Automation and the Welfare State: Technological Change as a Determinant of Redistribution Preferences

Abstract

Technological change is widely considered to be a key driver of the economic and occupational structure of affluent countries. Current advances in information technology have led to a significant substitution of routine work by capital, while occupations with abstract or interpersonal manual task structures are complemented or unaffected. We develop a simple theoretical framework for the reasons why individuals in routine task-intensive occupations would prefer public insurance against the increased risk of future income loss resulting from automation. Moreover, we contend that this relation will be stronger for richer individuals who have more to lose from automation. We focus on the role of occupational elements of risk exposure and challenge some general interpretations of the determinants of redistribution preferences. We test the implications of our theoretical framework with survey data for 17 European countries between 2002 and 2012. While up to now the political economy literature has emphasized other occupational risks, we find vulnerability to automation to be an important determinant of the demand for redistribution that should not be ignored.

Keywords

technological change redistribution preferences welfare state European Social Survey

Introduction

Technological change is widely regarded as one of the main drivers of long-term economic development (Romer, 1990). By complementing occupations with certain skill profiles while making others redundant, it structures employment and significantly shapes the occupational structure (Goldin & Katz, 2008; Oesch, 2013). Technological innovations can have far-reaching social implications that differ across occupations. For Marx, technology contributed to the creation of a “reserve army of the unemployed,” which allowed capitalists to reduce the wages of the working classes. More optimistically, technological change enables specialization and skill upgrading, which facilitates a move away from routine labor (Erikson & Goldthorpe, 1992; Iversen & Cusack, 2000; Wren, 2013).

Current technological innovations are strongly connected to computer-based information technology. Its momentous implementation in the last decades has been spurred by significant real price declines in computing power (Autor, Levy, & Murnane, 2003). Computers can perform routine tasks, which are well defined and repetitive, and can complement complex and more ambiguous abstract task structures. They, however, have a limited effect on interpersonal service tasks. An earlier literature on skill-biased technological change associated these developments to the significant increase in wage inequality between those with university degrees and those with a high school degree or less (Katz & Murphy, 1992). More recently, technological change is argued to promote a significant decrease in the share of routine occupations that occupy the middle of the educational and wage distributions. Information technology, therefore, does not lead to linear upskilling of work, but rather to a process of polarization (Autor, Dorn, & Hanson, 2015; Spitz-Oener, 2006).

Technological change is, therefore, connected to income inequality (Goldin & Katz, 2008; Michaels, Natraj, & Van Reenen, 2014) and, as such, it affects the political economy of industrialized democracies in important ways. Inequality is frequently invoked as an explanation for a number of crucial issues in political science. It is often considered a determinant of processes as diverse as the decline of electoral turnout (Nie & Kim, 1978; Rosenstone & Hansen, 1993), the increase in the support of extreme-right parties (Betz, 1994), or the likelihood of political conflict (for a review, see Lichbach, 1989). This article wishes to address one of the assumptions underlying most arguments about the influence of inequality on political outcomes. If inequality matters to individual political behavior, it seems reasonable to assume that it does so through the effects of income on redistribution and social policy preferences. These redistribution preferences may (or may not) then be reflected on party positions and, eventually, government policy. To begin at the beginning, the issue of whether technological change is a significant determinant of redistribution preferences is a topic in need of further analysis.

Given the pervasive substitution effects of information technology on routine occupations, we would expect individuals holding routine occupations to have strong preferences for nonmarket protection to insure against increased risk of employment and wage loss. The idea that insurance motivations are a significant determinant of preferences for redistribution has become prominent in the recent comparative political economy literature. Although some authors mention risks resulting from technological change,¹ the vulnerability of concrete occupations to technological change is rarely examined directly. To our knowledge, only Kitschelt and Rehm (2014) analyze the relationship between routine occupations and redistribution preferences, but they focus on education as the key defining factor for these occupations, rather than measuring routine task intensity (RTI) directly.

In this article, we focus on the risks associated to technological change, and argue that the degree of RTI in different occupations is an important determinant of redistribution preferences. We develop a simple theoretical framework in which risk-averse individuals support redistribution because of their desire to insure against occupational hazards. We also argue that insurance preferences resulting from technological risks will be accentuated by income. We argue that income plays an intermediating role, because individuals will have more to lose from automation when their income level is higher. These arguments aim to reexamine the effects of redistribution and insurance motivations on redistribution preferences, while emphasizing the importance of a factor that has received a remarkable amount of popular attention (see, for example, Martin Ford’s Rise of the Robots: Technology and the Threat of a Jobless Future).

The Argument

As suggested above, technological innovation entails occupational risks for those individuals whose occupations are susceptible to automation. Vulnerability to automation, in turn, is related to the routine tasks involved in any particular occupation. Individuals will, therefore, favor redistribution as insurance against the risk of automation when the RTI of their occupation is high. Below, we will further argue that this positive effect of RTI on preferences for redistribution is affected by income: RTI becomes a stronger influence on redistribution preferences when an individual has more to lose from automation (i.e., when her income is higher).

Technological Change as an Unequally Distributed Occupational Risk

Current innovations in information technology are generally viewed to have strong and dissimilar effects across occupations (Goldin & Katz, 2008; Oesch, 2013; Wren, 2013; Thewissen & Van Vliet, forthcoming). They complement individuals with abstract or personal tasks, while individuals in routine occupations face an increased risk of being substituted by capital (Autor et al., 2015). Routine tasks can be partitioned into step-by-step rules and do not require cognitive or service skills that are difficult to automate (Goos & Manning, 2007; Goos, Manning, & Salomons, 2014). It is important to emphasize that routine tasks susceptible to automation might well be complex and require extensive educational training (e.g., bookkeeping). Because of this, innovations in information technology do not affect occupations linearly across educational levels. In fact, routine occupations tend to lie in the middle of the educational and income distribution (Oesch, 2013).

Advances in information technology have been found to significantly affect the occupational structure of industrialized democracies in the last couple of decades. Oesch (2013) finds a decrease of relative employment between 29% and 41% in routine occupations in Denmark, Germany, Spain, Switzerland, and the United Kingdom from around 1990 to 2008, whereas employment in nonroutine analytical and interactive occupations increased by 23% to 41%. Michaels et al. (2014), using data for the United States, Japan, and nine European countries between 1984 and 2004, report strong polarizing effects of information technology, accounting for a quarter of the growth in relative demand toward nonroutine high-skilled labor. Goos et al. (2014) analyze the period between 1993 and 2010 in 16 Western European countries, and show that technological change and offshoring can account for three quarters of the observed increase in high-skilled nonroutine work and decrease in medium-skilled routine employment.²

RTI as Determinant of Preferences for Redistribution

In the traditional political economy literature, redistribution preferences are a function of material self-interest (Meltzer & Richard, 1981). The Meltzer–Richard (MR) model assumes that the preferences of the median voter determine government policy and that the median voter seeks to maximize current income. If there are no deadweight costs to redistribution, all voters with incomes below the mean maximize their utility by imposing a 100% tax rate. Conversely, all voters with incomes above the mean prefer a tax rate of zero. When there are distortionary costs to taxation, the MR model implies that, by increasing the distance between the median and the mean incomes, more inequality should be associated with more redistribution.

More recently, scholars have questioned the idea that material self-interest motivations should be limited to a measure of present income. This approach distinguishes an insurance component of redistribution preferences that incorporates an intertemporal element in material self-interest. Individuals will insure against uncertain future income levels and will therefore favor social protection when they are exposed to an increased risk of job or wage loss. As these forms of social security (such as unemployment benefits or social assistance) are redistributive,³ redistribution preferences for individuals exposed to these risks will be high (Moene & Wallerstein, 2001; Sinn, 1995; Iversen & Soskice, 2001, 2009; Rehm, 2009). Most insurance models of redistribution preferences integrate four elements: (a) the risk of job/wage loss, (b) the likelihood of regaining employment, (c) the degree of risk aversion, and (d) the presence of some policy that redistributes resources to those who experience the job/wage loss. Our main contribution in this article is to argue that the odds of becoming unemployed in (a) are a function of RTI.

Two articles have been particularly influential in the insurance approach to the determinants of social protection. We will contrast our reasoning to theirs. First, Moene and Wallerstein (2001) propose that insurance is a normal good, leading individuals to prefer more of it when their income rises. Assuming that individuals are sufficiently risk averse, so that the insurance motive dominates the MR redistribution motive, income will positively affect preferences for redistribution, holding risk and risk aversion constant. In this model, risk of job loss is a function of income: it is lower (or set to zero) for high-income than for low-income groups. Second, Iversen and Soskice (2001) argue that individuals with specific, as opposed to general, skills will favor insurance as protection against their investment in human capital. In their model, there is a homogeneous risk of job loss across the electorate, but the opportunities for reemployment are lower for individuals who have invested in specific skills. Holding income and risk aversion constant, an increase in the ratio of specific versus general skills will lead individuals to prefer higher levels of nonmarket insurance.

Our point of departure lies closer to the Iversen and Soskice model, as we explicitly recognize an occupational hazard, independent of the level of income, that translates into higher preferences for nonmarket protection. We part ways with their argument by emphasizing that the risk of job or wage loss depends on the occupational level of RTI, instead of focusing on the effects of skill specificity on reemployment possibilities. The implication is quite distinct: given a level of income and risk aversion, the level of RTI of an occupation positively affects preferences for redistribution.

Although technological change has not been recognized as an important determinant of redistribution preferences in the comparative political literature, it is germane to ask whether RTI is, in fact, related to more traditional occupational risks. We will show below that, empirically, the correlation between RTI and other occupational risks is low. Theoretically, they are distinct concepts as well.⁴

Kitschelt and Rehm (2014) mention routine occupations in their analysis of the relationship between occupational characteristics and redistribution preferences. As we show in more detail in Appendix A, however, their operationalization follows educational and income lines and does not capture the degree of occupational RTI. Kitschelt and Rehm, in fact, do not argue that individuals in routine occupations favor more redistribution as insurance against increased risks due to automation. Rather, elaborating on Oesch (2006), they are interested in occupations as the source of socialization profiles. They differentiate occupations based on discretionary disposal over own work (the “logic of authority”), and they hypothesize individuals with more discretionary space and authority over subordinate employees will find preserving material incentives to be important, and, therefore, will be against redistribution.⁵

As mentioned above, a common component of the more traditional conception of occupational risk is skills (Cusack, Iversen, & Rehm, 2006; Iversen & Soskice, 2001). Skill specificity reflects investments in human capital and, consequently, affects occupational risks. In the Iversen and Soskice approach, therefore, the distinction that matters is between general and specific skills, not whether a certain skill (be it specific or general) is routine, manual, or abstract. There are no a priori reasons to believe that the specificity of skills is related to the degree of occupational RTI. As an example, models, salespersons, and demonstrators have among the most general skills, whereas stationary-plant and related operators have very specific skills. In terms of RTI, however, these occupations are very comparable—both are average as we will show below.

It is also important to note that the outsourcing of production and its specific effects on certain occupations is significantly connected to risk (Grossman & Rossi-Hansberg, 2008). The crucial factor is the degree to which parts of the production process can be executed abroad, and how this offshorability is concentrated on particular activities. Walter and coauthors have explored how offshorability affects redistribution preferences (Dancygier & Walter, 2015; Rommel & Walter, 2014; Walter, 2010, 2017). But here again, we argue there is an analytic distinction between offshorable and automatable occupations (Autor et al., 2015; Goos et al., 2014; Oesch, 2013). There are occupations that can be executed abroad but require nonroutine cognitive skills that are difficult to automate (like those in architecture, software developing, or statistical analysis). And there are occupations that are routine and can be computerized but require spatial proximity (like security guards or customer service clerks). Moreover, studies analyzing the determinants of occupational structure find much weaker or insignificant effects of international trade and offshoring once the impact of technological change is accounted for (see Autor et al., 2003; Goos et al., 2014; Spitz-Oener, 2006).

The Mediating Effect of Income

The last part of our argument concerns a factor that can exacerbate the (positive) effect of RTI on preferences for redistribution. We argue that the importance of RTI as a determinant of nonmarket insurance demand will be increasing in the level of present income. If an individual has relatively more to lose from an occupational risk, then this risk will become more decisive in determining her preferences for nonmarket protection. This view is related but, again, deviates in significant ways from existing models of redistribution. As mentioned above, the MR model emphasizes current income as the determinant of redistribution demand and does not consider insurance motivations. Income plays a similar role for Iversen and Soskice (2001): depending on risk aversion and the nature of benefits, redistribution preferences are negatively associated to present income, and they experience a general increase when an individual possesses specific skills (because of insurance-related reasons). In fact, the effect of skill specificity in the Iversen and Soskice model is not income dependent, insurance motivations are expected to produce a similar increase in demand for protection whether an individual’s income is high or low.

The model of Moene and Wallerstein (2001) is also connected to the argument we are presenting, because it argues that insurance is a normal good that individuals will demand more of as their income goes up. Moene and Wallerstein, however, focus on the effects of a mean-preserving increase of macroinequality on individual demand for insurance. We emphasize the greater effect of RTI vulnerability on the demand for insurance promoted by increased levels of individual income. In their model, income is positively associated with demand for redistribution. In our model, however, income has a direct negative effect on preferred levels of redistribution, but it will positively influence the effects of RTI (our risk exposure variable) on redistribution preferences.

Our theoretical expectations (and our empirical model to be developed below), therefore, contain three components: (a) income, which in traditional MR fashion is associated with decreasing support for redistribution; (b) RTI, which captures risk exposure and insurance motivations (increasing RTI is positively associated with support for redistribution at any given level of income); and (c) the interaction between income and RTI (we argue that insurance motivations become more influential as individuals have more income to lose, which implies that the effect of RTI will increase as income increases). These intuitions imply that the relationship between income (in an x-axis) and redistribution preferences (in the y-axis) would be represented by a negative slope. The influence of RTI would shift the slope higher (the insurance motivation increasing support for redistribution at any given level of income). And, the negative slope for the relationship between income and redistribution preferences would be much flatter for high levels of RTI than for lower levels of RTI (reflecting the increasing influence of RTI as income grows). For the empirical model to be estimated below, (a) implies a negative direct effect for income, (b) implies a positive direct effect for RTI, and (c) implies a positive effect for the interaction between income and RTI.

Some scholars have argued that educational levels moderate the effects of offshoring on redistribution preferences, because high-skilled individuals benefit from globalization, whereas low-skilled individuals do not (Dancygier & Walter, 2015; Walter, 2010). Others have put forward country-level institutions as a moderating factor for the effects of skill specificity on preferences for insurance (Gingrich & Ansell, 2012).⁶ But, to our knowledge, the individual level of income has not been considered an intermediating factor for RTI effects in existing studies on redistribution preferences.

Data

RTI Across Occupations

In the theoretical section, we have argued that individuals holding routine occupations bear the risks of wage or employment loss from automation. We use the RTI index from Goos et al. (2014), who rely on Autor and Dorn (2013) and Autor et al. (2015). Goos, Manning, and Salomons distinguish between routine, manual, and abstract task inputs, derived per occupation from the Dictionary of Occupational Titles (DOT). The RTI index measures the log routine task input per occupation, minus the log manual and abstract task inputs, so that the measure is increasing in the relative importance of routine tasks vis-à-vis manual and abstract tasks. As the RTI index gauges the tasks structure of an occupation, the index is time and country invariant. Goos et al. rescale these measures to mean 0 and standard deviation 1. The index is available at the two-digit occupational International Standard Classification of Occupations (ISCO)-88 level.⁷

Two additional occupational measures of the degree of RTI are available. The first, from Oesch (2013), is again based on the differences between routine, manual, and abstract (or analytical and interactive) tasks. This RTI measure contains information at the four-digit ISCO-88 level, and distinguishes occupations into multiple nonroutine and routine occupations drawing on Spitz-Oener (2006). These occupational categories can be combined into a dummy equal to 1 if an occupation is routine, and equal to 0 if otherwise. This dummy indicator and the continuous variable from Goos et al. (2014) are highly correlated (.73). As we have more variation for the continuous RTI index from Goos, Manning, and Salomons, we use this one as our benchmark and use the Oesch (2013) dummy as a sensitivity test. The second and highly influential measure is from Frey and Osborne (2017). It maps the forward-looking probability that an occupation will be automated, where the susceptibility of an occupation for automation is approximated using an algorithm following expert reviews. The measure is available at the six-digit Standard Occupational Classification (SOC) System 2010 U.S. coding and has a correlation of .61 with the two-digit RTI measure from Goos et al. As the measure is forward looking, and because we have to apply multiple crosswalks to link it to the survey data described below, we prefer the Goos et al. measure and again use Frey and Osborne’s measure as a sensitivity test.⁸

The European Social Survey (ESS) provides us with pooled time-series cross-sectional data of individual redistribution preferences. It has a standardized occupational identifier at the four-digit ISCO-88 level for 2002 to 2010 and ISCO-08 for 2012. We recode the 2012 wave into ISCO-88 definitions using the International Labour Organization (ILO) four-digit correspondence table⁹ and use this occupational identifier to link individuals to the RTI index from Goos et al. (2014).Our analysis draws on ESS surveys between 2002 and 2012 for the 17 Western countries for which at least two waves are available.¹⁰

To obtain a better understanding of what type of occupations score high and low on the RTI index, we postpone our definition of redistribution preferences for a moment and first discuss our operationalization of education and income. We use measures of years of education and present income (using respondents’ answers to a survey question on household total net income). We transform the income bands in the survey’s show-cards into their survey-specific midpoints, following Rueda (in press). The highest income band, which has no upper limit, is assumed to follow a Pareto distribution (Hout, 2004; Kopczuk, Saez, & Song, 2010).¹¹ Self-reported household total net income is recoded into annual 2010 Purchasing Power Parity (PPP)-adjusted U.S. dollars using exchange rate information from Organisation for Economic Co-Operation and Development (OECD; 2014b). We equivalize the income level using the square root of the household size to account for differences in household size and economies of scale.¹²

Table 1 lists the occupations ranked by their level of RTI. It shows that on average nonroutine occupations have a higher wage and educational level. Yet, these relationships are not very strong: middle-income and middle-skill occupations have high values of RTI (see also Autor et al., 2015, or Goos et al., 2014). In general, there is a relatively low correlation between the RTI index and both equivalized income (–.13) and educational level (–.17). General managers have the least routine occupation, a profession with above-average wage and skill level, but the second-least routine occupation is drivers and mobile-plant operators (low skilled and low paid). The most routine occupations are customer service and office clerks, and precision workers.

Table 1.

Levels and Changes in Employment Shares and Income for Occupations Ranked by Their Level of RTI.

Occupation	ISCO	RTI	Education (years)	Equivalized income		Employment share
	ISCO	RTI	Education (years)	int-2002	% change 2002-2012	2002	% change 2002-2012
Nonroutine		–0.68	14.04	29,099	19.14	63.58	4.30
General managers	13	−1.52	13.39	28,560	12.24	3.36	−0.43
Drivers and mobile-plant operators	83	−1.50	11.13	20,574	23.30	4.07	−0.36
Life science and health professionals	22	−1.00	17.72	36,648	16.01	2.24	0.95
Physical, mathematical, and engineering science professionals	21	−0.82	16.19	36,564	13.29	4.50	1.06
Corporate managers	12	−0.75	14.82	40,690	7.27	7.07	1.55
Other professionals	24	−0.73	16.36	34,761	17.74	7.06	1.33
Personal and protective services workers	51	−0.60	12.26	21,094	21.52	10.06	1.40
Other associate professionals	34	−0.44	14.00	29,576	18.94	10.69	0.11
Physical and engineering science associate professionals	31	−0.40	13.87	26,983	29.30	5.36	−0.92
Life science and health associate professionals	32	−0.33	14.74	26,667	18.21	4.11	−0.22
Extraction and building trades workers	71	−0.19	11.24	21,339	23.03	5.04	−0.16
Routine		0.89	11.81	21,981	15.91	36.42	–4.30
Sales and services elementary occupations	91	0.03	10.47	18,270	8.22	5.20	0.32
Models, salespersons, and demonstrators	52	0.05	12.37	22,034	18.37	4.45	0.03
Stationary-plant and related operators	81	0.32	11.67	25,044	8.33	1.18	−0.30
Laborers in mining, construction, manufacturing, and transport	93	0.45	10.68	20,037	−2.36	2.17	0.20
Metal, machinery, and related trades workers	72	0.46	11.90	21,257	35.41	6.17	−2.09
Machine operators and assemblers	82	0.49	10.77	18,893	8.39	3.46	−0.41
Other craft and related trades workers	74	1.24	10.48	18,771	26.96	1.78	−0.49
Customer services clerks	42	1.41	12.71	25,344	5.18	1.93	0.56
Precision, handicraft, printing, and related trades workers	73	1.59	12.73	25,750	18.68	0.90	−0.21
Office clerks	41	2.24	12.98	25,325	21.68	9.18	−1.91

For nonroutine (negative RTI score) and routine (positive RTI score) occupations, bold figures show average weighted values for RTI, years of education, equivalized income, and income changes. They show the sums of employment shares and employment changes. Calculations are based on the countries for which information for both 2002 and 2012 is available (all except Austria, France, Greece, Ireland, Luxembourg, and Spain). RTI = routine task intensity; ISCO = International Standard Classification of Occupations.

As mentioned above, existing contributions in the labor economics literature illustrate the relationship between automation and wage/job risk for individuals holding routine occupations (Autor et al., 2003; Goos et al., 2014; Michaels et al., 2014; Spitz-Oener, 2006). Using ESS data, we can also explore these outcomes. Table 1 is consistent with previous findings, and shows that within the relatively short time period in our analysis (2002-2012), nonroutine occupations (with a negative RTI score) saw, on average, an increase in their employment share and a higher increase in income when compared with routine occupations (with a positive RTI score).

Redistribution Preferences

The ESS contains a question designed to directly capture what we aim to explain: whether or not an individual supports government redistribution. Respondents are asked whether they agree or disagree on a 5-point scale with the following statement: “The government should take measures to reduce differences in income levels.” This variable is recoded to capture support for government redistribution. This question is the only one tapping into social policy preferences available in all waves of the ESS, and it has frequently been used in studies seeking to explain redistribution preferences (Burgoon, 2014; Burgoon, Koster, & Egmond, 2012; Häusermann, Kurer, & Schwander, 2015; Kitschelt & Rehm, 2014; Rehm, 2009; Rueda, in press; Wren & Rehm, 2014).

To better illustrate the differences in redistribution preferences across occupations, we generate a binary measure for support for redistribution equal to 1 if an individual agrees or strongly agrees with support for redistribution. This variable has an overall mean of 0.65. In Figure 1, we rank the occupations on their level of RTI, again distinguishing between occupations with a negative RTI index score (nonroutine, N) and a positive one (routine, R).¹³ The figure reflects that individuals in routine occupations have higher levels of support for redistribution. In both groups, support for redistribution increased over time.

Figure 1.

Support for redistribution across occupations in 2002 and 2012.

Controls

We include a vector of individual-level controls common in the redistribution preferences literature (e.g., Burgoon, 2014; Rehm, 2009; Rueda, in press). We include measures for years of education, age, the degree of religiosity (scaled 1-10), and indicator variables for gender, (present or former) trade union membership, and whether an individual is unemployed.¹⁴ At the country level, we again follow previous studies by including social spending as a percentage of GDP (Burgoon et al., 2012; Rueda, in press) and the unemployment rate (Burgoon, 2014; Burgoon et al., 2012), both lagged 1 year. By including ex ante levels of social spending, we can account for possible diminishing marginal returns to redistribution. It could also be that higher levels of social spending affect the occupational distribution, for instance, by leading to higher levels of public versus private employment. Similarly, there are reasons to believe that individuals will favor higher levels of redistribution when unemployment is high and that unemployment might affect occupational patterns.

Analysis

To account for the fact that individuals are nested within countries, we estimate a multilevel model with random intercepts for countries, and we cluster standard errors at the country level. Our dependent variable is categorical and ordered. We could analyze its determinants with ordered probit or ordinary least squares (OLS) estimation techniques. In an ordered probit model, predicted probabilities are restricted to the range of the dependent variable, and potential heteroscedasticity resulting from the categorical nature of the dependent variable is corrected. Some authors, however, argue that interaction effects in nonlinear models cannot be directly interpreted (Ai & Norton, 2003; Greene, 2010). Moreover, in a multilevel framework, the models sometimes do not converge and marginal effects cannot be easily calculated when a model contains both random and fixed terms. A linear OLS model does not have these drawbacks, and we correct for heteroscedasticity by clustering our standard errors at the country level. We estimate our main equations using both techniques. Our results are very comparable and, therefore, we present findings from the OLS estimations, because the interaction effects are more intuitive.¹⁵ We apply design weights, which correct for the different probabilities that respondents have to be included in the sample due to the sampling design used, in all our analyses.¹⁶

Main Results

The results of our estimation of the effects of RTI on redistribution preferences are presented in Table 2. We present four models. The first two contain our main variables of interest (first RTI, then income) and the other two add an increasing number of control variables. Regarding the additional variables, the estimates are all consistent with previous findings in the literature. First, we find that poorer individuals favor higher levels of redistribution than richer ones. This is in line with the MR model. The coefficient in Table 2 implies that a 1 percentage point increase in individual income relative to the country- and year-specific mean is associated with a 0.002 decrease in expressed redistribution preferences.¹⁷ Thus, the model predicts that an individual with 1.5 times mean income has an, on average, 0.2 lower level of redistribution preferences compared with an individual with 0.5 times mean income, ceteris paribus. Furthermore, having less education, being older, female, unemployed, not religious, or a trade union member all increase the likelihood of agreeing that the government should reduce income disparities. Neither the country levels of social spending nor unemployment have statistically significant effects on individual redistribution preferences in our analysis.

Table 2.

RTI and Redistribution Preferences.

	(1)	(2)	(3)	(4)
	RTI only	+ individual income	+ individual controls	+ country-level controls
RTI	0.085*** (.000)	0.067*** (.000)	0.042*** (.000)	0.042*** (.000)
Income		−0.211*** (.000)	−0.180*** (.000)	−0.180*** (.000)
Years of education			−0.022*** (.000)	−0.022*** (.000)
Male			−0.208*** (.000)	−0.208*** (.000)
Age			0.003** (.014)	0.003** 0.017)
Trade union member			0.176*** (.000)	0.176*** (.000)
Degree of religiosity			−0.008** (.016)	−0.008** (.019)
Unemployed			0.137*** (.000)	0.135*** (.000)
Lag of social spending				−0.004 (.503)
Lag of unemployment				0.006 (.554)
Constant	3.743*** (.000)	3.977*** (.000)	4.193*** (.000)	4.246*** (.000)
Log likelihood	−93,913	−93,176	−92,414	−92,412
Intraclass correlation	.101	.104	.113	.106
N	64,639	64,639	64,639	64,639
Number of countries	17	17	17	17

Multilevel ordinary least squares (OLS) model with random country intercepts and standard errors clustered at the country level; p values in parentheses. RTI = routine task intensity.

p < .1. **p < .05. ***p < .01.

Moving on to our main variable of interest, the results in Table 2 indicate that RTI is positively associated with redistribution preferences. This is the case no matter the number of additional variables in the analysis. This result provides empirical support for our first hypothesis that individuals in routine occupations favor more redistribution to insure against the increased risk of job or income loss.

How robust are the results for RTI presented in Table 2? In Table 3, we explore the sensitivity of the effect of RTI on redistribution preferences to a number of different specifications and additional explanatory variables suggested in the literature. We start by exploring the robustness of our results to alternative measures of RTI. We use the Oesch (2013, p. 156) coding to generate a dummy variable for routine occupations (Model 1). Next, we apply the Frey and Osborne’s (2017) coding of risk of automation (Model 2). Our results are replicated with both of these measures tapping into risk of automation.¹⁸

Table 3.

Robustness Checks for Main Results.

Model		RTI coefficient	p
	Original result from Table 2, column 4	0.042***	.000
(1)	Dummy RTI from Oesch (2013)	0.071***	.000
(2)	Automation from Frey and Osborne (2017)	0.162***	.000
(3)	Skill specificity	0.039***	.000
(4)	Foreign ratio	0.048***	.000
(5)	Offshoring	0.057***	.000
(6)	Class	0.028***	.000
(7)	Occupational unemployment rate	0.024***	.002
(8)	Task groups from Kitschelt and Rehm	0.071***	.000
(9)	Deindustrialization	0.043***	.000
(10)	Labor market outsiderness	0.036***	.000
(11)	Public-sector employee	0.048***	.000
(12)	Left–right scale	0.039***	.000
(13)	All individuals	0.035***	.000
(14)	Market income earners	0.042***	.000
(15)	Imputed RTI missing groups	0.039***	.000
(16)	Eastern Europe	0.040***	.000
(17)	Excluding 2012	0.042***	.000
(18)	Before 2008 (excluding 2008-2012)	0.038***	.001
(19)	Binary dependent variable	0.018***	.000
(20)	Absolute redistribution	0.042***	.000
(21)	Gini market income	0.042***	.000
(22)	EPL index	0.041***	.000
(23)	UB replacement rate	0.042***	.000
(24)	OLS with country fixed effects	0.042***	.000

RTI = routine task intensity; EPL = employment protection legislation; OLS = ordinary least squares; UB = Unemployment benefit.

***

p < .01.

We then explore a number of the occupational risks discussed in the theory section of this article. A first alternative is skill specificity (Cusack et al., 2006; Iversen & Soskice, 2001). In Model 3, we use the measure of skill specificity in Rehm (2009). This is a time-invariant measure available at the two-digit ISCO-88 level.¹⁹ Burgoon et al. (2012) identify migration as an occupational risk. We follow their empirical strategy and include the number of foreign born (around the year 2000) as a percentage of the population at the two-digit ISCO-88 level (OECD, 2008). We find that individuals in occupations with higher ratios of foreigners have higher levels of redistribution preferences, as also found by Burgoon et al. More importantly, the significance of our variables of interest (RTI) in Models 3 and 4 is not affected by including these occupational hazards. We then turn to the effects of offshoring, relying on Walter’s binary index (Dancygier & Walter, 2015; Walter, 2010, 2017). This index is defined at the four-digit ISCO-88 level.²⁰ We argued above that RTI substantively differs from skill specificity and offshoring. This is reflected in modest correlations between these variables (.14-.19). More importantly, in Model 5, our main results are confirmed.²¹ Occupational risk is an important part of a general sociological conception of class and, as such, it needs to be accounted for in our empirical analysis. In Model 6, we include a control for class. We use an international comparative version of the European Socio-Economic Classification, based on the class categories in the Erikson–Goldthorpe–Portocarero Schema (Erikson & Goldthorpe, 1992). The variation of RTI within class categories is high enough to allow us to assess whether the results regarding routine task intensiveness are robust to controlling for the effects of class.²² Although the effects of RTI are reduced, we still find the model controlling for class to confirm our main results.

The levels of unemployment of specific occupations are directly connected to the insurance logic proposed in this article. In Model 7, we include the occupational unemployment rate from Rehm (2009).²³ This is a stringent test, because our argument is that RTI at the occupational level leads to higher levels of job risk. We lag the occupational unemployment rates by 1 year as information for 2012 is missing. Unfortunately, data are only available at the one-digit occupational level. The occupational unemployment rate and the RTI index are positively correlated (.23) and, as expected, including the occupational unemployment rate decreases the size of the RTI index coefficient on redistribution preferences. But our main finding is still robust to the inclusion of this additional variable.²⁴ Next, we introduce a measure for the routine occupations in Kitschelt and Rehm (2014). As we explain in more detail in Appendix A, this occupation operationalization follows educational and income lines and does not capture the degree of occupational RTI. Nevertheless, the results in Model 8 confirm our main findings. We continue by including a measure of deindustrialization (Model 9), as it might overlap with our RTI measure and is seen as a factor influencing redistribution preferences (see, for example, Iversen & Cusack, 2000). To measure this, we include a dummy for individuals working in the manufacturing sector and our RTI estimate is unaffected.²⁵ Authors like Rueda (2007) argue for the increasing importance of labor market “dualization.” In this framework, “insiders” have stable and protected employment, whereas “outsiders” have insecure jobs or no jobs at all. Because outsiderness is an important source of labor market risk, we address it in Model 10. Again, our main results hold when controlling for labor market outsiderness, which is proxied by a dummy for individuals with a fixed-term contract.²⁶

A general problem, not only for this article but also for other analyses in this literature, is that occupational categories capture very different factors. This is not fully resolved by the robustness tests described above. We attempt to further address this issue in Appendix C, dedicated to explore unobserved variation across and within occupations. In addition, it is clear that occupational categories capture not only insurance-related effects but also those related to socialization. We mentioned above how Kitschelt and Rehm (2014) argue in favor of the role of occupations as the source of socialization profiles. As Kitschelt and Rehm themselves recognize, it is difficult to estimate the effect exclusively due to socialization of occupational variables like RTI. We offer here some circumstantial evidence to show that the socialization (as opposed to the insurance) effect of RTI is not particularly strong. First, we look at age cohort effects (if socialization is strong, the effects of RTI should be stronger for older cohorts), then we do a couple of placebo tests (RTI should predict redistribution preferences, but not other attitudes).²⁷ In our age cohort analysis (unfortunately the ESS does not provide an indicator for job tenure, so we use age as a proxy), we include an interaction between age and RTI. This is done without any control variables and with all control variables (as in Table 3).²⁸ The interaction between age and RTI is never significant and the estimates for other variables do not change. Second, we conduct placebo tests, in which we regress trust in the police and trust in the legal system on RTI. We find that RTI is not a statistically significant determinant of these attitudes.

Having addressed a number of factors that are related to occupational and labor market risks, we continue by exploring some additional individual-level control variables. In Model 11, it makes little difference to our main findings if we add an indicator variable capturing whether individuals are public-sector employees. Our main analysis does not include the left–right position of individuals, as we think of redistribution preferences (like, for example, Rueda, in press) as a key element of ideology. But other authors have argued for ideology as an exogenous determinant of redistribution preferences (Margalit, 2011). When we include left–right self-placement as an independent determinant of redistribution preferences in Model 12, our main findings remain the same.

We then test the robustness of our results to the sample definition. In Model 13, we expand our sample by 65% by including all individuals for which information is available (we insert an additional dummy for people not active in the labor market). Ideally, we would use gross rather than net income, but as this information is not available in the ESS, we instead restrict our sample by focusing only on those individuals whose main source of income is wages, salaries, self-employment, farming, or capital income (Model 14). Our results are replicated.²⁹ An unfortunate element of the RTI measure we use is that information is missing for six two-digit occupational groups. Of these groups, legislators and senior officials, teacher professionals, and teaching associate professionals are classified as nonroutine by Oesch (2013) and also score low in automation probabilities in the Frey and Osborne coding. We give these occupations the average nonroutine score (–0.68, see Table 1). In Model 15, we then impute the average routine score (0.89) for market-oriented skilled agricultural and fishery workers, subsistence agricultural and fishery workers, and agricultural, fishery, and related laborers. These occupations are classified as routine by Oesch and have automatable probabilities above .67 in the Frey and Osborne coding. We find similar results, including the size of the coefficient. In Model 16, we test whether our results still hold when we include Eastern European countries for which at least two waves of data are available.³⁰ We also test whether leaving out 2012, which is based on another occupational coding scheme (Model 17), or leaving out all crisis years (Model 18) affects our results.³¹

By applying OLS to a categorical dependent variable, we implicitly make the proportional lines assumption that the effect of the independent variables is constant for each answer category of our dependent variable. This assumption can be relaxed by transforming our categorical dependent variable into a dummy equal to 1 when an individual prefers or strongly prefers redistribution (Model 19). This does not affect the sign and significance of our variable of interest.

We also test for robustness with country-level controls in Table 3. We again lag all these factors by 1 year. Support for redistribution might decrease when present levels of redistribution are high because of disincentive effects (Thewissen, 2014) or because actual levels of redistribution may act as a benchmark when answering questions about whether the government should reduce income differences (Rueda, in press). Alternatively, individuals may potentially favor more redistribution when existing levels of inequality are higher. We include a measure of absolute level of redistribution (Model 20) and the level of market income inequality from the Solt (2014) database (Model 21).³² Adding these factors does not affect the estimates of interest.

Two other country-level factors might be important as they could decrease the level of redistribution individuals favor by providing insurance (Gingrich & Ansell, 2012). We include the overall employment protection legislation (EPL) index and the summary measure of OECD unemployment benefit replacement rates (OECD, 2014a, 2014c) in Models 22 and 23. Once again, our main findings are confirmed.³³ They are also confirmed, finally, when we include country fixed effects in Model 24 to control for any unspecified time-invariant country-specific characteristic.

In the previous section we mentioned that, in Table 2, RTI was positively associated with redistribution preferences, no matter the number of additional variables in the analysis. The sensitivity tests in Table 3 leave no doubt about the robustness of our results. It is, however, not straightforward to get an intuitive impression of the substantive importance of RTI. We attempt to do this by explicitly comparing the effects of RTI with those of other explanatory variables. We first do this with two of the other occupational risks discussed in the theoretical section (skill specificity and offshoring) by calculating the effects of a standard deviation increase in each variable and comparing these effects. We then also include education to illustrate the effects of RTI. We do these comparisons for two reasons. On one hand, we wish to explicitly compare the effects of RTI with two of the most influential alternative approaches to the relationship between occupational risks and redistribution preferences. Iversen and Soskice’s argument that individuals with specific skills will favor insurance as protection against their investment in human capital has been highly influential in the comparative political economy literature on redistribution. Emphasizing the international dimension, the offshorability of parts of the production process and its concentration on particular activities has also become an important alternative occupational argument about demand for redistribution (Walter, 2010, 2017). The inclusion of education, on the other hand, moves away from specific occupational factors to address one of the most widely recognized determinants of redistribution preferences. Since Lipset (1960), education has been understood as a relevant determinant of left party support and more pro-redistribution preferences (see also Alesina & Giuliano, 2009). But the relationship between education and future expected income may make these effects more ambiguous. Benabou and Ok (2001) argue that the prospects of upward mobility would make more educated individuals less likely to support redistribution than their present level of income would suggest.

Focusing on the occupational effects in Table 4, the average level of RTI in our sample is −0.13,³⁴ and a standard deviation increase is roughly comparable with an occupational switch from extraction and building trades workers to machine operators and assemblers (–0.13 to 0.83). For the skill specificity variable, a standard deviation increase is approximately equivalent to an individual switching from physical, mathematical, and engineering science professionals to sales and services elementary occupations (4.2-7.5). And, for offshoring, it can be interpreted as an occupational switch from metal, machinery, and related trades workers to general managers (0.45-0.95). An increase equivalent to a standard deviation in education would move an individual from 13.3 years of schooling (the mean in our sample) to 17.2. Table 4 suggests that a one standard deviation increase of the RTI index is associated with an increase in redistribution preferences of about 0.05. Recall that respondents are asked whether they agree or disagree on a 5-point scale, and that the mean for this variable in our sample is 3.66. Our results, therefore, indicate that an increase in RTI equal to one standard deviation would increase by 1.4% an individual’s support for redistribution from the average in the sample. To put this increase in context, this is an effect that is more than two times stronger than a comparable increase in skill specificity on the favored level of redistribution. It is an absolute effect that is roughly comparable with a standard deviation increase in education (which is associated with a decrease in support for redistribution of about 0.07)³⁵ and in offshoring (associated with a decrease of less than 0.07). It is, therefore, the case that the effects of RTI emphasized in this article are more substantive in determining redistribution preferences than one of the most influential alternative approaches to occupational risks (skill specificity). They are similar to those of a variable generally recognized as essential to our understanding of political preferences (education) and to those of a variable that has received an increasing level of attention in academic and public policy circles (offshorability).³⁶

Table 4.

Substantive Effects of RTI.

	Effect of standard deviation change on redistribution preferences	95% confidence bounds
RTI	0.052***	0.039	0.066
Skill specificity	0.024***	0.015	0.033
Offshoring	−0.065***	−0.076	−0.053
Education	−0.073***	−0.092	−0.054

RTI = routine task intensity.

***

p < .01.

Evidence for the Mediating Effect of Income

Having found a positive effect of RTI on redistribution preferences, we now inquire whether this relation is intermediated by income. Following an insurance logic, we argued that income would exacerbate the effects of RTI, as richer individuals have relatively more to lose from job losses due to automation. As already stated, we also argued that income itself would be negatively associated with preferences for redistribution. The results from all three models in Table 5 support this line of reasoning, showing a negative direct effect of income, a positive direct effect of RTI, and a positive effect of the interaction between income and RTI on preferences for redistribution.

Table 5.

RTI and Income Interaction.

	(1)	(2)	(3)
	RTI × Income	+ individual controls	+ country-level controls
RTI	0.068*** (.000)	0.043*** (.000)	0.043*** (.000)
Income	−0.201*** (.000)	−0.172*** (.000)	−0.172*** (.000)
RTI × Income	0.032*** (.000)	0.026*** (.000)	0.026*** (.001)
Years of education		−0.022*** (.000)	−0.022*** (.000)
Male		−0.207*** (.000)	−0.207*** (.000)
Age		0.003** (.013)	0.003** (.016)
Trade union member		0.176*** (.000)	0.176*** (.000)
Degree of religiosity		−0.008** (.016)	−0.008** (.019)
Unemployed		0.139*** (.000)	0.137*** (.000)
Lag of social spending			−0.004 (.500)
Lag of unemployment			0.006 (.560)
Constant	3.744*** (.000)	3.988*** (.000)	4.042*** (.000)
Log likelihood	−93,162	−92,403	−92,402
Intraclass correlation	.104	.113	.107
N	64,639	64,639	64,639
Number of countries	17	17	17

Multilevel OLS model with random country intercepts and standard errors clustered at the country level; p values in parentheses. RTI = routine task intensity; OLS = ordinary least squares.

p < .1. **p < .05. ***p < .01.

To facilitate the interpretation of the interaction, we evaluate the effect of RTI on redistribution preferences at different levels of income in Figure 2. We use Model 3 in Table 5, with the highest number of controls, for our calculations. All continuous control variables are held at their mean and the dummies at their median value. The figure makes clear that the effects of RTI on redistribution preferences are monotonically increasing in the level of income. For individuals with a very low income (less than 25% of mean income), the association between RTI and redistribution preferences is insignificant. For the largest part of our sample, however, RTI is a positive and significant determinant of redistribution preferences, and this influence becomes more substantive as income grows.

Figure 2.

Effects of RTI on redistribution preferences conditional on income.

We can provide a more formal test for whether the effect of RTI on redistribution preferences differs for individuals with different income levels (and we will also use this framework for our interaction sensitivity tests below). To do this, we define meaningful values of low, average, and high individual relative income. We select 50%, 100%, and 175% of the country- and year-specific mean income. These are not extreme values, 12% of the observations lie below 50% relative income and 13% of the observations lie above 175% income (and 175% is almost equal to the median plus one standard deviation of relative income).

We can then calculate the effect of RTI on redistribution preferences for an individual with average income, having an income of 100% of the country- and year-specific mean keeping all other control variables at their mean or median values. Moreover, we can compare the effect of RTI on redistribution demand for an individual with low income with one with high income, holding everything else equal, and we can calculate a simple (unadjusted) chi-square postestimation test to see whether the effect of RTI on redistribution preferences is statistically different for a low versus a high-income individual.

Table 6 presents the effect of RTI at the different levels of income explained above. We reproduce the robustness tests described in more detail when analyzing the results in Table 3 (and with the same theoretical justifications summarized then). In all tests, the effect of RTI is significant at average levels of income. With one exception, the effect of RTI is less sizable for individuals with low income and higher for individuals with high income (the exception being the insignificance of RTI for low income individuals when controlling for occupational unemployment, which, as mentioned above, is a highly demanding test for us). The chi-square tests in the last column show that the effect of RTI on redistribution preferences for individuals with low income is statistically significantly lower than for individuals with high income, without exceptions.

Table 6.

Robustness Checks for Interaction Results.

Model		Effect of RTI at different income			χ²
		Average	Low	High	Low vs. high
	Original results	0.039***	0.026***	0.059***	12.05***
(1)	RTI from Oesch	0.069**	0.033***	0.122***	18.31***
(2)	Frey and Osborne	0.154**	0.127***	0.194***	3.49*
(3)	Skill specificity	0.036***	0.023***	0.057***	12.59***
(4)	Foreign ratio	0.045***	0.031***	0.067***	14.44***
(5)	Offshoring	0.055***	0.044***	0.072***	9.32***
(6)	Class	0.026***	0.013*	0.045***	15.35***
(7)	Occupational unemployment	0.021***	0.009	0.039***	12.04***
(8)	Task groups (K&R)	0.068***	0.057***	0.086***	11.06***
(9)	Deindustralization	0.041***	0.028***	0.060***	11.77***
(10)	Outsiderness	0.034***	0.019**	0.055***	11.45***
(11)	Public-sector employee	0.046***	0.033***	0.064***	10.76***
(12)	Left–right scale	0.037***	0.026***	0.052***	6.00**
(13)	All individuals	0.034***	0.019***	0.057***	19.14***
(14)	Earners	0.039***	0.027***	0.057***	8.74***
(15)	Imputed missing groups	0.037***	0.025***	0.056***	13.72***
(16)	Eastern Europe	0.037***	0.021***	0.062***	9.07***
(17)	Excluding 2012	0.039***	0.024***	0.061***	16.89***
(18)	Before 2008	0.035**	0.023***	0.054***	10.96***
(19)	Binary dependent variable	0.017***	0.012***	0.024***	7.72***
(20)	Absolute redistribution	0.039***	0.026***	0.059***	12.03***
(21)	Gini market income	0.039***	0.026***	0.059***	11.76***
(22)	EPL index	0.039***	0.025***	0.060***	14.18***
(23)	UB replacement rate	0.039***	0.026***	0.059***	12.01***
(24)	Country fixed effects	0.039***	0.026***	0.059***	12.04***

RTI = routine task intensity; EPL = employment protection legislation; K&R = Kitschelt and Rehm; UB = Unemployment benefit.

p < .1. **p < .05. ***p < .01.

Conclusion

Current innovations in information technology involve a substantial employment risk for individuals holding routine occupations by facilitating the ease of automation. We find that individuals in routine occupations respond to this risk by demanding higher levels of redistribution as a means of nonmarket insurance. Even though technological change is widely considered to be a key occupational driver with large distributive effects, whether it influences the preferred level of redistribution has not been subject of inquiry in the comparative political economy literature thus far. Indeed, our analysis suggests that, on average, the RTI of an occupation has a larger positive effect on the preferred level of redistribution than other risks described in the literature. We show our results to be robust to a large number of sensitivity tests.

In this article, we also show that the degree of RTI of an occupation becomes a particularly influential determinant of redistribution preferences when an individual has more to lose from automation, that is, when his or her income is higher. Even though richer individuals, on average, might favor lower levels of redistribution, the RTI of their occupation becomes a more important determinant of their demand for redistribution. Our findings, therefore, offer an interesting counterpoint to the traditional MR redistribution story. Although recent increases in inequality in industrialized democracies may promote more anti-redistribution attitudes from the affluent, increasing levels of automation risk could mitigate these effects. Our results, in fact, suggest the possibility of cross-class coalitions in support of a redistributive welfare state between low-wage individuals in nonroutine occupations and high-wage individuals holding routine occupations. This has potentially significant implications for our understanding of economic voting and insider–outsider politics in industrialized democracies.

This study’s empirical work is built on survey data, rather than an experimental design, where individuals could be randomly assigned to occupations. One might argue that individuals self-select into occupations, leading to possibly confounded causal interpretations of our results. This reasoning would imply that risk-averse persons who already have higher preferences for the provision of public insurance choose occupations less exposed to risk. Or it could be that individuals in routine occupations (with high redistribution preferences) lose their jobs because of automation and move to nonroutine occupations while keeping higher levels of preferred nonmarket protection. Although these arguments point to interesting extensions of our research, it is important to mention that they predict a negative association between the degree of RTI and the preferred level for redistribution, militating against this article’s statistically significant findings of a positive association.

Finally, in this article, we focus on the differences in automation risk (depending on the degree of RTI) across occupations. We devote less attention to country-specific patterns like, for instance, the amount of investment in research and development, or qualitative educational factors that potentially shape how individuals cope with technological change. This would also be an interesting line of future inquiry. More generally, our analysis only begins to explore how technological risks shape actual redistribution and the welfare state. Future research should consider whether exposure to risk of automation affects voting behavior, and party and policy agendas, and ultimately, actual welfare state policies.

Footnotes

Appendix A

Appendix B

Appendix C

Acknowledgements

Earlier versions of this article were presented at the 2014 European Political Science Association conference, the 2016 Midwest Political Science Association and Council for European Studies conferences, and the Comparative Political Economy seminar at Oxford University. We thank all participants, and Nils-Christian Bormann, Koen Caminada, Henning Finseraas, Aina Gallego, Jane Gingrich, Kees Goudswaard, Robert Hellpap, Desmond King, Lieke Kools, Thomas Kurer, Brian Nolan, Stefanie Reher, Max Roser, Margit Tavits, Olaf van Vliet, and three anonymous reviewers for their helpful suggestions. All errors remain ours.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support from the Leiden University Fund and the KETEL 1 scholarship fund is gratefully acknowledged. David Rueda thanks the National Research Foundation of Korea for partial support of this research (Grant Number: 2014S1A3A2044032).

Supplemental Material

Supplementary material for this article is available online at the CPS website

Notes

Author Biographies

Stefan Thewissen is a research fellow at the Overseas Development Institute (ODI) in London. Previously, he was a postdoctoral researcher at Nuffield College and the Institute for New Economic Thinking (INET) at the University of Oxford. He is interested in questions related to inclusive growth, intergenerational mobility, social protection and health, and has published in journals such as Review of Income and Wealth, Political Science Research and Methods, Socio-Economic Review, and Health Policy and Planning.

David Rueda is professor of Comparative Politics at the Department of Politics and IR and Professorial Fellow at Nuffield College, Oxford University. He is the author of Social Democracy Inside Out (Oxford University Press, 2007). His research focuses on the causes and consequences of inequality, the determinants of redistribution preferences, the role of the welfare state in times of crisis, and insider-outsider politics.

References

Norton

(2003). Interaction terms in logit and probit models. Economics Letters, 80, 123-129.

Alesina

A. F.

Giuliano

(2009). Preferences for redistribution (Technical report). Cambridge, MA: National Bureau of Economic Research.

Autor

D. H.

Dorn

(2013). The growth of low-skill service jobs and the polarization of the U.S. labor market. American Economic Review, 103, 1553-1597.

Autor

D. H.

Dorn

Hanson

G. H.

(2015). Untangling trade and technology: Evidence from local labour markets. The Economic Journal, 125, 621-646.

Autor

D. H.

Levy

Murnane

R. J.

(2003). The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics, 118, 1279-1333.

Benabou

(2001). Social mobility and the demand for redistribution: The POUM hypothesis. Quarterly Journal of Economics, 116, 447-487.

Betz

H.-G.

(1994.). Radical right-wing populism in Western Europe. New York, NY: St. Martin’s Press.

Burgoon

(2014.). Immigration, integration, and support for redistribution in Europe. World Politics, 66, 365-405.

Burgoon

Koster

Egmond

M. v.

(2012). Support for redistribution and the paradox of immigration. Journal of European Social Policy, 22, 288-304.

10.

Cusack

Iversen

Rehm

(2006). Risks at work: The demand and supply sides of government redistribution. Oxford Review of Economic Policy, 22, 365-389.

11.

Dancygier

Walter

(2015). Globalization, labor market risks, and class cleavages. In Beramendi

Häusermann

Kitschelt

Kriesi

(Eds.), The politics of advanced capitalism (pp. 133-156). New York, NY: Cambridge University Press.

12.

Erikson

Goldthorpe

(1992). The constant flux: A study of class mobility in industrial societies. Oxford, UK: Clarendon Press.

13.

Ford

(2015). Rise of the robots: Technology and the threat of a jobless future. New York, NY: Basic Books.

14.

Frey

Osborne

(2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254-280.

15.

Gingrich

Ansell

(2012). Preferences in context micro preferences, macro contexts, and the demand for social policy. Comparative Political Studies, 45, 1624-1654.

16.

Goldin

Katz

(2008). The race between education and technology: The evolution of U.S. educational wage differentials, 1890 to 2005. Cambridge, MA: Harvard University Press.

17.

Goos

Manning

(2007). Lousy and lovely jobs: The rising polarization of work in Britain. Review of Economics and Statistics, 89, 118-133.

18.

Goos

Manning

Salomons

(2014). Explaining job polarization: Routine-biased technological change and offshoring. The American Economic Review, 104, 2509-2526.

19.

Greene

(2010). Testing hypotheses about interaction terms in nonlinear models. Economics Letters, 107, 291-296.

20.

Grossman

G. M.

Rossi-Hansberg

(2008). Trading tasks: A simple theory of offshoring. American Economic Review, 98, 1978-1997.

21.

Häusermann

Kurer

Schwander

(2015). High-skilled outsiders? Labor market vulnerability, education and welfare state preferences. Socio-Economic Review, 13, 235-258.

22.

Hout

(2004). Getting the most out of the GSS income measures (Technical Report, GSS Methodological Report No. 101). University of California, Berkeley. Retrieved from http://gss.norc.org/Documents/reports/methodological-reports/MR101%20Getting%20the%20Most%20Out%20of%20the%20GSS%20Income%20Measures.pdf

23.

Iversen

Cusack

(2000). The causes of welfare state expansion: Deindustrialization or globalization? World Politics, 52, 313-349.

24.

Iversen

Soskice

(2001). An asset theory of social policy preferences. American Political Science Review, 95, 875-893.

25.

Iversen

Soskice

(2009). Distribution and redistribution: The shadow of the nineteenth century. World Politics, 61, 438-486.

26.

Katz

Murphy

(1992.). Changes in relative wages, 1963-1987: Supply and demand factors. The Quarterly Journal of Economics, 107, 35-78.

27.

Kitschelt

Rehm

(2014). Occupations as a site of political preference formation. Comparative Political Studies, 47, 1670-1706.

28.

Kopczuk

Saez

Song

(2010). Earnings inequality and mobility in the United States: Evidence from social security data since 1937. The Quarterly Journal of Economics, 125, 91-128.

29.

Lichbach

M. I.

(1989). An evaluation of “does economic inequality breed political conflict?” studies. World Politics, 41, 431-470.

30.

Lipset

S. M.

(1960). Political man: The social basis of modern politics. New York, NY: Doubleday.

31.

Margalit

(2011, April). Ideology all the way down? The impact of the financial crisis on individuals welfare policy preferences. Paper presented at political economy workshop, Merton College, University of Oxford, UK.

32.

Meltzer

Richard

(1981). A rational theory of the size of government. Journal of Political Economy, 89, 914-927.

33.

Michaels

Natraj

Van Reenen

(2014.). Has ICT polarized skill demand? Evidence from eleven countries over twenty-five years. Review of Economics and Statistics, 96, 60-77.

34.

Moene

K. O.

Wallerstein

(2001). Inequality, social insurance, and redistribution. The American Political Science Review, 95, 859-874.

35.

Nelson

(2012). Counteracting material deprivation: The role of social assistance in Europe. Journal of European Social Policy, 22, 148-163.

36.

Nie

N. H

Kim

J-o.

(1978). Participation and political equality: A seven-nation comparison. Chicago, IL: University of Chicago Press.

37.

Oesch

(2006). Coming to grips with a changing class structure: An analysis of employment stratification in Britain, Germany, Sweden and Switzerland. International Sociology, 21, 263-288.

38.

Oesch

(2013). Occupational change in Europe: How technology and education transform the job structure. Oxford, UK: Oxford University Press.

39.

Organisation for Economic Co-Operation and Development. (2008). A profile of immigration populations in the 21st century: Data from OECD countries. Paris, France: Author.

40.

Organisation for Economic Co-Operation and Development. (2014a). Indicators of employment protection. Paris, France: Author.

41.

Organisation for Economic Co-Operation and Development. (2014b). PPPs and exchange rates. Paris, France: Author.

42.

Organisation for Economic Co-Operation and Development. (2014c). Tax-benefit models. Paris, France: Author.

43.

Rehm

(2009). Risks and redistribution: An individual-level analysis. Comparative Political Studies, 42, 855-881.

44.

Romer

(1990). Endogenous technological change. Journal of Political Economy, 98, S71-S102.

45.

Rommel

Walter

(2014, August). The political consequences of offshoring. Washington, DC: Paper presented at the American Political Science Association.

46.

Rosenstone

Hansen

(1993). Mobilization, participation and democracy in America. New York, NY: Macmillan.

47.

Rueda

(2007). Social democracy inside out: Partisanship and labor market policy in advanced industrialized democracies. Oxford, UK: Oxford University Press.

48.

Rueda

(in press). Food comes first, then morals: Redistribution preferences, parochial altruism and immigration in Western Europe. The Journal of Politics, 80(1). Retrieved from http://www.journals.uchicago.edu/doi/abs/10.1086/694201

49.

Schwander

Hausermann

(2013). Who is in and who is out? A risk-based conceptualization of insiders and outsiders. Journal of European Social Policy, 23, 248-269.

50.

Sinn

H.-W.

(1995). A theory of the welfare state. Scandinavian Journal of Economics, 97, 495-526.

51.

Solt

(2014). The standardized world income inequality database (Technical report, working paper, SWIID version 5.0). Retrieved from http://fsolt.org/papers/solt2016_pre.pdf

52.

Spitz-Oener

(2006). Technical change, job tasks, and rising educational demands: Looking outside the wage structure. Journal of Labor Economics, 24, 235-270.

53.

Thewissen

(2014). Is it the income distribution or redistribution that affects growth? Socio-Economic Review, 12, 545-571.

54.

Thewissen

Van Vliet

(forthcoming). Competing with the dragon: Employment effects of Chinese trade competition in 17 sectors across 18 OECD countries. Political Science Research and Method. Retrieved from https://www.cambridge.org/core/journals/political-science-research-and-methods/article/competing-with-the-dragon-employment-effects-of-chinese-trade-competition-in-17-sectors-across-18-oecd-countries/780191280EDEE31B8452EDA7C8C1A2FB

55.

Walter

(2010). Globalization and the welfare state: Testing the microfoundations of the compensation hypothesis. International Studies Quarterly, 54, 403-426.

56.

Walter

(2017). Globalization and the demand-side of politics: How globalization shapes labor market risk perceptions and policy preferences. Political Science Research and Methods, 5, 55-80.

57.

Wren

(2013). Introduction: The political economy of post-industrial societies. In The political economy of the service transition (pp. 1-70). Oxford, UK: Oxford University Press, Wren, Anne.

58.

Wren

Rehm

(2014). The end of the consensus? Labour market developments and the politics of retrenchment. Socio-Economic Review, 12, 409-435.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

248.87 MB

0.00 MB