The Link between Health Condition Costs and Standard of Living Through Out-of-Pocket Health Expenditures and Labour Market Outcomes in Turkey: A Structural Equation Modelling
Eleftherios Giovanis - University of Verona
Oznur Ozdamar - Bologna University
ERF Workshop on The Economics of Healthcare in the ERF Region
Cairo, Egypt - May 23, 2016
www.erf.org.eg
Similar to The Link between Health Condition Costs and Standard of Living Through Out-of-Pocket Health Expenditures and Labour Market Outcomes in Turkey: A Structural Equation Modelling
Impact Evaluation Training with AERC: China Cash Transfer Programme Technical...The Transfer Project
Similar to The Link between Health Condition Costs and Standard of Living Through Out-of-Pocket Health Expenditures and Labour Market Outcomes in Turkey: A Structural Equation Modelling (20)
The Link between Health Condition Costs and Standard of Living Through Out-of-Pocket Health Expenditures and Labour Market Outcomes in Turkey: A Structural Equation Modelling
1. Oznur Ozdamar
Adnan Menderes University, Aydın Faculty of Economics, Department of
Econometrics, Turkey
Bologna University, Department of Economics, Italy
Eleftherios Giovanis
University of Verona, Department of Economics, Italy
The Link between Health Condition Costs and
Standard of Living Through Out-of-Pocket Health
Expenditures and Labour Market Outcomes in Turkey:
A Structural Equation Modelling
2. • Purpose
• Motivation of the Study
• Previous research in Standard of Living Conditions
• Methodology
• Empirical Results
Outline
3. Purpose of the study
• A case-control study exploring the disability costs
• The relationship of income and disability-health status
• Calculation of the costs associated with disability and
comparison with the costs of the non-disabled households
• Sample of Study: Household Budget Survey (HBS) and the
Income and Living Conditions Survey (ILCS) in Turkey
• The study employees the standard of living (SoL) approach using
a Structural Equation Modelling (SEM)
4. Motivation of the study
• Health status is one of the most important indicators of human,
capital, welfare and productivity
• Many researches and methods have been developed and
applied in order to measure and describe the financial costs of
the health conditions, illnesses and disability.
• The main proxy that has been traditionally used for exploring the
distribution and inequality of standard of living is the income.
• Nevertheless, this proxy is not perfect and it has some disadvantages: it is not
usually correlated with some other aspects of standard of living
• the same income level may represent different standards of living for different
people
5. • This study aims to give more insights on the additional costs
of households with disability and health problems
Motivation of the study
6. • Three principal approaches have been developed for quantifying
the economic costs of disability;
• direct survey approach (DSA)
• expenditure diary approach (EDA)
• standard of living (SoL) conditions approach
Literature Review
7. • DSA approach asks the individuals with a disability how much
are willing to spend extra on specific expenditure items.
• Advantage:
• It is an inexpensive and straightforward approach
• Disadvantage:
• It is very difficult for the respondents to estimate their real expenditures
in need leading to inaccurate estimates
Literature Review
8. • More specifically,
• an accurate report of the spending on particular goods should
be given by the respondents, which can be very difficult for
general expenditure items.
• second and more problematic issue is that the respondents are
asked what they would have spent in the absence of disability
or health problems
• the answers rely on hypothetical expenditures without any
recent experience of the no disability environment leading to
biased estimates
Literature Review
9. • The study by Martin and White (1998) using the Office of
Population Censuses and Surveys (OPCS) in Great Britain
estimated the annual disability costs at £580.
• Other studies employing the direct survey approach have
found that the associated costs with wheelchair are £14.13 per
week (Hyman, 1977) corresponding to £72.89 in 2010 prices,
while Stowell and Day (1983) found that shopping costs
disabled people an extra £3.36 per week, corresponding to £9
in 2010 prices.
Literature Review
10. • The second method includes the expenditure diary approach
(Matthews and Truscott, 1988; Jones and O’Donnell, 1995).
• This approach takes detailed measurements of the
expenditures of a sample of disabled people and then the
results are compared with the expenditures of non-disabled
people.
Literature Review
11. • This approach can be more accurate that the direct survey, however
there are major limitations and drawbacks.
• Firstly, the expenditures are measured on household and not individual
level; thus the additional costs faced by the disabled individual can be
cancelled out by the reduction of consumption and therefore of
expenditures by the other members of the household in order to meet
the total costs of the household.
• Secondly, it is assumed that disabled and non-disabled people face the
same prices of goods and services. Actually, this might not be true as
disabled people may face additional costs like using taxi for shopping.
• Finally, the level of consumption and expenditures is heavily determined
by the composition of the household (e.g. employment status, household
size, house quality and expenses) and the income level. Thus, disabled
persons may not live in typical average households.
Literature Review
12. • The final method refers on the indirect approaches as the standard
of living approach (SoL) (Tibble, 2005; Zaidi and Burchardt, 2005).
• Morciano et al. (2015) estimated the annual disability costs at
£4,800, while Cullinan et al. (2011) found the short-run and long-
run costs ranging between £4,900-£5,200 per annum.
• Zaidi and Burchardt (2005), found the annual disability costs in
United Kingdom for low, middle and severe limitations estimated
respectively at 1,500; 3,800 and 6,400
Literature Review
13. • This project contributes to the literature by the
following:
ofirst, the examination of the relationship between SoL
and disability costs in Turkey has not been examined
before.
osecond, the effects of health insurance coverage and the
association between them, disability and OOPEs is
explored.
Literature Review
14. • The study follows the Standard of Living Conditions (SoL)
approach developed by Berthoud et al. (1993) and applied
by Indecon (2004), Zaidi and Burchardt, (2005) and Cullinan
et al. (2011).
• These studies examine whether an individual has a disability
answering in type of question like “Do you have any chronic
physical or mental health problem, illness or disability?”
• SEM is applied which is proper when the SoL is latent and
unobservable
Methodology
18. • The SoL refers on the household’s economic growth. This approach is
useful in order to identify the effects of disability and health insurance
programs on household’s economic growth, as well as, on OOPEs and
productivity.
• In this case the SoL approach examines the reduction in the SoL that
disabled people attain from the income. In other words this method
implies that because of the expenditure allocated to disability related
expenses the standard of living for both the disabled person and the
household will be reduced
Methodology
19. • For example if the share of the disability related costs and the non-
disable related expenditures stay the same at every income level then
the disability related costs will be higher at absolute terms at higher
income levels. Thus, the disabled people can enjoy the same SoL but
they require a higher income for that.
• In figure 1 S0 represents the disability SoL at income level Y0
represented by the curve disabled households (D). S1
D is the SoL of
households with disabled members which is equal with S0
ND, which
denotes the SoL of households with non-disabled members (ND) and
how much is necessary to spend is order to equivilise the SoL of the
two types of households represented by income Y1 and curve ND.
• In other words in order for the households with disabled members to
enjoy the same level of SoL a higher income Y1 is needed.
Methodology
20. • From the previous figures it becomes clear that the SoL is similar to
Engel curves and equivalence scales.
• More specifically, while in the first graph the SoL-income relationship
is presented, in the second figure the health-income relationship is
presented.
• In other words the indifference curves may represent utility (which is
the average SoL or predicted SoL from the figure 1, and health
represents the expenditures similarly to Engel curves.
• SoL within a SEM framework and using items as material deprivation
will allows us to derive the first figure, with equivalence scales which
are useful to derive the disability costs.
Methodology
21. • Structural equation models (SEMs) with latent
variables provide a very general framework for
modelling of relationships in multivariate data
(Bollen, 1989).
• SEM is most commonly applied in studies involving
latent variables, such as life satisfaction and
happiness and they provide a parsimonious
framework for covariance structure modelling.
Structural Equation Modelling (SEM)
22. • The advantages of SEM include:
It is possible to study the complex patterns of relationships
in a conceptual or theoretical model.
the measurement of the unobserved or latent variables by
observed indicators can be modelled taking into account
the effect of the measurement error on the structural
relationships.
Structural Equation Modelling (SEM)
23. • SEM uses a two-step process:
the measurement model
the structural equation model.
• The measurement model specifies how the latent
(unobserved) variables or hypothetical constructs are
measured in terms of the observed variables.
Structural Equation Modelling (SEM)
24. • The observed variables and unobserved constructs are
linked by one of two factor equations for observations
i=1,…..,N results to the measurement and structural
equations:
Structural Equation Modelling (SEM)
25. d
iidi DΛd
s
iisi
SoLΛs
(1)
(2)
iii
uaDOOPE X'1
(3)
ii
vbD W' (4)
iiiii
eOOPEyDSoL Ζ')log( 321
(5)
26. • Model (1) relates d or di=(di1,……,diq)΄ to an n-vector of the latent variable
defining the disability status for n≤q, through the q×n factor loadings matrix
Λd.
• Similarly, model (2) relates the an m-vector of the latent variable
SoLi=(SoLi1,……, SoLim)΄ for m≤p, through the p×m factor loadings matrix Λs. The
vectors εi
d and εi
s are the measurement error terms, with dimensions q × 1 and
p × 1 respectively.
• Model (3)-(5) is the structural equation model,
• where (3) explores the relationship of disability and OOPEs controlling for
characteristics represented by vector X,
• (4) examines the determinants of disability controlling for characteristics in
vector W and
• (5) explores the relationship between disability D, OOPEs and SoL given
individual and household characteristics in vector Z. The last equation is the
main interest of the study which is used to estimate the disability costs.
Methodology
27. The disability costs is the marginal effects of disability status over
the marginal effects of net household income. More specifically, this
is expressed by the partial derivatives of SoL with respect to
disability over the partial derivative of SoL with respect to income
which is just the marginal rate of substitution (MRS).
Methodology
y
SoL
D
SoL
MRS
log
/
(6)
28. • Relationship between SoL, disability-health and labour outcomes.
Methodology
iii
uaDOOPE X'1
ii
vbD W'
iiii
cOOPEcDcWage S')log( 21
iiiii
eOOPEwageDSoL Ζ')log( 321
(7)
(8)
(9)
(10)
29. • SoL indicators: a colour TV; telephone; washing machine; personal car;
computer; internet; kitchen; internet connection; hot water; piped water;
mobile phone; refrigerator; dishwater; air conditioner.
• Additional indicators can be considered. More specifically,
• The first set includes situations that a household cannot afford: to pay for
arrears on mortgage, loan or rent; to pay on utility bills; to pay for arrears on
utility bills; to go for holiday at least one week from home every year; to face
unexpected expenses; a meal with meat or fish every second day and to keep
home warm.
• The third set consists of five housing indicators on whether there are: leaking
roof, damp walls or rot in window frames problems; problems such as darkness
of rooms or lack of day-light; shortage of space in the dwelling; no bath or
shower; no indoor flushing toilet for sole use of the household; spending more
than 40 per cent of income net on housing costs.
• However, the first set is chosen which is common for both surveys.
Methodology
30. • Two datasets will be employed for this study.
• The first dataset is the Income and Living Conditions Survey (ILCS)
and it is a panel survey and the period covered is 2009-2012.
• This dataset will be used in order to estimate SEM and the
relationship of disability, SoL, wage and labour hours lost.
Furthermore, the aim of this study is to explore also the relationship
between SoL, disability and out-of-pocket expenditures (OOPEs). For
this reason the Turkish Household Budget Surveys (HBS), available
from TUIK during the period 2002 to 2013, which is a time-series of
repeated cross sections, is employed.
Data
31. Table 1. SEM Estimates for SOL using
the Household Budget Survey (HBS)
32. Measurement Equation for Disability DV: Disability Structural Equations DV: OOPE DV: Disability DV: SoL
Limited in daily activities due to mental or health
problem
0.7779***
(0.0028)
OOPE -0.0104***
(0.0025)
Limited in activities related to work due to mental or
health problem
0.9714***
(0.0033)
Disability (High) 0.0181***
(0.0030)
-0.1197***
(0.0024)
Measurement Equation for SoL DV: SoL Log of Household Income -0.3176***
(0.0021)
0.4617***
(0.0032)
Bathroom 0.4811***
(0.0021)
Log of Pre-benefit Household Income -0.0067*
(0.0034)
Kitchen 0.2728***
(0.0030)
Gender (Male) 0.0187***
(0.0030)
-0.0288***
(0.0029)
-0.0638***
(0.0026)
Dishwasher 0.6086***
(0.0021)
Age 0.1032***
(0.0031)
-0.0767***
(0.0238)
-0.3626***
(0.0195)
Air Conditioner 0.2731***
(0.0029)
Age squared 0.2962***
(0.0175)
0.1312***
(0.0144)
PC computer 0.6715***
(0.0021)
Education (reference =Illiterate)
Internet 0.6220***
(0.0023)
Education -Literate – not completed a school 0.0124***
(0.0038)
-0.1067***
(0.0039)
-0.0724***
(0.0032)
Mobile phone 0.3636***
(0.0028)
Education -Primary School 0.0287***
(0.0051)
-0.2096***
(0.0053)
-0.0196***
(0.0044)
Refrigerator 0.2127***
(0.0030)
Education -Primary Education 0.0623***
(0.0044)
-0.1777***
(0.0048)
-0.0410***
(0.0040)
Piped water 0.3795***
(0.0029)
Education -Secondary School -0.4438***
(0.0633)
-0.1181***
(0.0037)
0.0235***
(0.0031)
Hot water 0.5703***
(0.0023)
Education -High School -0.6255***
(0.0569)
-0.1726***
(0.0044)
0.0560***
(0.0037)
Washing machine 0.5017***
(0.0026)
Education -Senior High School -0.8154***
(0.0649)
-0.1365***
(0.0038)
0.0392***
(0.0032)
TV 0.1564***
(0.0033)
Education -College -0.8942***
(0.0804)
-0.1036***
(0.0034)
0.0520***
(0.0028)
Car 0.1493***
(0.0033)
Education -University -1.082***
(0.0665)
-0.1479***
(0.0039)
0.1017***
(0.0034)
33. Measurement Equation for Disability DV: Disability Structural Equations DV: OOPE DV: Disability DV: SoL
No. Observations 52,818 Marital Status (reference=never married)
Log likelihood -221,009.51 Marital Status -Married -0.0804***
(0.0343)
-0.0877***
(0.0045)
0.0092**
(0.0036)
AIC 442,249.011 Marital Status -Divorced -0.1377**
(0.0667)
0.0666***
(0.0036)
-0.0965***
(0.0166)
BIC 443,048.173 Marital Status -Widowed -0.1324
(0.1010)
-0.0012
(0.0030)
-0.0330***
(0.0113)
Chi square/df 3.16 Health Insurance (Reference=Public)
CFI 0.878 Health Insurance-Private -0.0518
(0.0909)
0.0068
(0.0114)
-0.0091***
(0.0021)
TLI 0.825 Health Insurance-Green Card 0.1547***
(0.0246)
0.0136
(0.0128)
-0.2131***
(0.0026)
RMSEA 0.044 Health Insurance-No Insurance 0.0435
(0.0830)
-0.0023
(0.0030)
-0.1100***
(0.0024)
SRMR 0.066 Household Size -0 .0530***
(0.0057)
-0.0406***
(0.0032)
-0.2045***
(0.0026)
Deposits in bank -0.0021
(0.0028)
Homeowner -0.0035
(0.0029)
Employed (No) 0.1664***
(0.0272)
-0.0740***
(0.0026)
Urban Area -0.2758***
(0.0259)
-0.0127***
(0.0031)
0.1577***
(0.0025)
34. Estimated form of the SoL-Income
Relationship using the HBS and the Log
of Household Income
-1-.5
0
.5
1
SoLPredictedValues
0 50000 100000 150000 200000
Household Income
Disabled Households Non-Disabled Households
35. Estimated form of the SoL-Income
Relationship using the HBS and the
Household Income in Linear Terms
-.5
0
.5
1
1.5
SoLPredictedValues
0 50000 100000 150000 200000
Household Income
Disabled Households Non-Disabled Households
36. Estimated form of the SoL-Income
Relationship on wages using the ILCS
and Log of Household Income
-2-1
012
SoLPredictedValues
0 50000 100000 150000
Household Income
Disabled Households Non-Disabled Households
37. Estimated form of the SoL-Income
Relationship on wages using the ILCS
and Household Income in Linear Terms
-1
01234
SoLPredictedValues
0 50000 100000 150000
Household Income
Disabled Households Non-Disabled Households
38. HBS ILCS
DV: SOL DV: SOL DV: SOL DV: SOL
Disability -0.0578***
(0.0114)
-0.1197***
(0.0024)
-0.1279***
(0.0342)
Linear Income 0.2438***
(0.011)
0.0181***
(0.0030)
Log Linear Income 0.4617***
(0.0032)
0.4292***
(0.0033)
Log likelihood -284,664.16 -221,009.51 -178,148.1 -123,680.59
AIC 523,554.311 442,249.011 389,191.096 247,537.181
BIC 524,339.575 443,048.173 191,185.254 248,118.600
Table 2. Linear and Log Linear Income Specifications
Standard Errors within brackets, ***, denotes significance in 1% level.
39. Panel A: HBS
Mean T-statistic P-value
Variables Treatmen
t
Control
Gender 0.4586 0.4478 1.39 0.197
Age 36.365 36.772 -1.55 0.124
Education -Literate – not completed a school 0.12996 0.12844 0.26 0.794
Education -Primary School 0.31597 0.30177 1.76 0.079
Education -Primary Education 0.02489 0.0255 -0.22 0.824
Education -Secondary School 0.04215 0.03528 2.04 0.042
Education -High School 0.03879 0.042 -0.93 0.351
Education -Senior High School 0.02276 0.02245 0.12 0.906
Education -College 0.00855 0.00825 0.19 0.848
Education -University 0.01497 0.01542 -0.21 0.83
Marital Status -Married 0.54459 0.55056 -0.99 0.382
Marital Status -Divorced 0.20877 0.22083 -1.68 0.093
Marital Status -Widowed 0.01985 0.01772 0.9 0.368
Health Insurance-Private 0.06292 0.05895 0.95 0.342
Health Insurance-Green Card 0.20312 0.20922 -0.86 0.388
Health Insurance-No Insurance 0.1037 0.10064 0.58 0.564
Log of Household Income 9.3924 9.3832 1.35 0.186
Household Size 4.3845 4.3677 0.4 0.688
Urban Area 1.5796 1.5715 0.94 0.349
Table 3. T-test after the matching propensity score process
Panel B: ILCS
Mean T-statistic P-value
Variables Treatmen
t
Control
Gender 0.4537 0.46074 -1.18 0.253
Age 65.548 65.669 -0.28 0.778
Marital Status -Married 0.44429 0.4629 -1.56 0.118
Marital Status -Divorced 0.40791 0.39731 0.9 0.367
Marital Status -Widowed 0.01919 0.02292 -1.08 0.279
Education -Literate – not completed a school 0.15755 0.16213 -0.52 0.601
Education -Primary School 0.17216 0.16557 0.73 0.463
Education -Secondary, vocational secondary or
primary education school 0.03294 0.03094 0.48 0.634
Education -High school 0.00917 0.00773 0.65 0.513
Education - Vocational or technical high school 0.00487 0.00516 -0.17 0.865
Faculty/university, college or higher education
level 0.00401 0.00458 -0.37 0.714
Log of Household Income 9.3903 9.4106 -1.25 0.21
2 adults, no dependent children, both adults < 65
years 0.17388 0.167 0.76 0.445
2 adults, no dependent children, at least one adult
>65 0.00773 0.00802 -0.14 0.892
Single person with dependent children 0.03437 0.03781 -0.77 0.441
Two adults with two dependent children 0.02234 0.02378 -0.4 0.69
40. Structural Equation DV: SoL
Disability (High) -0.1345***
(0.0116)
Disability (High) Lagged -0.1251**
(0.0762)
Log of Household Income 0.4037***
(0.0119)
Log of Household Income Lagged 0.4204***
(0.0120)
Table 4. SEM Standardised Estimates for SOL with Income and Disability Lagged
using the Income and Living Conditions Survey (ILCS)
Standard Errors within brackets, *** and ** denote significance in 1% and 5%
41. Panel A: HBS (Table 3)
Percentage of Income 23.70%
Monetary Values per annum 4,200 TL
Panel B: ILCS (Table 4)
Percentage of Income 30.80%
Monetary Values per annum 4.600 TL
Panel C: ILCS (Table 7)
Short Run-Percentage of Income 33.30%
Short Run-Monetary Values per annum 5,100 TL
Long Run-Percentage of Income 29.70%
Long Run-Monetary Values per annum 4,500 TL
Table 5. Disability Related Costs using HBS and ILCS
42. Panel A: Link Between SoL, Disability, OOPEs and Wages
Structural Equations DV: OOPE DV: Disability DV: SoL DV: Wages
OOPE -0.0293***
(0.0049)
-0.0480***
(0.0047)
Disability (High) 0.0202***
(0.0061)
-0.0352**
(0.0165)
-0.2610***
(0.0108)
Log of Pre-benefit Household Income -0.0315***
(0.0076)
Log of Non-Wage Income 0.2880***
(0.0050)
0.0441***
(0.0052)
Log of Wages 0.1842***
(0.0060)
Log of Household Income -0.2251***
(0.0065)
No. observations 28,281 CFI 0.837
AIC 272,570.411 TLI 0.787
BIC 272,644.664 RMSEA 0.048
Chi square/df 3.58 SRMR 0.072
Disability Costs (Percentage of Wages) 19,0% Disability Costs (Monetary Values) 1,050 TL
Panel B: Link Between SoL, Disability, OOPEs and Weekly Working Hours
Structural Equations DV: OOPE DV: Disability DV: SoL DV: Weekly Hours
OOPE -0.0108***
(0.0011)
Disability (High) 0.0193***
(0.0045)
-0.0183**
(0.0079)
-0.0353***
(0.0043)
Log of Pre-benefit Household Income -0.0060
(0.0054)
Log of Household Income -0.2275***
(0.0046)
0.2395***
(0.0088)
0.0376***
(0.0043)
Weekly Hours 0.0106***
(0.0012)
No. observations 27,651 CFI 0.789
AIC 272,373.239 TLI 0.752
BIC 272,965.119 RMSEA 0.067
Chi square/df 5,23 SRMR 0.063
Disability Costs (Percentage of Income) 7.6% Disability Costs (Monetary Values) 1,100 TL
43. Panel C: Link Between SoL, Disability, OOPEs and Wages-Hours Productivity
Structural Equations DV: OOPE DV: Disability DV: SoL DV: Wages-Hours Productivity
OOPE -0.0254***
(0.0019)
-0.0451***
(0.0047)
Disability (High) 0.0845*
(0.0465)
-0.0192**
(0.0082)
-0.0104***
(0.0022)
Log of Pre-benefit Household Income 0.0083*
(0.0049)
Log of Household Income -0.2029***
(0.0069)
0.4060***
(0.0031)
Log of Non-Wage Income 0.0503***
(0.0053)
Wages-Hours Productivity -0.0377***
(0.0025)
No. observations 28,281 CFI 0.815
AIC 272,344.576 TLI 0.783
BIC 272,833.121 RMSEA 0.055
Chi square/df 6,15 SRMR 0.052
Disability Costs (Percentage of Income) 4.7% Disability Costs (Monetary Values) 1,020 TL
Panel D: Link Between SoL, Disability, OOPEs and Transportation Expenditures
Structural Equations DV: OOPE DV: Disability DV: SoL DV: Transportation Expenditures
OOPE -0.0916***
(0.0026)
0.2294***
(0.0025)
Disability (High) 0.0092*
(0.0049)
0.0114***
(0.0027)
Log of Pre-benefit Household Income -0.0081***
(0.0030)
Disability (High)*Transportation Expenditures -0.0108***
(0.0026)
Log of Household Income -0.3176***
(0.0027)
0.3930***
(0.0030)
-0.1796***
(0.0027)
Transportation Expenditures
No. observations 52,818 CFI 0.854
AIC 463,778.342 TLI 0.820
BIC 463,834.053 RMSEA 0.043
Chi square/df 8.59 SRMR 0.094
Disability Costs (Percentage of Income) 2.7% Disability Costs (Monetary Values) 470 TL
44. Status No disability t (%) Disability t (%)
No disability t-1 98.34 1.66
Disability t-1 27.19 72.81
Variables Coefficients Costs
Disability -0.2932***
(0.0223)
Percentage of Income 56.00%
Log of Household Income 0.5234***
(0.0084)
Monetary Values per annum 8,500 TL
Standard Errors within brackets, *** denotes significance in 1%
Table 7. Transitions in disability status
Table 8. Disability Related Costs using ILCS
Variables Differencing (m=2) Differencing (m=3) Differencing (m=5)
HBS 0.9915 1.1651 1.2645
Table 9. Test of Regression Equality (Yatchew, 1999; 2003)
IB Test 1.4864
[0.1590]
1.3293
[0.1832]
1.3586
[0.1721]