Health systems and healthcare have undergone significant progress and change in recent years. The cost of healthcare and health spending have dramatically increased due to these problems on a global scale. But it cannot be sustained. Governments have therefore been looking for measures to lower the price of services in the health sector. In this respect, the major goal of the study was to identify the determinants of health expenditure in Southeast Asia, taking into account ten out of eleven countries.. It was found that GDP per capita and the inflation rate (GDP deflator) were the most important factors affecting health expenditure in Southeast Asia (p < 0.05) and also, that school enrollment rate (tertiary level), unemployment rate and age dependency ratio had not statically significant effect on health expenditure (p > 0.05). This research is intended to focus on socioeconomic factors, therefore there may be other important factors that influence health expenditure that are not captured in the analysis. Thus, it is suggested that the following studies should also take a wider perspective into account. KEYWORDS: Health expenditure; GDP; educational level; inflation; unemployment; age dependent; urbanization
LITERATURE REVIEW
Health expenditure
Health expenditure, as defined by the World Health Organization (WHO), encompasses all costs associated with health services, family planning, nutrition activities, and emergency health aid, while excluding expenses related to drinking water and sanitation.
Our study focuses on per capita total health expenditure, which is defined as the total health spending per person, calculated using the average exchange rate for the respective year in US dollars This metric provides a clear view of health expenditure in relation to the population it serves, enabling effective international comparisons.
Government expenditure on health resources, access, and services, including nutrition, is influenced by a country's wealth and population While higher health spending is linked to improved health outcomes, particularly in low-income nations, there is no universally 'recommended' spending level Generally, countries with higher per capita income tend to invest more in health; however, some nations allocate significantly more or less than expected based on their income A lower proportion of total expenditure on health may suggest that a government does not prioritize health and nutrition.
Determinants of health expenditure
The COVID-19 pandemic has significantly increased global health expenditure, prompting countries to prioritize health-related issues Health spending is influenced not only by economic factors such as income and out-of-pocket expenses but also by sociodemographic elements like population, age, and location Our research specifically examines the connection between socioeconomic factors and total health expenditure.
In their 2020 study, Yetim et al analyzed the predictors of health expenditure in OECD countries from 2000 to 2017 using panel ordinary least squares (OLS) regression Their findings revealed that education level significantly and positively influences health expenditure Conversely, inflation was found to negatively affect total health expenditure, as higher inflation rates lead to decreased purchasing power Additionally, factors such as the unemployment rate and age dependency ratio were determined to have an insignificant impact on health expenditure.
A study by Bedado et al (2022) found a significant link between older age and urban living, leading to increased health expenditures Utilizing cross-sectional study design and logistic regression analysis, the authors determined that urban residents are nearly four times more likely to face out-of-pocket healthcare costs than those living in rural areas.
A study revealed that individuals aged 31–49 are approximately 2.58 times more likely to incur out-of-pocket healthcare expenses compared to those under 31 Similar findings were reported by Matteo et al (1998) and Lopreite and Zhu (2020), although their research primarily focused on individuals over 65 Matteo's analysis of pooled time-series data from Canada's provinces (1965–1991) indicated that the decline in the Canada Health and Social Transfer, coupled with an increasing elderly population, poses a significant challenge for provincial healthcare expenditures, as older adults typically require more specialized care Additionally, Lopreite and Zhu utilized Bayesian-VAR models to demonstrate a strong correlation between the elderly population and health expenditure per capita in China.
Age and urbanization significantly influence health expenditures, as demonstrated by Samadi et al (2013), who found a negative correlation between healthcare spending and the percentage of individuals under 15 and over 65 years old in the long term, as well as urbanization in the short term Countries with a higher proportion of these age groups tend to have healthier populations and, consequently, lower healthcare costs However, contrasting findings by Tian et al (2018) indicate that the relationship between an aging population and healthcare expenditure is not uniform Their study, which analyzed data from 1990 to 2012 using an instrumental variable quantile regression method, revealed that the impact of older populations on healthcare spending varies depending on the specific stage of per capita health expenditure growth in different regions This suggests that the effects of aging on healthcare costs are context-dependent and influenced by the developmental phase of healthcare spending.
Regarding the factor of the unemployment rate, Braendle and Colombier
Research from 2016 indicates a strong positive correlation between public healthcare expenditure and unemployment, which serves as a key indicator of socioeconomic conditions Analyzing data from 1970 to 2012 using dynamic panel estimation methods, the study found that the unemployed tend to have a higher likelihood of illness This inclusion of unemployment in the analysis provides valuable insights into its impact on the growth of healthcare expenditure.
In terms of GDP growth, both Hitiris and Posnett (1992) and Jakovljevic et al.,
In 2020, the importance of Gross Domestic Product (GDP) as a determinant of healthcare expenditure was confirmed Jakovljevic's panel regression analysis revealed that increases in GDP negatively affected current healthcare spending, measured as a percentage of GDP, per capita healthcare expenditure in Purchasing Power Parity (PPP) using constant 2011 international USD, and individual out-of-pocket expenditure (OOPS) per capita in PPP international USD The findings indicate that real GDP growth has a significant and adverse effect on current healthcare expenditure as a percentage of GDP and per capita healthcare spending in constant 2010 USD, both from statistical and economic perspectives.
McKinsey projects that by 2027, the annual US national health expenditure will increase by $370 billion due to inflation, surpassing pre-pandemic forecasts A January 2023 OECD report revealed that health spending across OECD countries rose by nearly 1% of GDP during the pandemic, as governments addressed unexpected public health expenses This crisis underscored the necessity for additional investments to enhance health system resilience, estimated at 1.4% of pre-pandemic GDP on average across OECD nations Furthermore, the report noted that the ongoing war in Ukraine has exacerbated rising energy costs, contributing to inflationary pressures globally.
OECD This has repercussions for the cost of health care, as well as the ability to maintain service levels and address the backlog of care due to the pandemic.
Turgut et al (2017) found a statistically significant positive relationship between the growth rate of total health expenditure and inflation rates, indicating that health expenditure growth outpaces inflation Additionally, health expenditures differ markedly between countries and fluctuate over time within individual nations This variation is largely influenced by factors such as demographics, life expectancy, infant mortality, socio-economic conditions, and the organization of health care systems (Payne et al., 2015).
METHODOLOGY & MODEL SPECIFICATION
Methodology
The research used panel data in ten South East Asian countries from 2000 to
2020 Among eleven South East Asian countries, Timor Leste is excluded from the research due to the shortage of data.
The analysis utilizes secondary data sourced from the World Bank's Open Data website, a reliable repository of information gathered from the statistical systems of member countries The data, collected in 2020, is deemed relatively current in comparison to the year 2023.
The dataset comprises 210 observations, with each observation representing a set of seven figures that correspond to the seven variables in the model Each value of the independent variables is calculated to reflect the status of its respective aspect.
The study deals with collected data using quantitative and qualitative methods.
● Quantitative method is used for analyzing secondary data It will go through a series of testing with Excel and STATA to claim the significance of the coefficients
● Qualitative method is used to make implications From the result of the quantitative method, the study proposes aspects for policymakers and future studies.
The study approaches the relationship between variables using the panel data of
210 observations from 10 South East Asian countries from 2000 to 2020
Panel data enhances the accuracy of parameter estimates by offering greater data variation Additionally, it allows for the direct control and estimation of individual-specific effects, which are not visible in cross-sectional data This ability to account for unobserved heterogeneity is crucial for accurately estimating specific causal effects.
Panel data estimation methods include Fixed Effects (FE) and Random Effects (RE) models The fixed-effect model assumes a single true effect size across all studies, attributing any observed differences in effects solely to sampling error.
The assumptions of the model include: (i) a fixed individual effect rather than a random one, (ii) independence of the independent variables from the individual effect, (iii) a homoscedastic error term that is uncorrelated with both the independent variables and the individual effect, (iv) a linear relationship between the independent variables and the dependent variable, (v) correct model specification with no omitted variables, and (vi) the absence of multicollinearity among the independent variables.
In the random-effects model, we acknowledge that true effect sizes may vary across studies, allowing for the possibility that while some studies may share a common effect size, others may exhibit different effect sizes.
The assumptions of the model include: (i) individual effects are normally distributed, (ii) they are independent of one another, (iii) they share the same variance, (iv) they are uncorrelated with the predictors, (v) the predictors are measured without error, and (vi) the model is correctly specified.
In Chapter III, the Hausman test is utilized to determine whether to use a fixed effects model or a random effects model A statistically significant difference in coefficients between the two models indicates that the fixed effects model is appropriate Conversely, if no significant difference is found, the random effects model may be employed.
Theoretical model specification
Our research identifies six key factors that influence health expenditure The relationship between these independent variables and health expenditure is represented by a linear function, expressed as: \[hexp = \beta_0 + \beta_1 \cdot uni + \beta_2 \cdot unemp + \beta_3 \cdot inf + \beta_4 \cdot agedependent + \beta_5 \cdot urban\]
In the regression model, the intercept, denoted as \$\beta_0\$, represents the slope when all independent variables are set to zero The coefficients \$\beta_1\$ to \$\beta_6\$ correspond to the regression coefficients for various factors: \$\beta_1\$ indicates the impact of the school enrollment ratio at the tertiary level, \$\beta_2\$ reflects the influence of the unemployment rate, \$\beta_3\$ represents the effect of the inflation rate, \$\beta_4\$ pertains to the age dependency ratio, \$\beta_5\$ relates to the urban population rate, and \$\beta_6\$ signifies the contribution of gross domestic product Additionally, \$u\$ denotes the disturbance term in the model.
For making an estimable regression, we used the logarithm form for the model: loghexp = ❑ 0 + ❑ 1 *loguni + ❑ 2 *logunemp + ❑ 3 *loginf + ❑ 4
The model consists of seven variables:
hexp : the dependent variable (regressand)
- Current health expenditure per capita (current US$)
- Measurement: hexp= Total valueof healthcare goods∧service consumed
- School enrollment ratio, tertiary (% gross)
- Measurement: uni= Number of students enrolled∈tertiary education
- Unemployment rate (% of total labor force) (national estimate)
- Measurement: unemp= Number of unemployed
- Inflation rate (i.e GDP deflator) (annual %)
- Measurement: inf = NominalGDP Real GDP ×100
- Age dependency ratio (% of working-age population)
- Measurement: age= Populationagesunder 15+ Population ages
- Urban population ratio (% of total population)
- Measurement: urban= Number of personsresiding ∈urbanareas
- Gross domestic product per capita (current US$)
- Measurement: gdp= ∑ value added by resident producers + product taxes −subsidies
Health expenditure plays a crucial role in enhancing a country's well-being by ensuring access to quality healthcare, fostering economic growth, and promoting social cohesion and equity Various factors influence health expenditure, and our research identifies six key elements that significantly impact this vital area.
The school enrollment ratio is positively linked to health expenditure, as education fosters healthy behaviors, enhances access to health information, and improves individual health decisions Over time, higher education levels lead to better-paying jobs and increased economic opportunities, ultimately contributing to improved health outcomes.
The unemployment rate negatively impacts health expenditure, as joblessness reduces access to employer-sponsored health insurance and the ability to afford medical services Unemployed individuals face a higher risk of poor health outcomes due to increased financial stress, inconsistent healthcare access, and deteriorating mental health.
The inflation rate is anticipated to negatively impact health expenditure, as rising inflation can elevate the costs of medical goods and services This increase may result in higher healthcare expenses for individuals and families with private health insurance, while also placing budgetary strains on public healthcare programs.
The age dependency ratio positively correlates with health expenditure, as nations with a higher ratio typically incur greater healthcare costs This is due to the increased medical services and complex healthcare needs of elderly individuals Consequently, these countries must invest in healthcare infrastructure and workforce to adequately support their aging populations.
The urban population is projected to positively influence health expenditure due to improved access to healthcare facilities Additionally, urban areas experience higher rates of chronic diseases and possess greater income and education levels, leading to an increased demand for healthcare services and subsequently higher health expenditures.
Countries with higher gross domestic product (GDP) typically allocate more resources to healthcare, reflecting a positive relationship between GDP and health expenditure As economies develop and populations become more urbanized, there is often a rise in chronic diseases like diabetes and heart disease, leading to increased healthcare spending.
Data description
Variables Measurements Description Data sources Expected signs hexp Current US$ Current health expenditure per capita WB 2000-2020 (+) uni % gross School enrollment, tertiary WB 2000-2020 (+) unemp
% of total labor force (national estimate)
Between 2000 and 2020, the World Bank reported a decline in unemployment rates and inflation, as indicated by the GDP deflator The age dependency ratio also decreased during this period, reflecting a healthier balance between the working-age population and dependents Additionally, the urban population grew, contributing positively to economic dynamics Overall, GDP and GDP per capita in current US dollars showed an upward trend, highlighting economic growth over the two decades.
Table: Summary of the variables in the model
The sum command was executed to analyze the dataset, yielding results that encompass the number of observations (Obs), the average value (Mean), the standard deviation (Std Dev.), and the minimum (Min) and maximum (Max) values for each variable presented in the table below.
The average health expenditure in Southeast Asia is around \$320, but the large standard deviation indicates significant inequality in growth among the countries in the region.
● School enrollment rate (tertiary level) - uni
The school enrollment rate ranges from the highest value of 93.13477% to the lowest value of 2.37991% The average school enrollment rate is 26.60532% with a standard deviation of 17.9081.
The unemployment rate has the highest value of 11.19% and the lowest value of 0.14% The average unemployment rate is 3.217429% with a standard deviation of 2.019133.
Inflation rate (GDP deflator) - inf
The inflation rate ranges from the highest value of 42.30327% to the lowest value of -22.09142% The average inflation rate is 5.199243% with a standard deviation of 7.331954.
The age dependency ratio ranges from the highest value of 85.75195% to the lowest value of 27.31119% The average age dependency ratio is 50.93853% with a standard deviation of 11.89551.
The urban population rate ranges from the highest value of 100% to the lowest value of 18.586% The average urban population rate is 49.14736% with a standard deviation of 24.13589.
The GDP per capita ranges from the highest value of 66859.3$ to the lowest value of 131.4674$ The average GDP per capita is 9696.818 with a standard deviation of 15530.46$
Running the corr command in STATA, we obtained the table of correlation matrix between variables:
The correlation matrix indicates a range of correlations among the variables, predominantly leaning towards high values, which may lead to multicollinearity issues To mitigate these problems, Karl Pearson's research suggests maintaining a correlation coefficient below 0.7 or 0.8.
● r(hexp, uni) is 0.7008, which is high The positive coefficient implies that school enrollment rate has a positive effect on health expenditure.
● r(hexp, unemp) is 0.1560, which is quite low The positive coefficient indicates that unemployment rate and health expenditure have a positive relationship.
● r(hexp, inf) is -0.2413, which is medium The coefficient is negative,suggesting that inflation rate has a negative effect on health expenditure value.
● r(hexp, agedependent) is -0.6161, which is quite high The negative coefficient implies that age dependency ratio and health expenditure have a negative relationship.
● r(hexp, urban) is 0.7742, which is high The coefficient is positive, indicating a positive relationship between health expenditure and urban population rate.
● r(hexp, gdp) is 0.9452, which is high The coefficient is positive, indicating a positive relationship between health expenditure and GDP per capita.
QUANTITATIVE ANALYSIS
Model selection: fixed effect model (FE) and random effect model (RE)
The first test is Breusch-Pagan check to see whether the model has the factor ai or not.
The p-value of 0.0000, which is below the standard significance level of 0.05, allows us to reject the null hypothesis, indicating that the model includes the factor ai Consequently, for quantitative analysis, the model should employ either the Fixed Effects (FE) or Random Effects (RE) model.
The second test we use is the Hausman test in order to decide one method between the FE and RE model
The result came out that we should use the FE model due to its p-value exceeding the significance level of 0.05, meaning that it will bring about better results.
Regression model
Along with the selection above, we have the following model: loghexp = ❑ 0 + ❑ 1 *loguni + ❑ 2 *logunemp + ❑ 3 *loginf + ❑ 4 *logagedependent + ❑ 5
The analysis utilizes the STATA software to execute the xtreg command, focusing on key logarithmic variables: loghexp, representing the logarithm of current health expenditure per capita; loguni, indicating the logarithm of the tertiary school enrollment rate; logunemp, which denotes the logarithm of the unemployment rate; loginf, reflecting the logarithm of the inflation rate; logagedependent, capturing the logarithm of the age dependency ratio; logurban, representing the logarithm of the urban population rate; and loggdp, which signifies the logarithm of GDP per capita.
From the above results, we have the following regression model: loghexp = -0.7487416 - 0.0531158*loguni - 0.0664905*logunemp -
Diagnosing the problems of the model
The study employs a fixed effects model (FE) to analyze the data, which inherently accounts for the influence of unobserved variables (ai) Consequently, if a significant variable is omitted from the model, it is also treated as an unobserved variable, eliminating the need for testing for variable omission.
To check for multicollinearity, we use command vif:
The analysis reveals that the Mean Variance Inflation Factor (VIF) is 5.46, indicating no multicollinearity issues, as it is below the threshold of 10 However, the VIF for the urban variable exceeds 10, suggesting a significant correlation that must be considered in subsequent testing.
Testing the normal distribution of u with the command sktest e in STATA, we have the results:
We see that: p-value = 0.3201 > 0.05 With this, we conclude that the model has normal distribution.
We will use the Modified Wald test for groupwise heteroskedasticity in fixed effect regression model:
According to the results of the test, p-value = 0.0000 < 0.05, so we reject H0. This means there is heteroskedasticity in the model
Heteroskedasticity in fixed effects regression models arises from data variability or model misspecification, resulting in biased coefficient estimates, inaccurate standard errors, and flawed hypothesis tests.
We will use Wooldridge test for autocorrelation in panel data:
According to the results of the test, p-value = 0.0004 < 0.05, so we reject H0. With this, it is concluded that the model has serial correlation.
We will use Pesaran's test of cross sectional independence:
The result shows that p-value = 0.3642 > 0.05 We cannot reject H0 and the model is implied to have no cross-section correlation.
Correcting model
The model has these two existing problems: heteroskedasticity and serial correlation In order to fix this, we will use the clustering method.
And here is the result of the new cluster model:
The cluster method has not altered the coefficients, indicating that heteroskedasticity impacts only the t and F-statistics, while the model effectively addresses the underlying issues.
Hypothesis test
According to the corrected model: p-value = 0.0000 < 0.05 (5% level of significance)
We reject the null hypothesis, indicating that the independent variables collectively account for the variation in the value of logexp Consequently, we conclude that the model is significant at the 5% level.
Looking at each of the independent variables’ p-values, we can also draw some other insights:
● Only the variables loginf and loggdp has p-value lower than 0.05, indicating a strong evidence that these factors are individually and significantly related to the outcome variable with is loghexp.
The analysis revealed that the variables loguni, logunemp, logagedependent, and logurban each had p-values exceeding 0.05, indicating a lack of strong evidence for a significant individual relationship with the dependent variable.
Result analysis
3.6.1 Statistical meaning of the regression coefficient
When controlling for factors such as tertiary school enrollment rate, unemployment rate, inflation rate, age dependency ratio, urban population rate, and GDP per capita, the average health expenditure per capita is projected to be -0.7487416, indicating the influence of external factors not accounted for in the model.
When the unemployment rate, inflation rate, age dependency ratio, urban population rate, and GDP per capita remain constant, a 1% increase in the tertiary school enrollment rate leads to a decrease of approximately 0.0531% in average health expenditure per capita.
When controlling for factors such as tertiary school enrollment rate, inflation rate, age dependency ratio, urban population rate, and GDP per capita, a 1% increase in the unemployment rate is associated with a decrease of approximately 0.0665% in average health expenditure per capita.
When the tertiary school enrollment rate, unemployment rate, age dependency ratio, urban population rate, and GDP per capita remain constant, a 1% increase in the inflation rate results in a 0.0318644% decrease in average health expenditure per capita.
When the tertiary school enrollment rate, unemployment rate, inflation rate, urban population rate, and GDP per capita remain constant, a 1% increase in the age dependency ratio leads to a 0.0671363% rise in average health expenditure per capita.
When the tertiary school enrollment rate, unemployment rate, inflation rate, age dependency ratio, and GDP per capita remain constant, a 1% increase in the urban population rate leads to a 0.820564% decrease in average health expenditure per capita.
When the tertiary school enrollment rate, unemployment rate, inflation rate, age dependency ratio, and urban population rate remain constant, a 1% increase in GDP per capita leads to an average health expenditure per capita increase of approximately 1.061579%.
3.6.2 The model’s coefficient of determination
The within R-squared value of 0.9443, or 94.43%, demonstrates that the independent variables account for a significant portion of the variation in the dependent variable within each group, indicating that the model effectively captures this variation.
The R-squared value of 0.9694, or 96.94%, demonstrates that the independent variables account for a significant portion of the variation in the group means of the dependent variable This indicates that the model effectively captures the differences in the dependent variable across the groups defined by the grouping variable.
The R-squared value of 0.9583, or 95.83%, demonstrates that the independent variables account for a significant portion of the variation in the dependent variable across all groups, indicating that the model effectively captures the overall variation.
High R-squared values indicate that the independent variables in the model effectively account for a significant portion of the variation in the dependent variable, both within and across groups, as well as in the overall analysis.
3.6.3 Interpretation of the estimated results obtained
According to the research results presented in the report, the following conclusions can be drawn.
In Southeast Asia, there is an inverse relationship between tertiary school enrollment and health expenditure per capita; as enrollment rates rise, health spending per person tends to decline, contrary to the anticipated positive correlation between the two factors.
An increase in the unemployment rate leads to a decrease in a country's health expenditure, and conversely, a decrease in unemployment results in higher health spending, assuming all other factors remain constant This relationship aligns with initial expectations regarding the variable.
● As the inflation rate increases, we will witness a decrease in the health expenditure per capita, and vice versa (with all other factors constant), which is as expected.
As the age dependency ratio rises, health expenditure per capita is expected to increase, contrary to the anticipated negative correlation between these two factors, assuming all other variables remain constant.
As urban populations grow, health expenditure per capita is expected to rise, contrary to the anticipated negative correlation between these two factors, assuming all other variables remain constant.
● As the GDP per capita increases, we will witness an increase in the health expenditure per capita, and vice versa (with all other factors constant), which is as expected.