nuclear weapons not because of their ideological attitudes but because of external pressure from other states. The opposite might be true for right-wing governments.

The paper proceeds as follows: Section 3.2 describes the data; section 3.3 outlines the empirical model, reports the main results and investigates their robustness; section 3.4 con-cludes.

3. 2 Data

3. 2.1 Data on voting in the UNGA

I use theraw UNGA voting dataset by Voeten (2013) which I match with the vote description data. Voting data is available for all UNGA members for the period 1946-2014. I use data only for the period 1975-2012 because data on government ideology is available only for these years. Figure 3.1 shows the number of UNGA resolutions over time and by different issue areas.2 The number of resolutions decreased after the Cold War for all types of resolutions.

The number of resolutions on arms control decreased, for example, from 23 per year during the period 1975-1990 to 15 per year during the period 1991-2012, whereas the number of

1On expressive voting and expressive behavior see Hillman (2010).

2The UNGA resolutions are grouped in the Voeten (2013) dataset into six issue areas that are non-exclusive, namely votes related to the Palestinian conflict, to nuclear weapons and nuclear material, to arms control and disarmament, to colonialism, to human rights, and to (economic) development.

89

010203040Number of resolutions

1975 1980 1985 1990 1995 2000 2005 2010

Year

Palestinian conflict Nuclear weapons

Arms control Human rights

Colonialism Economic development

Figure 3.1: Number of UNGA resolutions by issue areas

resolutions on nuclear weapons decreased from 20 per year during the period 1975-1990 to 12 per year during the period 1991-2012. The two issue areas, arms control/disarmament and nuclear weapons/material, are non-exclusive: a resolution can be classified as relating to both issue areas. On average, the UNGA members have voted on 15 nuclear weapons resolutions and 18 disarmament resolutions per year during the period 1975-2012. These two issue areas represent together about 35 % of all resolutions in the period 1975-2012 (see Figure 3.2 for the shares of resolutions by year and by issue areas).

A country can vote in favor of or against a resolution, abstain or not be present at the time of the vote. Based on this information, I construct for all countries the annual share of approving votes as follows: I weight abstentions3 and absences 4 by 0.5 and add these votes to the number ofYes votes; I then divide the resulting number by the total number of votes a country cast in a particular year. However, absences are not necessarily a sign of approval.

Government turnovers, for example, may result in countries not having a UN delegation for some time (Voeten 2013). For this reason, I use two alternative specifications of computing the share of approving votes. In the first one, I weight only abstentions by 0.5 and absences by 0. In the second, I exclude the observations when a country has not voted in the UNGA for the entire year and compute the share of approving votes by weighting both abstentions and absences by 0.5.

3Boockmann and Dreher (2011) and Hillman and Potrafke (2015) find that abstaining and voting against a resolution are two different choices that should not be bundled into one category, with voting against representing a much stronger disapproval than abstaining.

4Absences have been treated in the UNGA voting literature as equivalent to abstentions (Luif 2003; Dreher and Jensen 2013; Potrafke 2009a), as a distinct voting position (Barro and Lee 2005), or as missing values (Boockmann and Dreher 2011; Burmester and Jankowski 2014).

90

010203040Share of resolutions (in %)

1975 1980 1985 1990 1995 2000 2005 2010

Year

Palestinian conflict Nuclear weapons

Arms control Human rights

Colonialism Economic development

Figure 3.2: Share of UNGA resolutions by issue areas

Voting Yes for a resolution related to arms control or nuclear weapons implies that a country is, generally speaking, in favor of disarmament and reducing nuclear material. Given the hypothesis formulated in section 3.1, I expect that the share of approving votes will be higher under left-wing governments than under right-wing governments.

3. 2.2 Data on government ideology

To measure government ideology I employ the Database of Political Institutions by Beck et al. (2001). The variableexecrlc describes the orientation of the chief executive party with respect to economic policy and can take one of the three values: right (or 1), center (or 2), or left (or 3). According to Keefer (2012) right refers to parties that can be defined as

“conservative, Christian democratic, or right-wing”. Left refers to parties that can be defined as “communist, socialist, social democratic, or left-wing”. Center stands for parties that are

“centrist or when party position can best be described as centrist”. Data is available for the period 1975-2012 for most countries in the world.

Beck et al. (2001) assign a fourth category not applicable to parties that cannot be classified on a left-right dimension because their political agenda does not focus on economic issues or there is not enough information about a party’s program. In my main specification I consider these cases missing values. Dreher et al. (2015) contend, however, that these non-partisan governments fall in between left-wing and right-wing ideologies and assign them the value 2 orcenter. For this reason, I run robustness tests also using the modification proposed by Dreher et al. (2015). A fifth category in the Beck et al. (2001) classification contains countries that do not have a chief executive. I treat these observations as missing values.

91

Comparing government ideologies across countries is not straightforward because of the different political regimes in the world. This issue arises also when using the Beck et al. (2001) dataset. While the authors group social democratic and communist governments under the same left ideology, clearly a Christian democratic government might be ideologically closer to a social democratic government than a communist government. I therefore also consider the sub-sample of OECD countries in which governments are easier to compare from a left-right ideological perspective.

For robustness tests, I also use the government ideology index proposed by Potrafke (2009b). The Potrafke index is available for the period 1950-2014 but only for 23 OECD countries5 and places governments on a right to left scale, with values from 1 to 5. The index takes the value 1 if the share of governing right-wing parties in terms of seats in the cabinet and in parliament is larger than 2/3, and 2 if it is between 1/3 and 2/3. The index is 3 if the share of center parties exceeds 50 %, or if the left-wing and right-wing parties form a coalition not dominated by one side or the other. The index is symmetric and takes the values 4 and 5 if the left-wing parties dominate. I group right-wing governments (coded 1 and 2 by the Potrafke index) into one category and label itright or 2. The Potrafke index never assumes the value 5 in my sample. Hence the coarse version of the Potrafke (2009b) index takes the valuesright (or 2),center (or 3), andleft (or 4). The correlation coefficient between the Beck et al. (2001) and the Potrafke (2009b) indices is statistically significant at the 1 % level and amounts to 0.72.

3. 2.3 Correlations between government ideology and voting behavior Table 3.1 shows the average shares of approving votes for the two issue areas (nuclear weapons and arms control) under left-wing, centrist, and right-wing governments. The average share of approving votes is always larger under left-wing than under right-wing governments, regardless of the sample considered (all countries or OECD countries) or the index being used (Beck et al. 2001 or Potrafke 2009b). OECD countries vote, however, on average less in favor of disarmament and reducing nuclear material than other countries do, both under left-wing and under right-wing governments. The variance analysis shows that the differences of the means are always statistically significant at the 1 % level. The individual tests show that the differences in the means between left-wing and right-wing governments are also always statistically significant at the 1 % level for both issue areas.

Table 3.2 presents correlations between the share of approving votes and government ideology. The correlation coefficients indicate that these two measures are somewhat positively correlated, with stronger correlations for nuclear weapons (correlation coefficient: 0.2) than

5The OECD sample includes: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, U.S.A., and United Kingdom. Switzerland became a UNGA member only in 2002 and, for this reason, I do not include it in the set of OECD countries.

92

disarmament (correlation coefficient: 0.16). The correlation coefficients are slightly higher when including all countries than when including only the OECD countries.

3. 3 Analysis

3. 3.1 The econometric model

The baseline econometric model has the following form:

Share approving votesijtjIdeologyitjlogP opulationitjlogGDPit

+ ΣkζjkConf lictikti+t+uijt

where i = 1,...,133 describes the countries; j = 1 or 2 indexes the two types of issue areas (disarmament and nuclear weapons); and t = 1,...,38 describes the years (1975-2012). The dependent variable Share approving votesijt describes the annual share of votes in favor of disarmament or reducing nuclear material for country i and year t. The variable Ideologyit describes the government orientation as right-wing, center, or left-wing. I also include the logarithm of population size and the logarithm of a nation’s real per capita GDP. The data are taken from the World Bank (2016). Conflictikt is a set of conflict dummy variables that assume the value 1 when a particular type of conflict takes place in countryi and year t, and 0 otherwise. I employ theUCDP/PRIO Armed Conflict Dataset by Gleditsch et al. (2002), version 4-2015 (Pettersson and Wallensteen 2015), and distinguish between three types of armed conflicts: ‘interstate’, ‘internal’, and ‘internationalized internal’. ηi is a fixed country effect, tis a fixed period effect, anduitis the error term. Tables 3.8 and 3.9 in the appendix show descriptive statistics for all variables included in the model, separately for the two samples: all countries and OECD countries. Table 3.10 in the appendix presents definitions of all variables and data sources.

I estimate a fixed effects model with robust standard errors clustered on countries that are heteroskedasticity- and autocorrelation-consistent.6 The data might suffer from cross-sectional dependence as is often the case in microeconometric panel datasets. Because my panel is highly unbalanced, I cannot conduct a test for cross-sectional dependence for the entire sample of countries, i.e., the number of common observations across countries is not sufficiently large to apply the Pesaran cross-dependence test (Pesaran 2004). The Breusch-Pagan LM test of independence (Breusch and Breusch-Pagan 1980), applied to the sub-sample of OECD countries, shows however that there is cross-sectional dependence. To be on the safe side, I also estimate the fixed effects model using Driscoll and Kraay (1998) standard errors that are heteroskedasticity-consistent and robust to general forms of spatial and temporal

6The Wooldridge test for serial correlation in the idiosyncratic errors of a linear panel-data model (Wooldridge 2002) rejects the null hypothesis of no first-order autocorrelation.

93

Table 3.1: Share of approving votes under different ideologies All countries

Beck et al. (2001) ideology

Means Variance Individual tests

Right Center Left F-Test Right-Center Center-Left Left-Right

(1) (2) (3) (4) (5) (6) (7)

Share approving votes

Nuclear weapons 76.56 81.48 84.92 78.71*** 4.92*** 3.44*** 8.36***

(0.00) (0.00) (0.00) (0.00)

Disarmament 81.10 85.14 86.48 47.96*** 4.04*** 1.34 5.38***

(0.00) (0.00) (0.27) (0.00)

OECD countries

Beck et al. (2001) ideology

Means Variance Individual tests

Right Center Left F-Test Right-Center Center-Left Left-Right

(1) (2) (3) (4) (5) (6) (7)

Share approving votes

Nuclear weapons 58.36 64.20 64.07 13.47*** 5.84*** -0.14 5.70***

(0.00) (0.00) (1.00) (0.00)

Disarmament 67.98 73.37 72.67 12.61*** 5.39*** -0.69 4.69***

(0.00) (0.00) (0.90) (0.00)

OECD countries

Potrafke (2009b) ideology

Means Variance Individual tests

Right Center Left F-Test Right-Center Center-Left Left-Right

(1) (2) (3) (4) (5) (6) (7)

Share approving votes

Nuclear weapons 58.19 66.39 62.71 17.75*** 8.20*** -3.69 4.51***

(0.00) (0.00) (0.05) (0.00)

Disarmament 67.68 75.45 71.42 20.02*** 7.77*** -4.03 3.74***

(0.00) (0.00) (0.01) (0.00)

Notes:

The share of approving votes is computed by weighting both abstentions and absences by 0.5.

Column (4) presents the F-statistics and columns (5)-(7) the individual differences of the means. The p-values are reported in parentheses. *** implies to reject the null hypothesis at the 1% significance level.

The p-values in columns (5)-(7) refer to the test of Scheff´e (1953).

Table 3.2: Correlations between the share of approving votes and government ideology

All countries OECD countries

Ideology Beck et al. (2001) Beck et al. (2001) Potrafke (2009b) Share approving votes

Nuclear weapons 0.203 0.168 0.133

Disarmament 0.158 0.157 0.127

Observations 3646 813 831

Note: The share of approving votes is computed by weighting both abstentions and absences by 0.5.

dependence.7 I use a lag length of 3 years as the maximum lag length to be considered in the autocorrelation structure. The results are robust to changes in the selected lag length.

To test the suitability of the fixed effects model over a random effects model, I perform a Hausman test. I first use the alternative specification of the Hausman test proposed by Wooldridge (2002) that is robust to heteroskedastic errors. Based on this, Hoechle (2007) proposes an implementation of the Hausman test that is also robust to general forms of spatial and temporal dependence. Both tests reject the null hypothesis of no fixed effects at the 1 % level of significance indicating that pooled OLS estimation is likely to produce inconsistent coefficient estimates.

3. 3.2 Main results

Table 3.3 presents the baseline results of the fixed effects model when the sample consists of all member states of the UN General Assembly. Table 3.11 in the appendix presents a list of all countries included in the sample.8 Columns (1)-(3) present the results when the issue area is nuclear material and nuclear weapons; columns (4)-(6) present the results when the issue area is arms control and disarmament. The coefficient estimates for the ideology variable are always positive and statistically significant at the 1 % level, regardless of the type of standard errors being used (clustered on countries or Driscoll-Kraay) and of time fixed effects being included in the model or not.9 The numerical meaning of the coefficients in columns (1)-(3) is that an increase in ideology by 2 points (i.e., from a right-wing government to a left-wing government) increases the share of Yes votes in favor of reducing nuclear material by about 5 percentage points (or about 1/4 of a standard deviation). For resolutions related to arms control and disarmament, the effects are slightly smaller: switching from a right-wing

Table 3.3 presents the baseline results of the fixed effects model when the sample consists of all member states of the UN General Assembly. Table 3.11 in the appendix presents a list of all countries included in the sample.8 Columns (1)-(3) present the results when the issue area is nuclear material and nuclear weapons; columns (4)-(6) present the results when the issue area is arms control and disarmament. The coefficient estimates for the ideology variable are always positive and statistically significant at the 1 % level, regardless of the type of standard errors being used (clustered on countries or Driscoll-Kraay) and of time fixed effects being included in the model or not.9 The numerical meaning of the coefficients in columns (1)-(3) is that an increase in ideology by 2 points (i.e., from a right-wing government to a left-wing government) increases the share of Yes votes in favor of reducing nuclear material by about 5 percentage points (or about 1/4 of a standard deviation). For resolutions related to arms control and disarmament, the effects are slightly smaller: switching from a right-wing