Welfare state determinants of technology perception

Studies have long focused on the role the welfare state plays in mitigating risk perceptions and promoting risk-taking attitudes among citizens (Domar and Musgrave, Reference Domar and Musgrave1944; Bird, Reference Bird2001; Grant et al., Reference Grant, Koulovatianos, Michaelides and Padula2010). A generous welfare state that protects its citizens against various social risks is expected to foster positive attitudes toward economic changes that entail both risks and opportunities. Economic globalization is a prime example (Scheve and Slaughter, Reference Scheve and Slaughter2004; Hays et al., Reference Hays, Ehrlich and Peinhardt2005; Mayda et al., Reference Mayda, O’Rourke and Sinnott2007). In the context of trade liberalization, Schaffer and Spilker (Reference Schaffer and Spilker2016) indeed find that a strong social safety net increases the public’s ‘likelihood of thinking of globalization as something positive’.

Given these studies, I first hypothesize that citizens in more generous welfare states, on average, exhibit more optimistic attitudes toward technological change than citizens in smaller welfare states. The mechanism behind this baseline hypothesis is a broadly conceived risk-mitigating role of the welfare state. Welfare states protect citizens not only against job and income loss but also against various non-work-related risks such as hazardous impacts on human health.

Given that technological change is an important source of labor market vagaries (Autor, Reference Autor2015; Quintini et al., Reference Quintini2018), various labor market interventions by the welfare state can influence technology attitudes by attenuating individuals’ exposure to and/or perception of job and income insecurity. Three types of interventions are of particular importance as they focus specifically on addressing risks associated with labor market participation and remuneration.

First, measures that compensate the unemployed can mitigate the perception of economic losses among individuals who are at risk of being replaced by new technologies. Anderson and Pontusson (Reference Anderson and Pontusson2007), for instance, find that generous unemployment benefits reduce citizens’ subjective insecurity. Their counter-factual analysis shows that adopting the Swedish-style unemployment compensation alone would substantially reduce the job-loss anxiety of the average American.

Second, wage-setting institutions shape the perception of labor market risks via their effect on employer strategies in hiring, training, and technology adoption. The centralized wage-setting system, in particular, effectively serves to ‘force and enable’ firms to ‘make more efficient use of’ production factors including human capital (Streeck, Reference Streeck1989: 90). More specifically, such wage-setting institutions create an additional incentive for firms to increase the productivity of unskilled workers and enable the firms to be ‘the residual claimant of the increase in productivity due to technology adoption’ (Acemoglu, Reference Acemoglu2003: 128). Amid technological development, firms are thus incentivized to retain and retrain their workers and selectively adopt new technologies complementary to the existing workforce (Acemoglu, Reference Acemoglu2003). Such employer strategies, in turn, can reduce labor market insecurity faced by workers.

Third, the ‘capacitating’ role of the social investment welfare state (Morel and Palier, Reference Morel and Palier2012) is expected to positively influence technology attitudes of citizens facing new labor market risks that are increasingly difficult to predict (Hemerijck, Reference Hemerijck2018). Governments can proactively equip their citizens with necessary human capital by providing higher education, offering life-long learning, and upskilling active labor market policies. Such active interventions are considered to be distinctive and complementary to protective interventions of the traditional welfare state that are less able to address labor market exclusion (Morel et al., Reference Morel, Palier and Palme2012).

Importantly, the effect of the aforementioned labor market institutions may not be uniform among different groups of citizens. Previous empirical studies on trade attitudes find that government policies catering to the economic losers of globalization do mitigate the likely losers’ opposition to free trade (Ha et al., Reference Ha, Lee and Amri2014) yet might at the same time dampen the likely winners’ enthusiasm for free trade (Ehrlich and Hearn, Reference Ehrlich and Hearn2014). If so, it is plausible that the salutary effect of welfare state interventions on technology attitudes is observed mainly among the likely losers of technological transformation.

Who are such losers? Generally, the risk of job loss under the current trend of SBTC (Card and DiNardo, Reference Card and DiNardo2002) is greater among lower-skilled citizens (Acemoglu, Reference Acemoglu1998; Goldin and Katz, Reference Goldin and Katz1998; Michaels et al., Reference Michaels, Natraj and Van Reenen2014). The lower-skilled, on average, would thus be less sanguine about technological transformation. Yet because they are likely to be the beneficiaries of various welfare state interventions such as unemployment compensation and retraining, the risk-moderating effect of such interventions should be more pronounced in this group.

To summarize, the effect of the welfare state’s labor market interventions on technology attitudes would be conditioned by individual skill endowment. The effect is expected to be clearly salutary among the low-skilled, the likely losers of SBTC. The effect would be less positive or might even be negative among the high-skilled. The latter group might exhibit more positive technology attitudes under liberal labor market institutions where firms accelerate SBTC and the skill premium rises faster.

A key macro-level implication of Hypothesis 2 is that the labor market institutions of the welfare state would reduce the otherwise considerable skill cleavage over technological change. In societies with such institutions, the high- and low-skilled would be less divided over the technological transformation.

The technology attitude-welfare attitude nexus

The preceding subsection discussed how existing welfare state institutions shape the public’s perception of technological change. This subsection explores how, within a society, citizens’ perceptions of technological change shape their welfare state attitudes. Investigating this micro-level nexus allows us to predict the coalitions and fault lines over the welfare state forged under technological transformation. I propose two competing hypotheses.

On the one hand, one might expect that citizens who are critical of the new technologies’ economic consequences would want more redistributive state intervention that compensates for their losses. Those optimistic about technological transformation might perceive that new technologies allow them to more effectively address various social risks on their own. If so, they would be against the expansion of distributive public programs. This expectation is in line with findings from the literature on trade preference where the likely losers of trade support compensatory public policies whereas the likely winners oppose it (Ehrlich, Reference Ehrlich2010).

On the other hand, one’s acceptance of new technologies might foster, not undermine, her support for the welfare state. The expectation of large potential gains from technological advancements involves substantial uncertainties. Compared to those who defy new technologies, individuals more willing to adapt their behaviors in accordance with technological change might have stronger incentives to seek protection against future risks. Their ‘insurance motive’ for redistribution (Rehm, Reference Rehm2009: 858) is in line with the view that welfare states serve the future-oriented, risk-pooling needs of citizens who do not necessarily face any imminent economic disadvantage (Barr, Reference Barr2001; Rehm et al., Reference Rehm, Hacker and Schlesinger2012; Busemeyer and Garritzmann, Reference Busemeyer and Garritzmann2017).

Furthermore, pro-technology citizens might support welfare state expansion if distributive policies are seen as essential for sustaining and furthering technological advancements. The expectation is broadly consistent with Rickard (Reference Rickard2015) who finds that pro-trade legislators who represent more exporters are actually more likely to vote for expanding Trade Adjustment Assistance programs in order to broaden the pro-trade coalition.

To summarize, Hypothesis 1 concerns the direct effect of the welfare state on technology attitudes. Hypothesis 2 captures the conditioning effect of the welfare state on the micro-level nexus between technology-induced economic risk and technology attitudes. Lastly, Hypothesis 3 explores the effect of technology attitudes on welfare state attitudes.

Operationalization of variables

Technology attitudes

Technology attitudes are my dependent variable in Hypothesis 1 and Hypothesis 2 and the independent variable in Hypothesis 3. The WVS includes several questions pertaining to perceptions of science and technology. I created a composite indicator Pro-Technology Index (PTI) by aggregating responses to these questions using factor analysis.Footnote ⁴ More specifically, the index is based on the following ordinal questions: (1) whether putting more emphasis on the development of technology is desirable, (2) whether science and technology make lives easier and more comfortable, (3) whether science and technology are expected to bring more opportunities for the next generation, and (4) whether the world is better off because of science and technology. The list of questions and their WVS question ID number are reported in Table 1. More positive answers to these questions together capture respondents’ greater willingness to embrace technological change. The resulting index variable ranges from 0 to 2.7 with a mean of 2.0.Footnote ⁵ For the robustness check, I show in the appendix that my main findings also hold using geometric averaging (with equal weightings) as the aggregation method.

Table 1.

Attitudes toward technology and welfare state

Note that in creating the index I excluded questions that are explicitly about the normative implications of technology such as whether science conflicts with faith or whether science hinders moral judgments. While including these items do not substantially change my key findings, it unnecessarily stretches the index beyond the concept I seek to capture.

Welfare state institutions

My independent variables in Hypothesis 1 and Hypothesis 2 are welfare state institutions. As an indicator of overall welfare state generosity (relevant for Hypothesis 1), I employ total public social protection expenditure (as % of the GDP). The data are from the ILO’s World Social Protection Report 2014/15 (ILO, 2014).Footnote ⁶ The measure is referred to as Social Spending hereafter. It ranges from 0.2 in Pakistan to 26 in Sweden.

As discussed in the previous section, welfare states not only mitigate technology-induced job and income insecurity but can also mitigate other non-labor-market-related concerns such as new technologies’ hazardous impact on health and safety. So I expect the Social Spending variable to have a positive association with the PTI. As shown in Figure 1, the country means of PTI and Social Spending exhibit a weakly positive correlation in both developing and developed country groups. Of course, more rigorous investigation into the relationship requires controlling for other compounding factors at both country and individual levels, which will be the subject of the next section.

Figure 1.

Social spending and Pro-Technology Index.

For Hypothesis 2, I employ a series of indicators capturing the labor market interventions of the welfare state. First, I employ Unemployment Protection, which measures the effective coverage of unemployment benefits (as % of the unemployed population). The data are from the ILO’s World Social Protection Report 2014/15 (ILO, 2014).Footnote ⁷ The variable is highly right-skewed as most developing countries have very low coverage while a few developed countries feature broad-based coverage. I thus employ a log transformation of the variable.Footnote ⁸

Second, I use an indicator of the coordination in wage setting from the ICTWSS database (Visser, Reference Visser2016). The measure, Wage Coordination, is a 5-category ordinal indicator.^{Footnote 9} It is expected that in countries with more centralized wage-setting institutions the demand for lower-skilled workers does not decrease as much as in other countries with no such institutions (Acemoglu, Reference Acemoglu2003; Domeij and Ljungqvist, Reference Domeij and Ljungqvist2013). In line with this expectation, Wage Coordination is indeed negatively correlated with the skill premium of high-skilled workers. According to OECD Statistics (2019), in 2015, the relative earnings index^{Footnote 10} of the university educated was 117 in Sweden (the lowest among 38 sample countries) where, according to the ICTWSS database, ‘wage norms are based on centralized bargaining by peak associations’ (Wage Coordination = 4). The index was 174 in the USA with ‘fragmented wage bargaining, confined largely to individual firms or plants’ (Wage Coordination = 1).

Third, I employ spending on active labor market programs (ALMPs % GDP) from the OECD database. While the measure is only available for the OECD member states, it allows for capturing the effect of capacitating interventions. See the Appendix (Table A.2 in Supplementary material) for country-specific data availability.^{Footnote 11}

Measurement validity

Does PTI reflect ‘economic’ risks from technological change?

Given the general wording of the survey questions employed to construct the PTI (see Table 1), the index is subject to the criticism that it may not sufficiently reflect the respondents’ assessments about the economic consequences of technological change. For instance, one might exhibit a low PTI score primarily because she believes new technologies make people more cold-hearted or less religious. If so, PTI is not a valid indicator to test my hypotheses, which posit that the welfare state shapes technology attitudes by mitigating technology-induced economic risks.

To address the validity concern, Table 2 presents a set of preliminary models. If PTI is indeed a function of expected economic risks and losses from technological change, the indicator of skill endowment (Education) should have a positive and significant association with PTI because the high-skilled are less exposed to the risks of job and income losses under SBTC. The findings in Model 1 are consistent with the expectation. Importantly, the positive association between Education and PTI holds, controlling for a set of factors that drive the non-economic variation in PTI including (religious) Disbelief and social/political Trust.

Table 2.

Does PTI reflect exposure to technology-induced economic risk?

*P < 0.1; **P < 0.05; ***P < 0.01.

In Model 1, I also include the self-reported task routineness indicator (Routine) in addition to Education. This consideration addresses findings from the literature that technological change tends to decrease demand for ‘middling’ occupations performing cognitive but routine tasks relative to occupations with lower routineness (Autor et al., Reference Autor, Levy and Murnane2003; Goos et al., Reference Goos, Manning and Salomons2014). Thus, I expect Routine to be negatively associated with PTI. The variable Routine has the expected negative sign in Model 1; that is, individuals performing routine tasks tend to exhibit less positive attitudes toward technological change. The magnitude of the effect, however, is small compared to that of Education.

Task routineness is also expected to reduce the positive effect of skill attainment on PTI. I thus examine the interaction between Education and Routine in Model 2. The finding is visualized in Figure 2. I simulate the predicted values of PTI for those with vocational secondary education (Education = 5, in red) and those with a university degree (Education = 9, in blue) over a range of task routineness on the x-axis (from 1 = highly creative to 10 = highly routine). All other variables are set at the sample mean values. As expected, the negative slope of the blue line indicates that the positive effect of university education on PTI is dampened by task routineness. Due to the low PTI of highly educated routine taskers, the skill-based gap over PTI (captured by the distance between the two lines) is smaller among routine taskers than among creative workers.Footnote ¹⁵

Figure 2.

Effects of skill level and task type on technology attitudes.

Because my theoretical expectations for the welfare state’s effect on technology attitudes apply to both developing and developed countries, it is also important to show that the cleavage over new technology is not confined to the developed world. Model 3 thus includes the interaction term of Education and GDP per capita. While the lower-order term Education remains significant, the interaction term is not. The Education effect (capturing the skill cleavage) is observed at all substantively meaningful values of GDP per capita.

The analyses reported in this subsection together suggest that PTI reflects individuals’ perceptions of economic risks associated with new technologies in developing as well as developed countries. The following section tests my hypotheses concerning the nexus between the welfare state and PTI.

Tests for Hypothesis 1 and Hypothesis 2: Do welfare state institutions promote technology acceptance?

The models in Table 3 explore the association between welfare state institutions and the dependent variable PTI. In Model 4, the key independent variable is Social Spending. The coefficient estimate of Social Spending is significant and positive. The finding is consistent with Hypothesis 1, which expects that individuals in a more generous welfare state exhibit more positive attitudes toward technology than individuals in a smaller welfare state.Footnote ¹⁶

Table 3.

Effects of welfare state institutions on technology attitudes

*P < 0.1; **P < 0.05; *** P < 0.01.

Does the welfare state-technology attitude nexus vary between developing countries and developed countries? If so, how? Two competing expectations can be considered in this regard. First, the nexus might be weaker in developing countries where many social programs are rather new and less consolidated. Existing studies find that the extent citizens look to the welfare state as a viable means of protection depends on citizens’ ‘experiences with having lived in settings where state institutions govern social protection to address risks’ and that in emerging welfare states, citizens exposed to social risks do not necessarily turn to government protection (Lim and Burgoon, Reference Lim and Burgoon2018). If so, the effect of social spending on reducing perceived risks under SBTC might be weaker in developing countries than in developed countries with mature welfare state institutions.

Second, the nexus might be stronger in developing countries because the marginal utility of a one-unit income transfer via social spending tends to be greater for developing countries whose average living standard is lower than wealthier countries. Furthermore, it is plausible that citizens in rich and mature welfare states take the traditional measures of the welfare state for granted. Unlike the low-skilled in developing countries occupied with subsistence concerns, those in affluent countries might expect more from their welfare state. If so, the welfare state-technology attitude nexus in developed countries might be sensitive to the type of welfare state measures. Only those measures that directly address ’new’ labor market risks under SBTC might have a PTI-increasing effect.

These competing expectations are explored using the subsample analysis reported in Models 5 and 6. In the developing country subsample (Model 5), Social Spending has a significant and positive association with PTI. Social Spending is insignificant in the model with the developed country subsample (Model 6). The finding is in line with the second expectation.

This finding also brings us to the next step in the analysis, which directly employs the indicators of labor market interventions. As discussed in the previous section, I employ three measures of labor market interventions capturing distinct yet complementary functions of the welfare state: unemployment protection, wage coordination, and ALMPs.

First, Models 7–9 employ Unemployment Protection measured by the percentage of the population effectively covered by unemployment benefits. As Hypothesis 2 posits that the PTI-increasing effect of protective labor market institutions would vary by Education, I interact Unemployment Protection with Education. Model 7 is the full sample model. The interaction term coefficient is negative and significant, but the lower-order term Unemployment Protection is insignificant. The findings suggest that unemployment benefits reduce the PTI among the higher skilled while having no positive impact on the PTI among the lower-skilled.

The finding from the pooled model, however, can be misleading if the same welfare state intervention has distinct effects in developed and developing countries. Models 8 and 9 thus explore developing and developed countries separately. In the developing country subsample (Model 8), the lower-order Unemployment Protection term is positive and significant while the interaction term coefficient is not statistically significant. They together capture the positive effect of unemployment benefit coverage on technology acceptance across different skill levels in developing countries. In Model 9, only the interaction term is negative and significant. This suggests that in developed countries, unemployment benefit coverage has no PTI-increasing effect among the low-skilled and may even have a small PTI-reducing effect among high-skilled citizens. These findings are largely consistent with the findings on the effect of social spending (recall Models 5 and 6).

Why does Unemployment Protection have no positive association with PTI in developed countries? One might suspect that the measure based on the population coverage (breadth) has limited variation in the developed world where most industrial employees are covered by unemployment protection. To address this concern, I employ an alternative compensation measure: unemployment insurance replacement rate. Unfortunately, the data are only available for nine developed countries in my sample.^{Footnote 17} Random effect models produce biased estimates for higher-level indicators when the number of higher-level units is small.Footnote ¹⁸ I thus estimate a pooled ordinary least squares (OLS) model (Model 10) and a country fixed effects model (Model 11) including the interaction term of the country-level moderator and a micro-level independent variable (Möhring, Reference Möhring2012). Nevertheless, I do not find any evidence that the higher replacement rate is associated with more positive technology attitudes among the low-skilled in these developed economies.

As discussed earlier, the divergent findings in developing and developed countries might in part be attributable to the varying public expectations toward the role of the welfare state. Compensation alone would be insufficient to foster technology acceptance where the public is concerned about ‘exclusion’ (Morel et al., Reference Morel, Palier and Palme2012) and ‘marginalization’ (Gesthuizen et al., Reference Gesthuizen, Solga and Künster2010) in the labor market. These ‘new’ social risks are not adequately addressed by unemployment compensation. This explanation is also in line with my earlier finding that the overall size of social spending does not promote public technology acceptance in developed countries. It might be that interventions that are empowering and capacitating are needed to foster their technology acceptance in developed countries.

In Table 4, I thus move on to models utilizing additional indicators of the welfare state’s labor market intervention: (1) wage coordination systems incentivizing firms to retain and retrain low-skilled workers and (2) spending on ALMP. See the Appendix (Table A.2 in Supplementary material) for country-specific data availability.Footnote ¹⁹

Table 4.

Effects of welfare state institutions on technology attitudes II

* P < 0.1; ** P < 0.05; *** P < 0.01.

Table 5.

Determinants of welfare state attitudes

* P < 0.1; ** P < 0.05; *** P < 0.01.

Models 12–14 examine the effect of centralized wage setting. In Model 12, the baseline model without an interaction term, the Wage Coordination variable lacks statistical significance while the Education term is significant and positive. To test Hypothesis 2, Model 13 adds the interaction term of Wage Coordination and Education. The positive and significant coefficient estimate of the lower-order Wage Coordination variable captures the PTI-increasing effect of a centralized wage coordination system among the lowest skilled. The negative and significant interaction term coefficient points toward skill-varying effects of centralized wage setting that are consistent with Hypothesis 2. The substantive meaning of the interaction effect is visualized in Figure 3a in terms of predicted PTI. I vary Wage Coordination (on the x-axis; from 1 for the most fragmented to 5 for the most centralized) and Education while fixing all other covariates at their sample mean values. Again, the distance between the blue and red lines captures the skill-based gap in PTI. The gap is larger under a fragmented wage bargaining system (when Wage Coordination = 1) than under a coordinated system (when Wage Coordination = 5). The findings provide evidence that a strong Wage Coordination system might mitigate the skill-based gap in public technology attitudes.

Figure 3.

Effects of wage-setting institutions on PTI.

Even under a centralized wage bargaining system, some workers have limited access to collective bargaining. The growth of precarious employment in sectors that are difficult to organize has been a problem in many countries, and representation of irregular workers in collective bargaining has been difficult even in countries with strongly institutionalized peak-level tripartism such as the Netherlands (Boonstra et al., Reference Boonstra, Keune and Verhulp2012). Thus, workers exposed to ‘destructive effects of market forces’ would feel more shielded when they are unionized and the state intervenes to ‘back either collective bargaining or its collective agents (unions and employer associations)’ (Traxler, Reference Traxler2003: 144). Furthermore, the experience of taking part in and benefiting from inclusive and centralized wage coordination can have an enlightenment effect on low-skilled union members, which increases their perception of self-efficacy (Markowitz, Reference Markowitz1998) under challenging situations such as SBTC. Such experiences might create in workers ‘the sense that they can exercise some control over their working lives’ (D’Art and Turner, Reference D’Art and Turner2007: 108)

As such, Model 14 explores a three-way interaction effect of active union membership (Union), Education, and Wage Coordination. I expect the positive association between Wage Coordination and PTI among less-skilled workers to be more pronounced when they are union members. The finding is visualized in Figure 3b. The figure on the right(/left) is for those with(/without) active union membership. The difference in the slope of the red lines reveals that the positive association between Wage Coordination and the less-skilled workers’ PTI is stronger among active union members. Meanwhile, the slightly negative slope of the blue line on the right indicates that high-skilled union members are more pro-technology under a fragmented bargaining system allowing for greater wage inequalities and skill premiums. This outcome is consistent with my expectation in Hypothesis 2.

Lastly, I examine how government spending on ALMP shapes the skill cleavage over technology acceptance. It should be noted that only 12 countries could be included in the analysis due to limited data availability. I thus turn to pooled OLS (Models 15–16) and a country fixed effects model including the interaction terms of ALMP and Education (Model 17). I present the finding based on Model 16 in Figure 4. The effect of ALMP is broadly in line with the effect of Wage Coordination. Spending on active labor market policies seems to promote technological acceptance of the low-skilled (captured by the red upward-sloped line) and reduces the skill-based gap in PTI. The magnitude of the effect, however, is very small. Future empirical research should investigate the effect of the social investment welfare state on technology acceptance with more fine-grained data and in a larger sample of countries.

Figure 4.

Effects of ALMP spending on welfare attitudes.

Overall, my analyses lend partial support to Hypothesis 2. A protective intervention in the form of unemployment compensation was associated with a broader-based technology acceptance in developing countries. Such measures of protection, however, had no meaningful association with the public’s technology acceptance in developed countries. In the latter, I find first-cut evidence that empowering and capacitating workers via institutions for collective bargaining and ALMP might have potential to foster broader-based technology acceptance.

Tests for Hypothesis 3: How does technology perception shape welfare state attitudes?

This section explores the micro-level nexus between technology perception and welfare attitudes within a society (Hypothesis 3.1 and Hypothesis 3.2). Among the citizens with varying PTI, who are more likely to turn to the welfare state for the social embedding of technological change?

My dependent variable now is the PWI, and the key independent variable is PTI. Several macro-level factors such as welfare state generosity, country income, and level of economic openness are important determinants of PWI as well as PTI. I thus include these macro-variables as controls. Random country intercepts are also included. Table 5 reports the results.

Model 18 is the baseline model including only the macro-level control variables and PTI. The PTI variable has a positive and significant association with PWI. The finding is consistent with Hypothesis 3.2, which posits that pro-tech citizens would be more supportive of welfare states than techno-skeptics. Model 19 introduces a set of micro-level control variables. The PTI variable itself still has a positive association with the PWI. As reported in the Appendix, the finding holds when I employ an alternative measure of PWI or use the dominant sub-items from PWI (V137) and PTI (V192) instead of the composite index.

In Model 20, I explore the possibility that technology attitudes have a non-linear effect on welfare state attitudes. It might be that, compared to those with middle-level PTI, citizens with low PTI and those with high PTI both exhibit stronger support for the welfare state; the former wants compensation, and the latter is willing to offer such compensation to prevent any backlash against technological transformation. I thus employ a three-category nominal indicator of technology attitudes: Techno-Skeptic (PTI < 1st quartile), Techno-Optimist (PTI> 3rd quartile), and neutral as the reference category. I do not find evidence for non-linear effects. PWI is highest among the Techno-Optimist and lowest among the Techno-Skeptic. The findings together provide consistent support for Hypothesis 3.2.

As discussed earlier, there is more than one reason why pro-technology citizens are more favorable to redistributive welfare state policies than techno-skeptics. First, adapting to new technologies can bring about large potential gains yet, at the same time, involve risks and uncertainties. Being on board with technological transformation thus creates incentives for insurance-seeking via public risk-pooling. Second, individuals with a higher PTI might seek to preempt a popular backlash against the trend of technological progress. To the extent that individuals’ PTI reflects their perceived gains from technological change, individuals with a higher PTI have a vested interest in maintaining the change by compensating for the losers who might try to block the change.

The two motivations have distinct observable implications. If the insurance-seeking motivation explains the pro-tech population’s support for the welfare state, the positive PTI–PWI nexus would be stronger among the higher-risk-bearing segment of the pro-tech group (i.e., the low-skilled techno-optimists). If the intention is to preempt a popular backlash and secure their gains, the nexus would be stronger among those who enjoy disproportionate economic gains from technological transformation, that is, the high-skilled techno-optimists. To see whether and how the association between PTI and PWI varies by skill type, I interact PTI and Education in Model 21. The lower-order PTI coefficient is positive, but the interaction term coefficient is negative. In other words, the positive nexus between pro-technology attitudes and welfare state support is stronger among the lower-skilled.

The findings also hold when we examine developing countries (Model 22) and developed countries (Model 23) separately. I visualize the finding from Model 23 in Figure 5 in terms of predicted values for PWI. I vary the level of Education and PTI while setting all other variables at the sample mean values. The positive slope of the red line suggests PTI is positively associated with PWI among those with secondary education. The finding is consistent with the explanation emphasizing the pro-tech population’s insurance motive. Those facing higher risks under SBTC who nevertheless embrace the opportunities offered by new technologies exhibit the highest PWI. These citizens are stronger supporters of the social embedding of the technological transformation than both low-risk tech-enthusiasts (i.e., the university educated with high PTI scores) and high-risk techno-skeptics (i.e., the secondary-educated with low PTI scores).

Figure 5.

Effects of technology attitudes on welfare attitudes by skill level.