Supplement — Unemployment Insurance and the Family: Heterogeneous Effects of Benefit Generosity on Reemployment
Summary
S1 Online Supplement to Kuhn, Ursina, Debra Hevenstone, Leen Vandecasteele, Samin Sepahniya and Dorian Kessler, 2024, "Unemployment Insurance and the Family: Heterogeneous Effects of Benefit Generosity on Reemployment and Economic Precarity," Sociological Science 11: 649-679. Its contents list sections on data preparation and analytical methods, results on reemployment, results on financial precarity, job quality and references. The opening section assesses the discontinuity design around 18 months of contribution, including heaping due to voluntary quits and the relation between contribution months and socio-economic characteristics, with tables comparing groups by contribution month.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
- Issuer
- Source documents
- Document type
- Unemployment Insurance And The Family Heterogeneous Effects Of Benefit Generosity On Reemp
- Case
- Unemployment Insurance And The Family Heterogeneous Effects Of Benefit Generosity On Reemployment And Economic Precarity
Cite this document
“Supplement — Unemployment Insurance and the Family: Heterogeneous Effects of Benefit Generosity on Reemployment.”
Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/unemployment-insurance-and-the-family-heterogeneous-effects-of-benefit-generosity-on-reemployment-and-economic-precarity/
, SHA-256 253631941879b7e7eab3f4a5e1516b4af9ee8ba7fc6877d19f67b2a92ba8ef53.Full text
Supplement to:
Kuhn, Ursina, Debra Hevenstone, Leen Vandecas-
teele, Samin Sepahniya and Dorian Kessler. 2024.
“Unemployment Insurance and the Family: Hetero-
geneous Effects of Benefit Generosity on Reemploy-
ment and Economic Precarity.” Sociological Science
11: 649-679.
S1
Online Supplement
Unemployment Insurance and the Family:
Heterogeneous Effects of Benefit Generosity on
Reemployment and Economic Precarity
Ursina Kuhn, Debra Hevenstone, Leen Vandecasteele, Samin Sepahniya, Dorian Kessler
Contents
A1. Data preparation and analytical methods 2
A1.1 Discontinuity assessment 2
Potential manipulation of the assignment variable (contribution months) 2
Relation between contribution months and socio-economic characteristics 2
Anticipation effects 6
A1.2 Evolution of labor market status, welfare state benefits and precarity 7
A1.3 Linear probability model and logistic regression 9
A2. Results on reemployment 14
A2.1 Regression tables for reemployment (pooled analysis) 14
A2.2. Employment: Logistic regression 15
A2.3 Employment: by month after unemployment 15
A2.4 Employment: including individuals with 12 contribution months 16
A2.5 Employment: additional control variables 16
A2.6 Employment: adding other households 17
A3. Results on financial precarity 18
A3.1 Regression tables for financial precarity 18
A3.2 Precarity: Logistic regression 19
A3.3 Precarity: by month after unemployment 19
A3.4 Precarity: including individuals with 12 contribution months 20
A3.5 Precarity: additional control variables 20
A3.6 Precarity: including child allowances 21
A3.7 Precarity: alternative threshold 22
A3.8 Precarity: adding other households 23
A4. Job quality: employment income (only employed individuals) 23
References 25
1
A1. Data preparation and analytical methods
A1.1 Discontinuity assessment
Potential manipulation of the assignment variable (contribution months)
The validity of the discontinuity design relies on the assumption that individuals do no not select
into treatment (contribution month). We discuss the validity of the assumption regarding heaping
due to voluntary quits (1), relationship of contribution months with socio-economic variables (2),
and placebo tests for the first year of unemployment, when all unemployed in the sample have
benefit rights (3).
As unemployment benefits are considerably more generous after 18 months of contribution, em-
ployees who intend to quit have an incentive to wait until they have 18 months of contribution be-
fore doing so. That said, most voluntary quitters do not go on to unemployment insurance bene-
fits, but move directly to another job, because voluntary quitters have a waiting period before
they can receive unemployment benefit, which discourages voluntarily quits before having a new
job offer. Incentives to strategically wait to reach 18 months of contribution before quitting a job
exist only for employees who anticipate more than 12 months of unemployment. If individuals
feel confident of finding a job within a year, the difference in PBD is unlikely to affect the timing
in quitting their job. If the job loss is involuntary, employees have no control over the timing and
months of contribution. The same applies if unemployment occurs because a temporary contract
has ended or individuals do not immediately find a job after leaving education.
The distribution of contribution months does not show an accumulation just after the threshold
at 18 contribution months (Figure 1 in the main text). Although there is a slight increase in the
number of unemployed with between 17 and 18 contribution months (from 4,481 to 5,981), the
frequencies are continuously increasing with contribution months, with the most important jumps
at 23 and 24 months of contribution.
Relation between contribution months and socio-economic characteristics
As an additional assessment of the continuity assumptions, we compared various observable
characteristics by number of contribution months descriptively in Table A1 and using regression
models in Table A2. Several characteristics are associated with more contribution months: occu-
pation, tertiary education, Swiss citizenship, living with a partner (both married and cohabiting),
having dependent children, being female. Moreover, income levels increase with the number of
contribution months. This is the case for insured income, income in the year before unemploy-
ment, partner’s income before unemployment and household income. The descriptive analysis
shows a rather linear increase with no specific selectivity between contribution month 17 and 18.
2
16 (N=4200) 17 (N=4501) 18 (N=6007) 19 (N=6054)
Gender: female 2047 (49.0%) 2211 (49.3%) 2989 (50.1%) 2994 (49.9%)
Household type
Single 1324 (31.5%) 1427 (31.7%) 1867 (31.1%) 1822 (30.1%)
Breadw. No kids 240 (5.8%) 250 (5.6%) 333 (5.6%) 315 (5.2%)
Breadw. kids 341 (8.2%) 382 (8.5%) 506 (8.5%) 521 (8.6%)
Egalitarian no kids 474 (11.4%) 520 (11.6%) 707 (11.9%) 822 (13.6%)
Egalitarian kids 502 (7.1%) 494 (11.0%) 721 (12.1%) 732 (12.1%)
Sec. Earner no kids 221 (5.3%) 227 (5.1%) 281 (4.7%) 270 (4.5%)
Sec. Earner kids 323 (7.7%) 386 (8.6%) 502 (8.4%) 481 (8.0%)
Single parent 749 (17.9%) 795 (17.7%) 1047 (17.6%) 1091 (17.4%)
Married 1265 (30.1%) 1380 (30.8%) 1843 (30.9%) 1908 (31.8%)
Age 36.3 (7.8) 36.3 (7.8) 36.7 (7.8) 36.5 (7.7)
Nationality: Swiss 2018 (48.4%) 2265 (50.6%) 3198 (53.0%) 3270 (54.5%)
Partner’s income (/1000, 2.219 (3.2) 2.377 (3.9) 2.557 (3.7) 2.611 (3.9)
before unemployment
Insured employment in- 4.609 (2.2) 4.669 (2.2) 4.937 (2.4) 4.989 (2.4)
come (/1000)
Waiting days 4.3 (4.4) 4.5 (4.4) 4.8 (4.7) 4.8 (4.6)
Education
Low 907 (21.6%) 865 (19.2%) 1065 (17.7%) 1042 (17.2%)
Upper Secondary 2004 (47.7%) 2160 (48.0%) 2842 (47.3%) 3007 (49.7%)
Tertiary 1289 (30.7%) 1476 (32.8%) 2100 (35.0%) 2005 (33.1%)
Occupation
High 770 (18.3%) 891 (19.8%) 1253 (20.9%) 1251 (21.2%)
Intermediate 2590 (53.4%) 2836 (53.2%) 3993 (55.6%) 3997 (55.8%)
Low 1610 (38.3%) 1665 (37.0%) 2014 (33.5%) 2057 (34.0%)
Region
Central 290 (6.9%) 287 (6.4%) 668 (8.1%) 667 (7.6%)
East 484 (11.5%) 612 (13.6%) 1168 (12.6%) 1122 (12.9%)
Leman 1000 (23.8%) 1035 (23.0%) 2188 (23.1%) 2059 (22.0%)
Mittelland 899 (21.4%) 936 (20.8%) 1685 (19.1%) 1696 (19.6%)
Northwest 496 (11.8%) 505 (11.2%) 995 (11.6%) 1028 (11.5%)
Ticino 175 (4.2%) 186 (3.7%) 401 (4.4%) 334 (3.6%)
Zurich 856 (20.4%) 958 (21.3%) 1785 (21.1%) 1853 (22.8%)
Table A1: Descriptive statistics by contribution month
As a more formal assessment, we use socio-economic variables as dependent variables to test
whether there is a discontinuity between 17 and 18 contribution months, controlling for linear
and quadratic contribution month. This is the same approach as in the main analysis. As
3
dependent variables we test educational level (low, intermediate, high), occupational categories
(low qualified, intermediate, high), insured income (income level before unemployment), partner
income, and age. We find discontinuities between 17 and 18 contribution months for low and
high education (men only), occupation (low skilled men, medium skilled women), income level
(men and women), and partner income (men only). There are no apparent selection effects with
regard to age, women’s education, and partner income for women. When controlling for insured
income and partner income (as we do in the main analysis), the effects for education and different
occupations are no longer relevant, with one exception (intermediate occupations for women). As
this is only one effect out of 14 variables tested, we consider this to be random, and believe that
selection effects are taken into account.
That said, it is important to note that sensitivity analysis show that estimates for shorter PBD are
hardly affected by control variables in the model. For example, the marginal effects for employ-
ment rate amount to 6.0, 6.3 or 6.2 percentage points when PBD is shorter, depending on
whether no socio-economic controls, controls for income before unemployment or extended con-
trols are considered in the models.
4
Insured income Partner income Age
Men Women Men Women Men Women Men Women
Short PBD (Ref: Long PBD) 0.231*** -0.194*** -0.240** 0.018 -0.315 -0.433* -0.154 -0.363
Contribution months 0.107 0.091 0.006 0.190 -0.606** -0.830*** -0.699** -0.846***
squared -0.001 -0.001 0.000 -0.003 0.017** 0.022*** 0.017** 0.022***
Insured income 0.877*** 0.351***
Partner income -0.173*** -0.087***
Constant 3.9 2.983 1.752 1.179 42.671 439 396 42.9
Observations 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853
Low education Intermediate education High education
Men Women Men Women Men Women Men Women Men Women Men Women
Short PBD (Ref: Long PBD) 0.020* 0.010 0.010 0.003 0.014 0.001 0.001 -0.007 -0.034** -0.011 -0.012 0.004
Contribution months -0.014 -0.013 -0.010 -0.009 0.007 -0.015 0.012 -0.011 0.007 0.028 -0.002 0.020
squared 0.000 0.000 0.000 0.000 -0.000 0.001 -0.000 0.000 -0.000 -0.001 -0.000 -0.001
Insured income -0.039*** -0.037*** -0.042*** -0.039*** 0.080*** 0.077***
Partner income -0.004*** -0.005*** -0.012*** -0.005*** 0.016*** 0.010***
Constant 0.376 0.349 0.534 0.467 0.388 0.570 0.571 0.694 0.235 0.081 -0.105 -0.161
Observations 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853
Low occupation Intermediate occupation High occupation
Men Women Men Women Men Women Men Women Men Women Men Women
Short PBD (Ref: Long PBD) 0.034** 0.019 0.018 0.012 -0.012 -0.026* -0.015 -0.031* -0.022* 0.007 -0.003 0.019
Contribution months 0.007 0.008 0.013 0.012 -0.019 -0.021 -0.017 -0.018 0.012 0.013 0.004 0.006
squared -0.000 -0.000 -0.000 -0.000 0.001 0.001 0.001 0.000 -0.000 -0.000 -0.000 -0.000
Insured income -0.060*** -0.038*** -0.016*** -0.028*** 0.075*** 0.065***
Partner income -0.010*** -0.001 0.001 -0.005*** 0.009*** 0.005***
Constant 0.477 0.168 0.726 0.281 0.464 0.804 0.522 0.893 0.059 0.028 -0.248 -0.174
Observations 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853 29,080 28,853
Table A2: Discontinuity analysis for socio-economic characteristics
5
Anticipation effects
As all unemployed in the sample are entitled to benefits in the first year of unemployment, differ-
ences in potential benefit duration do not directly affect the income in the first year, but may have
anticipation effects. To test whether differences in PBD affect the probability of reemployment
and precarity in the first year after unemployment, Figures A1 and A2 show placebo discontinuity
plots.
For employment (Figure A1), there is a significant discontinuity between 17 and 18 contribution
month, which is considerably smaller than the discontinuity in the second year. There are differ-
ent potential explanations for this discontinuity. First, self-selection into contribution month is
unlikely as discussed above (no heaping at 18 months, no large incentives). Second, we cannot
exclude other unintended selection effects, for example in the attribution of contribution month
by institution or employers, there are no indications for this though. Third, anticipation effects
play a role. We know from previous studies, that differences in PBD do have anticipation effects
(Le Barbanchon 2016). For precarity, there is no discontinuity in precarity risk between 17 and 18
contribution months (Figure A2).
Figure A1: Probability of employment in first year following unemployment (month 1-11) by con-
tribution months.
6
Figure A2: Probability of economic precarity in first year following unemployment (month 1-11)
by contribution months.
A1.2 Evolution of labor market status, welfare state benefits and precarity
Figure A3 shows the status of individuals during the first two years following unemployment, dis-
tinguishing between long and short PBD. Generally, the share of individuals on unemployment in-
surance drops considerably at the end of PBD, but the decrease is more pronounced for the group
with shorter PBD (12 months) than for the group with longer PBD (18 months). At the end of PBD,
there is an increase in the number of people transitioning onto social assistance and no registered
income. Unemployment insurance receipt does not end after the PBD, as individuals can extend
their eligibility by working intermittently during their unemployment spell. Interestingly, these over-
all patterns are very similar for men and women. The transition to no registered income and social
assistance at the end of PBD are stronger when PBD is shorter. Transitions into employment are
less influenced by PBD. Figure A4 shows precarity status by months after unemployment.
7
Figure A3: Labor market status and income sources by month after unemployment
Figure A4: Probability of economic precarity by month after unemployment (descriptive, not mod-
elled)
8
A1.3 Linear probability model and logistic regression
For binary outcomes, both non-linear models and linear probability models (LPM) are commonly
used in empirical research. Non-linear models, most importantly logit and probit, became increas-
ingly popular in parallel with increasing computational power since the 1980s and have become
the standard approach to analysing dichotomous variables. The main arguments against using
LPM are that predictions are not bounded between of 0 and 1, the latent variable is assumed to
be linear, and heteroscedasticity (Long 1997). More recently, linear probability models regained
popularity, mainly due to their interpretability, fast estimation, robustness of OLS models due to
weaker assumptions required, and consistent results compared to logistic regression (Breen,
Karlson, and Holm 2018; Mood 2010). For longitudinal, interactions or multilevel models, non-
linear regression requires important computational power, often has conversion problems and
probability estimation is complex (Hippel 2015). Difficulties with the LPM do not imply that non-
linear regression is the correct specification of the probability model (Angrist and Pischke 2008;
Horrace and Oaxaca 2006 Chapter 3). Because the functional form between the predictors and
outcome probabilities are unknown, the assumption of an S function of logistic regression is as
arbitrary as a linear function (Jaccard and Brinberg 2021).
Although there is no agreement on the appropriate models to use for binary outcomes, various
criteria have been made for the appropriateness of the LPM or logistic model. A rule of thumb
suggests that linear probability is problematic when probabilities are close to 0 or 1. If the mod-
elled probabilities are mostly between 0.2 and 0.8, the linear probability should suffice (e.g. Long
1997). However, Ganzach et al. (2000) show that this rule does not apply when interactions are
involved.
Several scholars suggest that the LPM is most problematic if the main purpose of a model is to
predict probabilities. When the main interest is in marginal effects, and especially if the variable
of interest is categorical, the LPM can be a good choice (Hippel 2015, 2017, Pischke 2012). Most
importantly, it is stressed that the appropriateness of the model needs to be explored. For exam-
ple, Battey et al. (2019) suggest to check the number of fitted values outside the unit range, ex-
plore these cases and check whether their omission affect the conclusions.
Addressing the problem of noncollapsibility of the logistic model 1, Karlson and Jann (Karlson and
Jann 2023) suggest using marginal odds ratios rather than average marginal effects when re-
searchers are interested in relative differences and using logistic models. In contrast to marginal
effects, odds ratios do not depend on the distribution of the dependent variables.
Considering that there are no agreed guidelines on the appropriate models to be used in scien-
tific literature, we estimated and compared both linear probability models and logistic regression
for our application. In addition, we estimated stratified models using each household type as a
subsample (Tables A3, A4, A5, A6).
1
Estimates of regression coefficients change when additional variables are included in the model, even if they are
uncorrelated with the predictor.
9
We opted to show and discuss the results of the LPM models in the main text for different rea-
sons. First, the comparison of different models (short and long term, women and men) is a cen-
tral part of our study. Secondly, the previous literature reported effects of unemployment gener-
osity in percentage points, which allows us to relate our results to previous studies rather than
showing odds ratios. Thirdly, the comparison of both LPM and logistic models shows that the esti-
mates by the LPM are closer to the estimates in the stratified models, in particular for employ-
ment probability. Fourth, logistic regressions are preferable if modelling a continuous probability
function is central (Hippel 2015). As we are interested in estimates for specific groups, this argu-
ment is not of relevance for our approach. However, for the interpretations and conclusions we
consider the results as definitive only then the models agree.
The results were consistent for employment probability in the short and long term (Table A3 and
A4). For financial precarity in the short term, the main conclusions were consistent, but large dif-
ferences in a few marginal effects appeared. We were cautious in interpreting marginal effects
that were not robust across model specification. We conducted further investigations, which we
outline below, to identify the most plausible model to be used as a base for the Figure illustrating
marginal effects (Figure 4 and Figure 5). For financial precarity in the long term, results were in-
consistent.
Employment, short term Stratified Interaction
LPM Logit AME Logit MOR LPM Logit AME
Men
Single 6.4*** 6.4*** 1.3*** 6.1*** 6.4***
Single parent 4.0 4.1 1.5*** 8.6*** 8.8***
Main earner, no kids 5.6 4.8 1.3** 6.7** 6.4**
Main earner, kids 7.3** 7.2** 1.2* 4.5* 3.9*
Egalitarian earner, no kids 8.9** 9.1** 1.3** 5.6** 4.7**
Egalitarian earner, kids 3.3 3.2 1.3** 5.4** 4.5**
Secondary earner, no kids 8.7 8.7 1.4* 8.4* 8.6*
Secondary earner, kids 12.3 12.3 1.7*** 12.4*** 12.6***
Women
Single 2.1 2.2 1.2** 4.1** 3.7***
Single parent 4.8* 4.9* 1.3*** 5.7*** 6.1***
Main earner, no kids 9.2 8.8 1.2 4.7 4.1
Main earner, kids 6.9 5.8 1.3 5.9* 6.6*
Egalitarian earner, no kids 5.4 5.0 1.2* 5.3** 4.0**
Egalitarian earner, kids 5.9 5.7 1.4*** 7.4*** 7.6***
Secondary earner, no kids 7.6 7.6 1.3** 6.3** 5.6**
Secondary earner, kids 8.9** 8.9** 1.3** 5.4** 5.8**
Note: Stratified models are separate regression for each household type. * significant at 95%, **
significant at 99%, *** significant at 99.9%
Table A3: Marginal effects of less generous unemployment insurance (shorter potential benefit
duration) on reemployment probability in the short-term. AME (average marginal effect) MOR
(marginal odds ratio).
10
Employment, long term Stratified Interaction
LPM Logit AME Logit MOR LPM Logit AME
Men
Single 3.1 3.1 1.1* 2.4 3.0*
Single parent 2.1 2.1 1.2* 4.2* 4.4*
Main earner, no kids 3.6 3.6 1.2 3.7 3.3
Main earner, kids 1.9 1.7 1.1 2.9 1.9
Egalitarian earner, no kids 2.1 2.0 1.0 1.2 0.3
Egalitarian earner, kids 3.7 3.1 1.1 2.8 1.4
Secondary earner, no kids -3.2 -3.2 1.1 1.0 1.3
Secondary earner, kids 13.0 13.1 1.7** 12.0* 12.1**
Women
Single 0.3 0.2 1.1 1.3 1.1
Single parent 1.4 1.3 1.1 2.2 2.6
Main earner, no kids 7.0 6.9 1.0 0.9 0.6
Main earner, kids -0.8 -0.8 1.0 -1.1 0
Egalitarian earner, no kids 6.7* 6.5* 1.2 4.3* 3.0*
Egalitarian earner, kids 2.4 2.3 1.3** 4.8* 4.9*
Secondary earner, no kids 4.5 4.4 1.3 4.8* 4.3*
Secondary earner, kids 4.8 4.8 1.2 3.0 3.3
Note: Stratified models are separate regression for each household type. * significant at 95%, **
significant at 99%, *** significant at 99.9%
Table A4: Marginal effects of less generous unemployment insurance (shorter potential benefit
duration) on reemployment probability in the long-term. AME (average marginal effect) MOR (mar-
ginal odds ratio).
For financial precarity in the short term (Table A5), the main findings on the heterogeneity be-
tween households types are consistent across both logistic and LPM, with stronger effects for
high financial responsibility (in particular singles) and weak effects for low financial responsibility
(in particular egalitarian earners without children). There are some large differences in predicted
marginal effects and marginal odds ratios, though. The LPM model yields stronger differences be-
tween household types, and suggests that shorter PBD may even reduce precarity risk for some
categories of individuals with weak financial responsibility. In the logistic model, results suggest
insignificant or precarity-increasing effects of shorter PBD. The stratified models (separate models
for each household type to avoid interactions) do not give an indication on whether results of the
LPM or of logistic regression are more plausible, as both models differ considerably from the
stratified estimates. In terms of consistent regarding significance tests, both the LPM and the lo-
gistic models compare similarly to the stratified models.
11
Precarity, short term Stratified Interaction
LPM Logit AME Logit MOR LPM Logit AME
Men
Single 8.3*** 8.2*** 1.5*** 9.7*** 8.1***
Single parent 5.6* 5.6* 1.1* 5.6*** 3.4*
Main earner, no kids 2.5 2.3 1.3** 3.5 4.5**
Main earner, kids 1.5 1.6 1.2* 5.3*** 3.5*
Egalitarian earner, no kids 1.7 2.0 1.2 -3.4** 1.7
Egalitarian earner, kids 4.3 5.4** 1.2 -0.5 2.8
Secondary earner, no kids 1.2 2.4 1.2 -2.3** 2.0
Secondary earner, kids 7.4 8.0 1.2 0.7 3.6
Women
Single 7.8*** 7.7*** 1.4*** 8.8*** 6.7***
Single parent 2.5 2.8 1.1* 3.2** 3.2*
Main earner, no kids 2.2 3.6 1.1 2.6 1.9
Main earner, kids 5.7* 5.7 1.2 8.0** 4.6
Egalitarian earner, no kids -2.3 1.9 1.2 -0.7 2.2
Egalitarian earner, kids 2.5 2.0 1.2 1.3 2.6*
Secondary earner, no kids -0.3 -0.1 1.1 -0.6 1.1
Secondary earner, kids 4.1* 3.2* 1.4** 1.5 3.2**
Note: Stratified models are separate regression for each household type. * significant at 95%, **
significant at 99%, *** significant at 99.9%
Table A5: Marginal effects of less generous unemployment insurance (shorter potential benefit
duration) on probability of financial precarity in the short-term. AME (average marginal effect)
MOR (marginal odds ratio).
For financial precarity in the long term (Table A6), the differences between the LPM and the lo-
gistic models are even stronger. Findings were robust to excluding the cases with negative proba-
bilities in the LPM model, and adding additional control variables, but differences arose due to
the comparison of either probabilities or odds between the difference of the groups (see Gazach
et al. 2000 for an illustration). The logistic model and the linear probability model show different
conclusions for financial precarity in the long-term. In the linear probability model, less generous
unemployment insurance seems to reduce precarity risk in the longer term for men in egalitarian
earner couples and secondary earner men without children. At the same time, less generous un-
employment insurance heightens precarity risk for singles (both men and women). In the logistic
regression, none of these effects is significant. The stratified regression suggests that both pre-
carity reducing and precarity increasing effects could arise for some groups, but these effects are
not statistically significant from zero and estimated marginal effects of both the logistic and LPM
interaction model differed considerably from the stratified model. Most importantly, both the
stratified model and the logistic interaction model show no effects (with one exception) that differ
significantly from zero. Therefore, we did not visualize the effect with a Figure and conclude that
there are no clear long-term effects.
12
Precarity, long term Stratified Interaction
LPM Logit AME Logit MOR LPM Logit AME
Men
Single 8.3 1.1 1.1 3.1* 1.8
Single parent 0.1 0.2 0.9 -1.1 -2.5
Breadwinner, no children 1.7 -0.3 0.9 -1.8 -1.0
Breadwinner with children -5.8 -5.6 0.9 -1.2 -2.1
Egalitarian earner, no children 1.1 1.6 1.0 -4.2** 0.6
Egalitarian earner with children 0.6 2.2 0.9 -3.6* -1.0
Secondary earner, no children -5.7 -5.9 0.9 -5.6* -1.8
Secondary earner with children -8.7 -9.2 0.8 -5.5 -4.4
Women
Single 1.7 1.9 1.1 4.2** 1.7
Single parent 0.3 0.2 1.0 0.0 -0.1
Breadwinner, no children -1.3 -1.3 0.9 -1.2 -2.0
Breadwinner with children 9.5 2.6 1.0 2.2 -0.7
Egalitarian earner, no children -3.2 -4.1* 0.9 -2.7 -0.6
Egalitarian earner with children -0.5 -1.9 0.9 -2.5 -1.4
Secondary earner, no children 2.4 3.6 1.2 -0.9 1.5
Secondary earner with children 2.5 2.0 1.2 0.4 2.2
Note: Stratified models are separate regression for each household type. * significant at 95%, **
significant at 99%, *** significant at 99.9%
Table A6: Marginal effects of less generous unemployment insurance (shorter potential benefit
duration) on probability of financial precarity in the short-term. AME (average marginal effect)
MOR (marginal odds ratio).
13
A2. Results on reemployment
A2.1 Regression tables for reemployment (pooled analysis)
(1) (2) (3) (4) (5) (6)
12-18 19-24 12-18 19-24 12-18 19-24
month months month months month months
Men Men Women Women Gender Gender
Short PBD (Ref: Long PBD) 0.061*** 0.024 0.041** 0.014 0.059*** 0.023*
Household type : (Ref : Single)
Single parent -0.001 0.031** -0.081*** -0.045*** -0.025*** 0.008
Main earner, no children 0.062*** 0.072*** 0.010 0.002 0.045*** 0.049***
Main earner, with children 0.092*** 0.111*** -0.142*** -0.098*** 0.029*** 0.056***
Equal earner, no children 0.109*** 0.115*** 0.054*** 0.053*** 0.096*** 0.103***
Equal earner, with children 0.107*** 0.131*** -0.072*** -0.041*** 0.028*** 0.062***
Secondary earner, no children -0.008 0.031 0.005 0.008 0.042*** 0.056***
Secondary earner, with children -0.033 -0.022 -0.093*** -0.052*** -0.037*** 0.004
Interaction Short PCD*Houdhold type -0.001 0.031** -0.081*** -0.045*** -0.025*** 0.008
Short PBD * Single parent 0.026 0.019 0.016 0.009
Short PBD * Main earner, no children 0.007 0.013 0.006 -0.005
Short PBD * Main earner, with children -0.016 0.005 0.018 -0.025
Short PBD * Equal earner, no children -0.005 -0.011 0.012 0.029
Short PBD * Equal earner, with children -0.007 0.004 0.033 0.034
Short PBD * Secondary earner, no children 0.023 -0.014 0.022 0.034
Short PBD * Secondary earner with children 0.063 0.097* 0.013 0.016
Insured income 0.010*** 0.013*** 0.010*** 0.012*** 0.012*** 0.014***
Partner income 0.003 0.003 0.001 -0.001 -0.001 -0.001
Contribution months 0.040** 0.059*** 0.013 0.028 0.027** 0.043***
contribution months squared -0.001* -0.001*** 0 0 -0.001** -0.001**
Women (Ref : Men) 0.026** 0.018
Interaction Short PBD*Women -0.001 0.008
Constant 0.038 -0.072 0.381** 0.310* 0.180* 0.092
Observations 187,246 137,116 187,322 134,500 374,568 271,616
R-squared 0.017 0.025 0.020 0.016 0.013 0.015
Remarks: control for clustered standard errors (multiple observations per person for different
time points). Swiss registry data from 2012-2016. Regression output for Figures 1 and 2. * signifi-
cant at 95%, ** significant at 99%, *** significant at 99.9%
Table A7: Pooled OLS regression for reemployment probability in the short-term and long-term by
gender
14
A2.2. Employment: Logistic regression
Figure A5: Marginal effects of less generous unemployment insurance on reemployment by household type (logistic
regression)
A2.3 Employment: by month after unemployment
Figure A6: Marginal effects of less generous unemployment insurance on reemployment by household type (by
month after unemployment start)
15
A2.4 Employment: including individuals with 12 contribution months
Figure A7: Marginal effects of less generous unemployment insurance on reemployment by household type (includ-
ing individuals with 12 contribution months)
A2.5 Employment: additional control variables
The following control variables were included in the model: insured income, income of the part-
ner before unemployment, marital status, age, education (tertiary vs others), occupation (mana-
gerial, middle, low), region (Southwest, Geneva region, Northwest, Central flatlands, Zurich, East,
Central mountains, and Ticino), citizenship (Swiss vs. other), work percentage in desired job, wait-
ing days (which generally indicate a voluntary quit compared to a layoff), and sanctions indicating
non-compliance with case workers’ job search requirements.
16
Figure A8: Marginal effects of less generous unemployment insurance on reemployment by household type (addi-
tional control variables)
A2.6 Employment: adding other households
Figure A9: Marginal effects of less generous unemployment insurance on reemployment (including other house-
holds)
17
A3. Results on financial precarity
A3.1 Regression tables for financial precarity
(1) (2) (3) (4) (5) (6)
12-18 19-24 12-18 19-24 12-18 19-24
month months month months month months
Men Men Women Women
Short PBD (Ref: Long PBD) 0.097*** 0.031* 0.088*** 0.041** 0.059*** 0.005
Household type: Ref(=single)
Single parent 0.237*** 0.177*** 0.453*** 0.402*** 0.343*** 0.289***
Main earner, no kids -0.048*** -0.102*** -0.049*** -0.076*** -0.073*** -0.113***
Main earner, kids 0.065*** -0.012 0.041* 0.012 0.048*** -0.015*
Egalitarian earner, no kids -0.125*** -0.197*** -0.176*** -0.214*** -0.205*** -0.248***
Egalitarian earner kids -0.080*** -0.144*** -0.185*** -0.220*** -0.183*** -0.225***
Secondary earner, no kids -0.160*** -0.198*** -0.250*** -0.288*** -0.278*** -0.300***
Secondary earner, kids -0.107*** -0.127*** -0.262*** -0.290*** -0.274*** -0.289***
Interaction Short PBD*Houdhold type 0.237*** 0.177*** 0.453*** 0.402*** 0.343*** 0.289***
Short PBD * Single parent -0.042** -0.043* -0.057*** -0.040*
Short PBD * Main earner, no children -0.062*** -0.049* -0.062* -0.052
Short PBD * Main earner, kids -0.044** -0.043* -0.008 -0.019
Short PBD * Egalitarian earner, no kids -0.132*** -0.073*** -0.096*** -0.068***
Short PBD * Egalitarian earner, kids -0.102*** -0.067*** -0.075*** -0.066***
Short PBD * Secondary earner, no kids -0.121*** -0.087*** -0.095*** -0.050**
Short PBD * Secondary earner, kids -0.090*** -0.086** -0.074*** -0.037*
Insured income -0.010*** -0.014*** 0.001 -0.001 -0.002*** -0.004***
Partner income -0.044*** -0.032*** -0.046*** -0.040*** -0.047*** -0.038***
Contribution months -0.038*** -0.049*** -0.015 -0.028* -0.029*** -0.041***
contribution months squared 0.001** 0.001** 0.000 0.001 0.001** 0.001***
Gender (Ref: Men) 0.007 -0.012*
Short PBD *Women -0.033*** -0.011
Constant 0.914*** 1.082*** 0.663*** 0.835*** 0.838*** 1.018***
Observations 187,246 137,116 187,322 134,500 374,568 271,616
R-squared 0.161 0.124 0.410 0.364 0.278 0.232
Robust standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05
Table A8: Pooled OLS regression for probability of financial precarity in the short-term and long-term by gender
18
A3.2 Precarity: Logistic regression
Figure A10: Marginal effects of less generous unemployment insurance on precarity by household type (logistic re-
gression)
A3.3 Precarity: by month after unemployment
Figure A11: Marginal effects financial precarity of short potential benefit duration by household type (by months
after unemployment)
19
A3.4 Precarity: including individuals with 12 contribution months
Figure A12: Marginal effects of less generous unemployment insurance on precarity by household type (including
12 contribution months)
A3.5 Precarity: additional control variables
Figure A13: Marginal effects of less generous unemployment insurance on precarity by household type (additional
control variables)
20
A3.6 Precarity: including child allowances
The registry data used for the analysis do not include child allowances which are paid as a lump
sum per child as part of the salary. The minimum amount is set by cantonal legislation and
ranges from 200 to 280 CHF per child per month. As the amount is not related to income, child
allowances have a redistributive effect on income inequality. Moreover, the differences in equival-
ized household income between household types decrease when family allowances are included.
To test whether the omission of child allowances in our main analysis affects the results, we simu-
lated child allowances and recalculated the models. We assumed 250 CHF per month for each
child, which reduces the share of households in financial precarity from 28.1% to 26.2 % (using
the same income threshold) for the period 12 to 17 months after unemployment.
The main effects of short PBD are consistent, but the main effects of household types differ.
Households with children (single parents, main earners, egalitarian couples) have a lower risk of
precarity if child allowances are considered. This is to some extent a mechanical result, as the in-
come of households with children is higher when child allowances are included. There are only
marginal changes for households where the secondary earner becomes unemployed and for
households without children.
There are no notable differences in the interaction effects in either the short or the long term.
The precarity risk of single households remains higher with less generous unemployment insur-
ance in line with the reference model.
Figure A14: Marginal effects of less generous unemployment insurance on precarity by household type (including
child allowances)
21
A3.7 Precarity: alternative threshold
Results for setting the precarity threshold considerably lower (levels of absolute poverty used for
official statistics) at 2240 CHF per month (equivalized income). With this lower threshold, 18 % of
observations in the analytical sample (compared to 30 %) fall below the precarity threshold. The
results are mainly comparable, with one exception: in the longer term, single mothers have an in-
creased precarity risk in the longer term when unemployment insurance is more restrictive.
Figure A15: Marginal effects financial precarity of short potential benefit duration by household type
22
A3.8 Precarity: adding other households
Figure A16: Marginal effects of less generous unemployment insurance on precarity by household type (including
other households)
A4. Job quality: employment income (only employed individuals)
Figure A17: Average monthly wages by number of contribution months.
23
Figure A18: Average monthly wages over time since start of unemployment spell.
Figure A19: Marginal effects of less generous unemployment insurance on earnings of employed persons by house-
hold type
24
References
Angrist, Joshua D., and Jörn-Steffen Pischke. 2008. Mostly Harmless Econometrics: An Empiricist’s
Companion (1st ed.). Princeton University Press.
Battey Heather, David Cox, and Micha V. Jackson. 2019. "On the linear in probability model for bi-
nary data." Royal Society Open Science 6: 190067.
Breen, Richard, Kristian B. Karlson, and Anders Holm. 2018. "Interpreting and Understanding
Logits, Probits, and Other Nonlinear Probability Models." Annual Review of Sociology 44(1), 39–54.
Ganzach, Yoav, Ishak Saporta, and Yaacov Weber. 2000. "Interaction in Linear versus Logistic
Models: A Substantive Illustration Using the Relationship between Motivation, Ability, and Perfor-
mance." Organizational Research Methods 3(3): 237–53.
Hippel, Paul von. 2015. Linear vs. Logistic Probability Models: Which is Better, and When? Statisti-
cal Horizons. https://statisticalhorizons.com/linear-vs-logistic/
Hippel, Paul von. 2017. When Can You Fit a Linear Probability Model? More Often Than You Think.
Statistical Horizons. https://statisticalhorizons.com/when-can-you-fit/
Horrace, William C., and Ronald L. Oaxaca. 2006. "Results on the Bias and Inconsistency of Ordi-
nary Least Squares for the Linear Probability Model." Economics Letters 90(3): 321–27.
Jaccard, James, and Miriam Brinberg. 2021. "Monte Carlo Simulations Using Extant Data to Mimic
Populations: Applications to the Modified Linear Probability Model and Logistic Regression." Psy-
chological Methods 26(4): 450–65.
Karlson, Kristian B., and & Ben Jann. 2023. "Marginal Odds Ratios: What They Are, How to Com-
pute Them, and Why Sociologists Might Want to Use Them." Sociological Science, 10, 332–347.
Le Barbanchon, Thomas. 2016. "The effect of the potential duration of unemployment benefits on
unemployment exits to work and match quality in France." Labour Economics, 42, 16–29.
Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables (1st ed.).
Sage Publications, Inc.
Mood, Carina. 2010. "Logistic Regression: Why We Cannot Do What We Think We Can Do, and
What We Can Do About It." European Sociological Review 26(1): 67–82.
Pischke, Jörn-Steffen. 2012. "Probit better than LPM?" Retrieved from http://www.mostlyharmles-
seconometrics.com/2012/07/probit-better-than-lp
25
File and source
- File
- unemployment-insurance-and-the-family-heterogeneous-effects-of-benefit-generosity-on-reemployment-and-economic-precarity.pdf
- Size
- 1,433,023 bytes
- SHA-256
- 253631941879b7e7eab3f4a5e1516b4af9ee8ba7fc6877d19f67b2a92ba8ef53
- Original
- doi.org