Emotional Wellbeing of the Elderly in East Asia:

Cross-country differences in age gradients and other determinants

Jinkook Lee, Ph.D.University of Southern California & RAND

October 2015Mixed Emoti Con

University of Michigan

2

Outline

• Motivations• Our approach• Cross-sectional Findings from Within and Across Country Analysis• Next steps• Some preliminary findings from Longitudinal Analysis

3

China, Korea and Japan

• Very high suicide rates among their elderly

• We want to understand the factors contributing to the differences in emotional well-being of the elderly populations across three countries.

4

Suicides rates The suicide mortality rates (29.1) in Korea and Japan (20.9) is much higher than in other OECD countries (12.5)

KOR HUN JPN FIN POL AUT SWE NOR IRL DEU ISL PRT NLD ESP GBR ISR GRC0

10

20

30

40

50

60

70

80

90

100

110

120

Number of Suicides per 100,000 persons in 2009

All ages Ages 55-74 Ages 75+

5

6

Prior Literature: Relationship between Age & SWB

• By referring to SWB as instantaneous utility (instead of permanent utility), Deaton (2007) posited that SWB would have “an inverse U-shape,” rising at first as people accumulate human capital, self-knowledge and the ability to enjoy themselves – learn to be happy – and then eventually falling as the capacity to enjoy fails with “age” for health or economic reasons. • Socio-emotional selectivity theory argues that SWB increases with age through

successful adaption (Diener et al., 1999; Hendrie et al., 2006).

7

Prior Literature: Cross-country Variations in Relationship between Age & SWB

• For the U.S., Easterline (2006) observed an inverted-U shaped relationship from age 18 to 89 after controlling for birth year dummies. • For the U.K., Clark (2007) found a U-shaped relationship from age 16 to 64 after

controlling for birth-year effects. • Examining persons 16 to 91 years of age in the U.K. and Germany, Wunder et al.

(2009) and Baird et al. (2010) found a second turning point later in life.

8

Cross-country analysis of emotional wellbeing

China• Albert Park, HKUST• Xiaoyan Lei, Peking U.

Korea• Jinkook Lee, USC & RAND• Chulhee Lee, SNU

Japan• Hidehiko Ichimura, U of Tokyo• Yasuyuki Sawada, U. of Tokyo

Basic research question we want to address• Evaluate impact of 4 broad factors on emotional wellbeing• Basics (Age, Education, Marital status)• Economic (Consumption/Income, Asset, Pension)• Family-Social (Contact with Children, Living Arrangements, Transfers, Social

Participation)• Health (ADL, Specific diseases)

• Explore differences across three countries• If the measurements are different, then it would cause differences as

well• By using harmonized data, we can minimize effects of differing measures

10

HRS Family of Surveys

Health and Retirement Study

Mexican Health and Ageing Study

English Longitudinal Study of Ageing

Survey of Health, Ageing, and Retirement in Europe

Irish Longitudinal Study on Ageing

Korean Longitudinal Study of Aging

Japanese Study on Aging and Retirement

China Health, Aging, and Retirement Longitudinal Study

Indonesia Family Life Survey

Longitudinal Aging Study in India

China Health and Retirement Longitudinal Study (CHARLS)

•Biennial panel•Community-residing people aged 45 and older•Multi-stage PPS random sampling– Counties, Villages, Households, Persons

•Pilot survey in 2008: Zhejiang and Gansu•Baseline survey in 2011-2012: 10,257 households, 17,708 individuals, 450 villages– 150 counties in 28 provinces– Tibet was excluded– Hainan and Ningxia had no counties sampled

•Wave 2 in 2013

11

12

Korean Longitudinal Study of Aging (KLoSA)

• Biennial panel• Community-residing people aged 45 and older

- Everyone in the HH at age 45+

• Multi-stage PPS random sampling– Provinces & MSAs, Apartment versus other housing– Exclude Jeju island

• Baseline in 2006: 10,256 respondents- 2007 Work History interview- 5 waves completed

Japanese Study of Aging and Retirement (JSTAR)

• Biennial panel• Different sampling strategy to link to administrative data• 1st wave: 2007 in 5 municipalities• 2nd wave: 2009 in 5+2 municipalities• 3rd wave: 2011/2012 in 5+2+3 municipalities• Response rate is about 60% excluding no contact.• Data from all waves are now available for researchers from RIETI. See <

http://www.rieti.go.jp/en/projects/jstar/index.html>

• 4th wave: 2013/14 completed

13

Building a Comparative Data Base

• Time: • CHARLS 2011-12 Wave 1• KLoSA 2012 Wave 4• JSTAR 2011-12 Wave 3

• Age groups: • 54-59, 60-64, 65-71, and 72-78

• Regional controls: • county in China• province in Korea• city in Japan

14

15

Sample statistics China Korea Japan

N 9,720 5,614 3,687

Age

54 – 59 37.7 25.3 22.6

60 – 64 25.3 20.5 15.0

65 – 71 22.3 27.9 33.1

72 – 78 14.6 26.2 29.4

% Male 50.0 47.7 42.0

China Korea Japan

Education

Illiterate 47.8 9.7 0.0

Primary school 23.3 28.6 1.5

Middle school 17.0 19.2 34.2

High school 5.4 31.5 53.6

College+ 6.5 10.9 10.7

% Married 84.7 82.8 55.6

No of children 2.8 2.7 2.1

% Childlessness 2.8 2.5 10.3

16

Harmonized measures

• Wellbeing • a binary variable, indicating

whether CESD score is 10+ • CESD score is calculated using the

10 item, 4-point Likert scale (ranges 0 to 3)

36.7

26.5

15.5

% Clinically depressed

China Korea Japan

17

10 Common CESD Items Used

• Bothered by normal things• Depressed• Everything is effort• Hopeful about future• Fearful or afraid

• Sleep was restless• Happy, feel good• Lonely• Couldn’t get going; tired or low

in energy• Trouble concentrating, keep my

mind on things

18

Analytical approaches: Base econometric specification

A country-specific linear probability model of the determinants of whether individual i in country c has elevated depressive symptoms (Dic).

(Eq 1)

The covariates Xic include four sets of variables: basic (Bic), economic (Eic), social (Sic), and health (Hic).

• Basic model: age, gender, education, married, no of children, no of children sq, regional controls

• Basic + economic variables

• Basic + family/social variables

• Basic + health variables

• Full model: basic + economic + family/social + health

19

Age gradients in depression in China, Korea, and Japan

Base Base + econ Base + social base + health Full

China Age 54-59 (ref)

60 – 64 0.024** 0.032** 0.024* 0.007 0.019

65 – 71 0.038*** 0.041*** 0.032** 0.004 0.010

72 – 78 0.009 -0.002 0.006 -0.033** -0.029

Korea 60 – 64 0.003 -0.012 0.004 -0.002 -0.009

65 – 71 0.075*** 0.025 0.074*** 0.051*** 0.019

72 – 78 0.144*** 0.052** 0.146*** 0.104*** 0.046*

Japan 60 – 64 -0.070*** -0.122*** -0.082** -0.073*** -0.139***

65 – 71 -0.067** -0.155*** -0.080** -0.085*** -0.191**

72 – 78 -0.083*** -0.233*** -0.100** -0.104*** -0.250***

20

Linear Probability Model of Being Depressed

China Korea Japan F-statBase 3.66***Male -0.087*** -0.009 0.029 10.40***Age 54 – 59 (ref) 60 – 64 0.019 -0.009 -0.139*** 4.84***

65 – 71 0.010 0.019 -0.191*** 72 – 78 -0.029 0.046* -0.250***

Middle school (ref) Illiterate 0.090*** 0.068** - 2.04** Primary school 0.046*** 0.033* 0.142* High school 0.008 -0.013 0.023 College+ -0.016 -0.018 0.002

Married -0.066*** -0.084 -0.029 No of children 0.009 -0.024 0.025 No of children2 -0.002 0.003 -0.007

*** p<0.01, ** p<0.05, *p<.10

21

Linear Probability Model (Cont.)

Economic China Korea Japan 1.61**Working -0.028*** -0.103*** -0.067** 4.68***

Consumption q4 (ref) q1 0.033** 0.005 -0.015 0.85 q2 0.023 -0.029 -0.014 q3 0.020 -0.019 -0.020

Receive pension 0.048 0.018 0.027 1.21Ln (1+pension income) -0.013** -0.016 0.004 Expect to receive pension 0.005 0.021 -0.027 Own home 0.156*** 0.030 0.010 0.80

Ln (1+gross housing value) -0.016*** -0.007 -0.003

Ln (1+total debts) 0.005** -0.002 0.000 1.42Ln (1+total financial assets) -0.008*** -0.007*** -0.004

*** p<0.01, ** p<0.05, *p<.10

22

Linear Probability Model (Cont.)

Social China Korea Japan 2.10***Ln (1+amount of transfer given) -0.005** 0.003 -0.000 1.40

Ln (1+amount of transfer received)

-0.004** -0.002 -0.008

Frequent contact with children -0.085*** -0.019 -0.046 3.49***

Frequent social activities -0.043*** -0.092*** -0.074***

Living with partner (ref) Living alone 0.006 0.407** -0.011 2.14** with children -0.003 0.029 -0.018 with others 0.027 0.000 -0.028

Living nearby children 0.020 -0.028 -0.012

*** p<0.01, ** p<0.05, *p<.10

23

Linear Probability Model (Cont.) China Korea Japan F-Stat

Health 2.01***

Any ADL difficulties 0.214*** 0.296*** 0.171* 0.03

Hypertension 0.015 -0.012 0.045* 2.37*

Diabetes 0.040** 0.052*** 0.077* 0.12

Cancer 0.075 0.151*** 0.063 1.97

Lung disease 0.068*** 0.064 0.211 1.83

Heart disease 0.100*** 0.093*** 0.090** 0.06

Stroke 0.112*** 0.055 -0.043 3.76**

Arthritis 0.102*** 0.030 0.097** 4.77***

* *** p<0.01, ** p<0.05, *p<.10

24

Analytical approaches: Oaxaca decomposition

• Comparing the results for country 1 and country 2, the difference in predicted probabilities R is:

]

explained unexplained

where * are the coefficients from a pooled regression using data from both countries 𝛽• This decomposition formula explains the difference in depression as the sum of explained and unexplained

components

• The explained part of the difference is that which can be explained by the characteristics of the elderly

• The unexplained part is from differences in the coefficients of the two country-specific regressions, 𝛽1 and 𝛽2

• These explained and unexplained parts can be divided among the four categories of variables (B, E, S, H) or among individual covariates.

25

Oaxaca Decomposition

Mean Diff Exp Unexp ExpBasic

ExpEcon

ExpSocial

ExpHealth

Unexp Basic

Unexp Econ

Unexp Social

UnexpHealth Const

China vsJapan

0.284 -0.144

0.140 0.116(83%) 0.024 0.010 0.086 -0.023 0.044 0.048 -0.090 0.021 -0.005 0.050

Korea vsJapan

0.231 -0.144

0.087 0.049(56%) 0.038 0.027 0.029 0.012 -0.018 0.005 -0.087 0.013 -0.005 0.113

Korea vsChina

0.231 -0.284

-0.053 -0.107(202%) 0.054 -0.013 -0.041 -0.002 -0.050 -0.012 -0.014 0.029 -0.012 0.063

5% negative significance is in blue, positive significance is in red.

26

Analytical approaches: simulation

• We use the country-specific regression coefficients from estimating equation (1) to investigate how depression rates would change if the distribution of covariates were the same as another country.• For example, what would the depression rate be for Chinese elderly if

they had the same characteristics as Japanese elderly? • We use the 3 sets of country-specific coefficients and 3 sets of

country-specific distributions of covariates to calculate 9 expected depression rates

27

Simulation: mean predicted probabilities of depression when we apply coefficients from each country-specific regression to the covariates of the other countries

Coefficients

Obs China Korea Japan

Covariates

China 4679 0.300 0.360 0.429

Korea 3591 0.195 0.235 0.307

Japan 1276 0.160 0.272 0.147

28

Analytical approaches: Common support – matching analysis

The linear regression analysis is sensitive to the support and the distribution of regressors.

30

Analytical approaches: Common support – matching analysis

• In order to address this issue of different distribution of regressors, we apply the program evaluation method that assumes selection on observables.• Using matching analysis, we examine the effect of the variables

controlling for other variables

31

Common Support and Controlling for Distributional Differences

Lin Prob K vs J E(Y_k-Y_J|Z) s.e. Lin Prob

C vs J E(Y_C-Y_J|Z) s.e.

Male 0.108 0.092 (0.027) 0.033 0.058 (0.027)

Age 60-64 0.130 0.052 (0.039) 0.070 0.122 (0.045)

65-71 0.143 0.082 (0.036) 0.078 0.126 (0.043)

72-78 0.154 0.268 (0.175) 0.080 0.236 (0.181)

Middle School High School 0.107 0.064 (0.028) 0.048 0.068 (0.037)

College 0.110 0.072 (0.061) 0.047 0.006 (0.053)

Married 0.106 0.092 (0.023) 0.045 0.079 (0.023)

32

Common Support and Controlling for Distributional Differences (Cont.)

Lin Prob K vs J E(Y_k-Y_J|Z) s.e. Lin Prob

C vs J E(Y_C-Y_J|Z) s.e.

Work 0.108 0.032 (0.023) 0.055 0.071 (0.027)

Expect Pension 0.120 0.077 (0.020) 0.055 0.059 (0.022)

Home Ownership 0.116 0.087 (0.023) 0.070 0.097 (0.025)

Freq Contact 0.117 0.101 (0.022) 0.045 0.104 (0.023)

Freq Social Act 0.109 0.069 (0.022) 0.057 0.064 (0.023)

Living Near Child 0.111 0.087 (0.024) 0.055 0.102 (0.025)

33

Common Support and Controlling for Distributional Differences (Cont.)

Lin Prob K vs J E(Y_k-Y_J|Z) s.e. Lin Prob

C vs J E(Y_C-Y_J|Z) s.e.

Any ADL 0.148 -0.121 (0.124) 0.062 -0.254 (0.124)

Hypertension 0.103 0.085 (0.033) 0.045 0.093 (0.036)

Diabetes 0.109 0.108 (0.062) 0.044 0.060 (0.068)

Cancer 0.130 0.303 (0.101) 0.052 0.302 (0.128)

Lung Disease 0.092 0.067 (0.141) 0.029 0.054 (0.133)

Heart Disease 0.114 0.199 (0.064) 0.052 0.138 (0.060)

Stroke 0.138 -0.212 (0.104) 0.090 -0.143 (0.115)

Arthritis 0.096 0.019 (0.071) 0.051 -0.049 (0.074)

34

Next Steps

• Exploit panel data• Examine the effects on spouse

35

Seemingly Unrelated Regression (SUR) Results of Husbands’ and Wives’ Cognition and Emotion

VARIABLES Husbands Wives MMSE12 CESD12 MMSE12 CESD12

Actor effects MMSE10 0.3909*** -0.1200*** 0.1143*** 0.0035 (0.0251) (0.0372) (0.0266) (0.0381)CESD10 -0.0482** 0.4239*** 0.0185 0.1929*** (0.0206) (0.0306) (0.0219) (0.0314)Spouse effects MMSE10 0.1505*** 0.0296 0.4568*** -0.0800** (0.0260) (0.0386) (0.0276) (0.0395)CESD10 0.0046 0.0801*** -0.0253 0.3146*** (0.0202) (0.0300) (0.0215) (0.0307)

Source: 2010 & 2012 KLoSA, controlling for age, age2, work status, and family income

36

Results of Fixed Effects Models: Husbands’ and Wives’ Cognition and Emotion

VARIABLES Husbands Wives

MMSEt CESDt MMSEt CESDtActor effects MMSEt-1 0.505*** -0.137*** 0.078*** -0.009 CESDt-1 -0.051** 0.375*** 0.011 0.120*** Spouse effectsMMSEt-1 0.061*** -0.015 0.463*** -0.120*** CESDt-1 -0.016 0.116*** -0.050*** 0.350*** Source: 2006, 2008, 2010 & 2012 KLoSA, controlling for work status and family income

37

Thank You!

38

Harmonized measures & Cross-country comparisons: Economic variables

% working % pension eligible own home any debt0

10

20

30

40

50

60

70

80

90

100China Korea Japan

39

Harmonized measures & Cross-country comparisons: Economic variables

(PPP) CHINA KOREA JAPANHH FOOD CONSUMPTION

10TH 128 1,645 69350TH 938 3,431 3,257

90TH 2,814 6,140 6,930PENSION INCOME

10TH 211 1,447 5,02450TH 5,844 3,553 14,553

90TH 16,234 20,394 32,051GROSS HOUSING VALUE

10TH 2,705 54,824 43,31350TH 22,998 164,471 155,925

90TH 135,281 438,591 433,125TOTAL DEBTS

10TH 271 10,965 1,90650TH 4,058 38,377 25,988

90TH 24,351 153,507 173,250TOTAL FINANCIAL ASSETS

10TH 27 1,316 4,33150TH 271 16,447 51,975

90TH 6,778 114,422 272,869

40

Harmonized measures & Cross-country comparisons: Family & social variables

any transfer to children

any transfer from children

frequent contact with

chidren

frequent social activities

living nearby children

0102030405060708090

100

China Korea Japan

41

Harmonized measures & Cross-country comparisons: Family & social variables

living alone living with partner living with children living with others0

10

20

30

40

50

60

Living Arrangement

China Korea Japan

42

Harmonized measures & Cross-country comparisons: Doctor-diagnosed diseases & ADL

hyper-tension

diabetes cancer lung disease heart disease

stroke arthritis ADL

29.2

7.4

0.9

11.3

14.3

3

34.6

16

28.5

12.7

3.42.1

5.53.4

15.2

2.5

35.6

12.5

4.72.1

12

3.9

7.24.6

China Korea Japan