• Aucun résultat trouvé

Operationalisation & methods

Chapter 4. Workers’ preferences on flexicurity

4.3 Operationalisation & methods

7

Austria, Belgium, Switzerland, Germany, Spain, Finland, France, Italy, Netherlands,

Norway, Portugal, Sweden.

country analysis are arguably weaker, I proceed with the analysis although cautioning against reading its results like nothing but an exploratory, preliminary take on the issue that will need dedicated field research to be further confirmed.

Moving on to the NEB variable, it measures the agreement with the proposal of increasing income differences between employed and unemployed (again: ’very much agree’, ‘tend to agree’, ‘tend to disagree’, ‘fully disagree’). The ALMP variable does the same with increasing ALMP expenditure.

To increase results’ relevance and minimise pro-liberalising biases as much as possible, when testing preferences for single policies, I isolate the most positive answer ('very much agree' or 'fully disagree') to all three questions to generate a proxy for preference salience. Moreover, I take advantage of the new INVEDUC dataset collected by Busemeyer et al. (2018) to run consistency tests for ALMP and NEB policy preference structures.

Independent variables are built by combining the Eurobarometer’s occupational classification and the division drawn by classical political economy of labour market institutions between insiders and outsiders.

These are provided for currently employed individuals, as well as for the unemployed.

As for job market position, I am unable to exactly reproduce Rueda’s classic operationalization of insiders and outsiders since the Eurobarometer survey has no additional information that would allow me to draw lines among the employed. Instead, I substitute 'insiders' with those employed, while 'outsiders' are proxied by those without a job and actively seeking a position.

First, I conduct the analysis using occupation and labour market position as independent variables for all three policy fields (EPL, NEB, ALMP). I use logistic regressions since the possibility of running marginal effects tests perfectly fits my ultimate goal of comparing different probabilities of supporting policy positions within occupational and job position groups. I feature one model specification per IV-DV couple. Aside from my variable of interest, the model is completed with controls for gender, age, country and year fixed effects.

Given the idiosyncrasy of the questions used to operationalise my key variables, it seems necessary to run some sort of parallel test to broadly confirm or discard the findings obtained through the Eurobarometer. As stated above, there is no other source of equal quality that features fitting questions for all three policy dimensions, including at the same time so many countries while also giving the possibility to build IVs that match the frameworks to be tested. There are, however, some sources that do include at least NEB and ALMP (surveys on EPL preferences seem to be very uncommon for no apparent reason). Of these, I choose the most recent wave of the European Social Survey (ESS 8th wave - 2016). A special module with questions on NEB and ALMP has been included for a subset of countries that, although does not fully match my broad Eurobarometer selection, is broad enough to grant some insights through comparison. The selected NEB question investigates the degree of agreement with the following proposal: "Unemployment benefit should be cut if an individual turns down a job that requires a lower level of education of that held by the individual", answers ranging from "1: Should lose all unemployment benefit" to "4: Should keep all unemployment benefit".

The ALMP question inquiries about the degree of agreement (from 1="strongly against" to 4="strongly

in favour") with the following proposal: "Spend more on education for unemployed at cost of unemployment benefit". Both do differ from the original Eurobarometer questions, but that is part of the reason to choose them: by using different operationalisations for the same DVs, findings should be strengthened and enriched.

In a second phase, the operation is repeated but the dependent variable is changed. Now by using the combined version to determine the support for flexicurity and for liberalisation, I build two IVs.

‘Flexicurity’ measures the likeliness of the individual to support a policy package that would include EPL reduction, ALMP increase and the maintenance of NEB levels. ‘Liberalisation’, instead, refers to a reduction in all three dimensions. Recalling that the chosen Eurobarometer questions have four possible positions (1 = “strongly agree” and 4 = “strongly disagree”), I combine them into a five-level ordered variable for each policy package explained in the following table.

Table 4.1. Operationalising preferences for flexicurity and liberalisation: variable definition.

Flexicurity Liberalisation

5 Strongly agreeing with EPL reduction, ALMP increase;

strongly disagreeing with NEB reduction.

Strongly agreeing with EPL/NEB reduction;

strongly disagreeing with ALMP increase.

4 Two out of three of the following: Strongly agreeing with EPL reduction/ALMP increase, strongly disagreeing with NEB reduction. The remaining should be “mildly agreeing (EPL, ALMP)/disagreeing (NEB)”.

Two out of three of the following: Strongly agreeing with EPL/NEB reduction; strongly disagreeing with ALMP increase. The remaining should be “mildly agreeing (EPL, NEB)/disagreeing (ALMP)”.

3 One out of three: Either strongly agreeing with EPL reduction/ALMP increase, or strongly disagreeing with NEB reduction, while mildly agreeing on the other two.

The remaining should be “mildly agreeing (EPL, ALMP)/disagreeing (NEB)”.

One out of three of the following: Strongly agreeing with EPL/NEB reduction; strongly disagreeing with ALMP increase. The remaining should be “mildly agreeing (EPL, NEB)/disagreeing (ALMP)”.

2 Mildly agreeing with EPL reduction, ALMP increase;

mildly disagreeing with NEB reduction.

Mildly agreeing with EPL/NEB reduction; mildly disagreeing with ALMP increase. the opposite policy position of flexicurity is fairly concurrent with stringent dualisation (maintaining EPL, reducing NEB, not increasing ALMP), while liberalisation would be countered with full-scale protection (maintaining EPL and NEB, increasing ALMP). However, since this approach omits all those positions that do not identify with either of the respective extremes, the sample gets significantly reduced. This, together with the already mentioned pro-liberalisation bias introduced by the formulation of the EPL question, I must insist on the fact that the values that it produces are only relevant in relative terms (i.e. to be compared internally), and therefore should be read as such.

Results of all specifications are included in the Appendix. In the following Section I focus on interpreting predicted probabilities since its interpretation is much more straightforward as well as appropriate for my hypotheses.