3. Aim of the project
• The aim of this master thesis is to estimate the relationship between
informal caregiving and the labour-market.
• The overall expectation was to identify a negative relationship, that can
explain why people giving care have a lower labour-market participation
• However, when accounting for an intensive caregiving margin in the
model, the effect is expected to be less intensive and maybe non-existent.
• Why is this thesis relevant?
08-12-2016 3
4. Informal caregiving can play an
important role in the realisation of an
effect home care and welfare state!
However, meeting the care needs of
relatives can have an impact on the
labour-market!
“
5. What is the
Causal Effect
of Informal Caregiving
on Labour-Market Outcomes
Q1
10. The parental welfare function in the decision making
• The parent welfare function
𝛽𝑣 𝑍𝑖, 𝐼𝐶𝑖
• It is expected that there will be an
increase in the likelihood of
providing informal care, when only
having one parent left
• When a parent pass away, an
decrease in the likelihood of being
employed is expected
08-12-2016 10
11. Theoretical effects of informal caregiving
• Informal caregivers will decide to reduce work only when the substitution
effect outweights the income effect.
• An intensive threshold for informal caregiving.
• If the respite effect is dominating a positive relationship between informal
caregiving and working hours should be found.
08-12-2016 11
14. The Empirical Model
• Labour-market model
𝐿 𝑖𝑗𝑡 = 𝑓𝑗 𝐼𝐶𝑖𝑡, 𝐻𝑖𝑡, 𝑋𝑖𝑡, 𝑅𝑖, 𝜖 𝑖𝑗𝑡 , 𝑗 = 1,2 𝑡 = 1,2 (4.1)
where
𝐼𝐶𝑖𝑡 is informal caregiving
𝐻𝑖𝑡 is the individuals health-status
𝑋𝑖𝑡 is social-economic control variables
𝑅𝑖 is the institutional framework in the country
08-12-2016 14
16. Correlation between the endogenous
variable and the instrument (Relevance)
The instrument variable must be
exogenous (Exogeneity)
The unbiased estimator
Binary Choice model
Instrument Variable model
Fixed Effect model
Average Treatment Effect model
Empirical strategy
• 𝐶𝑜𝑣 𝑧𝑖𝑡, 𝜖𝑖𝑡 = 0
• 𝐶𝑜𝑣 𝑧𝑖𝑡, 𝐼𝐶𝑖𝑡 ≠ 0
• 𝛽𝐼𝑉 =
𝐶𝑜𝑣(𝑌,𝑍)
𝐶𝑜𝑣(𝑋,𝑍)
17. Strictly exogeneity in the fixed-effect
model
Allows for correlation between
unobserved variables and the error-term
The unbiased FE-estimator
Binary Choice model
Instrument Variable model
Fixed Effect model
Average Treatment Effect model
Empirical strategy
• 𝐸 𝑋𝑖𝑡 𝛼𝑖 ≠ 0
• 𝐸 𝑢𝑖𝑡 𝑋𝑖𝑡, 𝛼𝑖 = 0
• 𝛽 𝐹𝐸 = 𝑖=1
𝑁
𝑿𝑖
′
𝑿𝑖
−1
𝑖=1
𝑁
𝑿𝑖
′
𝑳𝑖
21. Table 6-2: Regression results of Informal Care & Intensive Care (>15) on labour-market participation
08-12-2016 21
Coeff S.E. Marginal S.E. Coeff S.E. Coeff S.E. Coeff S.E. Coeff S.E. Coeff S.E. Coeff S.E. Coeff S.E.
Hours of care - - - - - - - - - - - - - - - - - -
Informal care 0,036 *** 0,010 0,017 * 0,009 -0,007 0,011 - - - - 0,060 0,084 - - - - -0,007 0,155
Intensive care -0,175 *** 0,031 -0,085 *** 0,025 -0,016 0,028 - - - - -0,705 ** 0,354 - - - - 0,087 0,624
Age 0,090 ** 0,045 0,095 *** 0,037 0,124 *** 0,042 0,016 0,053 0,007 0,022 0,093 ** 0,046 -0,026 0,109 0,000 0,050 0,119 *** 0,043
Age Squared -0,001 ** 0,000 -0,001 *** 0,000 -0,001 *** 0,000 0,000 0,000 0,000 0,000 -0,001 ** 0,000 0,000 0,001 0,000 0,000 -0,001 ** 0,000
Number of children -0,012 *** 0,004 -0,011 ** 0,005 -0,012 0,012 -0,007 0,004 -0,004 ** 0,002 -0,013 *** 0,004 -0,008 0,023 0,002 0,011 -0,014 0,012
Number of chronical conditions -0,021 *** 0,006 -0,018 *** 0,006 -0,003 0,010 0,012 * 0,007 0,002 0,003 -0,021 *** 0,006 0,019 0,020 0,004 0,009 -0,003 0,010
Married -0,118 *** 0,012 -0,090 *** 0,014 -0,022 0,033 0,013 0,014 0,006 0,006 -0,114 *** 0,013 0,048 0,119 -0,027 0,039 -0,021 0,040
Self-rated health 0,169 *** 0,016 0,131 *** 0,014 0,064 *** 0,017 0,049 *** 0,016 -0,005 0,008 0,162 *** 0,017 0,011 0,038 -0,009 0,018 0,067 *** 0,019
Household's income percentile 0,040 *** 0,002 0,028 *** 0,002 0,011 *** 0,002 0,007 *** 0,002 0,000 0,001 0,040 *** 0,002 -0,001 0,005 -0,001 0,002 0,012 *** 0,002
EURO-D -0,015 *** 0,003 -0,013 *** 0,003 -0,009 *** 0,003 -0,001 0,003 0,002 * 0,001 -0,013 *** 0,003 0,003 0,007 0,001 0,003 -0,009 *** 0,003
Wave2 dummy 0,033 *** 0,010 0,029 *** 0,007 -0,049 ** 0,023 -0,039 *** 0,013 -0,014 ** 0,005 0,033 *** 0,011 -0,015 0,056 -0,016 0,025 -0,049 ** 0,023
Years of education 0,013 *** 0,001 0,017 *** 0,002 - - 0,008 *** 0,001 -0,001 * 0,001 0,012 *** 0,002 - - - - - -
Female -0,208 *** 0,010 -0,214 *** 0,012 - - 0,090 *** 0,011 0,034 *** 0,004 -0,193 *** 0,014 - - - - - -
Region B 0,083 *** 0,014 0,080 *** 0,019 - - 0,076 *** 0,015 -0,023 *** 0,008 0,070 *** 0,020 - - - - - -
Region C -0,007 0,014 -0,011 0,018 - - 0,169 *** 0,015 -0,015 ** 0,007 -0,017 0,024 - - - - - -
Region D 0,150 *** 0,015 0,152 *** 0,019 - - 0,256 *** 0,018 -0,031 *** 0,007 0,129 *** 0,032 - - - - - -
Constant -1,726 1,248 - - -3,126 ** 1,255 -0,525 1,467 -0,221 0,621 -1,805 1,287 1,115 3,200 -0,094 1,443 -2,995 ** 1,271
Parent 1 - - - - - - 0,001 0,014 -0,016 ** 0,006 - - 0,102 *** 0,034 0,009 0,015 - -
Mother Age - - - - - - 0,002 *** 0,000 0,000 ** 0,000 - - - - - - - -
Mother Health - - - - - - 0,078 *** 0,014 0,038 *** 0,006 - - 0,057 ** 0,029 0,027 ** 0,012 - -
Mother Distance - - - - - - -0,162 *** 0,015 -0,021 *** 0,005 - - - - - - - -
Father Age - - - - - - 0,063 *** 0,015 0,011 * 0,007 - - - - - - - -
Father Health - - - - - - - - - - - - - - - - - -
Father Distance - - - - - - - - - - - - - - - - - -
Observations 6479 6479 6479 6479 6479 6479 6479 6479 6479
Individuals 3423 3423 3423 3423 3423 3423 3423 3423 3423
F-Test first stage 37,33 *** 6,6 *** 1,6 * 0,69
Overidentification test 0,92 (p=0.82)
Durbin-Wu-Hausman exogeneity test 1,62 (p=0.20)
Hausman (Test for random effects) 225,6 ***(Reject)
Wald Chi2 3204 *** 1073 *** 258 *** 2794 *** 251 ***
R2 29% 6% 9% 3% 25% 0% 0% 5%
Correlation (u,X) 0,118 -0,089 -0,104 0,077
*** Significant at 1 %; ** Significant at 5%; *Signifiant at 10 %
a = Informal care, 1 for given care or else then 0, b=Intensive care if more than 15 hours of care
Note that the marginal effect calculation is based on a unmarried man from region A who are not given care in the sample (Female=0, Married=0, self-health=0, RegionB=0, RegionC=0,RegionD=0,Informal Care=0, Intensive Care =0, rest of the variables are at means)
First-stage (a) Second-stage First-stage (a)
Intensive Caregiving
LPM (R6) Pooled Probit (R7) LPM FE (R8) LPM IV
Second-stage
LPM IV
First-stage (b)
LPM FEIV
First-stage (b)
LPM IV (R9) LPM FEIV LPM FEIV (R10)
22. Table 6-6: Regression results of defined Intensive caregiving margin for full sample and between genders
08-12-2016 22
Marginal S.E. Marginal S.E. Coeff S.E. Coeff S.E.
Full Sample
Informal Care 0,019 ** 0,010 0,016 * 0,009 0,017 * 0,009 0,017 * 0,009
Intensive Care -0,037 ** 0,016 -0,050 ** 0,020 -0,085 *** 0,025 -0,086 *** 0,026
Observation 6479 6479 6479 6479
Individuals 3423 3423 3423 3423
Wald 1829,70 *** 1828,76 *** 1844,43 *** 1841,41 ***
Men
Informal Care 0,022 * 0,012 0,021 * 0,011 0,019 * 0,011 0,019 * 0,011
Intensive Care -0,031 0,026 -0,053 0,042 -0,059 0,062 -0,076 0,065
Observation 2846 2846 2846 2846
Individuals 1502 1502 1502 1502
Wald 356,64 *** 395,00 *** 354,45 *** 354,70 ***
Female
Informal Care 0,016 0,014 0,015 0,014 0,019 0,014 0,018 0,014
Intensive Care -0,025 0,019 -0,038 0,023 -0,078 *** 0,027 -0,074 *** 0,028
Observation 3622 3622 3622 3622
Individuals 1921 1921 1921 1921
Wald 1211,67 *** 1203,90 *** 1231,27 *** 1221,15 ***
*** Significant at 1 %; ** Significant at 5%; *Signifiant at 10 %
Note that the marginal effect calculation is based on a man from region A who are not given care in the sample and for the women regression an
unmarried woman (a man in the mens sample) in region A (Informal care = 0, intensive care =0, Married=0, self-Health =0, Wave=0, RegionB=0,
RegionC=0,RegionD=0)
5 Hours 10 Hours 15 Hours 20 Hours
Intensive care
27. 08-12-2016 27
Table 6-10: Results of the treatment effect of a parent passing
away on the labour-market participation
Robust Robust
Agecohort Coef. Std. Err. Coef. Std. Err.
51-58 ATET -0,019 0,019 -0,009 0,030
Mean 0,875 *** 0,010 0,153 *** 0,017
Observation (N): 1870 739
59-64 ATET -0,075 * 0,043 0,040 ** 0,022
Mean 0,668 *** 0,028 0,047 *** 0,011
Observation 601 618
***significant at 1%: **significant at 5%: *singificant at 10%
Not EmployedEmployed
1 = 0
1 = 1
1 = 0
1 = 1
28. • Overall, a small negative relationship between
informal caregiving and labour-market
outcomes
• An significant existence of a intensive margin
• That gender differencies exists
• That the assumption of exogeneity couldn’t
be rejected
• That the likelihood of ”early” retirement
increases for individual older than 58
experience a parent passing away
Conclusion
30. Figure 2-1: The Global model of care. It illustrates the three components determine the
demand and supply of care. Source: Own illustration
08-12-2016 30
31. Table 6-9: Regional regression results of the analysis of informal caregiving hours on working-hours
08-12-2016 31
Marginal
Robust
S. E. Marginal
Robust
S. E. Marginal
Robust
S. E. Marginal
Robust
S. E.
Hours of care -0,119 * 0,066 -0,021 0,065 0,047 0,043 -0,047 0,071
Age 1,484 5,104 -2,212 4,055 5,353 3,323 3,020 2,712
Age Squared -0,012 0,046 0,017 0,037 -0,050 * 0,030 -0,028 0,024
Number of children -1,859 ** 0,797 0,046 0,393 0,108 0,343 -0,278 0,287
Number of chronical conditions -0,804 0,931 1,106 * 0,578 0,206 0,442 -0,195 0,467
Married 0,087 2,284 -2,452 ** 1,134 -1,605 1,063 -2,413 *** 0,787
Self-rated health 3,583 * 2,092 0,013 1,251 -0,102 1,008 2,514 ** 1,125
Household's income percentile 0,136 0,264 0,328 ** 0,166 0,479 *** 0,168 0,494 *** 0,153
EURO-D 0,754 ** 0,373 -0,048 0,211 -0,207 0,166 -0,442 ** 0,198
Wave2 dummy 2,188 ** 1,072 1,617 ** 0,746 -0,587 0,565 -0,843 0,558
Years of education 0,219 0,202 -0,055 0,104 0,263 ** 0,111 -0,034 0,115
Female -5,925 *** 1,817 -9,171 *** 0,931 -11,022 *** 0,829 -5,256 *** 0,745
Observations 799 1201 1536 1119
Individuals 430 676 862 609
Wald 70,35 *** 119,8 *** 243,52 *** 114,5 ***
*** Significant at 1 %; ** Significant at 5%; *Signifiant at 10 %
Region A Region B Region C Region D
Hinweis der Redaktion
The model is based on the Nocera and Zweifel framework from 1996.
The model is based on the Nocera and Zweifel framework from 1996.
Natural experiments are naturally occurring situations where we want to know the effect of variable X on Y and there is a variable Z related to X, but not ε
Another way so say this is: Z effects Y only through X
This variable Z is called an instrumental variable
It can be shown that
is an unbiased estimator of β1 in large samples but not in small samples (bIV is consistent)
IV can thus be used to address the following important threats to internal validity:
• Omitted variable bias from a variable that is correlated with X but is unobserved, so cannot be included in the regression;
• Simultaneous causality bias (endogenous explanatory variables; X causes Y, Y causes X);
• Errors-in-variables bias (X is measured with error)
Instrumental variables regression can eliminate bias from these three sources.
A valid instrument must satisfy two conditions
FE modeller kontrollere for alle tidsinvariante forskelle mellem personer, så den estimeret koefficient af FE modellen kan ikke være biased grundet omitted time-invariant karakteristika
En downside, ved at bruge FE modeller er at de ikke kan bruges til at undersøge tidsinvariant udsving af de afhængige variabler. FE modellere er designet til at studere ændringer within a person.
NOT SIGNIFICANT!
INCREASING STANDARD ERRORS
Gender difference, but not significant different. However it seems likely that women is having a trade-off
To summarize the findings, only the older cohort experiences a significant effect on employment when a parent passes away.
Discussion
Selection bias?
More periods?
Measurement error and intrepretion of questions
Larger sample size
Instrument variables valid and good?