Comparative evidence on the receipt of minimum-income benefits discussion of empirical assessment of state dependence / "scarring effects" in benefit receipt
Call Girls Service AECS Layout Just Call 7001305949 Enjoy College Girls Service
The dynamics of social assistance benefit receipt
1. InGRID Summer school on ‘Advanced poverty research:
poverty and material deprivation dynamics’
Luxembourg, 8 July 2015
The dynamics of
social assistance benefit receipt
Sebastian Königs
OECD, Directorate for Employment, Labour and Social Affairs
(with Herwig Immervoll, OECD and Stephen Jenkins, LSE)
2. Across OECD countries,
recipient numbers of
unemployment benefits and
means-tested social assistance
have strongly risen during the
crisis, and are still above pre-
crisis level
an upward trend is often
observed also for disability
benefits
Note: The Czech Republic, Greece, Hungary, Iceland, Italy, Mexico and Turkey are omitted from the OECD Average due to incomplete
data. Source: SOCR Database, 2015.
The Great Recession has increased
demand for social protection
2
3. Depending on the country, …
• smaller number of recipients compared with ‘higher-tier’ benefits
• ‘difficult’ client populations: no consistent and intuitive policy story
• devolution to regional or local level, with heterogeneous policy
approaches and limited central policy authority difficult data
situation
The crisis is changing some of these factors:
• greater & increasing demand for benefit ‘floors’, notably in countries
where they are (largely) inexistent
• greater & increasing concern over work disincentives & ‘benefit
dependence’
• fiscal pressures to reduce or control spending and improve targeting
3
Prior to crisis, MIB were sometimes a
‘side issue’ in social policy debates
4. Improved policy-design and targeting require evidence on
the ‘risk factors’ of benefit receipt;
typical patterns of benefit receipt, including
the incidence of long-term receipt
the frequency of ‘churning’ (benefit “recidivism”)
the characteristics of long-term recipients
the magnitude (and sources) of ‘scarring effects’ from benefit receipt, and
the groups who are most affected
Series of studies on the dynamics of SA benefit receipt
country studies of benefit dynamics in the UK (Cappellari & Jenkins, 2008)
and Germany (Königs, 2013a), and on the incidence of ‘scarring effects’ /
‘state dependence’
cross-national study on benefit receipt dynamics in 8 OECD (accession)
countries: CAN, GER, NL, NOR, LUX, LVA, NOR, SWE, UK
Immervoll et al. (2014)
Recent OECD work on social assistance
benefit dependence
4
5. 1. Social Assistance benefits in OECD countries
2. Data, concepts and measurement
3. Individual-level patterns of benefit receipt
4. The empirical evidence for benefit dependence
Agenda
5
7. Minimum Income Benefits (MIB) / Social Assistance (SA) / ‘welfare’
benefits are broadly defined as
– public cash or in-kind transfers
– aimed at preventing extreme hardship (‘benefits of last resort’)
– based on a low-income criterion as the central entitlement condition
(means- or asset-tested and non-contributory)
This includes
• broad ‘non-categorical’ social assistance programmes
• unemployment assistance benefits that are not conditional on a
work or contribution history (e.g. in AUS, DEU, NZ)
• means-tested ‘last-resort benefits’ for single parents or families
Note: in some cases, these benefits can be received as a top-up to
earnings from low-paid work
Definition
7
8. 8
MIB: a broad range of benefits
Source: Immervoll et al., 2014
9. 9
Benefit levels and work disincentives can
be a concern…
Singles, no children, in % of median income (pre-crisis)
Source: Immervoll et al., 2014, calculations based on OECD tax-benefit models, www.oecd.org/els/social/workincentives
10. 10
Benefit levels and work disincentives can
be a concern…
Source: Immervoll et al., 2014, calculations based on OECD tax-benefit models, www.oecd.org/els/social/workincentives
Singles, no children, in % of median income (pre-crisis)
11. 11
But the “reach” of MIB can be very low…
Number of recipients, % of ‘working-age households’ (pre-crisis)
Source: Immervoll et al., 2014, calculations based on the OECD Social Benefit Recipient Database (SOCR), forthcoming
12. 12
… and result in low coverage of the poor
pseudo-coverage rates: % of income-poor ‘working-age households’ (pre-crisis)
Source: Immervoll et al., 2014, calculations based on the OECD Social Benefit Recipient Database (SOCR), forthcoming
14. An important challenge for studying SA benefit dynamics
are high data requirements:
Need for micro data with
– a panel structure (and ideally a long observation period);
– a sufficiently large sample size to identify benefit recipients;
– frequent and reliable information on the receipt of benefit payments;
– detailed information on household and individual characteristics.
Data requirements are a challenge for
studying SA benefit dynamics
14
15. Household panel surveys (BHPS, SOEP, SIPP)
follow individuals (recipients and non-recipients) over longer
periods of time
provide rich information on background characteristics for the
individual and other members of the household
can typically be easily accessed accessed for research purposes
Drawbacks:
✗ panel surveys with sufficient sample size exist in only few
countries
✗ information on benefit receipt (and individual characteristics) is
often available only at the annual level
✗ data reliability: reporting errors, attrition, ‘seam bias’
The two main data sources are household
panel surveys…
15
16. Panel data drawn from administrative records (tax
registers, databases from welfare offices)
much larger sample sizes (population?)
often (though not always) short observation intervals (monthly or
shorter)
typically high data quality with no unwanted attrition
Drawbacks:
✗ difficult to gain access
✗ non-recipients often not represented in the data
✗ often sparse information on individual and household
characteristics
… and administrative data
16
17. Social Assistance is typically a household-level benefit:
• the means test considers the financial situation of the household
• benefits are paid at the household level, and activity requirements
may extend beyond the claimant
• within a benefit-receiving household, the claimant may change
over time
the natural solution may seem to study the benefit dynamics of a
given household rather than those of the claimant alone
But: what should this look like in practice?:
How should the divorce of a couple, the death of a family member,
or the moving out of a dependent child be dealt with?
It may be exactly these events that trigger changes in the benefit
receipt status and that are worth studying…
It is not always easy to define the
appropriate unit of analysis (I)
17
18. The typical approach has therefore been to study the dynamics of
benefit receipt of individuals, but to define benefit receipt status at the
household level:
• an individual is categorised as a benefit recipient if any member of
the household reports benefit receipt during that period;
all members in a household have the same benefit receipt
status and are included in the analysis
[a few studies focus on the ‘household head’ or a single sampled
individual]
Limitations:
• the household as identified in the data (survey, admin) does not
always coincide with the benefit-receiving unit
• changes in household composition during the observation interval
(often a year) cannot be accounted for
It is not always easy to define the
appropriate unit of analysis (II)
18
19. Information on benefit receipt in panel data is often only
available at the annual level:
In household surveys,
• benefit receipt status typically assessed only at the time of the
interview;
• where monthly information is available, this typically comes from
retrospective questions
no corresponding information on individual / household
characteristics is available
‘seam bias’
Also in administrative data, information may only be available at the
annual level:
• tax records may provide information on the level of benefit
received over an entire year
Choice of the period of analysis is often
driven primarily by the type of data used (I)
19
20. Two standard approaches for defining a benefit variable:
• ‘point in time’ approach: did the individual / household report any
benefit receipt at the time of the interview?
• ‘benefit year’ approach: did the individual / household (as defined
at a given point in time) report any benefit receipt over the past
year?
The observed rate of benefit receipt is by construction higher for the
‘benefit year’ approach than for the ‘point in time’ approach.
[The difference between the two depends on the rate of turnover in
benefit receipt.]
This also has an impact on the econometric analysis of drivers of
benefit dynamics (Bhuller et al, 2014)
For a more detailed discussion, see Cappellari & Jenkins (2008) or Immervoll et al. (2014)
Choice of the period of analysis is often
driven primarily by the type of data used (II)
20
22. 22
There has been no uniform trend in
Social Assistance receipt rates
Source: Immervoll et al, 2014
Luxembourg
23. 23
Long-term SA benefit receipt is frequent
only in some countries
Country Period
Spell duration in months Share > 12
months
(in %)
Share > 24 months
(in %)Median Mean
Latvia 2006-2011 3 5 11 1
Luxembourg 1988-2010 15 31 59 38
Netherlands 1999-2010 9 19 42 25
Norway 1993-2008 2 4 6 2
Sweden 2001-2009 2 5 11 4
Source: Immervoll et al, 2014
Results are in line with those from the few earlier studies:
• Fitzgerald (1995) calculates median spell durations of 11-12 months for AFDC and
Food Stamps in the U.S.
• For a selection of European cities, Gustafsson et al. (2002) calculate median SA
spell durations of between 4 months (Gothenburg, Helsingborg) and up to
3 years (Lisbon)
24. The sampling procedure can have a huge impact on calculated
spell durations (see Bane & Ellwood, 1994):
Benefit durations are much longer for on-going spells than for
starting spells.
Source: Immervoll et al, 2014
24
Spell durations of current recipients are
often reported, but entirely misleading
Country
Median spell duration in months
On-going spells Starting spells
Latvia 6 5
Luxembourg 102 17
Netherlands 85 12
Norway 8 2
Sweden 15 2
25. 25
Repeat receipt (‘recidivism’) is more
likely where spell durations are short
Country
Number of benefit spells per individual
Jan 2001 – Dec 2008
Spell duration in months Share of individuals with
Median Mean 2 spells 5 spells
Luxembourg 1 1.4 29 0.5
Netherlands 1 1.4 26 0.5
Norway 3 4.2 71 34
Sweden 2 3.6 67 27
Source: Immervoll et al, 2014
The resulting total benefit duration is on average about twice as high in
LUX and NL than in NOR or SWE;
At very similar receipt rates, this implies that a much higher share of the
population in the Nordic countries receives benefits at any time:
• In NOR: 2.1% of all working-age adults receive benefits in any given
month, but 10.2% receive benefits at some point over the 8-year period
from 2001-2008
26. Potential drivers of those differences:
• SA generosity?: benefit levels tend to be relatively higher in NL and
LUX than in LVA, NOR and SWE (when measured against the
median hh income)
• design of the benefit system?: e.g., separate SA benefit for single
parents in NOR; comprehensive disability benefit programmes in
NL, NOR, SWE
transitions between different benefit programmes cannot be
accounted for
composition of recipient population might differ
differences in activation and support provided?
differences in labour market dynamics?
The drivers of those differences in benefit
receipt patterns are difficult to identify
26
28. Earliest studies from the U.S. (Blank, 1989; Bane & Ellwood,
1994):
• What’s the incidence of long-term benefit dependence?
• To what extend are relatively long AFDC spell durations indicative
of ‘scarring effects’ in welfare benefit receipt?
Similar rationale as for scarring in unemployment:
SA benefit receipt may
• be stigmatizing (Moffitt, 1983; Blank, 1989) / be perceived as a
signal for low labour productivity by employers
• have a ‘behavioural impact’ on the recipient, e.g. by inducing a loss
of motivation or feeling of helplessness (Bane & Ellwood, 1994)
• come with information costs that are reduced for those with a
previous history of benefit receipt
Since the late 1980s, there has been a
growing interest in studying ‘welfare traps’
28
29. Main difficulty for the econometric analysis of ‘scarring’ in SA benefit
receipt: lack of precise data on benefit spell lengths
Information often available only on benefit receipt at single points in
time (an annual interview) or over a longer time period (annual admin
data). Such data do not allow for estimation of event-history models.
A few exceptions: Studies on SA benefit duration dependence
• in the U.S.: Blank (1989); Sandefur & Cook (1998)
• Norway: Dahl & Lorentzen (2003)
• Sweden: Bäckman & Bergmark (2011); Mood (2013)
The less demanding (and therefore more standard) approach has
therefore been to model year-to-year transitions into and out of
benefits using dynamic discrete choice models.
In such models, ‘scarring effects’ are measured in terms of the
degree of state dependence in benefit receipt
Econometric analysis has mostly focused
on period-to-period SA benefit transitions
29
30. For instance, in Britain (1992-2008), the rate of SA benefit receipt was
• 71% among individuals who received benefits the previous year
(‘persistence rate’)
• 2.4% among those who did not receive benefits the previous year
(‘entry rate’)
A person who receives SA benefits today is much more likely to
receive benefits again next period compared to an individual who
does not receive benefits today.
Part of this differential is clearly due to differences in observables:
• individual characteristics: age, sex, level of education, work
experience, health status
• household characteristics: family status, spouse characteristics,
number of small children in the household, etc…
Substantial state dependence can be
observed in raw panel data
30
31. Heckman (1981a) distinguishes between two different sources of
state dependence conditional on observable characteristics:
• ‘spurious’: differences in benefit persistence and entry rates due to
persistent differences in unobservable characteristics
• ‘structural’: the causal impact of past benefit receipt on current
benefit receipt ( ‘welfare trap’)
This distinction has important implications for policy making:
• ‘spurious’ state dependence implies that recipients and non-
recipients differ in personal characteristics (e.g. ability)
increase labour market attachment of recipients (e.g. through
training)
• ‘structural’ benefit receipt itself has pervasive effects
keep people off benefits / change structure of the system
… and the question of interest is to what
extent these differences are ‘structural’
31
32. Dynamic random-effects probit model
where .
Additional assumptions:
DREP models have been the standard
choice for studying state dependence (I)
32
,...,1;,...,1for0'1
01
)1()1(
*
iittiti
itit
TtNiuxy
yy
itiitu
n]correlatioserial[no)1,0(~
0
model]in theterm[constant0
N
E
E
it
iti
i
33. In this model, the probability of benefit receipt is given by
αi captures persistent unobserved heterogeneity
λ is interpreted as measuring structural state dependence
Implicit assumptions
• first-order Markov process
• strict exogeneity of xi (i.e. no effect of past covariates on yit or
feedback effects from yit to future covariates)
DREP models have been the standard
choice for studying state dependence (II)
33
ititiiitiiit xyxyyyP )1()1(),1(0 ',,...,1
34. Initial conditions problem
– as in linear models, the individual-specific effect αi induces a
correlation between the error term and the lagged dependence
variable
ML-estimation inconsistent even for large N as T remains small
– for linear models: first-differencing and use of an IV approach
(Arellano and Bond, 1991)
– but: no comparable transformation available for non-linear models
Challenge: This model suffers from an
‘initial conditions problem’
34
0'1 )1()1( itititiit xyy
35. Wooldridge (2005) specifies a distribution for αi and integrates
out the unobserved heterogeneity:
A convenient choice for is
where
• yi0 is the observed outcome in the initial period, and
• is a vector of individual averages of all time-varying covariates.
This can be interpreted as a ‘correlated random effects’
specification à la Mundlak (1978) / Chamberlain (1984).
The simplest solution to the initial conditions
problem is due to Wooldridge (I)
35
iiiiiiiiTitiiiTit dxygxyyyfxyyyf ;,),;,,,...,();,,...,( 000
;,0 iii xyg
),'(~, 2
20100 aiiiii xyNxy
ix
36. The resulting joint density unconditional on the individual-specific effect
αi can be written as
This is equivalent to the standard random effects probit model with the
additional regressors yi0 and .
maximum likelihood estimation using standard software packages.
The simplest solution to the initial conditions
problem is due to Wooldridge (II)
36
da
a
xyxy
xyxy
aa
y
iititi
T
t
y
iititi
it
it
1
''(1
''(
)1(
201)1()1(
1
201)1()1(
ix
37. An earlier, widely-used approach to the initial conditions problem
was proposed by Heckman (1981b). It relies on approximating
the unknown density , to remove the conditioning of
the joint density on the value in the initial
period
Both approaches – by Wooldridge and Heckman – have been
shown to perform equally well for panels of moderate length
(Arulampalam & Stewart, 2009; Akay, 2012). Cappellari &
Jenkins compare both approaches in their study of SA benefit
dynamics in Britain and find that they give nearly identical
results. The Heckman approach is however computationally
more demanding.
An alternative solution by Heckman typically
gives very similar results
37
iii xy ,0
),;,,...,( 0 iiiiTit xyyyf
.0iy
38. The degree of estimated structural state dependence can be
summarized by the average partial effect of λ (the coefficient of
the lagged dependent variable)
.
Intuitively, the APE gives the absolute difference in predicted
(counterfactual) persistence and entry probabilities averaged
across all individuals (recipients and non-recipients alike) and
time periods.
The level of ‘structural’ state dependence is
given by the average partial effect of λ
38
iititiit
T
t
N
i iititiit
i
xyxyyP
xyxyyP
NT
APE i
,,,01ˆ
,,,11ˆ1
0)1()1(
1 1 0)1()1(
39. Now, what do these models tell us…?
39
Study Country (Data) APE (ppts) Comments
Cappellari & Jenkins
(2008, 2009, 2014)
Britain (BHPS) 14.4
larger effect for single
parents
Königs (2013a, 2014) Germany (SOEP) 14.1
larger APEs in Eastern
Germany and for
migrants
Hansen & Lofstrom
(2011)
Sweden (LINDA) 4.6 - 29.8
APEs larger for men, and
much larger for migrants
Bhuller & Königs (2011) Norway (FD-Trygd) 9.8
APE rises with a ‘broader’
definition of SA benefits
Königs (2013b) Netherlands (IPO) 28.3
larger APEs for men and
migrants
Hansen et al. (2014) Canada (SLID) 35.4
huge variation in APEs
across provinces
(22.0 ppts – 47.4 ppts)
Source: Immervoll et al, 2014
40. Estimated ‘structural’ state dependence is much smaller than observed
‘raw’ state dependence (i.e. the gap between observed benefit
persistence and entry rates).
observed and unobserved heterogeneity matter
All studies find substantial positive APEs indicating that there is
‘structural’ state dependence. The magnitude of this effect tends to be
higher (at least in absolute terms) for more disadvantaged groups:
– immigrants, and in particular refugees;
– single parents;
– regions with higher unemployment (e.g. Eastern Germany).
Methodological differences matter:
– APEs are lower for the ‘point-in-time’ than for the ‘benefit year’ approach
– a ‘broader’ definition of SA leads to higher state dependence
What are the likely causes of these
differences in estimated state dependence?
40
41. Unfortunately, these models are not very much suited to shed light on
how exactly past SA benefit receipt affects SA benefit receipt in the
future:
1. Markovian state dependence (as permitted by the annual model) may
describe different types of time dependence in the underlying process (see
Heckman & Borjas, 1980):
• duration dependence?
• occurrence dependence?
• lagged duration dependence?
This distinction has important implications for policy making.
1. State dependence in SA benefit receipt is difficult to disentangle from state
dependence in (un)employment, low pay, or poverty (Contini & Negri, 2007).
the measured ‘scarring effects’ need not be due to benefit receipt as
such
The channels through which past receipt
affects current receipt are largely uncertain
41
42. Extending the state space beyond the binary case can provide
insights on some of the possible sources of state dependence:
Wunder & Riphahn (2013) / Riphahn & Wunder (2014) estimate a dynamic
multinomial logit model on annual SOEP data to identify state dependence
across three states: SA benefit receipt, employment and ‘inactivity’ (which
includes unemployment).
They find structural state dependence in all three modelled states, but…
– the size of the APE is lower when welfare persistence rates are compared
against entry rates from ‘inactivity’ rather than from employment;
– the persistence rates in SA benefit receipt is not statistically different from
the entry rate from ‘inactivity’ to SA receipt;
– average predicted exit rates to employment are higher from SA than from
‘inactivity’ “welfare benefit recipients have stronger work incentives”
“[no] convincing evidence for the welfare trap hypothesis”
More complex specifications can help shed
light on some of these issues (I)
42
43. Bhuller et al. (2014) extend the baseline DREP model to allow for
• differences in the impact of observable and unobservable characteristics on
persistence vs. entry probabilities
• controls for the duration of the current spell and the number of previous spells
using monthly administrative panel data from Norway.
For a cohort of 18 year-olds, they find
1. strong evidence of duration dependence both on and off benefits:
• having been on benefits for 12 months raises the predicted persistence
probability by 25 ppts compared to the first month on benefits
• having been off benefits for 12 months reduces the re-entry probability by
6 ppts
2. evidence of occurrence dependence on benefit entries (but not persistence):
• having had a previous spell raises the re-entry rate by 0.5 ppts
More complex specifications can help shed
light on some of these issues (II)
43
44. • DREP models point to substantial ‘structural’ state dependence in SA
benefit receipt dynamics, even though the magnitude of this effect is
much lower than the observed ‘raw’ state dependence
evidence of ‘scarring effects’ / a ‘welfare trap’
• Unfortunately, these models are not very helpful when it comes to
evaluating possible drivers of state dependence in SA: behavioural
effects? stigma? state dependence in unemployment / poverty?
• Wunder & Riphahn show that dynamics of SA benefit recipients are
similar to those of ‘inactives’
no direct effect of benefit dependence (but possibly scarring in
‘inactivity’?
• Bhuller et al (2014) provide evidence of duration dependence
( behavioural effect…?) and occurrence dependence ( information
costs…? stigma…?)
Conclusions
44
45. 45
Contact: Sebastian.Koenigs@oecd.org
New Working Paper: NEET Youth in the Aftermath of the Crisis
OECD Directorate for Employment, Labour and Social Affairs: www.oecd.org/els
In It Together: Why less Inequality benefits All: http://www.oecd.org/social/inequality-and-poverty.htm
Society at a Glance 2014: www.oecd.org/social/societyataglance.htm
Pensions at a Glance 2013: www.oecd.org/pensions/pensionsataglance.htm
Thank you!
@OECD_Social
46. Akay, A. (2012), "Finite-sample comparison of alternative methods for estimating dynamic panel data
models", Journal of Applied Econometrics, 17, pp. 1189-1204.
Arulampalam, W. and M.B. Stewart (2009), Simplified implementation of the Heckman estimator of the
dynamic probit model and a comparison with alternative estimators, Oxford Bulletin of Economics
and Statistics, 71(5), pp. 659–681.
Bäckman, O. & Bergmark, ̊A. (2011). Escaping Welfare? Social Assistance Dynamics in Sweden.
Journal of European Social Policy, 21 (5), 486–500.
Bane, M. J. & Ellwood, D. T. (1994). Understanding Welfare Dynamics. In M. J. Bane & D. T. Ellwood
(Eds.), Welfare Realities: From Rhetoric to Reform (pp. 28–66). Cambridge, MA: Harvard University
Press.
Bhuller, M., Brinch, C. N., & Königs, S. (2014). Time Aggregation and State Dependence in Welfare
Receipt. Statistics Norway Research Department Discussion Paper, 771.
Bhuller, M. and S. Königs (2011), The dynamics of social assistance receipt in Norway, unpublished
report, Statistics Norway.
Blank, R. M. (1989). Analyzing the Length of Welfare Spells. Journal of Public Economics, 39(3), 245–
273.
Cappellari, L. and S. P. Jenkins (2008), "The Dynamics of Social Assistance Receipt: Measurement and
Modelling Issues, with an Application to Britain", OECD Social, Employment and Migration Working
Papers, No. 67, OECD Publishing, Paris, http://dx.doi.org/ 10.1787/236346714741
References
46
47. Cappellari, L. & Jenkins, S. P. (2009). The Dynamics of Social Assistance Benefit Receipt in Britain. IZA
Discussion Papers, 4457.
Cappellari, L. & Jenkins, S. P. (2014). The Dynamics of Social Assistance Benefit Receipt in Britain. In
S. Carcillo, H. Immervoll, S. P. Jenkins, S. Königs, & K. Tatsiramos (Eds.), Research in Labor
Economics: ”Safety Nets and Benefit Dependence”, Vol. 39 (forthcoming). Emerald Group
Publishing Limited.
Chamberlain, G. (1984). Panel Data. In Z. Griliches & M. D. Intriligator (Eds.), Handbook of
Econometrics, Vol. II (pp. 1247–1318). North-Holland: Elsevier Science Publishers.
Contini, D. & Negri, N. (2007). Would Declining Exit Rates from Welfare Provide Evidence of Welfare
Dependence in Homogeneous Environments? European Sociological Review, 23(1), 21–33.
Dahl, E. & Lorentzen, T. (2003b). Explaining Exit to Work among Social Assistance Recipients in
Norway: Heterogeneity or Dependency? European Sociological Review, 19(5), 519–536.
Fitzgerald, J. M. (1995). Local Labor Markets and Local Area Effects on Welfare Duration. Journal of
Policy Analysis and Management, 14 (1), 43–67.
Gustafsson, B., Mu ̈ller, R., Negri, N., & Voges, W. (2002). Paths through (and out of) Social Assistance.
In C. Saraceno (Ed.), Social Assistance Dynamics in Europe: National and Local Poverty Regimes
(pp. 173–234). Bristol: The Policy Press.
Hansen, J. and M. Lofstrom (2011), "Immigrant-native differences in welfare participation: the role of
entry and exit rates", Industrial Relations, 50(3), pp. 412-442.
References
47
48. Hansen, J., Lofstrom, M., Liu, X., & Zhang, X. (2014). State Dependence in Social Assistance Receipt in
Canada. In S. Carcillo, H. Immervoll, S. P. Jenkins, S. Königs, & K. Tatsiramos (Eds.), Research in
Labor Economics: ”Safety Nets and Benefit Dependence”, Vol. 39 (forthcoming). Emerald Group
Publishing Limited.
Heckman, J. J. (1981a). Heterogeneity and state dependence. In S. Rosen (Ed.), Studies in Labor
Markets (pp. 91–140). University of Chicago Press: Studies in Labor Markets.
Heckman, J. J. (1981b). The Incidental Parameters Problem and the Problem of Initial Conditions in
Estimating a Discrete Time-Discrete Data Stochastic Process. In C. F. Manski & D. McFadden
(Eds.), Structural Analysis of Discrete Data with Econometric Applications (pp. 179–195).
Cambridge: The MIT Press.
Heckman, J. J. & Borjas, G. (1980). Does Unemployment Cause Future Unemployment? Definitions,
Questions and Answers from a Continuous Time Model of Heterogeneity and State Dependence.
Economica, 47(187), 247–283.
Immervoll, H., S. P. Jenkins and S. Königs (2014), “Are Recipients of Social Assistance 'Benefit
Dependent'?: Concepts, Measurement and Results for Selected Countries”, OECD Social,
Employment and Migration Working Papers, No. 162, OECD Publishing.
http://dx.doi.org/10.1787/5jxrcmgpc6mn-en
Königs, S. (2013a), "The Dynamics of Social Assistance Benefit Receipt in Germany", OECD Social,
Employment and Migration Working Papers, No. 136. OECD Publishing, Paris, http://dx.doi.org/
10.1787/5k3xwtg6zknq-en
References
48
49. Königs, S. (2013b), "The Dynamics of Social Assistance Benefit Receipt in the Netherlands",
unpublished report.
Königs, S. (2014). State Dependence in Social Assistance Benefit Receipt in Germany Before and After
the Hartz Reforms. In S. Carcillo, H. Immervoll, S. P. Jenkins, S. Königs, & K. Tatsiramos (Eds.),
Research in Labor Economics: ”Safety Nets and Benefit Dependence”, Vol. 39 (forthcoming).
Emerald Group Publishing Limited.
Moffitt, R. (1983). An Economic Model of Welfare Stigma. The American Economic Review, 73(5),
1023–1035.
Mood, C. (2013). Social Assistance Dynamics in Sweden: Duration Dependence and Heterogeneity.
Social Science Research, 42(1), 120–139.
Mundlak, Y. (1978). On the Pooling of Time Series and Cross Section Data. Econometrica, 46(1), 69–85.
Riphahn, R. T. & Wunder, C. (2013). State Dependence in Welfare Receipt: Transitions Before and After
a Reform. CESifo Working Paper, 4485.
Sandefur, G. D. & Cook, S. T. (1998). Permanent Exits from Public Assistance: The Impact of Duration,
Family, and Work. Social Forces, 77(2), 763–787.
Wooldridge, J. M. (2005). Simple Solutions to the Initial Conditions Problem in Dynamic, Nonlinear
Panel Data Models with Unobserved Heterogeneity. Journal of Applied Econometrics, 20(1), 39–54.
Wunder, C. & Riphahn, R. T. (2014). The Dynamics of Welfare Entry and Exit amongst Natives and
Immigrants. Oxford Economic Papers, 66 (2), 580–604.
References
49