Unemployment Risks and the Origins of Unemployment Compensation

The so-called class analysis is intrinsically based on the Marxian zero-sum nature of social interests—“the interests of capital and the interests of wage labor are diametrically opposed to each other” (Marx 2001[1847]: 39). So, in a pure form of class analysis, it follows that the possibility of cross-class compromise is almost infeasible by assumption, and even if it happens, it is not a sustainable equilibrium. About the conditions under which the compromise can occur, see Przeworski’s seminal article, “Proletariat into a Class: The Process of Class Formation” (1985). The power resources theory, one of the dominant approaches in the welfare state literature, is also fundamentally predicated in class analysis, while it focuses more on the link between the state (leftist party in the parliament and the cabinet) and society (labor movement; e.g., Stephens 1979; Korpi 1983; Esping-Andersen 1990; Huber and Stephens 2001).

A systemic exploration about the role of unions is beyond the scope of this paper. Yet, one thing clear from Baldwin (1992) and Mares (2003a, b) is that unlike a standard class analysis or “laborist approach” that focus on well-organized union movements exclusively based on class solidarity, many historical cases show that unions’ positions on issues of unemployment compensation greatly varied across countries precisely for the extent of the division among “risk categories” within unions. A typical example is a chronic division between craft skilled-based unions and unskilled unions in the working class movement. Indeed, it is well conceived in the class formation literature that a “class in itself” does not automatically translate into a “class for itself” (see Przeworski 1985). Concerning the issue of an institutional design of unemployment compensation, European trade unions revealed a varying degree of preferences. For example, French trade unions demanded a compulsory unemployment insurance (Mares 2003b: 111–113), and the British union movement “condemned the introduction of a compulsory scheme” (Harris 1972: 304), while Norwegian union leaders opposed the new unemployment insurance system and boycotted its implementation (Alber 1981: 153), and unions had no position on this issue in Switzerland (Harrington 1998: 209).

The study of welfare policy as the public provision of insurance has been recently burgeoning. Barr (1998, 2001) provides a generic social insurance model that explains the state intervention when social risks are differentiated. Hindriks and De Donder’s model (2003) incorporates people’s different risks and incomes in the context of provision of social insurance. Baldwin (1992), Harrington (1998), and Mares (2003a, b), as mentioned in the text, historically trace the origins of European welfare states, focusing on the importance of social risks. Inspired by Foucault’s notion of “governmentality,” Francois Ewald (1991) presents a sharp analysis about the relationship between the social insurance technique and the emergence of concept of unemployment as a risk. In a similar vein, Walters (2000) traces the history of governing unemployment as risk management. Moss (2002) employs the theory of risk management and analyzes the history of American social security. See also Rosanvalon (2000) and Kim (2007a).

Notice that a homogeneous low-risk society may appear in societies other than an agricultural society. A society where firms have difficulty or are not willing to fire workers (due to tight government employment policies, enterprise-based compensation, or labor hoarding) shows low unemployment rates. But these practices stem from a society’s employment system, not from its industrial configurations.

The notion of unemployment was invented along with the advent of industrial society. In contrast to prior beliefs and practices that viewed unemployment as an individual risk, resulting from idleness or unwillingness to work due to moral shortcomings of the worker, the “modern” notion of unemployment stresses the social determinations of this risk and the fact that unemployment is a natural, if regrettable by-product of a particular organization of work in the modern enterprise (Piore 1987).

About the data sources, see footnote 5. The number of countries declines from 170 to 144 in the statistical analysis, since the most important independent variables—the inequality and magnitude of risks—have missing observations in 37 countries.

Empirical studies supporting the sectoral variation of unemployment are abundant. See among others Creedy (1981).

In addition to the Gini index, I have constructed other similar indices such as the Theil index and coefficient of variation (CV) and found they are almost identical.

Although it was officially adopted by many international organizations (the UN, ILO, FAO, and UNESCO) in 1968, this classification had been historically used long before. I have found that many historical data follows this classification of nine sectors with one- digit codes.

This data set is the most comprehensive and consistent for my purpose in that the number of employed, unemployed, and labor force (or economically active population) are all collected with a consistent classification of sectors (the nine sectors), the longest time coverage and the largest number of countries in a single data-file. The data I used to calculate sectoral unemployment rates covers 43 countries from 1970 to 2000.

One might wonder that the calculated unemployment rates instead of these scales should be assigned to capture degrees of risk among sectors. However, it is not certain that these degrees of risk can be universally applied in the earlier years. Since the main purpose of this operationalization is to establish the rank of risks by sectors, I minimize this uncertainty and standardize the degrees of risk, by using ordered cardinal values fall between 1 and 9 with one unit interval. Yet both scales produce substantively similar results.

An important aspect is that I have grouped the data of the nine sectors, so the above standard equation of the Gini index needs to be modified. Notice that here workers not sectors are arranged in ascending order among the whole sample, N. Although workers’ risk values are the same within the same sector, they are treated as if they are arranged in ascending order in calculating the Gini coefficient. What differentiates workers is the sectoral risk values which workers are entitled by being employed in their sectors (find that ascending orders are multiplied by the sectoral risk values). Thus, the modified equation with grouped data is as follows: \( {\text{Gini}} = 1 + \left( {1/N} \right) - \left[ {\left( {1/2} \right){N^2}\mu } \right] \times \left[ {{y_5}\sum\nolimits_{i = 1}^{{n^5}} { + {y_2}} } \right.\sum\nolimits_{i = {n^5} + 1}^{{n^2}} { + {y_3}} \sum\nolimits_{i = {n^2} + 1}^{{n^3}} { + {y_6}} \sum\nolimits_{i = {n^3} + 1}^{{n^6}} { + {y_8}} \sum\nolimits_{i = {n^6} + 1}^{{n^8}} { + {y_7}} \sum\nolimits_{i = {n^8} + 1}^{{n^7}} { + {y_9}} \sum\nolimits_{i = {n^7} + 1}^{{n^9}} { + {y_4}} \sum\nolimits_{i = {n^9} + 1}^{{n^4}} { + {y_1}} \left. {\sum\nolimits_{i = {n^4} + 1}^{{n^1}} {} } \right] \).

In the original ACLP, a regime type variable is coded 1 for dictatorships and 0 for democracies. I switched the value from 0 to 1 for democracies to interpret the results in terms of democracy.

To fully capture the movement, a variable of trade unions should be included instead. However, due to the paucity of data on unionization, which covers only several data points for advanced capitalist countries, I take this measure as a rough proxy for the strength of the working class movement. In fact, based on the limited union density figures taken from Stephens (1979: Table 4.8), I found a relatively high correlation between unionization and this variable in early western European countries, 0.69. Nevertheless, I acknowledge the deficiency of the measure and accordingly draw the inferences with caution until better indicators of unionization for developing countries can be found.

An alternative method is to use likelihood ratio tests that are done against the three hypotheses (i.e., whether the model is lognormal, Weibull, or exponential) regarding the relevant parameters from the generalized gamma model (Box-Steffensmeier and Jones 2004: 41). I found that none of the three null hypotheses can be rejected, suggesting any of the three models—exponential, Weibull, and lognormal—is a possibility.

The AIC is defined as: \( AIC = - {\text{2ln}}\;L + { 2}\left( {a + c} \right) \); where L is log-likelihood, a is the number of model covariates, and c the number of model-specific distributional parameters. I estimated the models with the second lowest values as well, and their results are substantively similar to the ones with the lowest values. The AIC values are as follows:

This method is better than the use of a series of dummy variables, since the number of years in my sample is too great, resulting in the loss of many degrees of freedom. Alternatively, I employed more flexible semiparametric models of the generalized additive model and the Cox model and found that the results remain intact. The results are available on request.

Specifically, for risk inequality, the estimated hazard rate increases by 0.197%, that is, \( \exp \left[ {\left( {{{\hat \beta }_{\text{RiskInequality}}} \times 2} \right) - \left( {{{\hat \beta }_{\text{RiskInequality}}} \times 1} \right)} \right]/\left( {{{\hat \beta }_{\text{RiskInequality}}} \times 1} \right) \). For risk level, the increase is 4.640 and 2.781 for democracy.

Risk level is set at 3.91, working class at 13.98%, income at $2,943.30, and time at 42.54.