The Political Consequences of Self-Insurance: Evidence from Central-Eastern Europe, the Caucasus and Central Asia

In economics and political science, the concept of self-insurance is often used to describe actions taken by individuals or households in order to reduce the size of economic uncertainty, or a potential income loss. The expectation is that households facing a situation of risk, such as loss of income, can redistribute income from a good state to a bad one (for example from employment to unemployment) as a form of assistance for possible ‘rainy days’ to come (Ehrlich and Becker 1972; Brooks 2014). Self-insurance therefore encompasses private, i.e. household-based, or social safety nets, and stands in contrast to public safety nets, or state-provided social insurance.

Existing scholarship shows that self-insurance has important implications for household-level economic outcomes. Monetary and non-monetary support from friends and family at home and abroad (Harknett 2006; Doyle 2015), assets such as housing (Ansell 2014) and religious networks (Popova 2014; Clark and Lelkes 2005), have been found to ameliorate household well-being, and increase protection against transitory income loss. In the political realm, self-insurance has been found to dampen individual-level demand for redistribution on the one hand, and aggregate-level social spending on the other (Doyle 2015; Ansell 2014). Moreover, recent research has documented a link between self-insurance and the decision to engage in politics. Drawing on empirical evidence from Brazil for example, Sarah Brooks (2014) has shown that citizens lacking the means to insure against livelihood risks are more likely to shy away from active political engagement. Yet, we still know relatively little about how private safety nets affect not only what citizens want from government and how they express their grievances, but also how they evaluate government performance in the first place. By studying the individual-level mechanisms that drive the connection between self-insurance and political support, this study hopes to expand our understanding.

At the aggregate level, comparative research advances the view that the ‘cushion’ provided by private safety nets has often ensured political stability in Sub-Saharan Africa, East Asia, Eastern Europe and Latin America (e.g. Cheng and Gereffi 1994; Diaz 1993; Meagher 1995, Greskovits 1998). In recent years, Ahmed (2012, 2017) has provided compelling evidence that far from empowering political mobilization (De la Garza and Hazan 2003), private safety nets in the form of unearned foreign income can reduce government turnover. While this literature recognises self-insurance effects on government stability, we still know relatively little about the micro-level mechanisms that drive the political consequences of self-insurance and could influence the connection between household safety nets and political support.

Our work finally takes cue from studies showing that the importance of economic dislocation in shaping political attitudes is not constant, but conditional on risk protection mechanisms. In the transition countries of Central-Eastern Europe, the Caucasus and Central Asia, for example, religion, just like unearned income received from abroad have been found to buffer the well-being impact of economic reforms on political assessments (Popova 2014; Tertytchnaya et al. 2018). In Western countries, employment protection legislation has also been shown to dampen the importance of individual-level risk in shaping policy preferences (Gingrich and Ansell 2012; Rehm et al. 2012). Similarly, statist policy regimes have been linked to whether people vote on the one hand, and to how they factor objective economic indicators in their vote calculus on the other (Carlin and Hellwig 2015; Pacek and Radcliff 1995; Singer 2011). We draw on these insights to explain how self-insurance could influence voter perceptions of economic performance and evaluations of government performance. Governments’ approval function, existing research shows, serves as important proxy for a continual accountability mechanism in democracies (Carlin et al. 2014; Carlin and Hellwig 2015) and non-democracies alike (Colton and Hale 2009; Frye et al. 2017; Treisman 2011), shaping governments’ electoral strategies, legislative agendas and policy choices.

We advance three claims. First, we argue that self-insurance influences how citizens form economic perceptions. Specifically, we expect that voters with access to private safety nets will report better assessments of household finances, or pocketbook economic conditions than their counterparts, everything else equal. This is because private safety nets offer households protection against economic risk and enable them to smooth consumption in the event of income loss. More specifically, we expect that by reducing household exposure to the vagaries of the business cycle, and allowing citizens to smooth consumption in the event of transitory income loss, private safety nets reduce economic insecurity—or the ‘psychologically mediated experience of inadequate protection against hardship-causing economic risks’ (Rehm et al. 2012; Hacker et al. 2014). Unemployment, or wage reductions for example, should be less feared when households know that in the event of a ‘rainy day’, they will be able to either draw on existing savings, or that they will have access to help from friends and family (Carroll 1997). Similarly, during periods of transitory income loss private safety nets could actively help households to smooth consumption (Ansell 2014), and keep up with utility payments (Gill and Ilahi 2000).

Second, we suggest that the economic optimism self-insurance generates at the household level translates into more positive assessments of the state of the national economy. As existing literature reminds us, variation in personal economic circumstances often drives heterogeneity in perceptions of national economic performance. Individuals or groups experiencing economic hardship, for example, have been consistently shown to report more pessimistic sociotropic evaluations than their counterparts (Anderson 2007; Weatherford 1983). In conclusion, much like partisanship has been repeatedly shown to influence people’s economic evaluations (e.g. Johnston 2006), we anticipate that self-insurance also influences the way citizens evaluate macro-economic performance.

Finally, we expect that self-insurance may ameliorate evaluations of government. Concretely, we expect that individuals with access to risk-hedging assets, savings and networks will exhibit more favourable attitudes toward government relative to their counterparts, an effect that operates through improved economic assessments. Compared to their counterparts who lack access to means of self-insurance, more economically secure individuals are more likely to report positive evaluations of incumbent performance, even when the economy deteriorates. In this context, voters who rely on their pocketbook to infer how well or poorly the national economy and the incumbent are performing, may be crediting incumbents for their own household’s decision to manage economic risk and uncertainty. This could be true of voters even in countries where insurance markets are incomplete (Carroll and Kimball 2008; Carroll 1997), or where state-provided social protection, in the form of unemployment insurance for example, is lacking (Gill and Ilahi 2000). This proposition is in line with studies on economic voting. As existing research shows, in developed and developing economies, voters often rely on their pocketbooks to infer how the government, or the incumbent, are performing (Campello and Zucco 2015; Rosenfeld 2018; Duch and Stevenson 2008; Healy and Malhotra 2012; Lewis-Beck and Stegmaier 2009; Tilley et al. 2018). Finally, the argument is in line with recent scholarship on patrimonial voting, which argues that the possession of high- or low-risk assets affects support for right and left-wing parties respectively (Nadeau et al. 2011; Lewis-Beck and Nadeau 2011; Persson and Martinsson 2016).

Yet, when voters infer that the national economy, and by extension the government, is performing well because they can buffer the effects of an ongoing downturn, accountability mechanisms could erode, especially in contexts we study here in which these mechanisms are rudimentary to begin with. Rewarding incumbents for household decisions to manage economic risk and uncertainty, or for adjusting monthly consumption habits so that it is possible to set money aside for a ‘rainy day’, may undermine the link between government action, voters and electoral sanctions (e.g. Campello and Zucco 2017). When a government enjoys high approval in the midst of an economic downturn, it may be less incentivized to invest in policies aimed at enhancing social welfare or providing social assistance to the most vulnerable (e.g. Tertytchnaya et al. 2018). Such a breach in the accountability mechanism comes with important implications. As existing literature reminds us, in more and less liberal regimes, economic voting, i.e. voters’ ability to reward or punish incumbents on the basis of how well or poorly the economy is performing, is a key mechanism to ensure responsive and accountable governance (Campello and Zucco 2015; Stokes 2001).

It is worth briefly considering a number of potential qualifications to the argument we propose. To begin with, a note on the connection between wealth and self-insurance is warranted. Wealth, or the ownership of financial and non-financial assets, such as savings (a form of liquid assets) stocks, houses or land, may be considered as a particular sub-type of self-insurance. Ansell (2014) for example argues that homeownership acts as a form of self-insurance against the welfare losses associated with job loss, especially when access to credit may be curtailed. Assets may be converted into income through sale or by borrowing against them as collateral in order to smooth consumption (Ansell 2014; Nadeau et al. 2011). Yet, if individuals are unable to borrow against or sell any of the material assets they possess, such as their home for example, they do not see the realised value of those assets as self-insurance. In other words, assets are an economic buffer, or a means of self-insurance, for individuals who are able to either borrow against their assets or to sell them in order to receive consumption-smoothing income.

We also assume that access to self-insurance is not exclusively available to government supporters alone. Whether certain households across countries or sub-national units in the same country have access to various private safety nets, such as help from friends and relatives at home and abroad, while others do not, is an outcome that national governments do not fully control. Access to private safety nets does not necessarily dependent on citizens’ political affiliations, or pre-existing support for government. Similarly, means of private insurance are accessible to poorer and richer households alike. While the amount and value of certain private safety nets, such as savings, could be endogenous to individuals’ economic circumstances (Carroll and Kimball 2008), access to risk-insuring private safety nets is largely independent of individuals’ temporary or labour market income (e.g. Ansell 2014, p. 385). Existing research shows that measures of individual or household income do not necessarily capture households’ ability to buffer the effects of economic dislocation (Hacker et al. 2013). In line with existing literature in economics, we also acknowledge that the extent to which it can protect households from economic dislocation depends on various factors. The nature of the downturn the economy is facing is one of them. Household savings will better protect individuals from income loss when economic deterioration is not accompanied by high levels of inflation (e.g. Ansell 2014). But while some types of private safety nets, such as precautionary savings, decline as a function of the duration and size of economic shocks households experience, other forms of self-insurance, such as monetary assistance from friends or family at home and abroad, could in fact be counter-cyclical to individuals’ economic conditions. These propositions are well in line with the theoretical core of our argument. Yet, given that we propose a general argument on the political consequences of self-insurance, investigating heterogeneity in types of private safety nets falls beyond the scope of this study.

To test the proposition that self-insurance increases support for government, and that such an effect stems from an economic voting mechanism, we rely on evidence from Europe, the Caucasus, and Central Asia.² Countries in the region were greatly affected by the Great Recession and provide a critical setting against which to explore the effects of private safety nets on political attitudes for several reasons.

First of all, self-insurance has been traditionally important in this part of the world. Focusing on Russia alone, Richard Rose and his co-authors for example, have shown that when the costs of economic transformation were at a peak in 1996, only 15% of Russians were relying solely on earnings from employment. This figure was no higher at the height of the economic boom a decade later (Rose et al. 2011). Moreover, in the period preceding the Great Recession, revisions of growth projections and elevated uncertainty coincided with the increase in precautionary saving rates across the region. During the Great Recession, households in Central Eastern Europe, the Caucasus and Central Asia extensively relied on private safety nets to weather the effects of income loss (ERBD 2011). Reliance on self-insurance resulted to a great extent from the fact that access to welfare provisions was limited for the majority of countries. The EBRD for example estimates that social assistance programs, income support and unemployment insurance covered just over 20% of all crisis-affected households in the region (ERBD 2011). Moreover, because of fiscal constraints, some governments had to implement measures that actually reduced, rather than increased unemployment insurance. Last Resort Social Assistance Programmes (LRSA), adopted in response to the downturn, only covered a small share of the population (World Bank 2011).

In terms of the empirical analysis, the fact that we focus on a period of economic contraction, helps us further isolate the political consequences of self-insurance. If we had tested how private safety nets influence economic and political assessments during a period of growth, it would have been more difficult to disentangle whether the positive returns of self-insurance on approval were not the result of the government’s positive management of state finances. Around the Great Recession, however, households invested in self-insurance and accumulated precautionary savings largely because they were concerned about the looming economic downturn, not because economic performance was good. This is directly relevant to our analysis, which proposes that by reducing voters’ exposure to the vagaries of the business cycle, self-insurance colours economic, and by extension political assessments. As countries in the region we focus on here were some of the worst hit by the Great Recession, the case selection makes it particularly difficult to actually find that private safety nets influence economic and political evaluations. According to the World Bank for example, in 2009 Latvia’s annual GDP growth was negative, at − 14.25. As an influential literature shows, the salience voters attach to the economy closely tracks changing conditions. Economic downturns make citizens more apt to judge incumbents based on performance and to hold them accountable for failures (Singer 2011; Singer and Carlin 2013). When economic dislocation is rife, it is very difficult for citizens to ignore.

In this section, we test our conjectures about the connection between self-insurance and support for government using cross-national survey data from the 2010 EBRD Life in Transition Survey (LITS II), and panel data from Russia. Summary statistics for the surveys used in this paper are provided in Tables A1 and A2 in the Appendix.

We begin by testing our conjectures about the political consequences of self-insurance using cross-national survey data from the 2010 Life in Transition Survey. While most countries in Central-Eastern Europe, the Caucasus and Central Asia were still experiencing the aftershocks of the Great Recession, the ERBD administered an identical set of nationally representative surveys in many countries in the region. As the surveys were designed to capture the effects of the Great Recession, they include a very detailed set of questions that tap into people’s economic and political preferences. These survey items allows us to design well-specified models.

The empirical analysis presented below focuses on households across 28 countries included in the survey (see footnote 1) and captures access to self-insurance by a survey question asking if ‘at the end of a typical month, households have anything left over to put into savings’ (coded as 1) or not (coded as 0).³ We estimate the effect of self-insurance, which we proxy here with the help of the savings indicator, on economic assessments and evaluations of government performance. The two economic variables we rely on ask respondents to evaluate whether on the whole, they are satisfied with their financial situation on the one hand, and with the present state of the economy on the other hand. Respondents are also asked to evaluate how the government is performing. All outcome variables are measured on a 5-point scale with higher values denoting greater satisfaction with economic conditions and government performance.⁴

In addition to standard demographic controls, such as age, gender, marital status, employment and education, models presented below control for household wealth. The wealth index considers whether households have access to tap water, electricity, fixed telephone line, central heating and to the internet, as well as whether households own a secondary residence, a car and a computer. The first set of items describe household characteristics, while the final three items capture durable assets. Combining these items, and employing Principal Component Analysis, we implement a weighting system for constructing the wealth index. From this, we extract the first component which we use for the analysis (e.g. Cordova 2009). Attitudes towards risk in general are also controlled for,⁵ as theories of choice under uncertainty suggest that risk-acceptant individuals are more likely discount economic risk (e.g. Brooks 2014), and possibly less likely to insure.⁶ Finally, models control for country economic, and democratic performance. Economic performance data on economic growth (GDP growth, annual percentage for 2010) come from the World Bank, while the democratic performance indicator, also for 2010, comes from the Freedom House political rights index. This indicator, which ranges on a scale from 1 to 40, assigns higher scores to more democratic countries, and captures whether there is political pluralism and participation in a country, as well as open and free private discussion (Freedom House 2010). Political pluralism and participation are of critical importance for enabling opposition forces and the media to articulate blame for an economic downturn.

The cross-national analysis proceeds in three steps. First, we investigate the effect of savings on household coping strategies in light of the recession. Second, we estimate self-insurance effects on assessments of the economy and government performance. Finally, to shed light on the full mechanism that links self-insurance to economic evaluations and from there to government approval, we employ a causal mediation framework (Becher and Donnelly 2013; Imai et al. 2010a, b; Rosenfeld 2018). This approach allows us to explore how much of the total effect of self-insurance on political assessments is mediated by economic evaluations. All models presented below use mixed effects regressions, suitable for multilevel data. Multilevel statistical techniques allow us to jointly model the macro-level and individual level determinants of household coping strategies, economic and political assessments, and to estimate separate variance structures in order to produce unbiased standard errors. Moreover, country specific random intercepts help to reduce the threat of omitted variable bias from unobserved country characteristics.⁷

To explore some of the individual-level mechanisms developed in the main part of the argument in greater detail, and examine whether self-insurance effects on economic and political assessments hold when we consider types of self-insurance beyond savings, we turn to panel surveys from Russia. Survey evidence from Russia is comparatively rich and provides an unusual opportunity for further testing the mechanisms discussed in this paper in a illiberal context. What is more, reliance on private safety nets, such as networks of friends and family, land produce and barter has been traditionally very important in Russia (Rose et al. 2011). Russia also represents a classic example of hybrid, or competitive authoritarian regime, featuring the co-existence of legitimate and public competition with an organization and institutional field that renders this competition unfair (Magaloni 2006; Frye et al. 2017). Thus, in examining the Russian case in more detail, we can also delve deeper into the individual-level mechanisms that link self-insurance, economic and political assessments outside of democracies. Evidence employed for this second part of the analysis comes from the Russian Longitudinal Monitoring Survey (RLMS) (National Research University, NRU Higher School of Economics, HSE Demoscope et al. 2011). The dataset we employ here consists of four consecutive waves of surveys that span the period 2009–2012.

In repeated time-points, the RLMS questionnaire asks respondents to assess ‘how long they think that they could live at the present level, without decreasing their expenditures and without any income, in the unpleasant situation in which all family members lost their income’. This question taps into respondents’ own estimates of how long their household could get by, without earnings, if their current savings stopped. While this is a composite of objective access to private-safety nets and subjective assessments of their ability to weather an economic downturn (which could create noise in the measure) it is a measure extensively used in existing literature (e.g. Hacker et al. 2013). As an indicator of household economic buffers the measure the RLMS provides us with is broader than single-measure indicators (be it household access to savings and/or specific household assets) and captures households’ ability to weather the effects of economic shocks by relying on a wide variety of risk-hedging assets. Concretely, respondents are invited to assess whether they could get by from less than a day to up to half a year. We recode this variable into a dummy. Coded as (1) are individuals who could get by for a month or more, and coded as (0) are those individuals who could only get by for less than a month.⁸ This measure is close to the 3-month threshold employed in comparative literature on the political consequences of self-insurance in the United States, yet the main results do not change if a continuous measure, which ranges from less than a week (1) to over a year (6) is used instead.

To examine how this measure of self-insurance relates to economic optimism, we rely on two additional measures that capture satisfaction with economic performance at the household and macro-economic levels. The first dependent variable is measured on a 5-point scale with higher values denoting more favourable assessments of household economic conditions. The second dependent variable relies on a survey item exclusively available in the 19th wave of the RLMS that was administered in 2010, and which asks individuals to assess, ‘by their opinion, what is the percentage of unemployed people in their region’. This is a measure that has been previously used in influential studies of economic voting as a proxy of households’ economic assessments (Ansolabehere et al. 2014). We use this item as an indicator of economic evaluations in general, and are not interested in individual deviations from the real unemployment rate in one’s regions, or in whether citizens offer an accurate estimate of the unemployment rate. Consistent with existing literature, we interpret higher unemployment estimates to indicate that respondents are more concerned about how the economy is performing in their region.

As with the LITS surveys, the analysis relies on mixed effects linear models, with individuals clustered within household/years (Models 1 and 2), or just households (Models 3 and 4). In the case of household satisfaction, controls for previous-year satisfaction with household conditions.⁹ We do not lag the dependent variable when considering assessments of the unemployment rate, the sociotropic variable, as this item is only asked in a single year. In terms of demographic controls, we include an economic shock indicator that invites respondents to report whether they saw their working hours, or salary decline. We code this variable as (1) if respondents experienced the shock and (0) if otherwise. As the risk-exposure indicator is specific to work-force individuals, these models do not control for respondents’ employment status. Table A2 in the Appendix provides summary statistics.

By controlling for previous-year economic satisfaction, Models 1 and 2 in Table 4 examine changes in people’s satisfaction with household economic conditions, rather than absolute satisfaction. Model 1 in Table 4 uses a dichotomous measure of self-insurance, while Model 2 estimates the same, using a continuous measure instead. Models 3 and 4 respectively, which estimate self-insurance effects on subjective assessments of the regional unemployment rate, also control for the regional unemployment rate in respondents’ regions. This indicator comes from Russia’s Federal State Statistical Service (ROSSTAT). Evidence provided below suggests that individuals who were able to ‘get by’ for a longer period of time, consistently reported greater levels of economic satisfaction on the one hand, and lower estimates of the unemployment rate on the other. Turning to the last set of models for example, we see that the regional unemployment estimate of an individual who can ‘get by’ for one month or more is by 2 percentage points lower than that of their counterpart, who is only able to get by for less than a month, everything else equal.

Across all models, the experience of household economic shocks has the opposite effect of that of self-insurance. Cuts in respondents’ salary, or working hours are robustly associated with more negative assessments of household economic conditions, and more pessimistic assessments of the regional unemployment rate. This implies that even in an illiberal political environment, as the one Russia featured this period, personal experiences of economic loss, could successfully translate into economically meaningful attitudes, and influence assessments of regional economic performance. Income, a classic predictor of economic evaluations and unemployment estimates in established democracies (e.g. Ansolabehere et al. 2014), fails to reach statistical levels of significance in both Models 3 and 4. This dovetails with existing research suggesting that outside of established democracies, individual heterogeneity in economic perceptions and political attitudes could be driven more by patterns of economic (in)security rather than economic disadvantage, as traditionally proxied by income (see also Brooks 2014).

Overall, results regarding self-insurance effects on economic optimism are consistent with cross-sectional evidence from the 2010 LITS surveys and in line with our theoretical propositions. They suggest that in Russia, just like in the rest of the region, the availability of a more robust portfolio of economic assets is robustly associated with economic optimism.¹⁰

In this study, we have theoretically argued and empirically substantiated that private safety nets ameliorate government approval, and that a large part of this effect operates through the economic optimism channel. Yet, in many countries in Central-Eastern Europe, the Caucasus and Central Asia clientelism and corruption are rife. If incumbent supporters are the only ones with access to private safety nets (either because they are more well connected, or because they are more likely to receive handouts) the relationship between private safety nets and approval could be spurious. Similarly, one may be concerned that self-insurance ameliorates political support among only for voters who supported the government in the first place. To alleviate such concerns to the best of our ability, we turn to new datasets, and perform additional analysis. We introduce our findings below and present the results in Section E of the Appendix.

First, we show that controlling for risk exposure and economic performance, self-insurance generates economic optimism and ameliorates government approval for voters who support the party in power and those who do not. As neither of the surveys examined so far included items that tap into respondents’ partisanship, or affiliations towards the regime, we rely on a third unusually large dataset from Russia. The Russian Public Opinion Foundation Survey (FOM) of February 2011 (VCIOM 2011) comprises of a regionally, and nationally representative sample of over 50,000 respondents who were asked a range of opinion questions on economic and political issues. Just like the Life in Transition Surveys this survey includes a question that asks whether households have any savings, monetary or otherwise, but also includes an item tapping into respondents’ vote intention. More specifically, respondents are asked to report ‘whom they would vote for if a Parliamentary election was to be held this upcoming Sunday’. We recode this vote intention variable into a dummy that takes the value of (1) if respondents would vote for the United Russia, the ruling regime party, and (0) if they would not. In doing so, we investigate whether self-insurance ameliorates economic and political assessments for both pro-government and opposition supporters. Overall, the analysis presented in Table E1 in the Appendix suggests that savings’ holders are more likely to be satisfied with economic performance in their region, and to approve of Vladimir Putin’s performance, regardless of whether they would support the party of power in upcoming elections or not. These results, we also show in Table E1B, are robust to nearest-neighbour matching. In sum, self-insurance colours economic and political assessments, and these effects hold for ruling party supporters, and the rest of the electorate.

To address concerns related specifically to clientelism, and self-insurance accumulated through vote-buying, Tables E2 and E3 in the Appendix presents the following analyses. First, and drawing on evidence documenting that vote-buying is greater in least democratic regions, we replicate the cross-sectional analysis by interacting the savings and democratic performance indicators. Results presented in Table E2 and Figure E1 in the Appendix suggest that self-insurance effects on approval differ very little across more or less liberal democracies. This finding suggests that in line with an economic voting mechanism, self-insurance encourages economic optimism and ameliorates approval across political systems and is in line existing literature documenting economic voting effects in democracies and autocracies (Lewis-Beck and Stegmaier 2008; Rosenfeld 2018; Chaisty and Whitefield 2012). Finally, we test for self-insurance effects on turnout. If households are able to set money aside because they receive handouts, or government payments in the form of positive inducements in order to vote, then they should also be more likely to report that they have turned out to vote for the parties who bought them off. Cross-sectional analysis from the Life in Transition surveys presented in Table E3 of the Appendix provides evidence against this hypothesis. Being able to set money aside does not predict higher turnout, as the self-insurance coefficient is negative and fails to reach statistical levels of significance. Altogether, these findings do not seem in line with the argument that private safety nets result from clientelistic or vote-buying exchanges alone, and challenge the assumption that private safety nets matter only for ruling-party supporters.