Econometrics report factors affecting human development index in asia from 2010 to 2021

Abstract The study uses a geometric approach to analyze and examine how the factors including PM2.5 air pollution, GNI per capita and life expectancy at birth affect the Human Development Index (HDI) in 48 Asian countries from 2010 to 2021. We employ HDI as the dependent variable while the three indicators PM, GNI, LE are explanatory variables. Statistical data was collected from mainstream websites of the World Bank, UNDP and helped show the index of HDI and its components in 48 Asian countries. Knowledge acquired through the econometrics course enabled our team to utilize the economic model and linear regression. In addition, we used the Ordinary Least Squares method to estimate the regression function. At first, we laid foundations and provided an overall view about the Human Development Index (HDI) and its indicators. Economic theories learnt in class helped us to make predictions of the effects of those three factors affecting HDI. Secondly, we sorted data from 48 developed countries from 2010 to 2021 on Microsoft Excel and processed fixedrandom effects models by using STATA software. After that, the estimated model was yielded, and hypothesis testing was conducted. We then used STATA to evaluate the impacts of those indicators and describe how they affect the variation of HDI. Lastly, we produced a complete research report, concluding that most of those factors, according to the study, contributed to the Human Development Index. Keywords: HDI, GNI, life expectancy at birth, PM2.5 exposure

FOREIGN TRADE UNIVERSITY

FACULTY OF INTERNATIONAL ECONOMICS

-*** -ECONOMETRICS II MIDTERM ASSIGNMENT

FACTORS AFFECTING HUMAN DEVELOPMENT INDEX IN

FACTORS AFFECTING HUMAN DEVELOPMENT INDEX IN ASIA FROM 2010 TO 2021

Faculty of International Economics, Foreign Trade University

Knowledge acquired through the econometrics course enabled our team to utilize theeconomic model and linear regression In addition, we used the Ordinary Least Squaresmethod to estimate the regression function At first, we laid foundations and provided anoverall view about the Human Development Index (HDI) and its indicators Economictheories learnt in class helped us to make predictions of the effects of those three factorsaffecting HDI Secondly, we sorted data from 48 developed countries from 2010 to 2021 onMicrosoft Excel and processed fixed/random effects models by using STATA software Afterthat, the estimated model was yielded, and hypothesis testing was conducted We then usedSTATA to evaluate the impacts of those indicators and describe how they affect the variation

of HDI Lastly, we produced a complete research report, concluding that most of thosefactors, according to the study, contributed to the Human Development Index

Keywords: HDI, GNI, life expectancy at birth, PM2.5 exposure

1 INTRODUCTION

The Human Development Index (HDI) has emerged as a valuable metric forevaluating a nation's progress beyond economic growth alone As countries in Asiawitness remarkable changes in the global economy, there is a growing need to understandthe factors that influence the Human Development Index in the region This researchpaper aims to examine the factors affecting the HDI in Asia from 2010 to 2021, sheddinglight on the multidimensional nature of development and its implications for policy-making and societal well-being

Understanding the determinants of the Human Development Index in Asia is ofparamount importance While GDP has traditionally been used as the primary indicator

of progress, it fails to capture the broader aspects of development, such as education,health, and income distribution By analyzing the factors that influence HDI,policymakers, researchers, and stakeholders can gain valuable insights into the keydrivers of socio-economic development This knowledge will enable more informeddecision-making and the formulation of effective policies aimed at improving the well-being and quality of life for people in the region

The main objective of this research is to identify and analyze the factors thatcontribute to variations in the Human Development Index across Asian countries from

2010 to 2021 By examining these factors, we aim to provide a comprehensiveunderstanding of the dynamics and determinants of development in the region.Additionally, we seek to offer insights into the potential policy implications forenhancing human development outcomes

This study focuses on the Asian region, comprising diverse countries with varyinglevels of economic development, social structures, and cultural contexts By examining awide range of nations, including both advanced and emerging economies, we aim tocapture the heterogeneity of factors influencing HDI in Asia The study encompasses aperiod of eleven years, from 2010 to 2021, allowing for an in-depth analysis of long-termtrends and capturing potential changes over time

This research paper is structured as follows to address the research objectivesand questions:

Chapter 1: Introduction

Chapter 2: Literature Review

Chapter 3: Methodology

Chapter 4: Data Analysis and Results

Chapter 5: Discussion and Implications

Chapter 6: Conclusion

By addressing these aspects, this research paper aims to contribute to the existingliterature on the factors influencing the Human Development Index in Asia and providevaluable insights for policymakers, researchers, and stakeholders in the region

2 LITERATURE REVIEW

The Human Development Report which is published annually focuses on human

as a central topic, which throughout the years has been modified to better reflect it.Researchers have been in a constant search for more accurate reflection of the problem,with the final three official dimensions being health, education, and standard of living.These three dimensions have been annually analyzed and reported in from 1990 onwards,with respective indicators being life expectancy at birth, years of schooling, and grossnational income per capita

Initial studies about HDI can be exemplified by the case in 1994 of “HumanDevelopment Index: Methodology and measurement” by Amatyr K Sen and SudhirAnand It focused differently on the same mentioned factors with much more complexcalculations, with topics of interest including human development in general, aggregateindicators and intrapopulation, income distribution and poverty, life expectancy, andlevel of education, … The formulas the calculate HDI back then were very muchdifferent and complicated, using the minimum and maximum value of each indicator,together with a number of complex assumptions and rules for each Throughout the years,

a more simplified approach has been acquired to carry out the research

Many other research also focus on these factors that further support the foundationlaid by Human Development Report, and at the same time, many more shed lights onother new indicators in the hope of discovering the influence of others factors on HDI,finding more and newer indicators, and testing its reliability Several examples of relatedresearch are listed below:

An article published in 2018 named “Analysis indicator of factor affecting HumanDevelopment Index (IPM)” by Windya Wahyu Lestari and Victoria Efrida Sanarindicates that life expectancy index, education index and income index all haveconsiderable impacts on Human Development Index with estimated result showing14.788% of the variation of each observation is the same

Mehran Alijanzadeh, Saeed Asefzadeh and Seyed Ali Moosaniaye Zare’s paper:

‘Correlation Between Human Development Index and Infant Mortality RateWorldwide’ posted in 2016, Biotech Health Sci find the correlation between humandevelopment index and infant mortality rate The paper discovered the relationshipbetween human development index with infant mortality rate It had the scope of 135nations analysed and derived from SPSS software The conclusion is the report was thatsocio-economic factors or human dimensions are correlated with mortality rate The percapita income, life expectancy, and education with r being -0,625, -0,925, and -0,843 Itindicated that these indicators are negatively correlated with the mortality rate (P< 0.01)

In 2017, the study "Quality of Life among General Populations of DifferentCountries in the Past 10 Years, with a Focus on Human Development Index: ASystematic Review and Meta-analysis" was published, which mainly use mainly usesrelevant factors to determine HDI relating to physical, psychological, social andenvironmental aspects to calculate the Quality of Life index in countries around theworld Conclusion was the extremely high HDI subgroup had the greatest overall QOLmean at 74.26 (CI=72.40-76.12) while the lowest mean score was observed in thepsychological domain (M=67.37; CI=66.23–68.52) Therefore, the former had the highestmeans of various QOL domains

"The dynamic association between healthcare spending, CO2 emissions, andhuman development index in OECD countries" published in 2020 and "EnvironmentalSustainability and Human Development: A Greening of Human Development Index"published in 2014 as this study will show, also very clearly explores the connectionsbetween Health and Environmental concerns and the Human Development Index (HDI).More specifically, research in 2014 shows that the close relationship between thedevelopment index of environmental sustainability and the HDI constitutes a U-shaped

relationship between the HDI and the EPI In addition, this study also considers and

calculates a new index, which is EHDI, between both environmental factors andwitnessing a huge change in order, which is said to be objective and ensure thedevelopment of the environment more sustainable And the 2020 study shows that: allthree key variables, health care costs, CO2 emissions and HDI all show a cause-and-effect relationship; A two-way causal relationship exists between health care costs andCO2 emissions, which suggests that CO2 emissions significantly increase health carecosts in OECD countries, similarly, head investment in health care also increases

emissions by using more energy; positive relationship of investment in health facilitieswith HDI and finally, CO2 reduction has a positive effect on HDI.

The implications of HDI are:

- Highlighting for policymakers, the media, and non-government organizations that thedevelopment of a country must be assessed by humans and their capabilities, noteconomic indicators

- Questions the policymakers about their choices; analyze countries with the samelevel of income but different human development outcomes

- Highlight differences within countries, between provinces or states, and acrossgenders, ethnicities and other socioeconomic groupings

3.1.2 Measurement

HDI a measure of a country's average achievements in three dimensions of humandevelopment:

- A long and healthy life – determined by life expectancy

- Education acquisition – determined by the median number of years spent in schoolfor persons aged 25 and older and by the predicted years of schooling for youngstersstarting school

- A reasonable standard of living – determined by Gross National Income per capita Therefore, based on these assumptions, the formula for calculating HDI is:

HDI = √3 LEI × EI × II

LEI: Life Expectancy Index

EI: Education Index

II: Income Index

3.1.3 Components

LEI: Life Expectancy Index

The mean number of years a newborn is expected to live if exposed to the sex- andage-specific death rates in effect at the time of his or her birth, for a specific year, in agiven country, territory, or geographic area This indicator is calculated as:

LEI = 85−20¿−20

LE: Life Expectancy at birth

EI: Education Index

Since 2010, the expected years of schooling for pupils under the age of 25 and theaverage number of years spent in school by adults have been combined equally (50%–50%) to calculate the education index In the years before 2010, the education index wasmeasured by the adult literacy rate (with a two-thirds weighting) and the combinedprimary, secondary, and tertiary gross enrollment ratio (with a one-third weighting).The largest updated formula of EI:

EI = MYSI+EYSI2

MYSI: Mean Years of Schooling Index (MYSI = MYS15 represent for maximum 15 years of schooling) EYSI: Expected Years of Schooling Index

(EYSI = EYS18 represent for maximum 18 years of schooling)

II: Income Index

The income index here differs from that used in HDI in that it incorporates asufficiency threshold below the HDI’s maximum value of $75,000 (2017$ PPP) This isbecause to achieve an income of $75,000 per capita is empirically incompatible withplanetary boundaries Nations with income over $60,000 have an average materialfootprint of 35t per capita (more than five times over the planetary boundary) and CO2emissions of 19t per capita (eleven times over the planetary boundary) These levels ofecological impact are highly destabilizing and cannot be universalized In this sense, theHDI income index effectively precludes nations from achieving very high HDI while atthe same time remaining ecologically sustainable

II = ln(75,000)−ln(100)ln(GNIpc)−ln(100)

GNIpc: Gross National Income at purchasing power parity per capita

As analyzed, the components of the HDI include income, education and lifeexpectancy Combined with the literature review ahead, this study will focus primarily onthe income and health dimensions, ignoring the education dimension Therefore, to studyHDI, we use 03 independent variables: GNI per capita, PM 2.5 exposure and lifeexpectancy at birth

3.2 GNI per capita

3.2.1 Definition

Gross National Income (GNI) is the sum of all the money that a country's citizensand enterprises have made It is used to gauge and chart a country's wealth through time.The sum of the country's gross domestic product (GDP) and its foreign-source revenue isthe figure

An estimate of the entire value of all products and services generated inside acountry for a specific time period, typically a year, is known as GDP, which is morecommonly used Gross national income (GNI), a substitute for the gross domestic product(GDP), is regarded by certain countries as being a more reliable estimate of a country'swealth

In fact, GNI may now be the most accurate reflection of national wealth giventoday's mobile population and global commerce

3.2.2 Measurement

As mentioned above, GNI and GDP are highly correlated, so to calculate GNI wehave the following equation:

GNI = GDP + (EXfs - IMfs)

GDP: Gross Domestic Product EXfs: Money flowing from foreign countries IMFS = Money flowing from foreign countries

3.2.3 Components

GNI consists of 02 main parts: GDP and (EXfs - IMfs) Explanation of these 2components:

GDP: Gross Domestic Product

A country's Gross Domestic Product, or GDP, is the total monetary or marketvalue of all the goods and services produced within that country's borders during aspecified period of time Which have been known as an extremely fundamental indicator

to measure the economic development of a country

Firstly, the basic formula:

Secondly, the income approach way:

GDP=Total National Income + Sales Taxes + Depreciation + Net Foreign Factor Income

Total National Income: Sum of all wages, rent, interest, and profits

Sales Taxes: Consumer taxes imposed by the government on the sales of goods and services

Depreciation: Cost allocated to a tangible asset over its useful life

Net Foreign Factor Income: Difference between the total income that a country’s citizens and companies generate in foreign countries, versus the total income foreign citizens and companies generate in the domestic country

EXfs - IMfs

EXfs: Money flowing from foreign countries

IMFS: Money flowing from foreign countries

These indicators will be compiled annually by the General Statistics Office ofVietnam

3.3 Life Expectancy at birth

3.3.1 Definition

The life expectancy at birth reflects the population's overall mortality rate Itprovides an overview of the mortality pattern that affects people of all ages, includingchildren, adolescents, adults, and the elderly The average number of years a newbornshould expect to live if they were subjected to the sex- and age-specific mortality ratesthat were in effect at the time of their birth for a particular year in a particular nation,territory, or geographic location

3.3.2 Measurement:

Life expectancies are calculated using (abridged) life tables presenting age specificmortality rates Life expectancy tables are calculated based on death probabilitiesaccording to Farr's death rate method:

qx= Mx Bx+ Mx

2

Mx = the number of deaths at the age of x to under x+1 years in the reported period

Bx = average population aged x to under x+1 in the base period

qx = death probability from age x to x+1

Farr's method of calculation of abridged lifetables assumes that there is a constantmortality within the age intervals and thus the years of life lived by a person dying in theinterval is (on average) half of the length of the interval

3.4 PM 2.5 exposure

3.4.1 Definition

The term fine particles, or particulate matter 2.5 (PM2.5), refers to tiny particles ordroplets in the air that are two and one half microns or less in width Like inches, metersand miles, a micron is a unit of measurement for distance There are about 25,000microns in an inch The widths of the larger particles in the PM2.5 size range would beabout thirty times smaller than that of a human hair The smaller particles are so smallthat several thousand of them could fit on the period at the end of this sentence

According to UNDARK: “the bottom scale shows the U.S EnvironmentalProtection Agency's current benchmarks specifically for PM2.5, which is measured inmicrograms per cubic meter of air - sometimes rendered as µg/m³ The higher the mass offine particulates in the air, the more dangerous it is to breathe” This is an indicator thatgreatly affects the life expectancy aspect of the HDI's lifespan Therefore, this index isincluded in the study to further analyze the level of influence, especially for developingcountries in Asia

3.4.2 Measurement

There are different ways to calculate PM 2.5, especially the “Direct MeasurementMethod” The BAM 1020 Beta Attenuation Mass Monitor is US-EPA designated forcontinuous PM2.5 monitoring and is used extensively in air quality monitoring networksworldwide In its standard configuration, the BAM 1020 will measure and then report PMlevels with high accuracy on an hourly basis

5 RESULTS

5.1 Data description

5.1.1 Data source

This study uses panel data from 48 countries in Asia from 2010 to 2021 with 576observations For this study, researchers intend to employ the econometric approach,utilizing 4 variables to reinforce our model: Human Development Index (HDI), Lifeexpectancy at birth (LE), PM 2.5 exposure (PM), and GNI per capita (GNI) In thisregression model, HDI is the dependent variable, and LE, PM, GNI are the independentvariables The data of all four variables in the model, which are human developmentindex (HDI), Life expectancy (LE), atmospheric particulate matter (PM), and GrossNational Income (GNI), originates from OECD, World Bank, and UNDP.

A statistical description can be seen in the table below – i.e., minimum, maximum, mean,and standard deviation

- HDI: HDI has a mean value of 0.7257708, the minimum value (0.448) in

Afghanistan (2019), and the maximum value (0.943) in Saudi Arabia (2019) with

a standard deviation of 0.116623

- LE: LE has a mean value of 73.21615, the minimum value (60.851) in

Afghanistan (2010), and the maximum value (84.56) in Japan (2020) with a

standard deviation of 5.393452

- PM: PM has a mean value of 34.6046 the minimum value (5.84) in Brunei

Darussalam (2010) and the maximum value (99.2) in the Philippines (2015) with a

standard deviation of 18.94641

- GNI: GNI has a mean value of 12107.4, the minimum value (390) in Afghanistan

(2021), and the maximum value (91130) in the Philippines (2013) with a standarddeviation of 16477.25

5.2 Model specification

As the standard deviation of the independent variables are extremely large (5.393452,18.94641, and 16477.25) compared to that of the dependent variable (0.116623), theauthors decided to apply natural logarithm to all variables as a countermeasure As a

results, the new vaiables are lnHDI, lnLE, lnPM, and lnGNI.

Set up the hypothesis based on the Breusch-Pagan test at the significance level of 1%:H0: no significant difference across units (no panel effects/no existence of ai)

H1: significant difference across units (panel effect/there is existence of ai)

Using commands “xtreg lnHDI lnLE lnPM lnGNI, re” and “xttest0”, we have the

result:

Breusch and Pagan Lagrangian multiplier test for random effects

lnHDI[code,t] = Xb + u[code] + e[code,t]

Estimated results:

Set up the hypothesis based on the Hausman test at the significance level of 1%:

H0: ai does not correlate with X

Cov (a i , x¿ )=0

H1: ai does correlate with X

Cov(a i ,x¿)≠ 0

Using the combination of commands:

xtreg lnHDI lnLE lnPM lnGNI, fe

est sto fe

xtreg lnHDI lnLE lnPM lnGNI, re

est store re

hausman fe re, sigmaless

The results are obtained as follows:

Test: Ho: difference in coefficients not systematic

Using the command “xtreg lnHDI lnLE lnPM lnGNI, fe”:

Based on the results, the initial estimated regression model is obtained as follows:

lnHD I¿=−4.5199+0.8702lnL E¿−0.0298lnP M¿+0.0646 lnGN I¿+ ^a i + ^u i

5.3 Diagnosing issues of the model

5.3.1 Multicollinearity

Multicollinearity refers to a high degree of correlation or linear relationship between two

or more independent variables in a regression model It indicates that the independentvariables are not independent of each other, but rather, they are highly interrelated.Multicollinearity can pose challenges in regression analysis and affect the interpretationand estimation of the regression coefficients

To test the multicollinearity, we use the “vif” command and get the result:

and misleading predictions Methods to address heteroskedasticity include graphicalanalysis, statistical tests, and techniques like robust standard errors or weightedregression

Heteroskedasticity in regression models can be caused by various factors Some commoncauses include:

 Outliers: Extreme values in the data can lead to heteroskedasticity Theseoutliers may have a disproportionate impact on the variance of the residuals,causing uneven variability

 Missing variables: Omitted variables that are correlated with both thedependent and independent variables can introduce heteroskedasticity If thesevariables are not included in the model, their effects on the error term canresult in varying residual variances

 Nonlinear relationships: If the true relationship between the dependent andindependent variables is nonlinear, but a linear model is used, it can lead toheteroskedasticity The spread of the residuals may vary systematically as thevalues of the independent variables change

 Measurement error: Measurement errors in the independent variables cancontribute to heteroskedasticity If the measurement errors are larger for certainvalues of the independent variables, it can lead to uneven variability in theresiduals

 Time-series data: In time-series analysis, heteroskedasticity can occur when thevariability of the residuals changes over time This can happen due to changingeconomic conditions, seasonality, or other factors that affect the variability ofthe dependent variable

To test the heteroskedasticity, our group uses the Wald test with the command “xttest3”

and got the result below:

Modified Wald test for groupwise heteroskedasticity

in fixed effect regression model

H0: sigma(i)^2 = sigma^2 for all i

Serial correlation, also known as autocorrelation, refers to the correlation betweenconsecutive observations or residuals in a time series data or panel data analysis Itindicates the presence of a systematic relationship or pattern among the residuals overtime.

In a time series, each observation is dependent on previous observations Serialcorrelation occurs when the errors or residuals of a regression model are correlated acrosstime periods, violating the assumption of independence This correlation can manifest as

a positive or negative relationship between the residuals at different lags

Autocorrelation in panel data, which refers to the correlation between the residuals ofdifferent cross-sectional units at different time periods, can have several causes:

 Omitted variable bias: If there are time-invariant variables that are omittedfrom the panel data model, they can introduce autocorrelation These omittedvariables may be correlated with the dependent variable and have a persistenteffect over time, leading to correlation in the residuals

 Time-dependent omitted variables: Autocorrelation can also arise if there aretime-varying variables that are omitted from the model These variables may

be correlated with both the dependent variable and the error term, causingcorrelation in the residuals

 Dynamic relationships: In panel data analysis, there may be dynamicrelationships where the dependent variable in one time period is influenced byits own lagged values or by the lagged values of the independent variables Ifthese lagged effects are not accounted for in the model, it can result inautocorrelation

 Individual-specific effects: Autocorrelation can occur due to specific effects that are not adequately captured by the model Theseindividual-specific effects may persist over time and lead to correlation in theresiduals

individual- Misspecification of the model: If the panel data model is misspecified, such asincorrect functional form or inadequate consideration of heterogeneity, it canintroduce autocorrelation in the residuals

Using Wooldridge test for autocorrelation in panel data with the command “xtserial lnHDI lnLE lnPM lnGNI”, the result is obtained as below:

Wooldridge test for autocorrelation in panel data

- In cross-sectional data, each observation represents a distinct entity or unit andincludes information about various characteristics or variables of those units.Cross-sectional correlation measures the extent to which these variables are related

or move together across the different units in the dataset

- Cross-sectional correlation, or the correlation between different cross-sectionalunits within a specific time period, can arise due to several factors Some commoncauses include:

 Common factors: Cross-sectional correlation can occur when there arecommon underlying factors that influence the variables of interest across thedifferent units For example, in a study of stock returns, the overall marketconditions or macroeconomic factors may impact the returns of differentstocks, leading to cross-sectional correlation

 Spatial or geographic effects: In datasets that involve geographic or spatialunits, cross-sectional correlation can arise due to proximity or sharedcharacteristics of units within a specific area For instance, housing prices inneighboring neighborhoods or regions may exhibit correlation due to similarlocal market conditions

 Group or cluster effects: Cross-sectional correlation can occur whenobservations within specific groups or clusters share common characteristics orare affected by similar factors For example, if individuals are grouped by theiroccupation or industry, there may be correlation in their income levels or job-related variables

 Endogeneity: Endogeneity, or the potential presence of reverse causality oromitted variable bias, can lead to cross-sectional correlation If there areunobserved variables that are correlated with both the dependent andindependent variables, it can induce correlation among the residuals, resulting

in cross-sectional correlation

 Data aggregation: Cross-sectional correlation can also arise when data isaggregated at a higher level, such as averaging or summarizing values acrossgroups Aggregating data can lead to correlation due to the grouping processand the inherent characteristics of the groups

To test the cross-section correlation, set up the hypothesis based on Pesaran test at the significance level of 1%:

H0: the model does not have cross-section correlation

H1: the model does have cross-section correlation

Running the command “xtcsd, pesaran abs”, the results is shown as below:

Pesaran's test of cross sectional independence = 21.704, Pr = 0.0000 Average absolute value of the off-diagonal elements = 0.562

Conclusion: p-value = 0.0000 < 0.01 → reject H0 Therefore, the model does have section correlation

cross-5.4 Estimated model and statistical hypothesis testing

5.4.1 Estimated model

There are 3 problems in the model: heteroskedasticity, serial correlation, and

cross-section correlation Use the command “xtscc lnHDI lnLE lnPM lnGNI” to remedy the

model deficiencies:

Regression with Driscoll-Kraay standard

5.4.2.1 Testing statistical significance of regression coefficient

Set up hypothesis H0, H1 at the significance level of 1%:

H0: Regression coefficients of independent variables do not have statisticalsignificance

βj = 0

H1: Regression coefficients of independent variables have statistical significance

βj ≠ 0

Based on the results of the model, we have:

● Constant coefficient: p-value = 0.000 < 0.01 → Reject H0 Therefore, theregression coefficient of constant coefficient has statistical significance at thesignificance level of 1%

● lnLE: p-value = 0.000 < 0.01 → Reject H0 Therefore, the coefficient isstatistically significance, which means life expectancy doesexplain for the variation of HDI

● lnPM: p-value = 0.000 < 0.01 → Reject H0 Therefore, the coefficient isstatistically significance, which means PM2.5 exposure doesexplain for the variation of HDI

● lnGNI: p-value = 0.000 < 0.01 → Reject H0 Therefore, the coefficient isstatistically significance, which means GNI per capita doesexplain for the variation of HDI

5.4.2.2 Testing overall significance of the model

Set up hypothesis H0, H1 at the significance level of 1%: