Mortality Trends and Variation in China

Yong Cai caiyong@unc.edu

University of North Carolina, Chapel Hill

May 15, 2019

Harnack Haus, Berlin

Satellite meeting

Mortality data and methodological approaches in

estimating mortality in developing countries

Sponsored by Max Planck Institute for Demographic Research

Background

Data sources & challenges

Mortality trends and variation

Overview

Economic growth

Changes in nutrition and consumption

Social transformation

Individual awareness and life style

Establishment of “new” social welfare system

Healthcare and delivery

Public health investment

Socioeconomic background

Main sources

Population census and surveys

Health surveillance systems

Death registration

Household registration (Hukou)

Social pension system

Main challenges

Data quality

Underreporting

Migration

Government evaluation

Data availability

Segmentation

Undocumented adjustment

Data sources and challenges

Chinese data: good and bad

Mortality trends: e0t

70

72

74

76

78

80

82

1995 2000 2005 2010 2015 2020

Unadjusted Survey

NBS Official

WPP2017

Mortality trends: qxt

0.0001

0.001

0.01

0.1

1

0 20 40 60 80

q2000 q2005 q2006

q2007 q2008 q2009

q2010 q2011 q2012

q2013 q2014 q2015

q2016 q2017

Mortality trends: lgt(lxt) compared to 2000

-2

-1

0

1

2

3

4

5

6

-3 -2 -1 0 1 2 3 4

2005 2006 2007 2008 2009 2010 2011

2012 2013 2014 2015 2016 2017

Mortality trends: lgt(lxt) compared to 2000

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0

0.2

0.4

0.6

0.8

1

1.2

1.4

2004 2006 2008 2010 2012 2014 2016 2018

BET

A

beta alpha Linear (beta) Linear (alpha )

Mortality trends: gender gap e0f - e0

m

0

1

2

3

4

5

6

7

1995 2000 2005 2010 2015 2020

Unadjusted Survey

NBS Official

WPP2017

Morality trends: Shanghai e0

70

73

76

79

82

85

88

1980 1985 1990 1995 2000 2005 2010 2015

HK e0f HK e0m

Morality trends: Shanghai e0

70

73

76

79

82

85

88

1980 1985 1990 1995 2000 2005 2010 2015

HK e0f HK e0m

NBS e0f NBS e0m

Survey e0m Survey e0f

Mortality trends: infant mortality rate

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Surveillance Adjusted Population Survey Surveillance Raw

Rising life expectancy fits well with the general pattern of mortality decline as it was driven by multiple factors happening in China: socioeconomic development, investment in public health, change of life style…

Mortality underreporting is a constant problem, and will remain to be.

While overall mortality decline can be collaborated with other data sources, ascertain of mortality level requires careful cross-examination with other population data.

Mortality trends: basic observations

Mortality trends by rural/urban (unadjusted e0t)

70.0

72.0

74.0

76.0

78.0

80.0

82.0

84.0

86.0

88.0

2000 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

cities towns villages

Mortality trends: rural/urban gap (unadjusted e0t)

comparing to e0t for villages

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

9.0

10.0

1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

towns

cities

Regional trends and variation, e0t

65

70

75

80

85Xizang

YunnanGuizhou

Qinghai

Xinjinag

Gansu

Jiangxi

Neime…

Shaanxi

Ningxia

Hunan

Hubei

Sichuan

Guangxi

ChinaHenan

ShanxiChong…

Anhui

Helong…

Hebei

Fujian

Hainan

Jilin

Guang…

Liaoning

Jiangsu

Shang…

Zhejiang

Tianjin

BeijingShang… NBS2000

Census2000

Census2005

Census2010

Census2015

Mortality variation: e0 (2000, unadjusted)

Percentile

Variable Mean S.D. Min 5 Med 95 Max

e0m 69.9 3.9 46.4 62.9 70.5 75.3 86.9

e0f 73.8 4.7 45.3 65.4 74.5 80.0 106.7

1q0m 25.7 23.4 0.0 4.6 22.9 91.1 376.7

1q0f 32.8 32.0 0.0 4.9 18.9 68.6 307.3

Mortality variation: county level e0 (2000, raw)

Brass (1971) relational model: linear relationship between lx’s of an observed population and of a standard population at logit scale

Consistency of mortality across different age groups

Choice of standard life table: “must be some kind of average Brass (1971)”. We use provincial average.

)()(s

xx llogitllogit

Brass Logit Model Smoothing

Empirical Bayes Estimates

Empirical Bayes Estimates (Efron and Morris 1973, 1975; Clayton and Kaldor 1987; Marshall 1991): pool information across areas to produce more stable and robust estimates using the Empirical Bayes methods.

Information from neighboring units serves as priori for the local estimate.

)(ˆ2

2

ni

i

nn

n

ni mm

n

m

n

mS

n

mS

m

Empirical Bayes Estimates

14.44

6.46

14.61

7.99

15.65

14.696.90

11.02

14.693.50 10.06

11.16

11.33

9.96

12.77

14.6314.82

13.89

15.65

4.30

16.42

6.84

22.62

15.61 6.3111.12

15.412.70 9.83

11.16

10.32

7.33

14.21

15.0715.22

14.68

Percentile

Variable Mean S.D. Min 5 Med 95 Max

e0m 69.8 3.7 46.6 62.9 70.5 74.5 79.3

e0f 73.4 4.2 45.5 65.3 74.3 78.6 84.0

1q0m 26.0 22.1 0.9 6.1 23.8 93.0 355.2

1q0f 33.6 31.3 1.2 6.6 19.2 66.9 267.7

Mortality variation: county level e0 (2000 smoothed)

Mortality variation: county level e0t (2000)

Legend46 - 6060 - 65

65 - 67.567.5 - 7070 - 72.572.5 - 7575 - 80.8

Mortality variation: county level e0f clusters (2000)

cluster, sig

low, <.10low, <.05low, <.01high, <.01high, <.05high, <.10not sig.

PredictorMale Model Female Model Difference

coef std beta coef std beta (Male - Female)

(Intercept) 61.526 1.9371 *** 64.099 2.3035 *** -2.573

GDP 0.6292 0.1119 0.1205*** 1.0133 0.1331 0.1676*** -0.384*

Urbanization 0.0046 0.0042 0.0296 -0.0014 0.0050 -0.0080 0.006

Employment 0.0179 0.0089 0.0435* 0.0312 0.0106 0.0654** -0.013

Health Care -0.6532 0.2782 -0.0518* -0.4184 0.3308 -0.0286 -0.235

Water 0.0044 0.0026 0.0337 0.0040 0.0031 0.0265 0.000

Lavatory 0.0058 0.0022 0.0448** 0.0069 0.0026 0.0458** -0.001

Bath Facility 0.0121 0.0042 0.0659** 0.0162 0.0050 0.0761** -0.004

Education 0.9515 0.0982 0.3524*** 1.0719 0.1167 0.3428*** -0.120

Discrimination 0.0465 0.1629 0.0060 0.0353 0.1937 0.0039 0.011

Ethnicity -0.0253 0.0029 -0.2152*** -0.0151 0.0035 -0.1107*** -0.010*

Marriage -0.0417 0.0133 -0.0677** -0.0705 0.0158 -0.0987*** 0.029

Family Structure 0.0117 0.0095 0.0229 0.0106 0.0113 0.0180 0.001

Latitude 0.0324 0.0177 0.0580 -0.0023 0.0210 -0.0035 0.035

Longititude 0.0064 0.0153 0.0167 0.0168 0.0182 0.0379 -0.010

Elevation -0.0003 0.0001 -0.0780 -0.0002 0.0002 -0.0415 0.000

Terrain -0.0009 0.0003 -0.0544** -0.0011 0.0004 -0.0548** 0.000

*** denotes p<.001; ** denotes p< .01; * denotes p< .05

Mortality variation: e0 2000 regression

Mortality variation: county level e0f (2000)

cluster, sig

low, <.10low, <.05low, <.01high, <.01high, <.05high, <.10not sig.

Huge regional variation: China is a world of its own!

Regional variation is relatively stable

Regional differences are largely driven by developmental factors

Mortality variation: observations

A model/system simultaneously estimating fertility, mortality, population age structure using all available survey/census data to consistently estimate demographic parameters.

Subnational evaluation

HMD standards/protocols

Next Steps

Thank you!