Economic Growth, Campaign Pledges and Election Outcomes Jongmin Lee Abstract This paper aims to figure out the economic factors which affect the electoral outcomes. Employing the CMP(Comparative Manifesto Project) data of 402 parliamentary elections over 49 countries, I examine whether good macroeconomic outcomes during the term in office raise the ruling party s winning prospect and whether a specific public commitment of the ruling party can help win the election. We find that the voter s are likely to reward the incumbent party for economic growth over the term in office. We also discovered the ruling party s public commitment mentioning welfare state is effective in the electoral campaigns. This effect is more obvious in the countries with high income/democracy level than in the countries with low income/democracy level. Keywords : Economic growth, Manifesto, Welfare state, Democracy, Fiscal balance, Political Business Cycle 1. Introduction The relationship between economic policy and electoral outcome is one of the most popular topic in the political economic context. Theoretical framework for this topic was given by Nordhaus(1975) and subsequent research papers. In their world, voters have retrospective voting rule, rewarding the incumbent leader or party for good macroeconomic situation. Knowing this, the incumbents
have incentive to manipulate it with fiscal policy for re-election. In the pre-electoral period, as a result, it is likely that the government increases its budget deficit to make artificial boom so that people misread the economic condition and vote for the candidate from the incumbent party. It is not surprising that two empirical question on this political business(budget) cycle has been raised repeatedly. The first one is about the existence of political business cycle: does the government indeed make pre-electoral fiscal expansion to increase probability of re-election? The other is about voters behavior: can the good macroeconomic outcomes or expansionary fiscal policies attract votes? This paper focused on the second issue. Most of previous empirical studies on the economic factors affecting election outcomes were conducted for single country or several developed counties. More recently, Brender and Drazen(2008), De Hann and Klomp(2013) construct dataset of large panel of countries to make comparative analysis at the national level. These authors commonly suggest that economic growth, and government budget balance are key variable that affect the re-election probability of the incumbents. Since voters are assumed to be rational client maximizing expected utility subject to their choice set, these result should not be interpreted as just reward or punishment for economic performance. Rather, it is more likely that they update their belief about the incumbents from this information, thereby determine their voting behavior (Rogoff and Sibert, 1988 ; Rogoff, 1990). Thus, macroeconomic factors that affect election outcomes can be interpreted as signal that help voters estimate competence of the candidates. One of the most important information with which the public can assess the incumbents other than previous economic performance is their announced stance for economic policy. Given this information, the voters can maximize their life-time expected utility by choosing the candidate from incumbent party when he(she) suggest the policy aligned with their interest. In other words, there are - 1 -
two part of information set that can affect choice of the voters. The first one is information on the ability of incumbent government: realized economic factors such as growth rate, inflation or budget deficit over the term in office. The other is information on the policy preference of the party: official statement such as public speech or campaign brochure. In this regard, It is quite strange that the effect of public commitment on the election outcome has hardly been researched. To make an empirical analysis for this topic, one need to quantify the officially stated election pledges first. Some recently constructed dataset provide useful quantities measuring the extent to which parties emphasize each issue in their election campaign. Based on the data for 402 parliamentary elections over 49 countries in 1960~2013, our analysis aims to figure out the factors affecting the ruling party s winning probability. We include some variables summarizing election manifestos of the party to identify the policy pledges that can attract voters successfully. 1) In more detail, we test whether (1) GDP growth and low inflation during the incumbent leader s term helps the ruling party to win the election; (2) pre-electoral macroeconomic movements has additional effect on election outcome; (3) welfare-oriented or market-oriented incumbent party has higher chance of winning. (4) our estimates depend on the national income level or democracy level. The remainder of this paper consists of the following. Section 2 introduce the previous empirical studies of economic voting and discuss how our contribution is related to them. Section 3 describes the empirical model and data used in this paper. Section 4 shows our basic estimation results and their implications while Section 5 presents the result of the model with alternative assumptions to check the robustness of the estimates. Section 6 offers conclusion and suggests further research. 1) We employed Comparative Manifesto Project(CMP) dataset, which classify the sentences in the parties election manifestos by 110 subjects. The variables we use indicate proportion of sentences stating a specific policy direction, thus get values from 0 to 1. - 2 -
2. Literature Review Most of early studies on the relationship between election outcome and economic situation do not find evidence that GDP growth increases the reelection probability of the incumbents. They employed data on time series of the election in a single country or small panel of old democracies. For example, Lewis-Beck(1988) looked into elections in 5 leading European countries, - United Kindom, France, Spain, Italy, Germany - while Paldam(1991), Powell and Whitten(1993) studied 19 and 17 developed countries, respectively. They commonly found that the effect of economic growth on election outcomes is insignificant. An exceptional result was reported for the Presidential elections in the United States. Fair(1978), Tufte(1978), Hibbs(1987), Alesina and Rosenthal(1995) found that good macroeconomic condition increases the probability of reelection significantly. 2) Since, the empirical conclusions from the early studies are considered to be limited to these countries, researchers tried to construct database of a broader cross section of countries to provide more general implication. Brender and Drazen(2008), who employ the data for 350 elections in 74 democracies over the period 1960~2003, found the positive effect of economic growth on reelection prospect in the developing countries rather than in the developed. Similarly, De Hann and Klomp(2013) analyzed elections in 65 countries from 1975 to 2005 and presented the estimates indicating that the GDP growth strengthen the electoral supports for the parties in government. Previous literatures on the impact of fiscal policy have offered conflicting result. Peltzman(1992) and Drazen and Eslava(2010) reported that voters punished the ruling party for large budget deficit in the US elections and the Colombian local elections respectively, while Aidt et al.(2011) found that expansionary fiscal policy increases the winning probability of Portuguese incumbent 2) For a less developed country, recently, Gupta and Panagariya(2014) studied parliamentary elections in India to conclude that high economic growth was advantageous for the candidates from the incumbent state parties. - 3 -
mayors. 3) The lack of consistent results led researchers to be interested in the factors that make this difference. More recently, a few studies provide comparative analysis at the national level. Brender and Drazen(2008) classify the countries by the income level as well as by the oldness of democracy. They showed that loose fiscal policy has a negative effect on reelection in countries with high income, especially in old democracies that are developed. De Hann and Klomp(2013) used two-step approach. In the first step, they tested the existence of pre-electoral expansion, thereby sort out countries with political budget cycles. In the second step, they examined for countries with pre-electoral fiscal manipulation, whether this politically motivated fiscal policy help the incumbent get more votes. Their estimates indicate that the answer is yes. An alternative explanation for the relationship between fiscal policy and election is that voters care about the welfare services, not the overall fiscal policy. Pacek and Radcliff(1995) found that the voting behavior was not influenced by macroeconomic outcome in countries with high levels of welfare spending. One can interpret this result that the economic fluctuation cannot shift voters expected utility remarkably when the state has well-working social security system. Armingeon and Giger(2008) suggested that electoral punishment for retrenchment of welfare spending depends on the political capability of the incumbent. They showed that the government parties that retrench the welfare state do not lose votes if social policy cuts are not put on the agenda of the campaign. This paper conducts a comparative analysis for a broad cross section of countries by adopting panel logistic regression approach as in Brender and Drazen(2008) and focuses on the political context in the election campaign as in Armingeon and Giger(2008). Since the information on overall fiscal policy is too complex for people without specialized knowledge to understand, it may not be 3) Bremder(2003) employed data for Israel, but their result was not robust. Alesina, Perotti, and Tavares(1998) studied the effect of fiscal adjustments on a political support in a cross section of 19 OECD countries to find that sound fiscal management has positive political effects at the national level. - 4 -
directly reflected in the voting behavior. Rather, the agenda-setting activity of parties running in the upcoming election, which is presented in the election manifestos, can play a crucial role. Thus, one can estimate the voter s preference for each economic policy more realistically by including quantified manifesto data in the regression model. Since there is no studies at the national level, of which we are aware, that take into account variables indicating officially stated election pledges, we expect this paper to contribute to discussion on the economic voting. 3. Empirical Framework 3.1 Data and Variables The sample of this paper consists of the observations on 402 parliamentary elections in 49 democracies over the period 1960-2013. The dataset used in this paper is the combination of informations on the macroeconomic and political condition of the countries; GDP growth, inflation, power structure, electoral system; data for each election; voter s turnout, the number of votes and seats, manifesto of the parties, and some additional political variables. We refine this information and define the variables to be compatible with our research purposes. A detailed explanation of the data source and the construction of the variables is presented below. The dependent variable GPwin is a binary variable receiving the value 1 if the ruling party win the election and 0 if not. the ruling party is defined as the party that the incumbent leader belongs to. 4) The information on leaders is collected from the World Political Leaders 1945 2013 database of Zárate s 4) We decide whether the president or the prime minister is the leader based on the classification by World Bank s database of Political Institutions (DPI). - 5 -
Political Collections and complemented by online encyclopedias when needed. This simple definition, however, cannot be applied globally due to the complexity of political structure and governmental organization. So we use alternative definitions of the ruling party as follows. First, if government is formed by a coalition of parties, the party which have the largest number of seats among them is regarded as the ruling party. Second, when the governmental power is not concentrated on the leader or the ruling party such as a multi-party federal directorial system of Switzerland, a plurality party of the council is defined as the ruling party. Third, for the cases in which a leader was substituted by another candidate within 6 months before the election, the party to which the substitute leader belongs is treated as the incumbent. We exclude the observations that the incumbent party, defiened as above, came to power less than two years prior to the elections as in Brender and Drazen(2008). The elections that the ruling party cannot be defined in the manner mentioned above are excluded as well. The key variable GPwin is formed based on the number of seats each party earn from the election. It receives 1 if the ruling party take the largest number of parliamentary seats as a result of the election. 5) In this paper, we consider a party as a core agent of agenda-setting activity and aims to look into the direct effect of the economic performance and policy proposal on the voter s preference for the party. 6) That s why we focus on the electoral support that parties in the campaign receive. 7) Our variables for macroeconomic performance is calculated on the basis of data from the World Bank s World Development Indicators(WDI). 8) We introduce 5) Previous studies focused on whether the incumbent leader is reelected and the ruling party succesfully regained power. See Alesina et al.(1998), Brender (2003), Brender and Drazen (2008), Alesina et al. (2011). 6) De Haan and Klomp(2013) use the key variable that indicate change of the number of incumbent party s seat. In this case, explanatory variables should be adjusted by difference for the compatibility. 7) Among the 402 observations, we have 240 elections that the ruling party won. 8) We employ data from International Financial Statistics(IFS) and The Public Purpose, when there is missing values in the WDI dataset. We indicate inflation rate with the Consumer Price - 6 -
two indicators, Growth_term and Inflation_term, both of which measures average annual rate between the current and previous election year. The former reflects the growth rate of GDP per capita and the latter is the average inflation rate. If there is change of government in the middle of the term, we calculate the average by setting the year of regime change as a starting point. 9) The additional effect of economic condition in pre-electoral period is captured by Growth_preelection, which is annual growth rate of GDP per capita in the previous year of election. 10) Our indicator for the policy proposals are characterized by Welfare_GP and Market_GP. Welfare_GP denotes a proportion of the sentences about welfare state in the election manifesto of the ruling party and Market_GP is that about market economy. Both variables are based on the data taken from Comparative Manifesto Project(CMP) dataset. For democratic elections over the world, it classifies the sentences in the manifesto of each party running in the election into 110 categories. Numerical figures given by the dataset indicate a proportion of the sentences about each subject. We use the aggregation of the values for several subjects that fall into the welfare state and market economy. 11) These variables represent the policy preference of the ruling party in our estimation. We include several variables indicating political and institutional conditions which can affect the election outcomes. We obtained the data for political variable mainly from the World bank s Database of Political Institution(DPI), and the Polity IV dataset at the University of Maryland (Polity IV). Index(CPI) for the observations from Germany, Ireland, Switzerland, New zealand, and GDP deflator for the rest. We choose the longer one between the two popular price indices to minimize a loss of information. This can cause a bias in pooled regression model, but it can be corrected if we control country fixed-effect. 9) There are two main sources for these situations. One is the substitution of the leader in the Presidential republic. The other is the change in the structure of the party competition in the Parliamentary system. 10) All the annual variables are adjusted to fit the fiscal year. 11) According to the classification of CMP, sentences about equality in distribution, expansion of social service come under welfare, and those about free market and economic orthodoxy are related to market. - 7 -
The Ideological stance of the candidate or the party can be a important criteria for choice of voters (Jenkins, 2000; Van der Brug, 2010). We introduce binary variables, Left_GP and Right_GP, indicating whether the ruling party is left-leaning or right-leaning. In this paper, the party is called left if it have, in its manifesto, more sentences categorized as leftist policies than sentences categorized as rightist policies and right otherwise. We also consider the effect of the quality of political institution. The level of democracy is a polity score in the polity IV which is scaled from 10 to 10. New democracy is a binary variable, receiving the value 1 for the election held within a 15 years after a democratization. 12) The original public support and political strength are controlled by introducing the variable GP_seats, which means the number of seats that the incumbent party holds in parliament in the election year. The effect of the electoral system is capture by a binary variable, for each country, receiving the value 1 in a country with a Majoritatian system. Finally, Voter's turnout is included as control variable. In Table 1 we show the descriptive statistics of variables used in the empirical model. 12) As in Brender and Drazen(2008), the period of democratization is defined as the year that a country with a negative polity value in the POLITY IV dataset shifted to non-negative values. - 8 -
Table 1 > Descriptive statistics All countries OECD Non-OECD Mean Std.dev Mean Std.dev Mean Std.dev Source Growth_term 2.35 2.86 2.28 1.88 2.57 4.99 WDI, IFS Growth _preelection 2.42 3.88 2.33 2.64 2.74 6.65 WDI, IFS Inflation_term 9.81 15.70 6.61 8.90 21.20 26.13 WDI Balance_term -2.54 3.84-2.75 4.00-1.58 2.77 Retrenchment _preelection -0.34 2.80-0.32 2.84-0.43 2.62 Welfare_GP 12.27 6.95 12.57 7.11 11.19 6.25 Market_GP 4.26 4.54 4.48 4.82 3.47 3.30 Democracy level IFS, OECD IFS, OECD Manifesto Project Manifesto Project 8.28 7.57 9.25 2.27 4.85 15.17 Polity IV Voter s turnout 75.57 13.43 77.37 12.70 69.17 14.05 IDEA Seats_GP 42.24 13.24 42.16 12.41 42.54 15.93 DPI Observations 402 314 88 Notes : Balance_term is the average value of the ratio of central government s budget balance to GDP over the term in office. Retrenchment_preelection is the change in the balance/gdp ratio in the pre-election year relative to the previous year. The number of observations that have information on government budget is 260 in the whole cross-section, 215 in OECD countries, 45 in non-oecd countries. 3.2 Empirical Method Since the key dependent variable is a binary variable receiving the value 1 if the ruling party win the election, we employ a binary response model for the regression. A basic formulation is introduced as following equation : - 9 -
y it = + x it + z it + e it e it = + it (3.1) y it denotes the latent variable for the election outcome, which is assumed to be positive when we observe the victory of the ruling party. x it is the vector of explanatory variables in t th election of country i, which contains macroeconomic factors and policy proposals of government party. z it is the vector of control variables introduced in the previous section, and denotes country-specific effects which are time-invariant. it is the stochastic error term with mean zero and no autocorrelation. The estimation model depends on the statistical assumption about and it. In general. We assume is a random variable with mean zero, which is independent of x it and z it, and it is a stochastic error exogenous given for each i and each t. Then, e it is treated as i.i.d(independently and identically distributed) random shock, whose distribution is Logistic. 13) With the above assumptions, the equation (3.2) describes a probability that the ruling party win in the t th election in country i. The log-likelihood function for the parameters calculated based on this probability is presented in the equation (3.3). We conduct maximum-likelihood estimation for,, from the equation (3.3). 13) In a binary response model, the error term is generally assumed to have logistic distribution(logit equation) or standard normal distribution(probit equation). In our estimation, logit and probit model yielded very similar results. - 10 -
Pr(g it 14) = 1) = Pr(y it 0) = Pr (-e it + x it + z it ) = ( + x it + z it ) 15) (3.2) = (,, ; x, z, g) [g itln ( + x it + z it )+(1-g it)ln{1- ( + x it + z it )}] (3.3) Our statistical assumption may seem to be too strong, in the sence that some of the estimates are possibly influenced by unobserved country characteristics that are correlated with both the election outcome and the explanatory variables. In our model, however, the fixed effect is partly controlled by some group dummy variables. Moreover, we should bear some losses of observations with fixed-effect estimation, 16) and the fixed-effect Logit estimator is not consistent due to the relatively large number of individuals and small number of observations for each country. 17) For these reasons, we adopt pooled Logit regression to estimate the parameters, then we check the robustness of our estimates with the fixed-effect Logit and random-effect Logit estimation. We also test whether our estimates are sensitive to the length of series or inclusion of additional variables indicating government s budget condition. 18) 4. Estimation Result The basic estimation result of the equation (3.3) is presented in Table 2. We 14) GPwin it introduced in the section 3.1. 15) ( ) denotes the cumulative distribution function of logistic model, which is a zero-centered symmetric function. 16) In the binary choice model with fixed-effect, individuals that have dependent variable receiving same number with a whole series are dropped. 17) See Neyman and Scott(1948), Consistent estimation from partially consistent observations, and Lancaster(2000), The incidental parameter problem since 1948. 18) The empirical approach of this paper relies largely on the methodology developed by Brender and Drazen(2008). - 11 -
show the effect of growth, inflation and policy proposals on the election outcomes using pooled Logit model. 19) The figures in the table are logit coefficients and the figures in parentheses are p-values. 20) Column 1 and 2 is the result from whole sample. As in the previous studies, we find that economic growth is positively related to the probability that the ruling party win the election. The coefficient of the average GDP growth over the term in office is positive and significant, which indicate that the ruling party has a higher chance of winning when GDP is growing rapidly. This result imply that voters are likely to consider economic prosperity to reflect the ability of the incumbent government. On the other hand, the additional effect of the growth in the preceding year of election is not significant. That is, pre-electoral boom does not help the ruling party gain the electoral support. The proportion of election pledges about welfare state also has a positive effect on the winning probability of the ruling party, indicating that the voters regard policy preference of the party as important criteria as well as the economic performance. This suggests that voters expected utility is increasing when the candidates planning expansion of welfare spending is elected. We find no significant effect of inflation rate, voter s turnout, electoral system and the level of democracy. The coefficient of the number of seats that the incumbent party holds just before the election is strongly significant and positive, which reflects the consistency of electoral supports. 19) The effect of budget expansion is discussed in section 5.2. 20) The coefficients are ordinary maximum-likelihood estimator for the equation (3.3), and the p-values are calculated on the basis of Huber/White/Sandwich s heteroskedasticity robust standard errors formula. - 12 -
Table 2 > The effect of growth and policy proposal on the election outcome classified by income level. Dependent variable : GPwin Growth_term Growth _preelection Inflation_term Welfare_GP Left_GP Market_GP Right_GP Democracy level Voter s turnout Seats_GP Constant All countries OECD Non-OECD (1) (2) (3) (4) (5) (6) 0.1830 ** (0.034) 0.0047 (0.924) 0.0120 (0.316) 0.0385 ** (0.046) 0.2100 (0.404) -0.0226 (0.930) 0.0143 (0.146) 0.0357 *** -3.5690 *** 0.1857 ** (0.032) 0.0034 (0.944) 0.0108 (0.369) -0.0394 (0.124) -0.2429 (0.330) -0.0439 (0.867) 0.0154 (0.121) 0.0347 *** -2.7695 *** (0.001) 0.3639 *** -0.0137 (0.814) -0.0246 (0.193) 0.0470 * (0.055) 0.1267 (0.697) -0.4085 (0.214) 0.0176 (0.180) 0.0500 *** -4.0912 *** 0.3737 *** -0.0169 (0.774) -0.0259 (0.184) -0.0254 (0.372) -0.2499 (0.424) -0.4424 (0.181) 0.0170 (0.189) 0.0496 *** -3.1146 *** (0.003) 0.0470 (0.678) 0.0695 (0.346) 0.0122 (0.420) 0.0025 (0.957) 0.0520 (0.918) 0.6281 (0.235) 0.0278 (0.165) 0.0246 (0.102) -3.4576 ** (0.043) 0.0302 (0.792) 0.0745 (0.329) 0.0117 (0.478) -0.1152 (0.144) 0.1953 (0.698) 0.7893 (0.144) 0.0309 (0.144) 0.0232 (0.112) -3.3128 * (0.056) Pseudo R 2 0.1042 0.0989 0.1486 0.1374 0.0882 0.1061 p( >chi 2 ) 0.0000 0.0001 0.0000 0.0000 0.4679 0.2822 Observations 402 402 314 314 88 88 Notes : The figures in the table are Logit coefficients and the figures in parentheses are p-values calculated on the basis of Huber/White/Sandwich s heteroskedasticity robust standard errors formula. OECD, Majoritarian system, New democracy is included as control variables. * : Significant at the 10 percent level. ** : Significant at the 5 percent level. *** : Significant at the 1 percent level. 4.1 Classified by the Level of Income In this section, we examine whether our estimates depends on the level of income. In column 3 and 4 of Table 2, we look into the developed countries, - 13 -
and column 5 and 6 present the equation for the less developed countries. 21) The coefficient of economic growth is positive in both developed and less developed, but there is a large difference in size and statistical significance. We find that an increase of 1 percentage point in the GDP growth raises the ruling party s winning probability by 8 9 percentage points in the developed countrie s. 22) In the less developed countries, by contrast, the effect is just 0.7~1 percentage points and statistically insignificant. These finding is quite surprising because previous studies suggest that the election outcome is influenced by economic growth in the developed rather than in the less developed. 23) It is possibly due to the sample selection and variable definition. Our sample of developing countries mainly consists of the former socialist economies in Eastern Europe and the former Soviet Union while previous studies mainly focused on the Latin American countries. Moreover, we set the dependent variable to indicate electoral support that parties in government receive rather than the reelection of the incumbents. Another possible explanation is the importance of the institution. The effect of economic performance may well vary according to the quality of institution not the level of income. We conduct the estimation distinguishing countries with high level of democracy from those with low level in section 4.2. Our estimates also show large difference in voter s preference for the policy proposal about welfare state between the developed and the less developed countries. In column 3 at Table 2, the coefficient of Welfare_GP is significant at 21) The developed denotes OECD member countries, which constitute about three quarters of the whole sample, and the less developed denotes non-member countries. 22) Numerical effect on probability is calculated based on the odd ratio. Let k denotes the proportion of observations that the dependent variable receive the value 1, and b is the maximum likelihood estimator of the coefficient. Then the effect is calculated as the formular b k (1-k). 23) G. Bingham Powell Jr. and Guy D. Whitten (1993), Paldam (1991), Kaare Strøm and Seymour M. Lipset (1984) Lewis-Beck (1988) suggest that there is no strong evidence of relationship between economic growth and the reelection prospect in the developed countries. Later, Brender and Drazen(2008), Gupta and Panagariya (2014) found that economic growth can raise the probability of reelection of the incumbents in the less developed. - 14 -
the 10% level, while in column 5 the coefficient is not significant and close to zero. These results indicate that voters are more likely to emphasize the role of government as a social service provider as the level of income increases. We can explain the reason why the pledges of welfare state does not attract voters in less developed countries as follows. First, the expansionary fiscal management plan is not credible when it semms that the national economy cannot afford it. Second, election outcomes in the less developed may not be well-explained by the economic condition or policy proposal due to political instability. Especially, in the former socialist countries, which constitute about two-thirds of the sample from the less developed, there are frequent changes in the political dynamics during early stages of transition. The explanatory power of the equation in column 5 and 6 is fairly low, which supports the speculation. Our findings also show how much the policy proposal affect the winning probability of the ruling party quantitatively. In the OECD member countries, raising the proportion of sentences about welfare state in the manifesto of the ruling party by 1 percentage point is related to a 1.1 percentage point increase in the probability of winning. 4.2 Classified by the Level of Democracy To test whether the effect of economic growth and campaign pledges varies according to the quality of institution, we distinguish countries with high level of democracy from the low level. The classification is based on the polity score of Polity IV, which measures how much the country have, in their political system, democratic elements and authoritarian ones. We treat observations, where the polity score is 10 as high level democracies and those with the score between 0 and 9 as low level democracies. 24) In column 1, 2, 5 and 6 at Table 3, 24) Among the 402 observations, 281 elections are classified as held in high level democracies - 15 -
we estimate the same equation as section 4.1 with the classification by the level of democracy. Table 3 > The effect of growth and policy proposal on the election outcome classified by the level of democracy. High Democracy Level Low Democracy Level Dependent variable : GPwin Growth_term Growth _preelection Inflation_term Welfare_GP Left_GP Market_GP Right_GP OECD * Growth_term OECD * Welfare_GP OECD * Market_GP Voter s turnout Seats_GP Constant (1) (2) (3) (4) (5) (6) (7) 0.3827 *** (0.001) -0.0185 (0.808) -0.0497 (0.130) 0.0455 * (0.079) 0.2164 (0.534) 0.0215 (0.114) 0.0383 *** (0.007) -4.2270 *** (0.001) 0.4009 *** (0.001) -0.0246 (0.745) -0.0501 (0.134) -0.0239 (0.459) -0.3254 (0.327) 0.0206 (0.125) 0.0375 *** (0.007) -3.4438 *** (0.007) 0.3949 * (0.051) -0.0184 (0.809) -0.0494 (0.138) 0.0455 * (0.079) 0.2177 (0.535) -0.0138 (0.950) 0.0215 (0.116) 0.0383 *** (0.007) -4.2753 *** (0.006) 0.3806 *** (0.001) -0.0193 (0.800) -0.0524 (0.102) 0.1347 (0.262) 0.2388 (0.494) -0.0924 (0.454) 0.0215 (0.113) 0.0387 *** (0.006) -4.9515 *** (0.002) -0.0014 (0.990) 0.0962 (0.179) 0.0092 (0.505) -0.0028 (0.950) 0.4628 (0.387) 0.0237 (0.288) 0.0383 ** (0.011) -3.6669 ** (0.040) -0.0061 (0.955) 0.0942 (0.191) 0.0099 (0.498) -0.1249 * (0.069) -0.1066 (0.833) 0.0214 (0.346) 0.0362 ** (0.013) -2.8448 * (0.086) -0.0077 (0.944) 0.0952 (0.194) 0.0100 (0.495) -0.1343 (0.160) -0.1005 (0.842) 0.0218 (0.861) 0.0210 (0.356) 0.0362 ** (0.013) -2.7988 * (0.095) Pseudo R 2 0.1480 0.1365 0.1480 0.1494 0.1152 0.1346 0.1348 p( >chi 2 ) 0.0000 0.0000 0.0000 0.0000 0.2701 0.0801 0.1012 Observations 281 281 281 281 101 101 101 Notes: The figures in the table are logit coefficients and the figures in parentheses are p-values. OECD, Majoritarian system, New democracy is included as control variables. * : Significant at the 10 percent level. ** : Significant at the 5 percent level. *** : Significant at the 1 percent level. and 101 as in low level democracies. 20 observations are excluded due to the lack of polity score. - 16 -
The effect of economic growth and pledges about welfare state is found to be positive and significant only in the countries with high level of democracy. One can argue, with this finding, that people in the countries with high level of democracy prefer welfarism and that promise of welfare state expansion is credible only when the democratic institution is well-established. The coefficient of MarketGP in column 5 is also significant, indicating that market-oriented ruling party has a higher chance of losing the election. We can infer that the market-friendly policies with poor political institution are likely to be unwelcom e. 25) Since the estimation result for the high level democracies is almost coincident with that for the developed, which reflects the strong correlation between income and democracy (Acemoglu et al., 2008), 26) we cannot identify the group-specific effect of explanatory variables with above results. In column 3 and 4 we test for the group of high level democracies whether there the coefficients of economic growth and the welfare pledges vary between developed and less developed countries, where we find no significant differences. 27) We find similar result for the coefficient of MarketGP in less democratized group. This results implies that the effect of economic growth and policy proposals on electoral outcome depends on the level of democracy rather than the level of income. 25) We find that an increase of 1 percentage point in the proportion of market oriented policy proposals in the manifesto of the ruling party lower the probability of winning by 3 percentage points. 26) Our sample consist of 259 elections held in the countries with high income and high democracy, 22 in the countries with low income but high democracy, 39 in the high income low democracies, and 62 in the low income low democracies. 27) In column 4, we find the coefficient of WelfareGP in the less developed to be larger than that in the developed. However, it is not statistically significant due to the multicollinearity problem. - 17 -
5. Robustness check 5.1 Fixed and Random Effect As, mentioned in section 3.2, we cannot exclude the possibility that the results are partly influenced by unobserved country characteristics. If the country-specific factors, which are omitted in the equations, affect both election outcomes and the explanatory variables, pooled Logit estimator is not consistent(yachew and Griliches, 1984). In this section, we check the sensitivity of our estimation results by introducing fixed-effect and random-effect Logit model. Table 4 show that our estimates are robust to these specifications. The figures in column 2 and 5 indicate that the effect of growth and policy proposals about welfare state is positive and significant in the countries with high level of democracy. In column 3 and 6, on the other hand, the effect of market-oriented election pledges is still negative but no longer significant. - 18 -
Table 4 > The effect of growth and policy proposal on the election outcome Fixed effect and random effect Dependent variable : GPwin Growth_term Growth _preelection Inflation_term Welfare_GP Left_GP Market_GP Right_GP Voter s turnout Seats_GP Constant All countries 0.3386 *** -0.0163 (0.778) 0.0291 (0.035) 0.0414 * (0.076) -0.0724 (0.807) -0.0174 (0.459) 0.0784 *** Fixed effect High democracy 0.5693 *** -0.0347 (0.692) 0.0265 (0.516) 0.0605 ** (0.037) -0.3324 (0.389) -0.0710 ** (0.037) 0.1234 *** Low democracy 0.1044 (0.422) 0.0485 (0.608) 0.0105 (0.509) -0.1224 (0.193) 0.0971 (0.875) 0.0714 (0.106) 0.0250 (0.233) All countries 0.1957 ** (0.011) 0.0040 (0.938) 0.0118 (0.251) 0.0377 * (0.056) 0.1889 (0.468) 0.0170 (0.102) 0.0428 *** -3.9893 *** Random effect High democracy 0.4037 *** (0.001) -0.0129 (0.876) -0.0445 (0.226) 0.0487 * (0.053) 0.1320 (0.710) 0.0210 (0.154) 0.0538 ** (0.011) -4.7928 *** (0.001) Low democracy -0.0061 (0.953) 0.0942 (0.198) 0.0099 (0.437) -0.1249 (0.128) -0.1066 (0.835) 0.0214 (0.347) 0.0362 ** (0.013) -2.8455 * (0.086) Constant 65.73 70.14 10.89 37.66 33.48 14.93 p( >chi 2 ) 0.0000 0.0000 0.2832 0.0000 0.0002 0.1346 Observations 385 279 82 402 281 101 Notes: The figures in the table are logit coefficients and the figures in parentheses are p-values. OECD, Majoritarian system, New democracy is included as control variables. * : Significant at the 10 percent level. ** : Significant at the 5 percent level. *** : Significant at the 1 percent level. 5.2 Estimation with Government budget balance According to the original PBC theory, the incumbents are likely to expand its spending seeking for reelection(nordhaus, 1975 ; Rogoff, 1990). This argument presuppose that the expansionary fiscal policy is advantageous to the ruling - 19 -
party in the election campaign. As reviewed in section 2, empirical results on the impact of fiscal policy are conflicting each other. Some of the studies present the evidences that support the original theory(aidt et al.,2011; de Haan and Klomp, 2013), while others suggests that voters are likely to reward the incumbents for fiscal austerity (Brender and Drazen, 2008 ; Drazen and Eslava, 2010). In short, fiscal balance can affect election outcomes in either direction. Moreover, budget management of the present government can provide the standard for voters to assess the proposals for fiscal policy. In Table 5, we examine whether the fiscal performance affect the probability that the ruling party win the election and whether the effect of growth and policy proposals is robust to inclusion of new variables. In column 1 and 5, the equation includes Balance_term, which denotes the average value of the ratio of government s budget balance to GDP over the term in office. 28) Pre-electoral fiscal expansion is captured by Retrenchment_preelection, the change in the balance/gdp ratio in the pre-election year relative to the previous year in column 2 and 6. We also take into account the possibility that the effect of realized fiscal policies and proposed ones can interact each other by introducing cross-product terms. We find no significant effect of the fiscal balance and the interactions in both high level democracies and low level democracies. The inclusion of fiscal variables reduces the size of the sample, because most of the countries have shorter series of data on government budget than that on election or macroeconomic condition. 29) Especially, in the former socialist countries, the observations for the early stage of the transition is excluded. Figures In column 5 and 6 show dramatical increase in explanatory power of our model compared with full sample analysis. 30) It implies that the effect of 28) We use the definition of term in office as in the Section 3. 29) The size of the sample decreased from 402 elections in 49 countries to 260 in 46. It consists of 188 elections in the high level democracies, 64 in the low level democracies and 8 undefined. 30) Pseudo R-square is almost doubled and p-value of chi-square statistics is decreased substantially. - 20 -
growth and election pledges can be detected more precisely when the observations for the politically confused period are excluded. We find the effect of growth and policy proposals on the election outcomes to be robust to this specification : the coefficients of Growth_term and Welfare_GP are positive and significant in the countries with high level of democracy, and statement of market economy has negative effect on the probability of winning in the less democratized countries. The largest difference is found in the effect of growth in the countries with low level of democracy. Unlike the results in previous sections, the coefficients of Growth_term is significantly positive, indicating that economic growth positively affect the electoral supports for the ruling party in the less democratized countries as well. That is, if we restrict the sample to recent data series and control the effect of government s fiscal balance, rapid economic growth can help the ruling party to gain the votes regardless of the level of democracy. - 21 -
Table 5 > The effect of growth, budget deficit and policy proposal on the election outcome. Dependent variable : GPwin Growth_term Growth _preelection Inflation_term Welfare_GP Left_GP Market_GP Right_GP Balance_term Retrenchment _Preelection Balance_term * Welfare_GP Balance_term * Market_GP Voter s turnout Seats_GP Constant High Democracy Level Low Democracy Level (1) (2) (3) (4) (5) (6) (7) 0.3365 ** (0.029) -0.0029 (0.976) -0.0045 (0.923) 0.0579 ** (0.047) 0.2304 (0.580) 0.0502 (0.296) 0.0187 (0.251) 0.0265 * (0.096) -4.1214 *** (0.009) 0.3646 ** (0.018) 0.0109 (0.914) -0.0097 (0.838) 0.0703 ** (0.015) 0.2714 (0.514) -0.0602 (0.308) 0.0193 (0.252) 0.0219 (0.159) -4.2353 *** (0.008) 0.3372 ** (0.028) -0.0033 (0.973) -0.0049 (0.917) 0.0584 * (0.064) 0.2344 (0.580) 0.0451 (0.666) 0.0004 (0.953) 0.0186 (0.252) 0.0267 * (0.093) -4.1324 *** (0.009) 0.4956 ** (0.015) 0.0067 (0.960) 0.1659 ** (0.046) 0.1334 *** (0.005) -0.6903 (0.197) -0.0013 (0.986) -0.0983 (0.113) 0.1381 *** 0.3064 * (0.065) 0.0327 (0.782) 0.0649 * (0.053) -0.2959 * (0.072) -0.5049 (0.541) 0.0724 (0.554) 0.0433 (0.185) 0.0437 ** (0.017) -5.3320 ** (0.022) 0.3525 ** (0.037) -0.0049 (0.969) 0.0780 ** (0.031) -0.2783 * (0.088) -0.5645 (0.487) 0.1862 (0.303) 0.0423 (0.172) 0.0415 ** (0.027) -5.3925 ** (0.014) Pseudo R 2 0.1247 0.1228 0.1247 0.2554 0.2633 0.2909 * (0.068) 0.0443 (0.727) 0.0624 * (0.051) -0.3347 (0.126) -0.5103 (0.531) 0.1277 (0.583) -0.0250 (0.743) 0.0440 (0.183) 0.0435 ** (0.017) -5.2325 ** (0.025) p( >chi 2 ) 0.0037 0.0065 0.0062 0.0000 0.0781 0.0931 0.0716 Observations 188 188 188 182 64 64 64 Notes : Balance_term is the average value of the ratio of government s budget balance to GDP over the term in office. Retrenchment_preelection is the change in the balance/gdp ratio in the pre-election year relative to the previous year. * : Significant at the 10 percent level. ** : Significant at the 5 percent level. *** : Significant at the 1 percent level. - 22 -
6. Conclusions The aim of this paper is to investigate the political and economic determinants of election outcomes. Using the data on 402 parliamentary elections in 49 democracies, we look into the effect of economic growth and policy proposals on the winning probability of the ruling party. We also examine whether this effect depends on the level of democracy or level of income. We adopt the panel Logit regression as in Brender and Drazen(2008). Our main findings are summarized as follows. First, economic growth over the term in office has a positive effect on the probabilities that the ruling party win the election. This effect is more obvious in the countries with high income/democracy level than in the countries with low income/democracy level. In the high level democracies, an increase of 1 percentage point in the GDP growth is associated with a 8 10 percentage point increase in the ruling party s winning probability. Second, we find no significant effect of growth in the pre-election year. Third, the ruling party s election pledges on welfare state also has a positive effect on its winning probability, which suggests that voters expected utility is increasing when the welfare-oriented candidates are elected. If the ruling party raise the proportion of pledges on welfare state in their manifesto by 1 percentage point, they have a 1.1 percentage point higher chance of winning. We cannot find this effect in the less democratized countries, which reflects the low level of trust in policies. Fourth, market-oriented ruling party has a higher chance of losing the election in the countries with low level of democracy. Lastly, we find no strong evidence that budget deficit increase or decrease the electoral support of the ruling party. As we stated in section 2, empirical results on the political business cycle and economic voting depend on the sample selection. Our analysis is based on the data on a large panel of countries but most of the sample is taken from - 23 -
the European countries with high income and high level of democracy. Thus, our estimation result are not be applicable to overall countries. More reliable study can be done if information on elections in the non-european developing countries is added to the dataset. In addition, variables indicating quality of institution can be introduced to investigate the political conditions under which a specific policy proposal is effective in the electoral campaigns. References Acemoglu, D., et al. (2008), Income and democracy, American Economic Review, 98(3): 808-842. Aidt, T. S., Veiga, F. J., and Goncalves Veiga, L., (2011), Election results and opportunistic policies: a new test of the rational political business cycle model. Public Choice, 148: 21 44. Alesina, A., and Rosenthal, H., (1995), Economic growth and National elections in the United States: 1915-1988, Partisan Politics, Divided Government, and the Economy, Political Economy of Institutions and Decisions series, Cambridge: Cambridge University Press. Alesina, A., Roberto P., and Jose T.. (1998). The Political Economy of Fiscal Adjustments. Brookings Papers on Economic Activity, 1: 197 248. Alesina, A., Carloni D., Lecce, G., (2011), The electoral consequences of large fiscal adjustments, National Bureau of Economic Research Working Paper 17655. Annika, Lacewell, Lehmann, A., Henrike, Merz, P., Promise, O., Regel, N., Schultze, Volkens, Werner, S., (2013), The Manifesto Data Collection, Manifesto Project (MRG/CMP/MARPOR), Version 2014a. - 24 -
Armingeon, K., & Giger, N., (2008), Conditional punishment: A comparative analysis of the electoral consequences of welfare state retrenchment in OECD nations, 1980 2003, West European Politics, 31(3), 558-580. Brender, A., (2003), The Effect of Fiscal Performance on Local Government Election Results in Israel:1989 998. Journal of Public Economics, 87(9 0): 2187 2205. Brender, A., and Drazen, A., (2005), Political Budget Cycles in New versus Established Democracies. Journal of Monetary Economics, 52(7): 1271 5. Brender, A., and Drazen, A., (2008), How do budget deficits and economic growth affect reelection prospects? Evidence from a large cross-section of countries, American Economic Review, 98(5): 2203-2220. De Haan, J., and Klomp, J.,(2013), Political budget cycles and election outcomes, Public Choice, 157: 245-267. Drazen, A., and Eslava, M., (2010), Electoral manipulation via voter-friendly spending: theory and evidence. Journal of Development Economics, 92: 39 52. Fair, R.C., (1978), The Effect of Economic Events on Votes for President., Review of Economics and Statistics, 6(2): 159 173. Giger, N., & Nelson, M., (2011), The electoral consequences of welfare state retrenchment: Blame avoidance or credit claiming in the era of permanent austerity?, European Journal of Political Research, 50(1): 1-23. Gupta, P., and Panagariya, A., (2014), Growth and Election Outcomes in a Developing Country, Economics and Politics, 26(2): 332-354. Hibbs, D. A., (1977), Political Parties and Macroeconomic Policy, American Political Science Review, 71(4): 1467-1487. - 25 -
Hibbs, D. A., (1987), The American political economy: macroeconomics and electoral politics in the United States, Cambridge: Harvard University Press. Jenkins, J.A., (2000), "Examining the robustness of ideological voting: evidence from the confederate house of representatives", Americal Journal of Political Science, 44(4): 811-822 Lewis-Beck, M. S., (1988), Economics and elections: the major western democracies, University of Michigan Press: Ann Arbor. Lewis-Beck, M. S., & Stegmaier, M., (2000). "Economic determinants of electoral outcomes." Annual Review of Political Science, 3: 183 219 M Woodridge, J., (2010), Econometric Analysis of Cross Section and Panel Data, MIT Press. Neyman, J., Scott, E.L., (1948), Consistent estimation from partially consistent observations, Econometrica, 16: 1-32. Nordhaus, W.D., (1975), "The Political Business Cycle", Review of Economic Studies, 42(2): 169-190. Pacek, A. C., & Radcliff, B., (1995), Economic voting and the welfare state: a cross-national analysis, The journal of Politics, 57(01): 44-61. Paldam, M., (1991), How robust is the vote function? A study of seventeen nations over four decades, Economics and Politics: The Calculus of Support, University of Michigan Press Ann Arbor, pp. 9 31. Peltzman, S., (1992), Voters as Fiscal Conservatives. Quarterly Journal of Economics, 107(2): 327 61. Powell, G. B., & G. D. Whitten., (1993), A cross-national analysis of economic - 26 -
voting: taking account of the political context, American Journal of Political Science, 37: 391 414. Rogoff, K., & Sibert, A., (1988). Elections and Macroeconomic Policy Cycles. Review of Economic Studies, 55(1): 1 6. Rogoff, K., (1990) Equilibrium Political Budget Cycles, American Economic Review, 80: 21-36. Shi, M., & Svensson, J., (2006), "Political budget cycles: do they differ across countries and why?" Journal of Public Economics, 90: 1367 1389. Strom, K. & M. L. Seymour., (1984), Macroeconomics and Macropolitics: The Electoral Performance of Democratic Governments. Tufte, E. R., (1978). Political control of the economy, Princeton: Princeton University Press. Van der Brug, W., (2010), "Structural and Ideological Voting in Age Cohorts", West European Politics, 33(3): 586-607 Zarate, Roberto O., (2013), Zarate s Political Collections (ZPC), 1996 2013. - 27 -