The Demographic Dividend or the Education Dividend? Evidence from China’s Economic Growth

Abstract

Developing countries face a significant challenge in sustaining their demographic dividend. However, there are few existing studies that approach this issue from a macroeconomic perspective or through empirical research. This paper aims to systematically analyze the impact of the demographic dividend and education dividend on economic growth. By utilizing China’s empirical evidence and employing the System GMM method, we explore how to improve both dividends. Our main findings can be summarized into three aspects. First, the demographic dividend does not depend on population size alone but also on the size of the labor force. Second, education can improve the demographic dividend and thereby prove the existence of an education dividend. Finally, the effects of the demographic and education dividends vary across regions and stages of development within developing countries. This research enriches the existing literature on education and population age in developing countries.

1. Introduction

The demographic dividend has long been as an important explanation for the high economic growth of developing countries [1,2,3]. Some research has found that developing countries have a high birth rate and a large young and middle-aged population, which leads to lower labor costs and makes their manufacturing industries more competitive in the international trade. Scholars have called this phenomenon the “demographic dividend” of developing countries. Studies have shown that the demographic dividend is widespread in developing countries in East Asia [4,5], South Asia [6], Latin America [7], and Africa [8]. However, with high economic growth, these developing countries are beginning to suffer from an aging population, rising wage costs, and declining fertility rates, which could lead to a reduction or even the disappearance of the demographic dividend [9,10]. The sustainability of the demographic dividend has become an important challenge for developing countries.

At the same time, the level of education in developing countries is gradually rising, which is considered to be an important factor in economic transformation and sustainable development [11,12]. Therefore, some researchers have suggested that the demographic dividend in developing countries has declined, but the development of education will contribute to economic growth by an increase in human capital [13,14,15], which is known as the “education dividend”. However, other researchers have argued that education does not always lead to economic growth [16]. The education dividend needs to occur in economies with a suitable industrial structure [17]. In developing countries especially, which have a limited demand for highly skilled labor, the development of education may lead to expensive education expenditure and structural unemployment [18,19].

In reality, demographic changes in developing countries often coincide with the development of education. Although studies have looked at the role of age structure and education on economic growth, there are still some research gaps [20,21,22]. In detail, the following questions deserve in-depth discussion. First, is it the demographic dividend or the education dividend that has an impact on economic growth for developing countries? Second, what are the theoretical economic mechanisms by which the demographic dividend and the education dividend drive economic development? Third, will the demographic and education dividends co-exist in developing countries?

To answer the above questions, we constructed a model of economic growth with the age structure and education factor to analyze the theoretical mechanisms by which the demographic and education dividends affect economic growth. China, as one of the world’s largest developing economies, is also facing the challenges of the aging problem. As shown in Figure 1, the proportion of China’s population over 65 years of age has risen sharply from 2002 to 2020 onwards, which has led to a number of problems such as rising labor costs, pension insurance and public health care.

At the same time, China’s educational standards were raised with the start of the universal compulsory education and university expansion policy at the beginning of the 21st century (Figure 2). For example, China’s high-school education rate has tripled from 17.16% in 2002 to 32.64% in 2020; its college education rate has tripled from 4.71% to 16.51%. By 2020, China already had the largest number of university students in Asia, which has resulted in a relatively cheap and highly qualified workforce compared to developed economies (e.g., Japan, Republic of Korea). Thus, the case study from China can provide an ideal case study for examining the relationship between the demographic dividend and the education dividend. In detail, by using the panel data of Chinese provinces from 2002 to 2020, we introduce a SYS-SMM empirical model to explore the impact of education on the demographic dividend.

This research could enrich studies on education and the age of the population in developing countries. In addition, empirical studies on China’s economic growth will provide insights into the handling of population and education policies in developing countries. The subsequent chapters are organized as follows: Section 2 is a literature review on related research. Section 3 is the theoretical framework. Section 4 describes the method and data used in this paper. In Section 5, we will introduce provincial panel data from China to test the theoretical hypotheses in Section 3. Furthermore, we will further discuss the spatial and temporal heterogeneity of the demographic dividend and the education dividend in Section 6. Section 7 is the conclusion and policy implications.

2. Literature Review

2.1. Studies on Economic Growth and the Demographic Dividend

In recent years, as population issues have become a hot topic globally, the relationship between the demographic dividend and economic growth in developing countries has attracted much attention. The demographic dividend plays an important factor in the rise and fall of developing countries.

On one hand, the demographic dividend refers to the demographic shift caused by a decrease in birth rates and death rates, resulting in a period where the number of employed workers exceeds the number of non-employed individuals [23,24]. The increase in the number of employed workers can promote economic development and improve national income levels [25,26]. Research has shown that the demographic dividend is one of the important driving forces for economic development, leading to high-speed economic growth [1,2,3].

Developing countries, as one of the regions with the world’s largest population, have enormous potential for demographic dividend [27]. For example, China implemented the “family planning” policy in the 1980s, which led to a demographic shift and a period of demographic dividend. In the early decade of the 21st century, the Chinese economy grew at an average rate of about 10% per year, becoming one of the fastest growing economies in the world [28,29]. An important reason for China’s higher economic growth during this period was the manufacturing cost advantage of a large labor force [30,31].

However, on the other hand, the demographic dividend is not unconditional and its realization requires the government to formulate reasonable policies related to population, education, labor force, etc. [16,17]. For example, India has faced serious employment issues while experiencing rapid population growth. The government has formulated policies to promote education and improve labor skills to better utilize the demographic dividend [32,33,34]. In addition, with the aggravation of population aging and the decline in birth rates, the demographic dividend gradually disappears and may even turn into a population burden [28]. Therefore, developing countries need to respond early to this trend, promote industrial upgrading, innovate technology, improve labor productivity, and maintain economic growth and social stability [9,10]. Therefore, how developing countries can prolong the demographic dividend window and eliminate the negative impact of the declining demographic dividend has become a common issue of concern for development economics, demographic economics, and macroeconomics at present [35]. All the above theoretical framework can be summarized as Figure 3.

2.2. Study on Demographic Dividend and Education Dividend

With the rapid growth of the world’s population, education has gradually become a hot topic globally. In this context, the relationship between the demographic dividend and education has attracted much attention. This article provides a literature review on the relationship between the demographic dividend and education from multiple perspectives.

Firstly, education is one of the key factors in unlocking the potential of the demographic dividend [11,12]. By raising the knowledge and skill levels of the educated population, the quality of the labor force can be improved, thus promoting economic development [15,16]. For example, studies show that improving basic education in Nigeria has allowed the country’s demographic dividend to transform into more sustainable economic growth [36,37].

Secondly, the level of education also has a crucial impact on the realization of the demographic dividend [11]. Countries with higher levels of education benefit more from the demographic dividend [1]. For example, China implemented its family planning policy in the 1980s and simultaneously promoted basic education, which provided a good education background for its employed population and laid a solid foundation for its rapid economic development [28,29].

Thirdly, education can also promote social fairness and inclusive growth. Education can eliminate inequality, expand opportunities, and improve social welfare [38,39,40]. For example, after promoting universal education in different regions and encouraging female education, India achieved a more equitable and inclusive demographic dividend [41,42].

However, the contribution of education to the demographic dividend is not always consistent. On the one hand, the demographic dividend requires a better education system to meet labor demand [18]; on the other hand, population burden may lead to excessive pressure on the education system and affect education quality [19]. At the same time, the relationship between education and the demographic dividend is also influenced by many other factors, such as government investment, technological innovation, vocational training, etc. [18,26,43].

In summary, there is an inseparable connection between education and the demographic dividend. Improving and enhancing education can promote the realization of the demographic dividend, thus promoting sustained economic development and social equity progress. However, it should be noted that the relationship between education and the demographic dividend is complex and requires joint efforts from the government and all aspects of society to achieve more sustainable demographic dividend benefits.

2.3. Existing Research Gaps and Marginal Contributions of This Paper

Demographics, education levels, and economic growth are important factors in developing countries, which have received extensive attention from relevant scholars. However, existing studies have some shortcomings and gaps in these issues and we summarize them in the following aspects.

First, one of the problems in existing studies is the lack of in-depth research on demographic change. Although studies have shown that population aging has a negative impact on economic growth, the causes and mechanisms behind demographic changes are still under-researched. For example, population aging is the result of a combination of factors such as declining fertility, increasing longevity, and population migration, which need more mechanism analysis.

Second, existing studies remain controversial in exploring the relationship between education levels and economic growth. On the one hand, some studies suggest that the improvement of education level can promote technological innovation and economic growth. However, on the other hand, some studies suggest that the contribution of education level to economic growth is not significant and may even have a negative impact. These controversies may be explained by the different definitions and measurements of the concept of “educational attainment” and the lack of understanding of the complex relationship between education and economic growth. Different theoretical perspectives require more empirical evidence.

Finally, existing studies have often ignored the influence of macroeconomic environment and institutional factors when exploring the relationship between demographics, educational attainment, and economic growth. For example, factors such as labor force structure adjustment and industrial structure transformation may contribute more to economic growth than changes in demographic structure and education level. Therefore, researchers should pay more attention to the analysis of institutional factors and the macroeconomic environment in order to explore the relationship between demographic structure, education level, and economic growth more comprehensively.

In order to fill these research gaps, this paper has constructed a model of economic growth with the age structure and education factor to analyze the theoretical mechanisms by which the demographic and education dividends affect economic growth. What is more, by using the panel data of Chinese provinces from 2002 to 2020, we introduce a SYS-SMM empirical model to explore the impact of education on the demographic dividend. This research could enrich studies on education and the age of the population in developing countries. In addition, empirical studies on China’s economic growth will provide insights into the handling of population and education policies in developing countries.

3. Theoretical Framework

3.1. Affect Mechanisms of Demographic and Education Dividends

Before constructing the theoretical mechanism of education affecting demographic dividend, the concept of demographic dividend studied in this paper should be clearly defined. In this paper, demographic dividend is defined as the traditional demographic dividend, which is the influence of population age structure on economic growth and the demographic dividend obtained by making full use of population opportunities through matching policies within the window of population opportunities.

In the 20th century, especially since the 1960s, the demographic structure shifts taking place in almost every country in the world have coincided with major educational expansions. Although there are more or less regional differences, all these differences show a general improvement in education [28,29]. The effect of the age structure of the population on economic growth is likely to be influenced by the increase in the education level of the population. From the perspective of the basic principle of population economics, only the employed population can create value and can constitute the source of “dividends”. Therefore, the demographic dividend is created by the working population of the working-age population after the opening of the population opportunity window. The improvement of the demographic dividend through education is mainly through the following ways:

The positive effects of education on health can improve the population age structure. Many scholars have found that there is a strong positive correlation between health and years of education. People with higher education levels are more likely to invest in the health of their spouse, children, and themselves, resulting in the improved survival of their children and longer life expectancy of their spouse and themselves. Meanwhile, the rate of dementia and restriction of activities of daily living (ADL) is lower without obvious deficits of cognitive decline. In addition, according to the general equilibrium fertility rate and human capital model, human capital investment can significantly reduce the mortality rate of young people, thus promoting demographic transition and economic growth. These all prevent adults from leaving early and minors from being unable to enter the working-age population, thereby increasing the working-age population proportion.

Education can be seen as a key trigger for declining fertility. From the perspective of unified growth theory, fertility rate and education dynamics are triggered by changes in skill demand rooted in economic and technological environment. The decline of fertility rate is closely related to the improvement of educational level, and the mutual causal relationship between the improvement of education level and the decline of fertility rate has been verified empirically [30,31]. Higher levels of education in the adult population, especially among women, generally enable them to better pursue their own preferences. They may have greater autonomy in reproductive decisions, have more knowledge and access to contraception and delay marriage and childbearing, and place greater emphasis on the quality rather than the quantity of children, leading to lower fertility rates. Bloom and Canning et al. found that the decline of fertility rate would increase the participation rate of the female labor force [21], and used the panel data of 97 countries from 1960 to 2000 to investigate the impact of the decline of fertility rate on women’s participation in the labor market, finding that every time a woman gave birth to one less child, labor participation time can be increased by two years.

Education can increase the labor force employment rate by facilitating labor mobility. Labor mobility is an important way to realize the optimal allocation of labor resources, and in order to seek a higher utility level, the mobility decision made by labor—which is based on cost-benefit comparison before and after mobility—may be affected by education. First, it can increase the ability of labor mobility. The flow of the labor force usually has a certain threshold of human capital. It is a common occurrence that workers are willing to flow but subject to the threshold of human capital. Education is an important form of human capital investment, which plays an extremely important role in the improvement of the human capital level of the labor force. The improvement of education level can promote the mobility of labor by improving their human capital level, and then match jobs in a broader labor market. Second, it can increase the expected income of labor flow. In most countries, the level of education tends to influence or to some extent determine the level of income. This is based on the human capital theory, which highlights that human capital is the key factor determining individual income in a perfectly competitive labor market [44]. The expected income level is the most direct factor to attract labor mobility; the increase in expected income level can promote labor mobility. Third, it can reduce the cost of labor mobility. Relevant studies show that the psychological cost of labor generally decreases with an increase in education level [45]. The decrease in flow cost may also promote the flow of the labor force, which is conducive to the market-oriented allocation of labor force factors, and thus promotes the improvement of the labor employment rate.

3.2. Mathematical Modeling

In order to describe the effect of the demographic dividend and the education dividend to economic growth, we set a macroeconomics growth model referring to Gregory et al. (1992) [46] and Benhabib et al. (2005) [47]. The basic assumptions of the model are the production of a single, good, full employment, and the production function is the ordinary Cobb–Douglas Production Function. Technology is combined with human capital (labor-increasing type) and returns to scale are constant. The specific production function is shown in Equation (1).

$$Y_{it}=K_{it}{}^{\alpha } {[A_{it}{H}_{it}]}^{\beta }$$

(1)

In Equation (1), $Y_{it}$ is the total economic output, $K_{it}$ is the input of capital, $A_{it}$ is the technical level, $H_{it}$ is the human capital ($H_{it}=L_{it}\times G\left(E\right)=L_{it}\times e^{\phi E}$, $L\left(t\right)$ is the normal labor force, E is the average years of education, $\phi$ is the rate of return on education), $\alpha$ is the output elasticity of capital, and $\beta$ is the output elasticity of effective labor. According to the assumption of the constant scale return, $\alpha +\beta =1$. The basic model is extended as:

$$Y_{it}=K_{it}{}^{\alpha } {[A_{it}{L}_{it}\times e^{\phi E}]}^{\beta }=K_{it}{}^{\alpha }{A}_{it}{}^{\beta }{L}_{it}{}^{\beta }{e}^{\phi E\beta }$$

(2)

Then, dividing both sides by $L_{it}$, we can obtain the GDP per labor force:

$$y_{it}=k_{it}{}^{\alpha }{A}_{it}{}^{\beta }{e}^{\phi E\beta }$$

(3)

Noting that the per capita GDP is $y_{it}{}^p$, the relationship between $y_{it}$ and $y_{it}{}^p$ is as follows:

$$y_{it}{}^p=\frac{L_{it}}{N_{it}}\times y_{it}=\frac{W_{it}}{N_{it}}\times \frac{P_{it}}{W_{it}}\times \frac{L_{it}}{P_{it}}\times y_{it}$$

(4)

wherein $\frac{W_{it}}{N_{it}}$ represents the proportion of working-age population in the total population and is the proxy variable of population age structure; $\frac{P_{it}}{W_{it}}$ is the labor force participation rate; and $\frac{L_{it}}{P_{it}}$ is the employment rate of the labor force. Putting Equation (4) into Equation (3), taking the logarithm of both sides of the new formula at the same time, and calculating the difference, we can obtain:

$$\mathrm{\Delta }ln{y}_{it}{}^p=\alpha \mathrm{\Delta }ln{k}_{it}+\beta \mathrm{\Delta }ln{A}_{it}+\mathrm{\Delta }\phi E\beta +\mathrm{\Delta }ln{L}_{it}-\mathrm{\Delta }ln{N}_{it}$$

(5)

According to the research of Loening (2004) [43] and Akhvlediani et al. (2020) [48], the technical level of a region $A_{it}$ is regarded as a linear function of the GDP per labor force of one period. Referring to Benhabib and Spiegel (1994, 2005) [47,49], it is assumed that the improvement of human capital stock (average years of schooling) is the main factor promoting technological innovation. Then, by extending $\mathrm{\Delta }ln{A}_{it}=\delta +\mu ln{y}_{i,t-1}+\rho E_{i,t-1}=\delta +\rho E_{i,t-1}-\mu ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)-\mu ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)-\mu ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)+\mu ln{y}_{i,t-1}{}^p$, and substituting it into Equation (5), we can obtain:

$$\begin{gathered} \mathrm{\Delta }ln{y}_{it}{}^p=\alpha \mathrm{\Delta }ln{k}_{it}+\beta \delta +\beta \rho E_{i,t-1}-\beta \mu \mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)-\beta \mu \mathit{ln}\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)-\beta \mu \mathit{ln}\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right) \\ \hspace{1em}\hspace{1em}+\beta \mu ln{y}_{i,t-1}{}^p+\mathrm{\Delta }\phi E\beta +\mathrm{\Delta }ln{L}_{it}-\mathrm{\Delta }ln{N}_{it} \end{gathered}$$

(6)

Further, we can introduce the interaction terms in Equation (6) to describe the interplay between the demographic dividend and the education dividend, and we finally construct the theoretical model shown as Equation (7).

$$\begin{array}{ll}\mathrm{\Delta }ln{y}_{it}{}^p=\alpha \mathrm{\Delta }ln{k}_{it}& +\beta \delta +\beta \rho E_{i,t-1}-\beta \mu \mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)-\beta \mu \mathit{ln}\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)-\beta \mu \mathit{ln}\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)\\ & +\beta \mu ln{y}_{i,t-1}{}^p+\beta \phi \mathrm{\Delta }E+\mathrm{\Delta }ln{L}_{it}-\mathrm{\Delta }ln{N}_{it}+\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\\ & +\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }E+\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times \mathit{ln}\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)\\ & +\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times \mathit{ln}\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)+\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }E\times \mathit{ln}\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)\\ & +ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }E\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)\end{array}$$

(7)

According to Equation (7), we can analyze the age structure of the population and the impact of education level on economic growth. In the next section, we will construct the corresponding empirical model based on Equation (7) and test the existence of the demographic and education dividends through the positive and negative coefficients in the model.

4. Method and Data

4.1. Empirical Method

Based on the analysis in Section 2, we constructed corresponding empirical models to test the role of the demographic dividend and the education dividend in China’s economic growth. Meanwhile, we introduced the interaction terms in order to answer the question of whether the demographic dividend and the education dividend co-exist. The empirical model is set up as follows:

$$\begin{array}{ll}ln{y}_{it}{}^p={\vartheta }_0& +{\vartheta }_{1}ln{y}_{i,t-1}{}^p+{\vartheta }_2\mathrm{\Delta }ln{k}_{it}\\ & +{\vartheta }_3\mathrm{\Delta }ln{L}_{it}+{\vartheta }_4\mathrm{\Delta }ln{N}_{it}+{\vartheta }_5{E}_{i,t-1}+{\vartheta }_{6}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)+{\vartheta }_{7}ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)+{\vartheta }_8\mathit{ln}\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)\\ & +{\vartheta }_9\mathrm{\Delta }{E}_{it}+{\vartheta }_{10}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}+{\vartheta }_{11}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}+{\vartheta }_{12}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\\ & \times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)+{\vartheta }_{13}\mathit{ln}\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times \mathit{ln}\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)+{\vartheta }_{14}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\\ & \times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)+{\vartheta }_{15}ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)+\gamma ln{X}_{it}+{\mu }_i+{\theta }_t+{\epsilon }_{it}\end{array}$$

(8)

In Equation (8), i represents the region, t represents the current time, t − 1 represents one time period lag. The explained variable $y_{it}{}^p$ is regional total output value per capita. $y_{i,t-1}{}^p$ is the first-order lag of real regional total output value per capita. Explanatory variables include capital input ($k_{it}$) measured by per capita capital stock. Because there are no official data on capital stock in China, we have calculated it based on the perpetual inventory method (PIM) according to Dey-Chowdhury (2008) [50], Wu et al. (2014) [51], Passas (2023) [52]. $L_{it}$, $N_{it}$, $E_{i,\mathrm{t}}$ represent labor force size, population size, and education level, respectively. ${\mu }_i$ is the fixed effect of region; ${\theta }_t$ is the fixed effect of time; ${\epsilon }_{it}$ is the random disturbance term; and ${\vartheta }_i$, $\gamma$ are the parameters to be estimated in the model.

When calculating partial derivatives of $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$ (population age structure) on both sides of Equation (8), the marginal influence of population age structure on economic growth (demographic dividend) can be written as:

$$\begin{gathered} \frac{\partial ln{y}_{it}{}^p}{\partial ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)}={\vartheta }_6+{\vartheta }_{10}\times E_{i,t-1}+{\vartheta }_{11}\times \mathrm{\Delta }{E}_{it}+{\vartheta }_{12}\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right) \\ +{\vartheta }_{13}\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)+{\vartheta }_{14}\mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)+{\vartheta }_{15}\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right) \end{gathered}$$

(9)

In Equation (9), ${\vartheta }_{10}$ is the direct influence coefficient of education level on the demographic dividend; ${\vartheta }_{11}$ is the direct influence coefficient of education level improvement on the demographic dividend; ${\vartheta }_{12}$ is the coefficient of education level influence on the demographic dividend through labor participation rate; ${\vartheta }_{13}$ is the coefficient of education level influence on the demographic dividend through labor employment rate; ${\vartheta }_{14}$ is the coefficient of education level improvement speed influence on the demographic dividend through labor participation rate; and ${\vartheta }_{15}$ is the coefficient that affects the demographic dividend through labor employment rate.

The regressions include lagged terms for variables such as GDP, population, and education levels, and these variables are correlated in such a way that using a static panel regression would make the estimated coefficients biased. Therefore, with reference to the existing studies [53,54,55,56], the systematic GMM (SYS-GMM) approach was used for estimation in this paper.

4.2. Data

In this paper, the panel data of 30 provinces (Tibet was excluded due to lack of data) of the China mainland from 2002 to 2020 were used for research. Data of GDP, population size, and labor force size came from <China Statistical Yearbook>, <China Statistical Yearbook of Population>, <China Statistical Yearbook of Population and Employment>. We used average years of schooling as a proxy variable for educational attainment, which came from <China Statistical Yearbook of Education> and provincial statistical yearbooks.

The control variable groups ($X_{it}$) include: financial expenditure ratio ($financ{e}_{it}$) measured by the proportion of financial expenditure in GDP; trade openness ($trad{e}_{it}$) measured by the ratio of total import and export volume to GDP, scientific, and technological development level; and ($te{c}_{it}$) measured by the number of domestic invention patent applications accepted (items). Control variables were from <China Statistical Yearbook> and <China Statistical Yearbook of Science and Technology>. Considering the impact of inflation on nominal GDP, all the price-related variables have been taken from 2002 as the base period to eliminate the price effect. The statistical characteristics of each variable are shown in

Table 1. Descriptive Statistics.

Variable	Mean	Std. Dev.	Min	Max
$ln{y}_{it}{}^p$	9.956	0.711	8.035	11.752
$ln{y}_{i,t-1}{}^p$	9.919	0.704	8.035	11.738
$\mathrm{\Delta }ln{k}_{it}$	0.124	0.133	−0.97	0.48
$\mathrm{\Delta }ln{L}_{it}$	−0.036	0.094	−0.671	0.214
$\mathrm{\Delta }ln{N}_{it}$	0.006	0.015	−0.057	0.101
$E_{i,t-1}$	8.671	1.037	6.04	12.782
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$	4.29	0.05	4.15	4.429
$ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	4.344	0.131	3.986	4.677
$ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	4.569	0.008	4.538	4.605
$\mathrm{\Delta }{E}_{it}$	0.095	0.244	−0.747	1.032
$lnfinanc{e}_{it}$	3.02	0.427	2.093	4.328
$lntrad{e}_{it}$	2.92	0.983	−0.271	5.142
$lnte{c}_{it}$	8.554	1.774	3.689	12.285

.

5. Empirical Results

5.1. Baseline Results

A dynamic panel can better solve the endogeneity problem of a traditional panel data fixed-effect model. Therefore, the SYS-GMM model was adopted in this paper. Multicollinearity is a common problem in macro-econometric models. Considering the impact of multicollinearity on the accuracy of regression results, we carried out first-order difference and VIF tests on variables and show the correlation coefficients. The test results show that the VIF of all variables after the first-order difference is less than 30 (most are less than 10), indicating that multicollinearity has been alleviated to a certain extent (see Appendix A).

The baseline regression results are shown in

Table 2. SYS-GMM Regression Result.

Variable	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
$ln{y}_{i,t-1}{}^p$	0.973 ***	0.964 ***	0.963 ***	0.994 ***	0.998 ***	0.975 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$	0.228 ***	1.129 **	1.116 **	1.699 ***	2.190 **	2.556 **
	(0.001)	(0.035)	(0.029)	(0.000)	(0.011)	(0.021)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$		−0.106 *	−0.110 *	−0.414 ***	−0.426 ***	−0.413 ***
		(0.054)	(0.055)	(0.000)	(0.000)	(0.002)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$		−0.114	−0.113	−0.080	−0.097	−0.091
		(0.357)	(0.357)	(0.227)	(0.172)	(0.577)
$\mathrm{\Delta }ln{L}_{it}$	0.411 ***	0.407 ***	0.407 ***	0.365 **	0.244 *	0.244 *
	(0.000)	(0.000)	(0.000)	(0.012)	(0.085)	(0.084)
$E_{i,t-1}$	0.058	0.015	0.020 *	0.024 **	0.025 ***	0.021 ***
	(0.175)	(0.114)	(0.088)	(0.084)	(0.001)	(0.007)
$\mathrm{\Delta }{E}_{it}$	0.022	0.045	0.157	0.086	0.048	0.050
	(0.662)	(0.575)	(0.803)	(0.792)	(0.313)	(0.294)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$			−0.081 *	−0.072 *	−0.026 *	−0.013 *
			(0.078)	(0.054)	(0.084)	(0.058)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$				−0.495 ***	−0.557 ***	−0.483 ***
				(0.000)	(0.000)	(0.005)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$					−0.015	−0.011
					(0.508)	(0.964)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$						−0.086
						(0.606)
Obs	510	510	510	510	510	510
AR(1)	(0.000)	(0.000)	(0.000)	(0.001)	(0.002)	(0.001)
AR(2)	(0.432)	(0.586)	(0.468)	(0.428)	(0.732)	(0.584)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes

Note: *, ** and *** are significant at the level of 10%, 5% and 1%, respectively, with p values in brackets.

. Model (1) is the regression result without introducing interaction terms, and Models (2) to (6) are the regression result with the introduction of interacting terms $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$, $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$, $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$, $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$, $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$, $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right),$ and other interaction variables on the basis of Model (1). According to the results of the Arellano–Bond test, the AR (1) of each dynamic panel model rejects the null hypothesis that the disturbance term has no autocorrelation at the significance level of 1%. In addition, the AR (2) cannot reject the null hypothesis, which indicates that the disturbance term has first-order autocorrelation, but there is no second-order autocorrelation, thus satisfying the prerequisite of dynamic panel estimation.

Based on the results of the baseline regression, we can obtain three conclusions on the demographic dividend and the education dividend. First, the results show that the demographic dividend and the education dividend exist in developing economies. In all six models, the coefficients of $\mathrm{\Delta }ln{L}_{it}, E_{i,t-1}, ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right), ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right), ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$ are all positive, indicating that education reduces the influence of population age structure on economic growth. In particular, $\mathrm{\Delta }ln{k}_{it}$ and $lnfinanc{e}_{it}$ have no significant impact on economic growth. The reason why capital accumulation speed has no significant impact on economic growth may be that the law of diminishing marginal returns of capital gradually weakens the driving force of capital accumulation on economic growth. It can be clearly seen from the influence coefficient of education level $E_{i,t-1}$ and scientific and technological development level $lnte{c}_{it}$ that economic growth has been transformed from factor-driven to human capital-driven and innovation-driven; the effect of fiscal expenditure on economic growth is not significant, which may be related to the decreasing driving effect of government purchase on economic growth. To a certain extent, it indicates that the government should pay more attention to the improvement of the efficiency of fiscal expenditure when expanding the scale of expenditure. Meanwhile, in all the six models, the coefficient of $\mathrm{\Delta }ln{N}_{it}$ is negative, indicating that the population growth rate has a negative impact on economic growth when other conditions remain unchanged, which is related to the use of per capita GDP index to measure economic growth in this paper.

Second, from the perspective of the direct impact of education on demographic dividend, the coefficient of the interaction term $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$ is significantly negative, indicating that the improvement of education level can significantly alleviate the negative impact of the decline in the proportion of working-age population on economic growth; that is, it can improve the demographic dividend, which indicates that there is an interaction between the demographic dividend and the education dividend. Nevertheless, the coefficient of the interaction term $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$ is negative, but not significant, indicating that education directly affects demographic dividend mainly through the improvement of education level.

Third, from the perspective of education indirectly affecting demographic dividend through labor participation rate and employment rate, the coefficients of the interaction terms of the three variables $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right), ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$ are all significantly negative, indicating that the improvement of education level can significantly alleviate the negative impact of the decline in the proportion of working-age population on economic growth through labor participation rate and labor employment rate. The interaction coefficients of the three variables $ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right), ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$ are all negative but not significant, indicating that the intermediary mechanism of education affecting the demographic dividend through labor participation rate and labor employment rate is mainly realized through the improvement of education level.

5.2. Robustness Test

This paper used the replacement regression method for robustness testing. Although static panel models could be biased in the estimation of coefficients, they are still a widely used method for estimating panel data. Hausman test results showed that fixed effect is better than random effect. Fixed effect can alleviate the endogeneity caused by ellipsis variable error to a certain extent. The benchmark regression results in this paper are shown in

Table 3. Robustness test.

Variable	Model (7)	Model (8)	Model (9)	Model (10)	Model (11)	Model (12)
$\mathrm{\Delta }ln{k}_{it}$	0.039	0.059	0.052	0.038	0.040	0.039
	(0.390)	(0.187)	(0.241)	(0.379)	(0.362)	(0.381)
$\mathrm{\Delta }ln{L}_{it}$	0.067 **	0.038 ***	0.042 *	0.066	0.067	0.067
	(0.042)	(0.002)	(0.056)	(0.237)	(0.233)	(0.231)
$ln{N}_{it}$	−0.121	−0.130	−0.096	−0.231	−0.237	−0.248
	(0.161)	(0.741)	(0.805)	(0.549)	(0.542)	(0.525)
$E_{i,t-1}$	1.251 ***	2.689 ***	3.408 ***	2.231 ***	2.273 ***	2.263 ***
	(0.000)	(0.000)	(0.000)	(0.001)	(0.001)	(0.001)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$	0.506 **	0.548 **	0.441 ***	0.685 ***	0.750 **	0.752 **
	(0.015)	(0.024)	(0.000)	(0.009)	(0.011)	(0.013)
$ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	0.415 ***	0.412 ***	0.491 ***	0.725 ***	0.812 ***	0.846 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
$ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	1.811 *	2.015 **	1.723 *	1.456 ***	1.641 ***	1.798 ***
	(0.067)	(0.014)	(0.079)	(0.000)	(0.000)	(0.000)
$\mathrm{\Delta }{E}_{it}$	0.128 ***	0.116 ***	0.137 ***	0.142 ***	0.161 **	0.303 *
	(0.000)	(0.000)	(0.000)	(0.000)	(0.035)	(0.053)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$		−0.052 ***	−0.499 ***	−0.354 ***	−0.383 ***	−0.488 ***
		(0.000)	(0.001)	(0.000)	(0.000)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$		−0.241	−0.234	−0.023	−0.015	−0.014
		(0.266)	(0.174)	(0.955)	(0.972)	(0.531)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$			−0.054 ***	−0.061 ***	−0.063 ***	−0.064 ***
			(0.002)	(0.000)	(0.000)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$				−0.995 ***	−1.001 ***	−1.023 ***
				(0.000)	(0.000)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$					−0.019	−0.019
					(0.631)	(0.642)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$						−0.415
						(0.525)
$lnfinanc{e}_{it}$	0.003	0.038	0.011	0.014 *	0.015	0.015
	(0.086)	(0.645)	(0.519)	(0.058)	(0.406)	(0.399)
$lntrad{e}_{it}$	0.040 **	0.056 **	0.064 ***	0.077 ***	0.078 ***	0.079 ***
	(0.035)	(0.040)	(0.001)	(0.000)	(0.000)	(0.000)
$lnte{c}_{it}$	2.267 ***	2.263 ***	2.243 ***	2.231 ***	2.231 ***	2.230 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
_cons	−4.375	−5.552	−3.551 ***	−6.959 ***	−8.575 ***	−3.904 ***
	(0.437)	(0.356)	(0.000)	(0.000)	(0.000)	(0.000)
Obs	540	540	540	540	540	540
R-sq	0.604	0.608	0.747	0.759	0.759	0.759
Provincial fixed	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes
Hausman test	103.54	106.26	95.08	90.09	89.48	89.37

Note: *, ** and *** indicate significance at the levels of 10%, 5% and 1%, respectively. and the figures in parentheses are p values.

.

The regression results of fixed effect (Table 3) and SYS-GMM (Table 2) showed that there was no significant change in the direction of each variable and interaction coefficient, except the difference in influence size, which verified the robustness of regression results.

6. Further Discussion: The Spatial and Temporal Heterogeneity

6.1. Temporal Heterogeneity

According to previous studies, the decline of the demographic dividend in some economies is closely linked to population growth. In 2010, the proportion of the population aged 15–64 years reached 74.5% and then showed a downward trend in China. So, is the impact of education on the demographic dividend heterogeneous before and after such a turning point? To answer this question, we divided the data into two parts based on the node of 2010 for comparative analysis, and the regression results are shown in

Table 4. Time-Segment Sample Regression Results.

	2002–2010	2011–2020
	SYS-GMM	SYS-GMM
$ln{y}_{i,t-1}{}^p$	1.012 ***	0.837 ***
	(0.000)	(0.002)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$	3.335 **	2.102 *
	(0.032)	(0.058)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$	0.297 **	−0.532 ***
	(0.025)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$	0.062	−0.097
	(0.414)	(0.381)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	0.008	−0.017 **
	(0.179)	(0.021)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	0.372 *	−0.492 ***
	(0.095)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	0.004	−0.013
	(0.213)	(0.106)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	0.032	−0.087
	(0.521)	(0.628)
AR(1)	(0.000)	(0.000)
AR(2)	(0.639)	(0.742)
Control variables	Yes	Yes

Note: *, ** and *** indicate significance at the levels of 10%, 5% and 1%, respectively. and the figures in parentheses are p values.

.

The results of time-division regression show that: as for the demographic dividend, the influence of population age structure on economic growth from 2011 to 2020 was weaker than that from 2002 to 2010, and the influence coefficient decreased from 3.335 to 2.102, indicating that with the deepening degree of China’s aging and the gradual decline of the proportion of working-age population, a series of measures to deal with the decline of labor supply (such as a vigorously developing digital economy, promoting the transformation of the demographic dividend into a talent dividend, etc.) weakened the impact of demographic structure change on economic growth.

As to the education dividend, there is obvious heterogeneity. The coefficients of all interaction terms were negative during 2011–2020, while they were positive during 2002–2010, demonstrating that education can strengthen the positive impact of the rising proportion of working-age population on economic growth and thus increase the demographic dividend in years when the proportion of working-age population is on the rise. At the same time, when the proportion of working-age population gradually declines, education can alleviate its negative impact on economic growth and produce the effect of improving the demographic dividend. In particular, most of the interaction coefficients from 2002 to 2010 were not significant, possibly because before 2010, the advantage of population age structure was one of the main driving forces for economic growth, and the effect of education on demographic dividend had not been highlighted. Another possible reason is the heterogeneity of population age structure in different provinces before 2010.

6.2. Spatial Heterogeneity

There are large differences in the age structure of the population among provinces caused by migration in China [57]. In general, the eastern regions have a higher migration of young people with higher levels of education. In contrast, regions such as the north-east and the west have a severe labor force loss. We are therefore interested in whether the demographic and educational dividends differ between the various regions of China. We divided the regions according to the rising and falling proportion of working-age population in the same period 2003–2007, and the regression results are shown in

Table 5. Regression Results by Region.

	Rising Proportion of Working-Age Population	Falling Proportion of Working-Age Population
	SYS-GMM	SYS-GMM
$ln{y}_{i,t-1}{}^p$	0.907 ***	0.443 ***
	(0.000)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)$	3.361 *	3.142 **
	(0.071)	(0.043)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}$	0.324 ***	−0.458 ***
	(0.002)	(0.000)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}$	0.086	−0.105
	(0.121)	(0.339)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	0.024 ***	−0.021 **
	(0.005)	(0.032)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times E_{i,t-1}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	0.381 ***	−0.247 ***
	(0.000)	(0.006)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{P_{i,t-1}}{W_{i,t-1}}\right)$	0.008	−0.016
	(0.217)	(0.121)
$ln\left(\frac{W_{i,t-1}}{N_{i,t-1}}\right)\times \mathrm{\Delta }{E}_{it}\times ln\left(\frac{L_{i,t-1}}{P_{i,t-1}}\right)$	0.041	−0.039
	(0.374)	(0.156)
AR(1)	(0.000)	(0.003)
AR(2)	(0.581)	(0.642)
Control variables	Yes	Yes

Note: *, ** and *** indicate significance at the levels of 10%, 5% and 1%, respectively. and the figures in parentheses are p values.

.

It can be found that in the regions where the proportion of working-age population has increased, the interaction coefficients of education affecting the demographic dividend are all positive, while the interaction coefficient of the region whose proportion of working-age population has declined is all negative. It shows that education can improve the demographic dividend: on the one hand, it can improve the negative impact of the decline in the proportion of working-age population on economic growth. On the other hand, it can improve the positive impact of the rising proportion of working-age population on economic growth, which verifies the theory of this paper.

7. Conclusions, Implications, and Limitations

The issue of population and economic growth has always been a global and strategic concern for developing countries. This paper aims to systematically analyze the impact of the demographic dividend and education dividend on economic growth. We constructed a theoretical framework for the impact of these dividends and utilized a dynamic panel SYS-GMM model to conduct empirical analysis on panel data from China’s provinces between 2002 and 2020. Based on China’s empirical evidence, we explored ways to improve the demographic and education dividends. The main findings and policy implications of this paper can be summarized in the following three aspects:

First, the demographic dividend does not come from the size of the population itself but from the size of the labor force. According to empirical results, although an increase in the size of the labor force leads to economic growth, the effect of population size on economic growth has been negative. This finding provides evidence for some developing countries regarding the existence of a demographic dividend [4,5,8]. The findings of this paper suggest that for developing countries, population does not necessarily lead to economic growth; the central factor is whether population can be transformed into low-cost labor.

Second, education can improve the demographic dividend, which is proof of the existence of the education dividend. The improvement of education level can directly alleviate the negative impact of the decline in the proportion of working-age population on economic growth, or indirectly alleviate the negative impact of the decline in the proportion of working-age population on economic growth through labor participation rate and labor employment rate.

The findings of this study may provide some insight into education policy. From a macroeconomics perspective, an increase in the level of education allows the population to be converted into a higher level of labor force, which is conducive to economic growth. Then, for some developing countries where the demographic dividend is disappearing, more education subsidies can be provided to reduce the pressure of population growth on the economy.

Third, the effects of the demographic and education dividends are heterogeneous across regions and stages of developing countries. This study may explain why existing studies dispute the effects of the demographic and education dividends [18,19,26,44], because these economies are at different stages. Heterogeneity analysis illustrates that the improvement of education level can improve the negative impact of the decline in the proportion of working-age population on economic growth. However, it can also enhance the positive impact of the rising proportion of working-age population on economic growth. In particular, for developing countries, the economic environment is changing rapidly, and these countries need to adapt their population and education policies to the different stages of national development.

Although this paper provides a systematic discussion of the education dividend and demographic dividend in developing countries, there are still some theoretical and empirical limitations. For example, China has a higher level of investment in education compared to other developing countries. Therefore, it is unclear whether the education dividend exists in developing countries with lower levels of investment in education, such as Africa. Additionally, developing countries have different industrial structures. Thus, it remains unclear whether the effects of the demographic and education dividends are consistent across economies with varying industrial structures. These phenomena deserve further discussion in future studies.