Gallego and Marx Online Appendix

Gallego and Marx Online appendix

A. Survey and sample

The analyses are based on an original conjoint survey experiment conducted in April and May of 2014 in Spain by Netquest, a commercial company. The sample size was 1,600, of which 1,508 responses have complete information for all variables. The company employs a quota sample drawn from their pool of respondents recruited on different websites. Respondents accrue points for their participation, which they can then exchange for money. The exact amount varies depending on responses in behavioural measures. The survey contained quotas for age, sex, education, and region. The aim of this study is not to produce descriptive frequencies of variables in the population but to examine causal treatment effects across different groups and, hence, a fully representative sample is not required. While the sample is more left-wing than the population, we have sufficient variation across relevant socio-demographic and political dimensions to compare treatment effects between groups.

Table A1 reports some socio-economic and political characteristics of the sample that were asked in similar ways in a survey by the Spanish Center for Sociological Research (CIS), conducted in May 2014.

Table A1: Comparison of the online survey and CIS

Online survey May 2014 / CIS survey
May 2014
% working (permanent or temporary contract) / 43 / 40
% unemployed / 20 / 25
If working, not at all/not very likely to lose job / 75 / 75
If unemployed, not at all/not very likely to find job / 80 / 64
% with households income < 3000 €/month / 63 / 60
Average position in left-right scale / 3.8 (on 0-10 scale) / 4.6 (on 1-10 scale)

Importantly for the purposes of this research, we have a large number of unemployed respondents and ample variation in the risk of unemployment across respondents. Fully 20 percent of the sample is unemployed, of which 53 percent have been so for more than two years, 20 percent between one and two years, 14 percent between six months and one year. The official unemployment rate in Spain at the time of the fieldwork was greater than 25 percent. We don’t expect full agreement for several reasons: the sample also contains inactive population such as pensioners or students; and a non-negligible number of officially unemployed people work on the black market such that the numerator is overestimated in the official figure but not in our survey, which included the option of working without a contract (4 percent of respondents). Unemployed respondents were pessimistic about their prospects of finding a job, as 70 percent report that the likelihood of finding a job is small or nil.

Among those in employment (47 percent of the sample) 72 percent have unlimited permanent contract, 19 percent have a temporary contract, and 9 percent have no contract. As can be expected from the dualized nature of the Spanish labour market, a large part of those who are currently employed perceive their employment to be relatively secure: 75 percent think that it is not at all or not very likely that they will lose their job. Past experiences of unemployment are relatively common. Only 46 percent of those in employment have never been unemployed before.

When compared to the CIS survey, respondents in our survey are similar or slightly more left-wing. The mean score in our survey is 3.8, and the mean score in the 2014 CIS survey is 4.5, but the rating scales differ. While it ranges from 0 to 10 in our survey, it ranges from 1 to 10 in the CIS survey.

B. Question wording of the conjoint experiment

The conjoint experiment asked respondents to choose between pairs of policy reforms that could differ along five dimensions described in the main text. Each of the five dimensions had several attribute levels and these were fully randomized, implying that some pair of proposals could have the same attribute (e.g. both proposals could have the same cost), while varying on others. The experiment allows us to assess the influence of different possible characteristics of unemployment policies on support for a proposal of policy change. This approach allows for the non-parametric estimation of the treatment effects of each attribute, and those effects are all estimated on the same scale.

The experiment was the second question the respondents encountered in the questionnaire. They could read “There is some talk about reforming current unemployment policies. Suppose there are two proposals with the following characteristics” and after reading the two proposals, they were asked “Which proposal would you prefer?” The attributes of each dimension were:

Generosity of the subsidies:

-  No change

-  Extends the 426 euros subsidy for 4 additional months

-  Increases the benefit by 20% for the three first months of unemployment

-  Maintains the benefit after the first six months

Coverage:

-  All residents in Spain

-  Spanish citizens

-  Households with an income lower than the mean salary

-  Households in which no one receives a salary or pension

More training programs:

-  No changes

-  Run by the public employment agency

-  In companies

-  Run by trade unions

Additional costs:

-  100 million euros

-  1,000 million euros

-  2,000 million euros

-  3,000 million euros

How to fund the reform:

-  Increasing IVA (consumption tax)

-  Increasing the income tax

-  Increasing public debt

-  Cuts in education and health

-  Cuts in police, external affairs and defence

Table B1: Sample pair of choices in the conjoint experiment (Example)

Proposal 1 / Proposal 2
Change in benefits / Extend the 426 euros subsidy for 4 additional months / Increase the benefits in the first three months of unemployment by 20%
More training programs for the unemployed / No change / Run by trade unions
Recipients / No one in household has a salary or pension / All Spanish residents
Cost of the program / 3,000 million euros / 100 million euros
How to pay for the reform / Cuts in education and health / Increasing the income tax

C. Individual differences: variables and results

Our measure of unemployment risk is employment status. We collapse the response options into four categories: employed with permanent contract, employed with temporary contract, unemployed, or in other situations (we only report for the first three categories). Given the dualism of the Spanish labour market, the sample contains a large number of temporary workers (9 per cent) and unemployed (20 per cent). This makes the sample well-suited to study the preferences of these groups.

Following Rehm (2009), we also assess the unemployment risk at the occupational level by assigning respondents to the unemployment rate of their current or last occupation, calculated at two-digits level of the International Standard Classification of Occupations.[1] In addition, we asked respondents about their perceived risk of becoming unemployed. Among those currently employed, 75 per cent think that it is not at all or not very likely that they will lose their job and 25 per cent think that this is very or quite likely.

Net household income serves as an indicator of socio-economic status. The question was asked offering nine response categories, which we collapse into three in some analyses. Our final self-interest variable is education, which is measured in six categories.

To capture ideology, we use respondents’ self-placement on the left-right scale (0-10). Spanish citizens are one of the most left-wing populations in Europe, according European Social Survey data. This sample is typical for Spain with the average left-wing score being 3.8. In some analyses, we distinguish between left-wing respondents (0-4), centrists (5), and right-wing respondents (6-10).

Three percent of respondents do not have the Spanish nationality. The results do not vary substantively when we exclude them.

Table C1 presents the results of linear probability models interacting respondents’ socio-economic situation with attributes of policy proposals. As explained in the methods section, the dependent variable is choice for one of the proposal which we model as a functions of the proposal’s attributes. For each model, the results are reported in two different columns. The first column displays the coefficient γ of the individual-level ideological measure zi and the vector of coefficients β or the “main” effects of the attributes X. The second column displays the vector of interaction coefficients δ discussed in Equation (2).

1

Table C1: Self-interest as a moderator of treatment effects

Model (1) / Model (2) / Model (3) / Model (4) / Model (5) / Model (6)
Permanent-Temporary / Permanent-Unemployed / Occupational risk / Subjective risk / Household income / Education level
Self-interest variable / -0.064 / -0.128* / 0.003 / 0.033 / 0.018* / 0.034* / -0.064
(0.096) / (0.072) / (0.004) / (0.047) / (0.011) / (0.020) / (0.096)
Generosity
Longer 426 euros subsidy / 0.088*** / 0.087 / 0.088*** / 0.064 / 0.118*** / 0.001 / 0.245** / -0.047 / 0.126*** / -0.009 / 0.189*** / -0.022
(Ref. no change in benefits) / (0.031) / (0.066) / (0.031) / (0.050) / (0.044) / (0.003) / (0.097) / (0.033) / (0.038) / (0.008) / (0.057) / (0.014)
Increase at start / 0.024 / 0.107* / 0.024 / -0.001 / 0.033 / 0.001 / 0.147 / -0.036 / 0.074** / -0.008 / 0.103* / -0.016
(0.030) / (0.061) / (0.030) / (0.048) / (0.043) / (0.003) / (0.090) / (0.031) / (0.037) / (0.008) / (0.055) / (0.013)
Constant benefits after 6 months / 0.062** / 0.028 / 0.062** / -0.076 / 0.029 / 0.002 / 0.070 / 0.001 / 0.037 / 0.002 / 0.073 / -0.007
(0.031) / (0.063) / (0.031) / (0.051) / (0.044) / (0.003) / (0.096) / (0.033) / (0.039) / (0.008) / (0.056) / (0.014)
Coverage
Recipients only Spanish citizens / 0.012 / -0.004 / 0.012 / 0.002 / -0.008 / 0.003 / 0.048 / -0.009 / 0.029 / 0.001 / 0.065 / -0.007
(Ref. all residents) / (0.031) / (0.068) / (0.031) / (0.051) / (0.043) / (0.003) / (0.102) / (0.035) / (0.037) / (0.008) / (0.056) / (0.014)
Below median salary / 0.010 / 0.041 / 0.010 / 0.033 / 0.014 / 0.001 / 0.327*** / -0.107*** / 0.064* / -0.007 / 0.080 / -0.012
(0.032) / (0.067) / (0.032) / (0.049) / (0.044) / (0.003) / (0.105) / (0.035) / (0.037) / (0.008) / (0.056) / (0.014)
No salary in household / -0.031 / 0.053 / -0.031 / 0.075 / -0.021 / 0.002 / 0.089 / -0.036 / 0.023 / -0.006 / 0.141** / -0.034**
(0.030) / (0.066) / (0.030) / (0.048) / (0.043) / (0.003) / (0.106) / (0.036) / (0.037) / (0.007) / (0.058) / (0.014)
Human capital intensity
Training by public services / -0.027 / -0.046 / -0.027 / 0.060 / -0.028 / 0.001 / -0.167* / 0.045 / 0.069* / -0.019** / 0.041 / -0.016
(Ref. no change in training) / (0.031) / (0.068) / (0.031) / (0.050) / (0.044) / (0.003) / (0.100) / (0.034) / (0.038) / (0.007) / (0.055) / (0.013)
Training in companies / -0.054* / 0.042 / -0.054* / 0.066 / 0.050 / -0.006* / -0.065 / 0.012 / 0.061 / -0.011 / -0.039 / 0.010
(0.031) / (0.069) / (0.031) / (0.050) / (0.043) / (0.003) / (0.105) / (0.035) / (0.037) / (0.008) / (0.058) / (0.014)
Training in trade unions / 0.014 / 0.003 / 0.014 / 0.026 / 0.066 / -0.002 / -0.060 / 0.033 / 0.073* / -0.006 / -0.022 / 0.016
(0.030) / (0.065) / (0.030) / (0.050) / (0.044) / (0.003) / (0.101) / (0.034) / (0.038) / (0.007) / (0.055) / (0.014)
Costs
1000 million / -0.024 / 0.079 / -0.024 / 0.005 / 0.009 / -0.002 / 0.047 / -0.020 / -0.014 / -0.001 / -0.005 / -0.003
(100 million) / (0.032) / (0.072) / (0.032) / (0.051) / (0.043) / (0.003) / (0.106) / (0.036) / (0.038) / (0.008) / (0.058) / (0.014)
2000 million / -0.028 / 0.029 / -0.028 / 0.033 / 0.045 / -0.005* / 0.220** / -0.089** / -0.032 / 0.001 / -0.066 / 0.010
(0.029) / (0.067) / (0.029) / (0.050) / (0.041) / (0.003) / (0.102) / (0.035) / (0.036) / (0.007) / (0.054) / (0.013)
3000 million / -0.014 / -0.038 / -0.014 / 0.051 / 0.065 / -0.007** / 0.071 / -0.038 / -0.003 / -0.001 / 0.009 / -0.007
(0.031) / (0.068) / (0.031) / (0.051) / (0.044) / (0.003) / (0.110) / (0.037) / (0.038) / (0.008) / (0.057) / (0.014)
Sources of funding
Income tax / 0.082** / 0.049 / 0.082** / 0.094* / 0.082* / 0.003 / 0.011 / 0.027 / 0.136*** / -0.006 / 0.205*** / -0.022
(Ref. Indirect taxes) / (0.035) / (0.081) / (0.035) / (0.057) / (0.048) / (0.003) / (0.124) / (0.041) / (0.044) / (0.008) / (0.063) / (0.015)
Public debt / 0.096*** / -0.055 / 0.096*** / 0.009 / 0.097* / -0.001 / -0.030 / 0.037 / 0.088* / -0.003 / 0.133** / -0.013
(0.035) / (0.081) / (0.035) / (0.060) / (0.050) / (0.004) / (0.114) / (0.038) / (0.045) / (0.009) / (0.067) / (0.016)
Education and health / -0.207*** / -0.056 / -0.207*** / 0.039 / -0.187*** / -0.000 / -0.336*** / 0.047 / -0.202*** / -0.001 / -0.122** / -0.019
(0.034) / (0.075) / (0.034) / (0.055) / (0.048) / (0.003) / (0.113) / (0.038) / (0.042) / (0.008) / (0.060) / (0.014)
Police and defence / 0.169*** / -0.064 / 0.169*** / 0.085 / 0.161*** / 0.002 / -0.003 / 0.058 / 0.167*** / 0.005 / 0.212*** / -0.003
(0.035) / (0.074) / (0.035) / (0.059) / (0.049) / (0.004) / (0.112) / (0.038) / (0.044) / (0.009) / (0.066) / (0.016)
Constant / 0.463*** / 0.463*** / 0.375*** / 0.347** / 0.337*** / 0.271***
(0.044) / (0.044) / (0.059) / (0.138) / (0.054) / (0.081)
Observations / 2,576 / 3,224 / 4,020 / 2,836 / 4,532 / 6,032
R-squared / 0.082 / 0.087 / 0.083 / 0.084 / 0.082 / 0.085

*** p<0.01, ** p<0.05, * p<0.1. The entries are logistic regression coefficients with clustered standard errors in parentheses estimated from six regression models. Columns 1 and 2 present the results of a single regression model. Column 1 displays the uninteracted or “main” effects of the variables of interest and column 2 displays the coefficients of the interaction between self-interest variables and each attribute (e.g. longer 426 euros subsidy*type of contract). Columns 3 and 4 replicate the same analysis with being unemployed as the moderating variable, and so on. See the methods section for a discussion of the models.