Does economic development widen or narrow the gap of public health service utilization among different subgroups of internal migrants? Evidence from the China Migrant Dynamic Survey in 2017

China is promoting the equalization of National Essential Public Health Services (NEPHS) for the internal migrants (IMs). Studies have analyzed the impact of migration range (MR) and regional economic development level (REDL) on the IMs' NEPHS utilization (NEPHSU), but no studies have explored the interaction of MR and REDL on it. Data from the China Migrant Dynamic Survey of 2017, involving 122656 IMs. Per capita GDP was set as the indicator for REDL, and 28 provinces were divided into three groups according to REDL: auent, medium and poor. The MR was divided into inter-province and intra-province, and social capital (SC) was distinguished into cognitive (CSC) and structural social capital (SSC). Awareness of NEPHS (ANEPHS) and establishment of health record (EHR) were selected as indexes of NEPHSU. Multiple line charts and hierarchical logistic regression were used to investigate the interaction of MR, REDL and SC on NEPHSU.


Background
Over the past three decades, social change has led to unprecedented massive internal migration in China. An increasing number of people leave their original places of residence to work and live in other provinces or cities to improve their lives, such people are called internal migrants (IMs). The number of IMs exceeded 240 million in 2017 [1]. IMs are vulnerable to social exclusion in their destinations [2], which prevents them from accessing local National Essential Public Health Services (NEPHS) as equitably as local residents [3]. Social exclusion has become an important risk affecting the health of IMs [4].
However, the situation has been improving in the past decade. China initiated the NEPHS projects in 2009, which required Community Health Centers to provide NEPHS to all residents in their area, including IMs that have lived there for more than six months [5]. Since then, the National Health Commission (NHC) has successively introduced major measures to strengthen the equalization of NEPHS for the IMs [6-8].
Although NEPHS deployment level has risen rapidly in recent years [9], there is still a signi cant gap in the national target, and the underlying factors need to be revealed. Some recent studies have con rmed that gender, education, community type, migration range (MR) and regional economic development level (REDL) have signi cant impacts on the IMs' NEPHSU [10][11][12][13][14][15].
MR and REDL are important variables re ecting regional differences in the NEPHSU of IMs. Some studies have pointed out that the NEPHSU level of trans-provincial migrants is low [9,10,[13][14][15], and some studies have suggested that the NEPHSU level of migrant population in a uent areas is low [12][13][14][15].
However, there is no study to nd out whether the interprovincial migrants in a uent areas have lower NEPHSU level, and the interaction between MR and REDL has been ignored. Existing studies on MR and REDL have two doubts: First of all, existing studies have divided China into three geographical regions: the eastern, central and western region. Data show that the eastern region has the lowest NEPHSU level of migrant population. However, most provinces in the eastern region are economically developed, so these studies infer that the NEPHSU level of the migrant population in the richer regions is lower, which is not rigorous. Secondly, when comparing the NEPHSU levels of intra-provincial and inter-provincial migrant population, existing studies did not exclude Beijing, Tianjin and Shanghai. However, these three cities only have an inter-provincial migrant population, which may lead to bias if they were not excluded.
In addition, the analysis shows that the demographic variables are limited in explaining the variation in the NEPHSU of the migrant population, and new variables must be introduced.
Social capital (SC) is an important social determinant of health, and access to health services has been suggested as a pathway by which SC in uences health outcomes [16]. Different types of SC affect health services utilization by in uencing the availability of health services in communities, the availability and effectiveness of outreach resources between health-care providers and communities they serve, and careseeking behavior of individuals in those communities [17]. Migration means a loss of the original social network and a reduction of social participation in the new environment [18]. However, the SC that migrants have in their destination is more important for their access to local health services [19,20]. SC is more often used to discuss its in uence on IMs' health outcomes in China [20][21][22][23][24][25], while ignoring its value in the accessibility of health services. Guo et al. [26] had pointed out that lack of information is the biggest obstacle to IMs accessing NEPHS in China. From the social resource perspective, SC can in uence health information through three mechanisms: increased information exposure, enhanced seeking abilities, and reinforced health culture or norms embedded in social networks [27]. Therefore, SC may play an important role in IMs' access to NEPHS.
Few studies have made explorations in this area. Hou et al. [28] revealed that individual structural SC has a signi cant positive effect on NEPHSU, while Zeng et al. [29] pointed that SC can moderate the negative effect of social discrimination on IMs' health service utilization. After reviewing the existing studies, there are at least two limitations that need to be improved: (1) Some studies have pointed out that the SC of IMs is inferior to that of local people [30,31], but few studies have revealed the differences of SC among different subgroups of migrant population, and our understanding of the distribution of SC of IMs is still not enough. (2) Not everyone has access to the same sources of SC and not everyone will bene t in the same way [32]. Carpiano and Moore [33] suggested that SC and health knowledge base can be better served by asking three foundational questions of (a) how, (b) for whom, and (c) in which contexts does SC work or operate, rather than focusing primarily on whether SC provides some universal health bene t. However, there is still a lack of research on question b and c in China.
Based on the existing research analysis, we believe that there may be a signi cant interaction effect between the MR, REDL, SC and NEPHSU of the IMs. This study focuses on testing three hypotheses: (1) Cheng et al. [34] suggested that socioeconomic status (SES) between registered population and migrant population as well as within the migrant population are highly heterogeneous. This difference may be in uenced by the interaction between MR and REDL. (2) Since MR and REDL can signi cantly affect an individual's SES, There is a strong correlation between SES and SC, so the social capital of the migrant population may also be in uenced by the interaction between MR and REDL. (3) Given that the relationship between SC and health can be moderated by SES, there may be a signi cant interaction among MR, REDL, SC and NEPHS utilization. Our ndings could help deepen the understanding of the relationship between SC and access to public heath services for IMs, and also provide a reference to the improvement on relevant policies of the Chinese government and other developing countries.

Data
The data was obtained from the China Migrant Dynamic Survey (CMDS) in 2017 provided by the Migrant Population Service Center. CMDS is an annual national sample survey of the internal migrants organized by the NHC from 2009, with an annual sample size of approximately 200,000 households. CMDS adopts the layered, multi-stage, and proportional to scale PPS (Probability proportional to size) sampling method. This study adopted the individual questionnaire A of CMDS, which was uniformly printed and distributed by the NHC. The questionnaire A includes basic information about respondent's demography, perception of the destination, the state of social interaction, and utilization status of NEPHS, etc. Full-time investigators collected the questionnaire data through household interviews, and each respondent gave informed consent before commencing the interview. Dates were entered through the migrant population health and household planning dynamic monitoring system, input data was subjected to multiple checks to ensure quality. The respondents consisted of IMs aged 15-59 living in the destination for more than one month.
In this study, the inclusion conditions were set as "18-59 years of age, residence duration more than one year. Because Beijing, Tianjin and Shanghai do not have an interprovincial migrant population, it is impossible to compare the differences between the intra-provincial and interprovincial groups, so the samples from these three cities were excluded. After the quality audit, 122665 people were nally included in this study. In addition, we introduce GDP per capita to re ect the REDL of each provincial region, and GDP per capita is based on 2017 data from the National Bureau of Statistics.

Measurement
Utilization of NEPHS Awareness of NEPHS (ANEPHS) is a prerequisite for NEPHSU [26]. ANEPHS was set as a dependent variable. The question was "Have you heard of the NEPHS" and the answer was "yes or no". Another outcome variable is establishment of health records (EHR). The question was "Have you established health records at the destination" and the answer was "yes or no". EHR is one of the service priorities and re ects the actual use of services by the migrant population.

Demographic variables
Demographic variables included in this study included gender, community type, MR, and REDL. The community types were divided into urban and rural areas. The MR was divided into group inter-province and intra-province. According to the provincial GDP per capita division published by the National Bureau of Statistics in 2017, the top 10 were classi ed as group a uent (Beijing, Tianjin and Shanghai were excluded), the middle 10 as group medium, and the last 11 as group poor.

SES
Education and household income are often used as indicators of SES. In this study, education was divided into four categories according to years of education: ≤6 years, 7-9 years, 10-12 years and >12 years. Considering that relative income can better re ect the impact of income gap when regional economic development levels are unbalanced [35], the household income in this study is in the form of relative income, that is, the annual household income divided by the per capita GDP of the locality. SC SC refers to the resources and bene ts received through connections with others, either as individuals or groups, it can be distinguished into two dimensions: SSC and CSC [16]. SSC refers to the presence of formal opportunity structures or activities in which individuals build or strengthen their social connections; CSC generally refers to individuals' perceptions, beliefs, and attitudes toward their social surroundings, with corresponding measures focused mainly on the concepts of generalized and particularized trust [36]. In this study, SC was limited to the destination, and it was a localized SC that re ects the social resources available to the migrant population there.
The SSC of this survey included civic participation and social participation in the destination. Questions of the former were: "since 2016, have you made suggestions to your unit/community/village or supervised the unit/community/village affairs management", "since 2016, have you participated in property donation, blood donation, volunteer activities, etc.", "since 2016, have you reported the situation/put forward policy suggestions to relevant government departments in various ways ", "since 2016, have you posted online comments on national affairs and social events or participated in related discussions", "since 2016, have you participated in party/youth league organization activities and party branch meetings". The answer to each question was "1= no, 2 = occasionally, 3 = sometimes, and 4 = often", the score ranged from 5 to 20. The question of social participation was "Have you participated in any of the following activities in the past year: trade unions, volunteer associations, homecoming associations, fellow-students association, home town chamber of commerce, others", the score ranged from 0 to 7. According to the distribution characteristics of scores, civic participation was integrated into 5 levels: 1 (5 points), 2 (6 points), 3 (7 points), 4 (8 points), 5 (9-20 points). Social participation was also integrated into 5 levels: 1 (0 points), 2 (1 points), 3 (2 points), 4 (3 points), 5 (4-6 points). The spearman correlation coe cient of civic participation and social participation was 0.366 (p<0.001), and the SSC grade can be obtained by adding the two grades, and 7 grades were: 1 (2 points), 2 (3 points), 3 (4 points), 4 (5 points), 5 (6 points), 6 (7 points), 7 (8-10 points).
The CSC generally refers to IMs' perceptions, beliefs, and attitudes toward their destination, with corresponding measures focused mainly on the concepts particularized trust. There were 5 questions in the survey: "I like the city/place I live now", "I am concerned about the changes in the city/place I live now", "I am very willing to blend with the local people and become a part of them", "I think the local people are willing to accept me as a part of them", "I feel locals look down on outsiders". The answer to each question was "1=totally disagree, 2=disagree, 3=basically agree, and 4=totally agree". The α of CSC was 0.779. According to the distribution of scores, CSC was divided into 7 levels: 1 (5-14 points), 2 (15 points), 3 (16 points), 4 (17 points), 5 (18 points), 6 (19 points), 7 (20 points).
Statistical analysis SPSS 22.0 was used for data analysis. First of all, we calculated the mean value and standard deviation of the included continuous variables, and counted the frequency distribution of different subgroups of classi ed variables (Table 1), which comprehensively described the basic situation of the sample. Secondly, we took household income, education and SC as dependent variables, MR and REDL as factors to conduct interaction analysis ( Table 2) to verify hypothesis 1 and 2. Thirdly, we used univariate analysis to verify whether gender, education, community type, MR, REDL and SC had signi cant impacts on ANEPHS and EHR one by one. The veri cation methods included cross-table, chi-square test and independent sample t test. Fourthly, we used multiple line charts to visually present the cross relationship among MR, REDL, SC and NEPHSU (Figure 1-4). Finally, we used gender, education, community type and household income as the control variables, MR and REDL as the moderating variables, and ANEPHS and EHR as the dependent variables for a hierarchical logistic regression analysis (Table 3-4) to discuss the degree and direction of the interaction among MR, REDL, SC and NEPHSU.

Characteristics of the sample
The IMs' mean residence duration was (7.01 ± 5.796) years of the sample. The average household income coe cient was (1.54 ± 1.201). The level of NEPHSU of the IMs was not high, and the rate of ANEPHS was 62.2%, while the rate of EHR was only 32.1%. Demographically, the proportion of group male, low education, urban and intra-province was higher, while the regional distribution of the sample is more balanced. The SC of the IMs is characterized by high CSC and low SSC. The proportion of CSC in the low group (Level 1) is only 15.6%, while that in the high group (≥16 points) was 57.0%. The proportion of SSC in low subgroup (2 points) reached 38.3%, this means that more than one third of IMs had neither civic nor social participation in the past year.

Impact of MR and REDL on SES, SC and NEPHSU
We analyzed the impacts of MR and RED on IMs' SES and SC (Table 2). In the three groups of REDL, the average educational level of the inter-provincial IMs was lower than that of the intra-provincial IMs, and the gap in group a uent was the largest (-0.32) while the difference in group poor was the smallest (-0.13). The gaps of household income between the inter-province and intra-province group were different in the three groups of REDL, and gaps of the group a uent, medium and poor were -0.08, 0.38 and 0.39, respectively. Multivariate ANOVA showed that MR and REDL had signi cant effects on education (F=114.377, p<0.001) and household income (F=382.699, p<0.001). Among three groups of REDL, the SC of group inter-province was lower than that of group intra-province. The gaps between the group interprovince and intra-province in the three REDL groups were -0.83, -0.04 and -0.36, respectively. The gaps in SSC were -0.32, -0.19 and -0.13. The interactions of MR and REDL on CSC (F=369.360, p<0.001) and SSC were signi cant. The rate gaps of ANEPHS in group a uent, medium and poor were -10%, -3% and -4%, respectively. The corresponding gaps in EHR were about -10%, -2% and -5%, respectively. The interactions of MR and REDL on ANEPHS (F=68.081, p<0.001) and EHR (F=67.457, p<0.001) were signi cant.  In order to further explore the interaction between MR, REDL, SC and IMs' PBS, we took CSC and SSC as the abscess axis, and rate of ANEPHS and EHR as the vertical axis to draw the variation trend diagrams of group intra-province (black solid line) and group inter-provincs (red dotted line) (Figure 1-4). The results showed that there are obvious differences between the two curves in different regions. On the whole, the two curves are always closest to each other in the group medium, while the two curves are furthest apart in the group a uent.

Results of multivariate analysis
The interaction between MR, REDL and SC had a signi cant impact on ANEPHS ( Among the four groups of interaction, only SSC * Poor* inter-province was not signi cant, and the OR values of the other three interactions were all signi cantly greater than 1. effects on EHR, and the interactions of all four groups were signi cant. OR values of CSC * Medium* Inter-province, CSC * Poor* Inter-province, and SSC * Medium* Inter-province were signi cantly greater than 1, while OR value of SSC * Poor* Inter-province was smaller than 1.

Discussion
There were three main ndings: (1) The socioeconomic status (SES) and social capital (SC) of interprovincial IMs were signi cantly lower than those of intra-provincial IMs, and the gap was most prominent in a uent areas.
(2) From low to high, the NEPHSU of inter-provincial IMs was ranked as a uent, medium and poor, while the corresponding order of intra-provincial IMs was medium, poor and a uent areas, and the gap of NEPHSU between inter-provincial and intra-provincial IMs was the largest in a uent areas. (3) SC could signi cantly promote the IMs' NEPHSU, but there was a gap between interprovincial and intra-provincial IMs, and the gap was the largest in a uent areas.
The household registration status still affects and determines the urban class structure and social integration to a large extent [37]. Despite the government's great efforts in recent years, the urbanization of people and the citizens of the migrant population has not been truly realized, and the social and economic conditions between the registered population and migrant population as well as within the migrant population are highly heterogeneous [34]. This study con rmed the hypothesis that there are signi cant differences in SES within the migrant population, and these differences were in uenced by the interaction of MR and REDL. Data showed that inter-provincial migrants were indeed at a worse disadvantage than intra-provincial migrants in a uent areas, with lower scores for both education and family income. However, the picture was very different in medium and poor areas, where the family income of inter-provincial migrants exceeded that of intra-provincial migrants, although their average education level was still lower. Therefore, we believe that the situation of inter-provincial migrants in a uent areas is more di cult, while that of inter-provincial migrants in medium and poor areas may not be so bad.
As pointed out by Palloni et al.
[18] that migration means a loss of the original social network and a reduction of social participation in the new environment. Previous study [31] has pointed out that the SC of migrant population is inferior to that of local people, but no research has explored the gaps among subgroups within the IMs. We found this gap and it was in uenced by the interaction of MR and REDL. On the whole, the SC of the inter-provincial migrants was lower than that of the intra-provincial migrants, and this gap was the largest in the a uent areas. This trend is basically consistent with the abovementioned trend of SES, which supports the credibility of this study conclusion. Dialect, customs and culture vary widely among provinces in China, so intra-provincial migrants in a given province have an advantage in social adaptation and social network construction compared to inter-provincial migrants. Therefore, intra-provincial migrants have a higher level of SC than inter-provincial migrants. At the same time, we also noticed that the variation trend of SC with REDL was different between the inter-provincial group and the intra-provincial group, and these complex interaction relationships need to be explored by further studies.
We agree with the point that NEPHSU of the inter-provincial IMs is lower [9,10,[13][14][15], but we do not support the view that the NEPHSU level of the IMs in the a uent region is lower [12][13][14][15]. Compared with the existing studies, we have three differences: (1) Regional classi cation based on per capita GDP; (2) Three cities, Beijing, Tianjin and Shanghai, were excluded; (3) The interaction between MR and REDL was considered. We found that the trend of NEPHSU level changing with REDL was different between the inter-provincial and intra-provincial migrants. The former ranks (from low to high) a uent, medium and poor in the order of NEPHSU level in three regions, while the latter ranks medium, a uent, and poor. Of particular concern is that the gap in NEPHSU of IMs between the inter-provincial and intra-provincial groups was the largest in a uent regions. Due to China's nancial transfer payment system, with the nancial support of the central government, the density of grassroots health service resources and the per capita supply of NEPHS in medium and poor areas are higher than those in rich areas [38]. Some studies attributed the lack of NEPHS for IMs in a uent areas to this [9,15]. We believe that service supply may have an impact, but it is not enough to explain the phenomenon of higher NEPHSU level among IMs in a uent provinces. The lack of enthusiasm of the migrant population to participate in the NEPHS is also a signi cant factor [3], as evidenced by the fact that the EHR is much lower than ANEPHS, because the EHR is a voluntary program [5].
Although some studies have pointed out that SC can promote the utilization of health services [18-20, 39], there has been no consensus on this [40]. SC may be more salient for the health and well being of certain sub-populations and in particular contexts [33]. Uphoff et al. [32] have pointed out that the relationship between SC and health is moderated by SES, but the direction of regulation may be enhanced or weakened. This study suggests that SC has a signi cant positive promoting effect on NEPHSU by the IMs, but this promoting effect is moderated by the combination of MR and REDL. As we can see from the line chart, the distance between the dotted and the solid line is always the largest in the a uent group and the smallest in the medium group, which indicates that the relationship between SC and NEPHSU was signi cantly different with different MR and REDL. On the whole, even though the SES and SC of the inter-provincial migrants were lower, they still bene t more from SC when utilizing NEPHS, regardless of the type of region. This is in line with the rst case proposed by Uphoff et al. [32], that is, a more signi cant social capital bene t on the health of disadvantaged persons in society, and no effects or limited health bene ts for those in positions higher up in the social ladder.
Two things have to be pointed out. First of all, the selection of SSC measurement content in this study has strong Chinese cultural characteristics, such as homecoming associations, Party branch of the Chinese Communist Party and Communist Youth League, etc., which is di cult to nd counterparts in the research based on other cultures, which may affect the comparability of different research conclusions to a certain extent. Secondly, because of China's NEPHS system of transfer payments, the cost of NEPHS in economically developed areas is mostly borne by themselves, while the cost of NEPHS in economically underdeveloped areas is mostly borne by the central government. As a result, NEPHS resources in poor areas are more abundant than those in rich areas. Such a situation may not be applicable in other developing countries.

Conclusions
This study reveals that the IMs' SES, SC and NEPHS utilization were in uenced by MR and REDL. REDL can signi cantly affect the gaps of SES and SC among subgroups of IMs with different MR. Interprovincial migrants in a uent areas have relatively lower SES and SC, which may be a major barrier to their access to NEPHS, and the government should pay special attention to this vulnerable group when promoting equalization of NEPHS for IMs.  Figure 1 Interaction of MR, REDL, and CSC on ANEPHS.

Figure 2
Interaction of MR, REDL, and SSC on ANEPHS.

Figure 3
Interaction of MR, REDL, and CSC on EHR.

Figure 4
Interaction of MR, REDL, and SSC on EHR.