3.1 Economic globalization and the center of gravity: Static perspective

We define economic globalization as the transborder flow of goods and investments and conceptualize it as a multidimensional phenomenon [2]. These trade and capital flows are facilitated by low tariff and non-tariff barriers. We take a disaggregated perspective based on the premise that there is no uniform globalization effect because different components of economic globalization have different consequences for party systems. Our analysis centers on the consequences of the flow of imports and exports. Compared to capital flows, trade flows are the most apparent component of economic globalization and lead to greater scrutiny by voters and parties [15]. The decision is supported by existing research finding no significant effect of foreign direct investment (FDI) on party positioning [8]. Below, we will introduce capital flows in the form of FDI as a robustness test. For the outcome, we focus on the economic dimension because of empirical evidence that the policy space is not well-described by a single left-right dimension and that an economic dimension is central to party competition in virtually all Western European countries ([16], chapters 5, 6).

The sum of total trade relative to the economic output is regularly used as a measure of economic globalization (KOF index [17]). In contrast, we argue and then test the assumption that it is necessary to decompose trade into imports and exports because they have different economic consequences for countries. These economic differences need to be examined and understood because political parties holding different positions toward the national economy respond differently to these national economic effects of globalization. We argue that the actual responses of parties to this incentive differ due to the initially dissimilar positions that parties occupy on the economic dimension. In the following discussion, we adhere to the convention that smaller values on a dimension, as well as a negative correlation between a covariate and the center of gravity, represent more leftish positions and changes.

The central premise of our hypotheses is that parties perceive imports negatively and exports positively. The difference in the perceptions of imports and exports can be explained in three ways, the first being derived from a standard political economy model of political support of organized groups [18, 19], which has previously been effectively applied to tariff making and trade cooperation [20, 21]. Higher imports increase competitive pressure on domestic companies producing goods that compete with imported goods (henceforth, “import-competers”). At the same time, they benefit domestic producers who require imports to serve as inputs to their production process (henceforth, “input-users”) and consumers by lowering the prices of the good. The unilateral reduction of tariffs might be welfare-enhancing, but it creates political backlash because of the organized opposition of import-competers. In contrast, input-users support higher import levels because it is to their advantage. Exporters, the third organized group that is relevant here, are also interested in rising import levels in their home countries. They follow the idea of reciprocity to leverage enhanced domestic market access for easier foreign access and, thus, rising imports at home and abroad.

We argue that parties confronting this constellation of domestic groups target those groups’ interests with specific policy proposals. The negative perception of imports creates an incentive for sheltering import-competers from imports by creating non-tariff barriers to trade as well as supporting measures such as subsidies. A study by Kono [22] corroborates this reasoning by showing that reduced tariff levels coincided with an increase in less visible non-tariff barriers to trade. An example of the use of subsidies to reduce import competition is the agricultural sector in many developed countries. Parties from both the left and right supported subsidizing agricultural producers for an extended period of time to shelter them from import competition, at least to some degree. In the 1980s, for example, the European Community and the United States were waging a “subsidy war” in the wheat sector that was costly to both sides and came at the expense of exporters from third countries ([23] p. 153). With regard to the consequences for party position taking, we argue that the incentive to favor protectionist leftist policies is stronger, the more parties are positioned to the right of the spectrum because parties on the left are already endorsing the idea of sheltering national producers. Import levels should therefore be negatively correlated with the COG.

The second mechanism works via the expectations of voters who feel more exposed to the pressure of rising imports. Voters with jobs that are vulnerable to rising imports expect parties to diminish their vulnerability, for example, by spending more on job training programs, etc. or by increasing spending (compensation hypothesis [24]). A counterargument is that left parties might also move more to the right of the economic dimension, reflecting their belief that spending levels cannot be maintained under conditions of economic globalization (efficiency hypothesis [25]). Our reasoning on the incentives for party positioning that work via voters does not require that welfare spending or spending more generally increases, which is the traditional focus of the compensation hypothesis. Instead, we argue that parties have incentives to offer policies that shelter voters from import competition, which might work in the same way as for import-competers, i.e., by pursuing statist industrial policies such as granting subsidies to industries. Trade policy making of the current Trump administration is an example of the protection of workers in import-competing sectors without financial compensation or increased welfare spending. The steel sector, which faces strong competition, is simply protected by increased tariff levels that reduce the level of imports and (supposedly) save jobs in the US steel industry. Following this argument, parties that are located on the left have already assumed a position that is in accord with the demand for higher compensation, whatever form it might take. The further a party is located to the right, the stronger the incentive to adjust its position and make a move to the left to compete for compensation-demanding voters.

A third possible mechanism is a mercantilist understanding of trade policy making. Mercantilism is long outdated among economists [26], but one can observe mercantilist thinking over the past 200 years until the present. Countries that import more than they export run current account deficits, which are traditionally considered to be undesirable because they are thought to be a sign of economic weakness [15]. Current account deficits and surpluses have been a salient issue since the beginning of modern trade relations in the 19^th century [27] and were also central to the recent Euro crisis. Germany has been running an account surplus since the mid-2000s and has lauded this as a sign of Germany’s economic strength and a role model for struggling countries such as Greece and Portugal [28]. This reasoning reflects the view that export surpluses equal strength and that high levels of imports are bad for an economy (which does not make sense in economic terms, but that is another matter). The policy implications of mercantilism are to make import-competers more competitive because the stronger they are, the less goods will be imported.

We argue that the mechanisms should hold in times of increasing international division of labor and a shift from inter-industry to intra-industry trade. If one perceives imports as an indication of domestic economic weakness or as a threat to the support of import-competers, it does not matter whether goods are traded on an inter-industry or intra-industry level. For example, this is reflected in the mercantilist move by the current Trump administration, the arguments of pro-Brexit groups who wanted to retake control over trade relations and imports and the support of German car manufacturers by the German government, that repeatedly and with different partisan composition prevented the regulation of emissions in the EU Council of Ministers.

Hypothesis 1: Increasing levels of imports are associated with a leftward shift of the center of gravity and decreasing levels of imports with a rightward shift on the economic dimension of party competition.

In contrast to imports, we find that exports are welcomed by all parties because they benefit exporters, who increase their support for parties strengthening the exporter’s competitiveness [29]. Workers employed by exporters also benefit from higher exports because that secures their jobs and increases their income. The agricultural industry is a case in point for these arguments because, for decades, developed countries have implemented export promotion programs to support their producers ([30], p. 116). From a mercantilist perspective, exports are desirable simply because they contribute to a current account surplus.

These points at least hold for developed economies that export manufactured products and services, although it might differ for developing countries predominantly exporting primary goods or those that export rare goods such as the minerals needed in the telecommunications industry. However, our theoretical arguments are confined to developed economies. Exports have been interpreted this way from the inception of the trade doctrine of mercantilism until the present day [15], including the Euro crisis wherein the ailing states were blamed for a lack of economic competitiveness and weak export industries. We assume that parties agree on the need to support exporters via appropriate policies, whereas the exact policies that are favored depend on whether a party is positioned on the left or right side of the political spectrum. Following theories on partisan ideology and economic policy making [31, 32], we argue that parties located more to the right of the spectrum endorse supply-side policies such as lower corporate taxes and deregulation of the labor market. Parties on the left embrace demand-side policies that are unsuitable for the support for exporters. Instead, they seek to shelter domestic producers from import competition via tariff and non-tariff barriers because of their belief that less competition at home means enhanced competitiveness abroad [33]. Another means by which parties can promote exports are subsidies, which are also an instrument of the political left, either in the form of export subsidies or subsidies specifically targeted at domestic producers [34]. In contrast to lower taxes and less regulation, subsidies represent an intervention of the state into the market and are not an element of supply-side politics.

Following this reasoning, leftist and rightist parties are already located on the proper sides of the spectrum, “proper” meaning that the current position is largely in line with the incentives created by exports. When exports increase, we expect parties located on the left to either remain stationary or move more to the left, on average, while parties on the right either stay put or shift further to the right. We therefore hypothesize that the rightward shifts of rightist parties and leftward shifts of leftist parties cancel each other out, with the consequence that the ideological center of gravity remains in place.

Hypothesis 2: Changes in the level of exports are unrelated with a party system’s center of gravity on the economic dimension of party competition.

We do not argue that all parties held uniform views on economic globalization during these periods (the same holds for the hypotheses we develop in the following section). However, we believe that it is possible to develop arguments applicable to many, in particular, the major European parties that have more weight for a party system’s COG.

3.2 Economic globalization and the center of gravity: Dynamic perspective

Economic globalization has been a contested issue among the public, politicians and organized groups since the beginning of the modern globalization era in the mid-19^th century [15]. We expect the two static hypotheses to hold in an aggregate analysis pooling the data from multiple decades because these should be, on average, a relationship between imports, exports and the COG. However, we also believe that it is important to take a dynamic perspective based on the assumption that the parties’ view on and responses to imports and exports have changed over time. One could choose any four decades from 1860, which is usually seen as the starting point of modern trade relations [35], and would observe at least one major change in economic globalization and how countries perceived and responded to it (see for example, [15]).

This general empirical insight motivates the formulation of multiple dynamic hypotheses for our period of analysis, which ranges from 1974 to 2015. We divide it into three stages according to our expectation of Western European parties’ responses to imports and exports. The expectations and hypotheses are derived from existing research on partisan behavior during these four decades and our own perspective on the empirical developments. We are unable to precisely determine the breakpoints between the three periods because policy beliefs seamlessly supersede one another. However, we think that it is possible to determine the approximate beginning and end of a stage. The empirical analysis of the dynamic hypotheses is a moving-window analysis that does not require a sharp differentiation between the three periods.

The first stage ranges from the 1960s to the mid-1980s. This period is characterized by the economic problems of the 1960s and 1970s, which became manifest in the budget and currency problems of the United States [36], the end of the Bretton Woods system, the two oil crises and the economic stagnation combined with high inflation that existed until the early 1980s. Keynesian demand policies, which are inward-oriented and seek to stimulate domestic demand via public deficit spending, prevailed during the 1970s among left parties [37]. At the same time, parties on the right of the spectrum started leaning toward monetarist policies. The Reagan administration initiated the Tax Reform Act of 1986, which set the precedent for cutting the top marginal income tax rates in many Western European countries [38]. On the international level, it would be appropriate to include the GATT Tokyo Round, but that only achieved moderate progress because states were cautious about pushing forward liberalization in difficult economic times [39]. Taken together, we argue that the positions of leftist and rightist parties in the 1970s and early 1980s were driven by the motivations that we previously laid out for the static hypotheses.

Hypothesis 3: Levels of exports are not associated with the center of gravity between the 1970s and the mid-1980s.

Imports were seen negatively by all parties in difficult economic times, who sought to counter them with leftist, protectionist measures and account for a negative association between imports and the COG. For example, Western countries used their superior political and economic power to protect their textiles industry against the upcoming competition of developing countries by negotiating import quotas and voluntary export restraints for textile exports [40].

Hypothesis 4: Levels of imports are negatively associated with the center of gravity between the 1970s and the mid-1980s.

The late 1980s and 1990s represented a turning point because left parties, social-democratic and socialist parties in particular, started embracing monetarist and supply-side policies [41]. One can date the peak of this period of moderation to the late 1990s and early 2000s ([42], p. 801), when leftist governments in Europe pursued “The Third Way”, lowered corporate taxes and engaged in cutbacks of the state in a manner not previously conceivable. For example, in a series of reforms, Labour raised the bar for the receipt of welfare state benefits to increase pressure on the unemployed to find a job ([43], p. 84). Regardless of party ideology, parties endorsed the idea of privatizing state-owned monopolists in a wide range of sectors, including aviation, postal services and telecommunication [44, 45].

Because parties on the left and right endorsed supply-side policies, liberalization and increased economic globalization during this period, we expect that the link between imports and exports and the COG was most pronounced between the mid-1980s and early 2000s. Parties on the right also reflected a changed attitude towards globalization by taking more radical positions and pushing for stronger retrenchment of the state and lower taxes than did the left parties [46]. For the period between the mid-1980s and early 2000s, we hypothesize that a positive association between exports and the COG existed because both left parties and right parties pursued supply-side policies with the goal of stimulating them.

Hypothesis 5: Between the mid-1980s and the early 2000s, the association between levels of exports and the center of gravity is positive.

With regard to imports, we expect that the negative association is stronger in the second period than in the first. A strengthening negative effect is a direct consequence of economic liberalization and globalization because it stimulated exports and economic integration. Since trade liberalization is usually reciprocal, countries have to buy an increase in exports by accepting higher levels of imports [33]. Import-competers feel increased pressure from rising imports and voters become increasingly vulnerable to tougher competition, making them susceptible to more radical parties that propose protectionist policies (economic nationalism, [47, 48]). We argue that left and right parties did not become blind to producers and voters exposed to import competition and endorse policies that at least partially compensate for the increased exposure that parties expected to follow due to import competition [24]. As Kono [22] shows, measures of protection and support that are less visible than tariffs increase when tariffs decrease. Tariff cuts and higher levels of imports more generally therefore do not automatically imply that parties stand by and watch when competition increases for import-competers. Two examples for a demand-side policy that is aimed at workers that falls right in the high days of economic integration and liberalization in the second period is the introduction of a minimum wage in the United Kingdom (1999) and Ireland (2000) [49]. We therefore theorize that higher imports shift the COG to the left in the second period and that the effect is stronger than in the first period.

Hypothesis 6: Between the mid-1980s and the early 2000s, the association between levels of imports and the center of gravity is more negative than in the first period.

The stock market crash and economic downturn of the early 2000s constrained national budgets and marked a turning point in attitudes toward economic liberalization among the public and parties. On the international level, the adverse economic development and demise of liberalization were foreshadowed by public protests against a continuation of economic globalization, for example, in Genoa in 2001, and the failure to successfully conclude the so-called Doha Round [50]. The comparatively low-interest rate policies of central banks aimed at supporting economic growth in Western countries then contributed to the second stock market crash in the late 2000s [51], which lead to increased skepticism to economic liberalization and deregulation. We argue that, during that decade, leftist parties reversed their attitudes and turned again to demand-side policies and away from supply-side policies ([52], chapter 6). Parties in the right spectrum could ignore neither the changed economic environment nor public attitudes toward globalization and we expect that they had to moderate their position as well. Take the case of Denmark, for example: In November 2001, the right-wing liberal Venstre party won the general election, formed a minority government with the Conservative People’s Party and governed until 2011. Under two prime ministers, Anders Fogh Rasmussen and Lars Løkke Rasmussen, there were no significant reforms of the welfare state, and their public spending exceeded that of the Social Democratic previous government [53].

We expect the financial crisis that started in 2007 further affects the COG in three ways that are specific to this period. First, the need to bail out banks to avoid bankruptcy meant that countries nationalized or subsidized them ([52], chapter 6), which are traditionally left issues. This was a quick response of parties regardless of their ideology, which we believe to be reflected in the positions of parties in the late 2000s and early 2010s to some degree. Altogether, this should move all parties more to the left. Second, in the medium run, the financial crisis spurred attempts to more tightly re-regulate the financial sector and at least partially reverse the liberalization measures taken in the 1990s and early 2000s. Third, the financial crisis of 2007, bailouts of banks and austerity policies in some countries contributed to the formation of new and strengthening of existing populist parties on the left and right of the spectrum. Left-wing and right-wing populist parties differ in many respects, but have in common that they favor economic nationalism. A key element of economic nationalism is to reduce imports by raising barriers to trade as classic left policy positions [54]. While economically nationalist parties on the left and right have protectionism and economic radicalism in common [55], we expect them to differ in their attitudes toward exports. Exports are compatible with economic nationalism because it is not equivalent with autarchy and the left-right differences between left-wing and right-wing parties should matter here in the ways we have theorized.

Because of these developments, we argue for the third period that the relationship between imports, exports and the COG is again becoming more similar to the relationship in the first period. Existing, non-populist parties were confronted with the consequences of the financial crises of the early 2000s and 2007, seen as a result of the liberalization progress in the previous decade, and returned to what could be called their traditional policies. This development is complemented by the formation and rise of economically nationalist parties that seek protectionist measures against imports.

Hypothesis 7: Since the early 2000s, the association between levels of exports and imports and the center of gravity is positive and attenuating compared to the second period.

Table 1 summarizes the expected effects of exports and imports on the ideological center of gravity of party systems.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Hypotheses—Expected effects of globalization variables on the COG.

https://doi.org/10.1371/journal.pone.0212945.t001

4.1 Dependent variable

The dependent variable is derived from the Manifesto Project Database (MARPOR) data [56]. MARPOR is based on human codings of party platforms and assigns quasi-sentences to fifty-six policy categories, plus one category for non-assignable quasi-sentences. These are divided by the number of all quasi-sentences to measure the percentage of all quasi-sentences for each category. We use the percentages to estimate the party position on an economic dimension. In Table 2, we compare existing constructions of the economic dimension of party competition [57–61].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Comparison of economic dimension of party competition variables.

https://doi.org/10.1371/journal.pone.0212945.t002

The comparison shows a large intersecting set, with all categories being used belonging to MARPOR domains four, five and seven. Not one of the 21 categories that is coded right by one approach is coded as left by another, or vice versa. However, the different choices of categories for the measurement of party positions reflect that theory does not dictate the choice of one specific set of items. Because of this uncertainty, we decided to follow a maximum approach toward the measurement of party positions on an economic dimension. The maximum approach builds the union of all 21 categories that have been used in one of the five studies to measure party positions. Conceptually, we find this approach defensible because each of the 21 categories can be plausibly linked to the economic dimension of party competition.

We follow customary practice and derive a position estimate by subtracting left from right percentages. A drawback of building the difference is that the estimate is not “independent of irrelevant alternatives” because changes in the number of quasi-sentences not feeding directly into the economic position estimate might affect that estimate ([61], p. 91). We address this problem and standardize the difference between right and left number by dividing it with the sum of the right and left percentages: (R-L)/(R+L) [62].

A simple example illustrates the advantage of the ratio measure: For a party manifesto that only consists of 60 right and 20 left quasi-sentences, the right and left percentages are 75 and 25 and the position estimate 50 on a scale from -100 to +100. The standardized measure is 0.5 on a scale ranging from -1 to +1. Now imagine the party adds 20 new quasi-sentences that are neither right nor left on the economic dimension. The non-adjusted difference is 40 (60–20), meaning the party would be estimated to be slightly more centrist, although neither the right nor left content of the manifesto changed. In contrast, the ratio is unaffected by the 20 new sentences because their number remains constant and we again estimate a position at 0.5. We decided against a logit transformation of right and left quasi-sentences [63]. Our procedure takes the same problem into account as the logit transformation, but we find it more intuitive. Because of the importance of the economic dimension and the relatively high number of quasi-sentences underlying it, there is also no need to smoothen estimates to address low numbers of quasi-sentences.

Under certain circumstances, the use of the ratio measure can generate undesired results: The fewer the number of included items is, the easier it can lead to very polarized scales. A manifesto with one or only few right issues and no left issues receives the extreme score of +1. To take that scenario into account it is reasonable to include the highest number of issues that are clearly attributable to either left or right economic policy positions. We therefore continue with our “broad” definition of the economic party positions. Table 3 shows the correlations of equally-constructed economic left-right measures that are based on the different selections of categories shown in Table 2. The strength of the correlations ranges from moderate to very high levels. In a robustness analysis we report below, we test whether the empirical results depend on the choice of measurement approach.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Correlation matrix for different ratio measures.

https://doi.org/10.1371/journal.pone.0212945.t003

We calculate the COG of a country at time t by weighting the position of each parliamentary party on the economic dimension, derived with the maximum approach, with its vote share and taking the sum over the weighted positions. The range of the center lies between -1 and +1 from left to right. Fig 1 presents the development of the COG for the 15 countries and period of analysis that we cover with our study. The figure demonstrates that, both within and between countries, the COG fluctuates over time, which is the variation we seek to exploit. The mean COG across all observations is -0.48, which partially reflects that the economic dimension includes more left than right categories. The line plot might overstate the number of observations per country, which is equal to or less than nine for nine countries out of fifteen. Detrending the imports and exports variables is therefore not warranted [64].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Center of gravity for 15 countries for our period of analysis.

https://doi.org/10.1371/journal.pone.0212945.g001

4.2 Main covariates

We operationalize ‘economic globalization’ via measures of capital and trade flows and use data extracted from the World Bank’s “World Development Indicators” [65]. We measure the export and import dimension with exports and imports of goods and services standardized with the gross domestic product (GDP) of a country. We lag the globalization variables by one year to account for a temporal lag of the potential effects.

4.3 Additional covariates

We add a public opinion measure as a control variable because parties have been found to be responsive to public opinion and the median voter [66], which might have follow-up effects for the COG. We calculate the mean self-reported left-right position of respondents in the Eurobarometer as a measure for the mean voter position. Empirical studies show that the mean voter and mean citizen position are almost perfectly correlated [67], allowing their synonymous use. Since there is no time-series cross-section data allowing us to calculate the mean voter position on the economic dimension, we must rely on public opinion data that locate respondents on the main left-right dimension. The economic dimension and main dimension are strongly correlated in some countries and only weakly in others ([16]), chapter 4, 5], meaning that the public opinion variable is likely to contain measurement error. The public opinion variable is lagged by one year. If more than one Eurobarometer wave was available per year, we pooled data from multiple waves before calculating the mean voter position. We use GDP growth as a control for a country’s economic development. The willingness to implement supply-side reforms increases, the worse the economic situation. GDP per capita controls for the level of economic development because more developed countries might have a more leftist center due to an increased focus on postmaterialist values in the electorate as opposed to materialism [68]. All control variables are lagged by one year.

In total, our data comprises 129 observations from elections in 15 countries from 1974 to 2015. We aimed to maximize the spatio-temporal coverage of our data with a focus on Europe to have sufficient power for the moving-window analysis. The main data constraint derives from the availability of the economic variables and the Eurobarometer data. An overview of the variables, operationalization and sources as well as a list of included elections is included in the supporting information (see S1 and S2 Tables as well as S1 Fig). Descriptive statistics for key and control variables are shown in Table 4. See S7 Table for descriptive statistics of the KOF indices that we use below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Descriptive statistics.

https://doi.org/10.1371/journal.pone.0212945.t004

4.4 Empirical strategy

We first focus on the time-invariant hypotheses with ordinary least-squares regression (OLS). A test for first-order serial autocorrelation rejects the null of no autocorrelation. A test for second-order autocorrelation fails to reject the null. Two F-tests for country fixed-effects and year fixed- effects fail to reject the null of no fixed effects. Based on the diagnostics, we decided to estimate an OLS regression with a lagged dependent variable and standard errors clustered by countries and years. The test for first-order serial autocorrelation when the lagged dependent variable is included fails to reject the null hypothesis of no autocorrelation. Although the F-tests indicate that there is no unobserved heterogeneity between or within countries, we opted for the two-way clustering of standard errors because the panel structure of the data makes it plausible to argue that errors are correlated on both dimensions [69]. Clustered standard errors have been shown to be superior to fixed effects when the unobserved effect varies instead of being constant, which we find plausible in an analysis covering four decades [70]. The following analysis is completely documented in a Stata log-file (S1 File).