3.1 Measuring sanction reforms across space and time

To measure sanction reforms across countries and over time, I draw on a novel comparative dataset on the conditionality of unemployment benefits (Knotz and Nelson, Reference Knotz and Nelson2015; see also Knotz, Reference Knotz2018). The dataset covers 21 advanced democracies (Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, South Korea, Sweden, Switzerland, and the United Kingdom) over the period between 1980 and 2012.Footnote ² It provides information on the statutory strictness of sanctions for unemployment benefit claimants who a) become voluntarily unemployed, b) refuse an offer of suitable employment once, twice, three times, and ‘repeatedly’, c) fail to attend appointments with employment service providers or the benefit administration, and d) fail to report sufficient evidence of job-search activities.Footnote ³

The data are based primarily on original historical documents, in particular historical legislation, regulations, guidelines, and policy manuals, which were retrieved from legislative databases, national archives, and libraries. This very labour- and time-intensive strategy was chosen since secondary sources like previous studies or social policy compendia and databases by international organisations and think tanks do either not provide sufficiently precise information on aspects like sanction rules or provide this information only for certain countries or years. The data collection proceeded as follows: as a first step, each country's last fundamental revision of its main unemployment benefit statute (e.g. the United Kingdom's Social Security Act) prior to 1980 was identified.Footnote ⁴ The next step was to locate the relevant paragraphs or sections in these statutes. Subsequently, all changes that were made to these provisions were traced by inspecting all relevant amendments and revisions up until 2012. In some cases, certain rules and requirements were either not specified in primary legislation or were only very generally worded (Grubb, Reference Grubb2000). In these cases, it was necessary to retrieve and inspect regulations, guidelines and manuals, which contain more specific and detailed information, to reconstruct the development of these provisions over time.

Since interpreting statutes is of course generally not a straightforward task, great care was taken to avoid misinterpretations or overlooking reforms. First, where possible, existing cross-sectional data or country-specific studies (Ministry of Finance Denmark, 1998; Hasselpflug, Reference Hasselpflug2005; Clasen and Clegg, Reference Clasen, Clegg, Clasen and Siegel2007, Reference Clasen and Clegg2011; Venn, Reference Venn2012) were used to guide the data collection and to cross-check the findings. Second, at least one expert reviewer per country was asked to inspect the collected data and to check them for accuracy and completeness.Footnote ⁵

To make the qualitative information analysable, a quantitative measure of the strictness of sanctions was constructed. The procedure chosen here generally follows the example of a similar indicator produced by the OECD (Venn, Reference Venn2012), some modifications notwithstanding. The indicator was constructed by first scoring the strictness of four types of sanctions (sanctions for voluntary unemployment, for the first refusal of an offer of suitable employment, for repeated refusals of offers of suitable employment, and failures to report sufficient evidence of job-search activities). Modifications, which concern the number of steps on the scales used to rate the strictness of these provisions as well as the location of cut-off points, were applied. This was done in order to take into account the greater variation in my data, which have a temporal dimension as opposed to the purely cross-sectional data generated by earlier studies. Despite these modifications, my data are still strongly correlated with the data generated by Venn, as is shown in the Supplementary Materials. This indicates that the design of the coding scheme does not affect the shape of the data in a very substantial way.

The numerical scores were then summed up and finally divided by the theoretical maximum to obtain an indicator of the overall strictness of sanction rules that ranges between 0 (no sanctions) to 1 (very harsh sanctions). More detailed information on the aggregation procedure is also provided in the Supplementary Materials.

While the indicator on sanction rules is used as the dependent variable in the large-N analysis, the main predictors used in the analysis are a) the unemployment rate and b) the government's budget balance, both being key measures of pressure experienced by countries due to socio-economic problems.Footnote ⁶ A positive effect of unemployment on the strictness of sanctions would indicate that sanction rules are tightened in times of high unemployment and would hence provide support for my argument that sanction reforms are linked to economically difficult times. Arguably, even more important is the effect of the budget balance because a significant effect of budget deficits on the strictness of sanctions, especially with unemployment controlled for, would suggest that it is really concerns about deficits that are a key driver of these reforms.

3.2 Sanction rule reforms in the advanced democracies since 1980

I start the analysis by providing descriptive evidence for the relationship between sanction reforms and unemployment and budget deficits. Figure 1 depicts the varying strictness of sanctions (solid line) as well as the unemployment rate (dashed line) in each country in the dataset. The first thing to note is that countries clearly differ with regard to how often and how significantly they reform their sanctioning rules. The United Kingdom, Ireland, Finland, Sweden, Austria and Italy introduced significant reforms in the period between 1980 and 2012, as indicated by strong increases in the sanctions indicator. Japan, Belgium and Denmark, on the other hand, introduced only a few reforms, and Greece has introduced no significant reforms since 1980. It is also interesting to see that sanctions are occasionally relaxed. Germany in the early 2000s is one example, as are New Zealand and Australia. Relaxations are, however, overall not as common as restrictions, even in economically better conditions, which corresponds to the prevailing finding in the literature that coercive instruments were generally strengthened rather than relaxed in recent decades. This pattern also confirms the presence of the above-mentioned ratchet effect. It hence seems as if policymakers do indeed have other priorities in times of affluence than to reverse harsh measures introduced under economically adverse conditions.

Figure 1.

Sanctions and unemployment since 1980

Notes: The scale on the right refers to the unemployment rate.

Figure 1 also shows that there is covariation between sanction reforms and the unemployment rate. This is particularly clear in the case of New Zealand, where the strictness of sanctions follows the unemployment rate surprisingly closely but there are also a number of cases where increases in the unemployment rate led to restrictive sanction reforms as indicated by increased strictness scores (for instance, in Sweden and Finland in the early 1990s, in the United Kingdom in the late 1980s and mid-1990s, and in Ireland in the 1980s and in 2009). Similar, albeit less pronounced, developments are visible in Canada and Switzerland in the early 1990s, in Portugal in the mid-2000s, Spain in the early 1990s, and France both in the early 1980s and the mid-2000s. This being said, there are, as always, also some cases that do not fit the predicted pattern. In a few cases, countries restricted sanctions after sustained decreases in unemployment. These include Belgium and Norway in the early 1990s, and Sweden and Finland in the late 1990s.Footnote ⁷

The fact that sanctions tend to be tightened in times of high unemployment is in line with my argument. In most cases, however, it took at least some years of increased or increasing unemployment before sanctions were eventually tightened. Finland, for instance, experienced several years of rising unemployment in the early 1990s before reforms were eventually introduced. This pattern is certainly not implausible since smaller fluctuations in the unemployment rate will arguably not provide a strong enough impulse to trigger a political reform process. For reforms to happen, a more significant and enduring increase in unemployment is needed.

Figure 2 depicts the relationship between sanction reforms and the budget balance. Although not as clearly visible at first sight, the two variables co-vary as well. The significant budget deficits experienced by Sweden and Finland in the 1990s, for instance, were followed by the introduction of tougher sanctions. Particularly illustrative are also the recent reforms in New Zealand and Ireland in 2009. Both countries’ budget balances deteriorated significantly after the onset of the 2008 financial crisis, which in both cases coincided with reforms that tightened sanctions. Austria's budget balance also turned slightly negative in the early 1990s, which was then followed by one significant and one smaller sanction restriction. The case of Spain in the early 1990s is a final fitting case.

Figure 2.

Sanctions and budgetary crises since 1980

Notes: The scale on the right refers to the budget balance.

While these descriptive analyses indicate that sanctioning rules are tightened in response to adverse economic conditions, a multivariate regression analysis is necessary to provide a more thorough test of whether there really are effects of rising unemployment and budget deficits on sanction reforms and also how large these effects are. To provide such a test, I estimate several error-correction models (ECM). ECMs are a type of dynamic regression model that is ideally suited to my analysis because they allow me to estimate not only the instantaneous but also the long-term effects of changes in the predictors (De Boef and Keele, Reference De Boef and Keele2008). This is crucial because the descriptive analysis above indicated that, in particular, the effect of unemployment may only materialise in the long term. For related reasons, Podestà (Reference Podestà2006) also suggested that the ECM should be the preferred model specification for regression models of welfare state reforms. A generic ECM specification looks as follows: the current change in the dependent variable is regressed on both the current changes and the lagged levels of the predictors, as well as on the lagged level of the dependent variable:

$$\begin{equation*} \Delta {Y_t} = {\alpha _0} + {\alpha _1}{Y_{t - 1}} + {\beta _0}\Delta {X_t} + {\beta _1}{X_{t - 1}} + {\varepsilon _t} \end{equation*}$$

The coefficient of the current change of a predictor (β₀) captures this predictor's instantaneous or short-term effect, while the coefficient of the lagged level (β₁) can be used to compute the overall long-term effect of a change in X. The coefficient of the lagged dependent variable (α₁) indicates how quickly this latter effect dissipates over subsequent time periods. The total long-term effect of a predictor, finally, can be calculated as the ratio of the coefficients of its lagged level (β₁) and the lagged level of the dependent variable (α₁; see also below).

In addition to including the unemployment rate and the budget balance as the two main predictors, I also control for a number of potentially relevant variables: government partisanship (the share of left parties in government), the strictness of dismissal rules for permanent contracts, and the presence of veto points in the political system (the effective number of parliamentary parties, and dummy variables for bicameralism and federalism).Footnote ⁸ All controls are introduced as both current changes and lagged levels except for the controls for bicameralism and federalism, for which only lagged levels are included. Since these institutional variables are very stable over time, their changes and levels are necessarily very highly correlated. Introducing both into the model thus creates problems of multicollinearity.

The controls for the features of the political system account for the potential distorting influence of institutional and partisan veto players, which may affect how strongly governments respond to economic downturns (Wlezien and Soroka, Reference Wlezien and Soroka2012). Government partisanship is included since there are good reasons to assume that the partisan composition of governments is simultaneously linked to economic downturns and sanction reforms, making it a potential confounder. That partisanship is linked to social policy reforms is, of course, one of the central hypotheses of power resources theory (Korpi and Palme, Reference Korpi and Palme2003).Footnote ⁹ Partisanship is also linked, however, to downturns since recent research has found that the electoral prospects of left parties are hurt by economic downturns (Lindvall, Reference Lindvall2014). It is therefore possible that any effects of economic downturns on sanction reforms may simply result from the fact that economic downturns favour right parties, who in turn are more likely to introduce tougher sanctions. The strictness of dismissal rules is controlled for to account for the fact that dismissal rules affect the distribution of economic vulnerability in the population, which may in turn affect how countries respond to increasing unemployment (Saint-Paul, Reference Saint-Paul1998; Rueda, Reference Rueda2014).

Finally, I also control for period-specific common shocks by introducing a set of year-dummies and use panel-corrected standard-errors to account for panel-heteroscedasticity (Beck and Katz, Reference Beck and Katz1995). Note that the findings are robust to various different specifications of the statistical model, including using a simple linear time trend instead of period dummies or not controlling for time at all.Footnote ¹⁰

The main results are presented in Table 1. Starting with the first model, which focuses on the effect of unemployment, the short-term effect of unemployment is not statistically significant and changes its sign across model specifications. Short-term fluctuations in the unemployment rate have, in other words, no significant effect on sanction reforms. This corresponds to the patterns shown in Figure 1 above. The lagged level of the unemployment rate, however, has a strong, significant and robust effect, which indicates that there may indeed be a significant long-term effect of increases in unemployment on sanction rules. This effect is, as expected, positive, which also confirms the finding from the descriptive analysis that sanction restrictions follow enduring increases in the unemployment rate.

TABLE 1.

The effects of unemployment and budget deficits on the strictness of sanctions

Panel-corrected standard errors in parentheses; * p < 0.05

The effect of the budget balance is estimated in the second model. In contrast to unemployment, budget deficits have a significant and negative short-term effect. This indicates that sanction reforms tend to be introduced rather swiftly when countries struggle with budget deficits. The fact that the lagged level is also significant and goes in the same direction suggests that there is an additional long-term effect of enduring budgetary problems, albeit a weaker one.

The third model includes both the unemployment rate and the budget balance as predictors. Although multicollinearity could of course be a problem here, it turns out that the budget balance and the unemployment rate are not overly strongly correlated (ρ = −0.27; p < 0.000), which suggests that there is not much risk in including both predictors in the same model. Doing so does change the results slightly: the coefficient for the lagged level of the budget balance becomes smaller and is estimated less precisely, and the coefficient for the current change in the unemployment rate becomes negative. The overall conclusion remains the same, however.

While the main predictors have rather clear effects, the effects of the controls are very weak. In fact, most have no significant effects at all. Particularly striking is the (not only statistically, but also substantively) insignificant effect of partisanship. This, however, resonates with recent criticisms of the partisanship-argument (Häusermann et al., Reference Häusermann, Picot and Geering2013) and, more generally, the recurrent finding that left governments have on many occasions tightened the access to unemployment benefits in recent decades (e.g. Arndt, Reference Arndt2013). Only bicameralism has a statistically significant effect but this effect turns out to be not robust across model specifications.Footnote ¹¹ The reason why these usually strong political predictors perform so weakly may be that, across all the advanced democracies, there is very little societal support for unemployment benefit claimants who breach benefit conditions.Footnote ¹² Hence, left parties and other potential advocates have few political incentives to defend this group and, in the absence of actors with an incentive to oppose sanction rule reforms, the presence of the means to effective opposition (political veto points) makes little difference.

The fact that only the lagged unemployment rate but not its contemporary change has a significant effect indicates that the effect of unemployment works only in the longer run. How strong exactly this long-term effect is can be calculated via the long-run multiplier (LRM)

$$\begin{equation&#x002A;} k = \frac{{{\beta _1}}}{{\left| {{\alpha _1}} \right|}} \end{equation&#x002A;}$$

where, as above, β₁ is the lagged level of the predictor and α₁ is the lagged level of the dependent variable (De Boef and Keele, Reference De Boef and Keele2008: 191-192). The long-run multiplier is the total effect that a change in a given predictor has over all subsequent time periods, starting at t+1.

The long-run multipliers for the unemployment rate and the budget balance, as well as their standard errors, are presented at the bottom of Table 1. All long-run multipliers are statistically significant (all are greater than 1.96 times their standard errors).Footnote ¹³ An increase in the unemployment rate has a robust and strong long-term effect on the strictness of sanctions. Depending on the model, an increase in the unemployment rate will result in an around three- to four-fold increase in the sanctions-indicator. This corresponds to the patterns shown in the descriptive analysis: Sweden, for instance, experienced an increase in the unemployment rate of around eight per cent between 1990 and 1994, which resulted in a reform that increased Sweden's score on the sanctions-indicator by a bit more than 20 points (see Figure 1).

The long-run multiplier of the budget balance is less stable across specifications and also not as precisely estimated. Controlling for unemployment in model 3 reduces the magnitude of the long-run multiplier of the budget balance considerably. In this specification, the long-term effect of the budget balance is only a bit more than half the size of the long-term effect of unemployment. This suggests, again, that the effect of unemployment unfolds over the longer run, whereas budget deficits have a more immediate impact.