In the era of IT, broadband is vital to modern society, particularly for BCP’s strategic positioning. By enhancing broadband access and mobile communication, BCP reduces information transmission time, lowers information asymmetry, and boosts public service efficiency [36,37]. Internet health care proved invaluable during the COVID-19 crisis, aiding epidemic control, reshaping physician-patient interactions, and facilitating resource use [38-40]. Information infrastructure development in medical systems alters service patterns, enhancing efficiency [41,42]. The internet empowers patients, tackles inequality, and controls medical costs. In this context, a hypothesis can be proposed:

China’s current resource allocation model favors the supply side, aggravating regional resource imbalances. Specifically, in health care, most top hospitals are concentrated in high-income regions, concentrating on high-quality medical services [43,44]. Tertiary hospitals, comprising 4% of all hospitals, employ >26% of medical staff and deliver more than 20% of medical services [45]. Eastern coastal regions have abundant medical resources, whereas western areas lack resources, infrastructure, and education [46,47]. Although the BCP has improved nationwide broadband coverage, regional discrepancies persist. Broadband progress is notable in central and eastern regions, narrowing the gap with the West, which still lags behind in key areas [48,49]. From this, we propose the following hypothesis:

Medical resources in China tend to concentrate in densely populated and economically developed top cities, especially in western regions [46]. The BCP has had a more significant impact on improving medical resource allocation and service quality in top cities than in ordinary prefecture-level cities, with top cities experiencing a 4.2% and 4.8% increase in human and material resources, respectively [50]. Nearly three-quarters of registered internet hospitals in China are concentrated in the top 11 large cities, highlighting their advantages in obtaining medical resources and providing services [51]. Implementing the BCP in top cities can improve local residents’ health status and promote health care services’ accessibility, generating spillover effects on surrounding areas, and promoting public health progress at a macrolevel. Thus, we propose the following hypothesis:

In recent decades, global youth populations have faced challenges accessing medical resources and have been neglected in health and social policies. However, the rapid development of digital media and broadband technology presents new opportunities for this group, especially in urban areas with a higher concentration of adolescents [52]. Adolescents are highly active on the internet, with up to 95% using it frequently and having access to computers and tablets at home. This makes the internet an effective platform for adolescent self-care, education, and health-related services [53-55]. In addition, adolescents’ susceptibility to social problems and their increased demand for medical resources and health information further highlight the significance of digital medical platforms and internet health resources in cities with a high proportion of youth populations [56]. Therefore, implementing network medical strategies and resource layouts (BCP) targeted at adolescents in these cities is crucial and more likely to be effective. On the basis of this, we propose the following hypothesis:

The degree of MM in cities is the dependent variable in this study. A previous study defined resource mismatch at the enterprise level [59], stating that optimal resource allocation results in consistent marginal output across different enterprises, thus maximizing total output and efficiency. Conversely, resource mismatch distorts the input-output allocation ratio among different enterprises, leading to a loss in the total output and efficiency. Following their approach, we calculated the degree of MM as follows:

In this equation, τ_kjt represents the degree of capital mismatch in city j in year t, where a higher value indicates a greater distortion in the marginal output of capital. τ_ljt represents the degree of labor mismatch in city j in year t, where a higher value indicates a greater distortion in the marginal output of labor; α_j and β_j represent the output elasticities of medical capital and medical labor in cities, respectively, with α_j+β_j≠1 [60]; σ represents the elasticity of substitution between outputs of different cities, and this value generally falls between 3 and 10 [61]. To obtain a conservative estimate, in reference to the existing studies [62], we set σ=3; R represents the cost of capital use. In reference to the existing studies [59], we set R as 0.1 in this study; P_jtY_jt represents city medical expenditure, measured by health expenditure; ωL_jt represents the income of medical labor in city j, ω represents the average income in the medical industry within city j, and L represents the number of medical personnel in the city; and K_jt represents the amount of medical capital in the city, measured by the total number of hospitals and beds in the city [8]. To facilitate interpretation of the results, we take the logarithm of the calculated results and obtain the dependent variable in this study, the degree of MM in cities.

The core independent variable in this study is the BCP dummy variable. When a city is selected as a pilot city for the BCP in a particular year, the variable is assigned a value of 1 for that year and all subsequent years, whereas it is assigned a value of 0 for other cities. This variable (BCP) serves as the core independent variable in this study.

To minimize the potential bias caused by omitted variables, we selected 4 variables—economic level, government intervention, population structure, and urbanization rate—as control variables based on existing studies [26,63,64] and the actual situation.

The first is the economic level (Gdp). Because broadband infrastructure construction is positively correlated with regional economic levels, it is necessary to control for the economic level to identify the impact of broadband infrastructure construction on resource misallocation [65]. This study measures the economic level using per capita GDP.

The second is government intervention (Gov). Government intervention is an important factor leading to resource misallocation because the government may suppress the effectiveness of market resource allocation, resulting in market failure. The higher the proportion of fiscal expenditure to regional GDP, the stronger the government’s intervention capacity and thus the higher the level of resource misallocation [66]. This study measures government intervention using the proportion of government fiscal expenditure to regional GDP.

The third is the population structure (Pop). The population structure has a significant impact on labor force allocation. The population dividend optimizes resource allocation efficiency through the flow of labor, and the unlimited supply of labor offsets the diminishing marginal returns of capital [67]. This study measures population structure using the average number of employed persons per year divided by the average total population.

The fourth is the urbanization rate (Urb). The agglomeration effect generated during the process of urbanization can promote the concentration of labor and capital within a region, expand the economic scope and labor market of urban centers and their peripheries, and improve the matching degree between capital, labor, and hospitals, thereby facilitating the flow of resources within the region, enhancing resource allocation efficiency, and reducing resource misallocation [66]. This study measures the urbanization rate using the proportion of the nonagricultural population to the total population. The major variables and their definitions are summarized in Table S1 in Multimedia Appendix 1 [59].

To test hypothesis 1, we conduct an empirical test based on equation 1. To demonstrate the effects of different factors, a stepwise regression is used. The specific regression results are presented in Table 2.

Column 1 shows the ordinary least squares regression results without any control. The regression coefficient of the BCP on MM is –0.134, which is significant at the 1% level, indicating that the implementation of BCP has reduced the MM by 13.4%. However, it should be noted that this only reflects a correlation. Column 2 adds control variables to control for the impact of other observable factors. The implementation of the BCP has reduced the MM by 13.3%. Column 3 further adds year fixed effects to control for the influence of unobservable factors in different observation years. The result remains unchanged, with only a slight increase in the SE. Column 4 further adds city fixed effects to control for the influence of unobservable factors in the different cities. The implementation of BCP has reduced the MM by 13.2%. Column 5 further uses robust SEs to control for the heteroscedasticity of the error term. The result is the same as before, with the implementation of BCP reducing the MM by 13.2%.

Due to the more complete control of the confounding effects of other factors, the baseline regression result of this study is based on column 5, and hypothesis 1 is clearly supported. The results were similar to those of existing studies [68], which studied the impact of the BCP on urban labor allocation and obtained similar results.

Looking at the control variables in column 5, Gdp reduces the MM by 0.86%, Pop reduces it by 12%, Gov increases it by 13.3%, and Urb reduces it by 7.92%, which is consistent with the previous expectations.

To conduct a DID regression, the parallel trends assumption must be satisfied between the control group and the experimental group. Otherwise, the regression results may be driven by preexisting differences between the groups. Drawing on the event study methodology commonly used in the field of finance, the following dynamic regression equation is constructed for testing:

In this equation, event represents the event period. In the period when the policy is implemented, event is set to 0. In the period before the policy implementation, event is set to –1. In the period after the policy implementation, event is set to +1, and so on for subsequent periods. Following the practice in mainstream literature, event=–1 is chosen as the base period. The interpretation of other variables remains the same and will not be repeated. The specific regression results are shown in Figure 1.

Before the policy implementation (ie, when event ≤–1), the coefficient of the BCP is not statistically significant, indicating that there are no significant preexisting differences between the pilot areas of the BCP and nonpilot areas, thus satisfying the parallel trends assumption. In the periods after the policy shock, the coefficient of BCP is significantly negative and decreases over time before stabilizing. Overall, the BCP has significantly reduced the degree of MM each year.

To ensure the robustness of the benchmark regression results, a series of robustness tests were conducted in this section, and the specific results are presented in Table 3.

In the previous calculation of the dependent variable MM, the elasticity values of medical labor and medical capital were measured using the Olley-Pakes method. In this section, following the existing studies [69], the Solow residual method was used for recalculation. Simultaneously, the substitution elasticity of urban output was replaced with an upper limit value of 10. The newly obtained dependent variable MM_solow was used, and the specific regression results are shown in column 1: the impact coefficient of the BCP on MM_solow is –0.218, which is significant at the 1% level, indicating that the implementation of BCP has reduced the degree of mismatch of medical resources by 21.8%, and the results remain robust.

Considering that the pilot cities for the BCP may not have been arbitrarily selected and that there may be inherent differences between pilot and nonpilot cities, it is precisely these differences that drive the benchmark regression results. To address this concern, this study uses the propensity score matching–DID method to match pilot and nonpilot areas in a 1:1 ratio based on propensity score matching and then conducts a reregression. The results are shown in column 2: the impact coefficient of the BCP on MM is –0.123, which is significant at the 1% level, indicating that the implementation of the BCP has reduced the degree of MM by 12.3%, and the results remain robust. It should be noted that the premise of using the propensity score matching–DID method is that there are no significant differences in covariates between pilot and nonpilot areas. We conducted a covariate balance test; the specific regression results are presented in Table S2 in Multimedia Appendix 1.

Considering the potential endogeneity issue between the BCP and the degree of MM, it is possible that cities with a lower degree of MM are also cities with higher levels of IT and economic development, making them more likely to be selected as pilot areas for the BCP. In addition, this study may also suffer from omitted variable bias and measurement error issues that could lead to endogeneity. Although the DID approach used in this study partially mitigated the endogeneity problem, to enhance result reliability, we further implemented an instrumental variable (IV) approach to address potential endogeneity issues, following the methodology of a previous study [68].

In reference to existing studies [68], we used the topographic fluctuation of cities as an instrument. On the one hand, cities with larger topographic fluctuation not only increase the cost of broadband construction but also lead to a decrease in the signal quality of broadband networks. Therefore, cities with larger topographic fluctuations face greater difficulties in broadband construction, making them less likely to become pilot cities for the Broadband China initiative. On the other hand, topographic fluctuation is a geographical condition with strong exogeneity and does not affect labor allocation distortion. For ease of understanding, we took the reciprocal of topographic fluctuation as IV1. The specific regression results are presented in column 3: the impact coefficient of BCP on MM is –0.133, which is significant at the 1% level, indicating that the implementation of BCP has reduced the degree of MM by 18.3%, and the results remain robust. Furthermore, the results pass the overidentification test (Kleibergen-Paap) and weak identification test (Cragg-Donald).

In reference to existing studies [70], we used the distance from each city to Hangzhou as an instrument. On the one hand, in 2004, the Alibaba Group launched the Alipay service in Hangzhou, which promoted the development of digital finance. Therefore, cities closer to Hangzhou are more likely to have the foundation for digital infrastructure construction and application, making them more likely to be selected as pilot areas for the BCP. On the other hand, this IV is a historical distance variable, satisfying the exogeneity assumption [71,72]. Thus, we obtain IV2. The specific regression results are presented in column 4: the impact coefficient of BCP on MM is –0.119, which is significant at the 1% level, indicating that the implementation of BCP has reduced the degree of MM by 181.9%, and the results remain robust. Furthermore, the results pass the overidentification test (Kleibergen-Paap) and weak identification test (Cragg-Donald).

To determine which channels the BCP reduces the MM, this study explores and identifies 2 potential channels: the innovation effect and the integration effect. The mechanism test is conducted using a 3-step method, and the specific regression results are presented in Table 4.

Research has shown that enhancing broadband infrastructure can elevate urban innovation levels, with a reported increase of 1.94% in innovation for every unit of internet access [73]. It is evident that broadband infrastructure construction plays a crucial role in advancing China’s regional innovation level by directly improving innovation levels and indirectly impacting innovation through the acceleration of human capital accumulation, financial development, and industrial upgrading [74]. The expansion of broadband networks not only broadens the spatial range of network space functions and innovation resource allocation but also offers more opportunities and development space for innovation activities and the evolution of innovation models [75]. Furthermore, there exists a substantial correlation between innovation efficiency and resource allocation efficiency, with higher innovation efficiency leading to enhanced resource allocation efficiency [76]. Hence, it is anticipated that the BCP can alleviate resource mismatches by fostering regional innovation.

In reference to the existing studies [77], this study measures urban innovation levels based on the number of invention patents. Patent numbers are divided into 2 categories: patent applications and patent authorizations. Considering that policy shocks will have an impact on urban innovation levels in the current period and that there is a certain time lag from patent application to authorization, this study uses the number of invention patent applications in each city (Patent) to measure urban innovation levels. The number of city patent applications comes from the China Patent Public Announcement System on the website of the China National Intellectual Property Administration. The regression results are shown in columns 1 and 2: the coefficient of the BCP’s impact on Patent is 0.225, and it is significant at the 1% level, indicating that the implementation of the BCP significantly improves urban innovation levels. The coefficient of Patent’s impact on MM is –0.0249, and it is significant at the 1% level, indicating that the significant improvement in urban innovation levels reduces the mismatch of medical resources, confirming the theoretical expectations.

Market segmentation restricts the free flow of resources. The development of broadband internet can promote market integration and improve resource mismatches [78]. Improving information technology infrastructure construction can achieve resource integration, enhance resource management levels, and improve production efficiency [79]. The internet accelerates production transformation and market integration. It optimizes resource flow between the upstream and downstream through information mechanisms, gradually reducing resource waste and promoting resource growth. In general, the BCP can improve resource mismatches in target areas through market integration [80,81].

Following the methodology of a previous study [82], this study uses (number of regional hospitals/regional medical expenditure)/(number of national hospitals/national medical expenditure) to measure regional integration (Integration). The regression results are shown in columns 3 and 4: the coefficient of BCP’s impact on Integration is 0.0786, which is significant at the 5% level. This indicates that the implementation of BCP significantly improves urban integration levels. The coefficient of Integration’s impact on MM is –0.0764, which is significant at the 1% level. This suggests that the significant improvement in integration levels significantly reduces the MM, confirming the theoretical expectations.

This section tests the heterogeneous hypotheses proposed earlier and conducts an empirical P value test for intergroup coefficient differences. The specific regression results are presented in Table 5.

Hypothesis 2 expects that there are differences in the effects of BCP between different regions, with a larger effect in the central and eastern regions compared with the western region. According to the regional classification standards of the National Bureau of Statistics of China, we classified Sichuan, Yunnan, Guizhou, Tibet, Chongqing, Shaanxi, Gansu, Qinghai, Xinjiang, Ningxia, Inner Mongolia, and Guangxi as the western region, while the rest are classified as the central and eastern regions. The group regression results are shown in columns 1 and 2: the BCP reduces MM by 15.9% in the central and eastern regions, but it does not have a significant impact in the western region. The empirical P value indicates a significant difference in the coefficients between the 2 groups. This confirms the expectation of hypothesis 2.

Hypothesis 3 expects that there are differences in the effects of BCP among different population sizes, with larger cities experiencing a greater effect. On the basis of the median total population at the end of each city, the sample is divided into 2 subsamples: a high population size and a low population size. The group regression results are shown in columns 3 and 4: the BCP reduces medical mismatch by 14.9% in the high population size sample, while it reduces it by 11.8% in the low population size sample. The empirical P value indicates a significant difference in the coefficients between the 2 groups. This confirms the expectation of hypothesis 3.

Hypothesis 4 expects that there are differences in the effects of BCP among different population structures, with cities having a higher proportion of the young population experiencing a greater effect. Due to the lack of data on the proportion of the young population in the China Urban Statistical Yearbook, considering that the ratio of the young population receiving primary and secondary education is >90% and the ratio of those receiving higher education is close to 60% [83], we use the number of students in ordinary primary, secondary, and higher education institutions as a proxy for the young population. We divide the sample into 2 groups based on the median proportion of these students to the total population: high youth proportion and low youth proportion. The group regression results are shown in columns 5 and 6: the BCP reduces medical mismatch by 14.4% in the high youth proportion sample, while it reduces it by 13% in the low youth proportion sample. The empirical P value indicates a significant difference in the coefficients between the 2 groups. This confirms the expectation of hypothesis 4.