Neighborhood and community effects in East and Southeast Asia, a systematic review and meta-analytical exploration of publication bias

Neighborhood and community eﬀects studies have informed urban policies in the West since three decades. Since about ten years, this research line is seen increasingly in East and Southeast Asia as well. As an emerging ﬁeld, the literature has yet to be critically reviewed and its body of literature provides a unique opportunity to study the eﬀects that diﬀerent research communities might have on its development. This systematic review collects 165 studies and gives a critical appraisal of this literature, speciﬁcally focusing on publication bias. Findings show that “true ” neighborhood eﬀects might be overestimated. Health research shows greater publication bias than human geography and general social science. Studies by only local scholars are more prone to bias than studies from collaborative teams or only nonlocal scholars, suggesting that this ﬁeld is relatively early in its life-cycle or that publication pressure is much higher in Asia compared to the West.


Introduction
The international literature on neighborhood and community effects is heavily focused on the United States and Europe. The majority of studies come from those two regions and the majority of theory about neighborhood and community effects is based on studies from those two regions. To illustrate the strong Western focus neighborhood effects research, out of all 88 studies that were conducted on neighborhood effects on educational outcomes in developed countries until 2011, only one was conducted with non-Western (i.e., Taiwanese) data . The research line revolves around the idea that where people grow up or live affects their life in a variety of ways (from educational attainment to health, and from unemployment to happiness), and that this effect remains after taking into account all other factors that could influence these outcomes ( Jencks & Mayer, 1990 ). Neighborhood and community effects research lays its focus on one specific context (i.e., people's residential environment) of people's lives, and tries to tease out the importance of it. This idea took off with the publication of Wilson's The Truly Disadvantaged ( Wilson, 1987 ), in which he described the workings of U.S. inner-city ghettos and the effects of concentrated poverty. This book gave rise to a surge of studies on urban poverty and race in the United States, and quickly found its way across the Atlantic as well. However, thus far, the locus of neighborhood effects research has been chiefly based in the United States and (Northern and Western) Europe, and is only slowly finding its way to other regions of the world.
ing socio-spatial segregation within cities, accessibility of services, and the social inclusion of rural-to-urban migrants ( Baker & Gadgil, 2017 ). Such topics have been studied in the Western literature on neighborhood and community effects, making the field a suitable candidate to also study these topics in East and Southeast Asia.
Because of the burgeoning amount of literature being produced about neighborhood and community effects in the East and Southeast Asia context, this topic lends itself well for a systematic analysis of the literature, including a meta-analytical view into the reliability of this literature. A meta-analytical review into the reliability of this literature will help to assess the state and maturity of this field. As mentioned, the Western neighborhood literature had a head start and is currently mostly refining research methods in order to get to causal relations ( Galster & Sharkey, 2017 ). However, the later adoption in East and Southeast Asia may mean this literature is still in a explorative and descriptive phase. To assess the overall reliability of the literature, special attention is paid to publication bias. Publication bias is a widespread phenomenon in research in general ( Ioannidis, 2005 ) and in research on neighborhood effects specifically ( Nieuwenhuis, 2016 ), and can contribute to overestimating the importance of neighborhood effects in East and Southeast Asia. By reviewing the literature on neighborhood and community effects in East and Southeast Asia quantitatively, I aim to give a comprehensive overview of the reliability of this literature.

The rise in East and Southeast Asian neighborhood and community effects studies
As is evident from Fig. 1 , East and Southeast Asian studies on neighborhood and community effects lag behind somewhat on the Western literature on this topic. One of the reasons for this is that research output is correlated with a country's GDP . This explanation has not so much to do with the topic of neighborhood effects specifically, but more with academic output in general. Whereas historically, the Western world led the charts, many East and Southeast Asian countries have seen rapid economic development over the past few decades ( Roser, 2013 ). This growth in GDP in the region partly explains the increase in the output of neighborhood effects studies, as well as its lag on the Western literature.
However, since East and Southeast Asian societies are not a monolith, many differences between the countries are to be expected. Economically, countries like Japan, South Korea, and Singapore, developed earlier than their neighbors ( Roser, 2013 ), suggesting neighborhood and community research started earlier in those countries. Fur-thermore, differences between countries in spending on research and development (R&D) and universities leads to differences in research output ( Meo et al., 2013 ). Increased R&D and university spending has several benefits for academic research, for example, more and better quality data can be collected, which can make studying East and Southeast Asian societies more accessible to local, as well as, foreign scholars. But also, if increased spending comes in the form of student mobility funding (such as the China Scholarship Council funding program), students can study at Western universities. For example, in the US, foreign student numbers show that, over a 36-year period (1979/'80 -2016/'17), East and Southeast Asian countries China, South Korea, and Vietnam have grown substantially as origin countries. Taiwan and Japan also send a fair amount of students, but already did so in the '80s ( Zong & Batalova, 2018 ). Upon returning to their own countries, these international students will bring with them academic social networks and research topics, including neighborhood effects, potentially stimulating research and research collaboration in this direction. Countries with high research spending include China, South Korea, Japan, and Singapore. Countries like Indonesia, Brunei, Cambodia, and the Philippines have very low spending. Countries with a lot of universities are Japan, China, Indonesia, and Malaysia, while only a few universities are in Brunei, Singapore, Mongolia, and Cambodia ( Meo et al., 2013 ). This suggests that most neighborhood research output will come from China, South Korea, and Japan, while studies from countries with low spending, few universities, or few students abroad will be sparser.
Besides economic developmental reasons for the general increase in scientific output, there may be political reasons for the increasing output on neighborhood and community effects specifically. The idea of neighborhood effects is, by some, argued to be an inherently neoliberal invention to avoid dealing with the negative externalities of austerity. Instead of studying why people are so poor they can only afford a house in the most undesirable neighborhoods, neighborhood effects research offers the thesis that it is actually the neighborhood or community that causes people's poverty and that clustering poor people causes neighborhood decline ( Slater, 2013 ). This argument potentially blames poor people for their own poverty, and dissolves policy makers from the task of elevating people from poverty, handing them the excuse to focus on (investor-led) neighborhood upgrading instead. This then leads to the well documented cycle of gentrification, where the upgraded neighborhood becomes too expensive for the original residents, who are forced to move out to another poor neighborhood while wealthier middle-class families move in ( Lees et al., 2008 ). East and Southeast Asian countries have been much more reluctant than Western countries to adopt Notes: Results are based on a search in Scopus (see Data section for details). These are results from pre-cleaning data and the exact numbers should not be taken at face value. The exact numbers as described elsewhere in this article are correct for South and Southeast Asia. JID: AJSS [m5GeSdc;August 1, 2022;22:41 ] neoliberalism ( Hill et al., 2012 ;Springer, 2017 ). If we place the idea of neighborhood effects in the neoliberal tradition, the later adoption of neoliberalism in East and Southeast Asia could explain the delayed interest in neighborhood effects research and policy. An addition to the above argument could be that policy makers in many East and Southeast Asian countries may have limited political resources, making it very difficult to address the problem of urban poverty satisfactory. Therefore, researchers and policy makers may have tapped into the belief that is commonly held surrounding neighborhood and community effects, in that they can bring about social change. In the Western context, policy makers have turned to neighborhood effects as a relatively cheap alternative to actually alleviating poverty, relying on volunteers and market-investments. For example, community enterprises are used as a policy to battle neighborhood deprivation and other neighborhood-based problems ( Mellor & Affleck, 2006 ;OECD/EU, 2019 ). This is a relatively cheap way to address local problems in the face of a receding welfare state, often sold to community volunteers under the guise of democratic engagement. Thus, shifting the burden of solving local problems from governmental bodies to community members. Where policy makers lack the financial resources to address poverty, they may turn to alternatives that rely on local residents to sort an effect.

Publication bias
In order to assess the reliability of the literature on neighborhood and community effects in East and Southeast Asia, I study patterns of publication bias within this field of research. Publication bias is a form of selection bias, where studies containing statistically significant results are more likely to be "selected " for publication than studies without significant results. Because systematic reviews and meta-analyses usually deal with published papers, publication bias is a potential problem when assessing the reliability of the aggregated results from the selected articles.
An overall view on publication bias in this literature is crucial to understand the current state of affairs in this literature. That publication bias is a severe problem for academic research is well understood ( Easterbrook et al., 1991 ). When studies that present positive results (i.e., significant neighborhood and community effects) are systematically more likely to be published than studies with negative results (i.e., no proof of a neighborhood or community effect), then, a summation of the published literature will give us a biased understanding of the true relation between neighborhood characteristics and individual outcomes. Hypothetically, a full set of studies, published and unpublished or nonacademically published (i.e., the "grey literature "), might leave us more hesitant about this relation. Important mechanisms behind the systematical bias for positive results are found in the academic publication practice. First, journals are more likely to publish positive results than negative results ( Dwan et al., 2013 ), because validating new ideas and existing hypotheses is likely to attract more citations, which will reflect positively in the, perhaps unwarrantedly but by many so highly valued ( Vanclay, 2012 ), impact factor of the journal. A second factor that contributes to publication bias is the competitive "publish or perish " culture many researchers are faced with in academia, in which they are evaluated based on the number of publications. Therefore, in order to attain easier publishable positive results, researchers might try many different model specifications in search of significant results ( Fanelli, 2010 ). Both mechanisms are interrelated and cause an overrepresentation of positive results and an underrepresentation of negative results in the published literature. Therefore, understanding the extend of publication bias in the literature on neighborhood and community effects in East and Southeast Asia will help when assessing the reliability of this field.
If publication bias is present and articles with significant results are indeed more likely to be published by journals, it is likely that researchers do an effort to find significant results ( Fanelli, 2010 ). This means that researcher can try to test additional models specifications or sample selections when their results fall just outside of the typically used, but arbitrary, 5% significance standard. This practice will push results just below the threshold, making them "significant " and thus more publishable. Studies that fail to find significant results might end up in the "file drawer " and never get published. However, when assuming that published articles are a random draw from a sampling distribution, there should be an equal number of studies in two equal-sized intervals just above and just under the p = .05 threshold ( Gerber & Malhotra, 2008 ). When the number of studies just below is much larger than the number of studies just above the arbitrary significance threshold, it means there is publication bias in this literature. Since publication bias was found earlier in neighborhood effects studies ( Nieuwenhuis, 2016 ), it is likely that it is also present in this sample. Therefore, the hypothesis is as follows: -In the literature on neighborhood and community effects in East and Southeast Asia, there are significantly more results just below than just above the 5% significance threshold (hypothesis 1).
Besides a general view of publication bias in this literature as a whole, the varied set of studies in this review allows to study publication bias in more detail. The studies stem from health science, human geography, and a broader social science category, and were written by researchers from a wide variety of countries. Such information yields the possibility to study the importance of research communities and collaborations, as well as the life stage which this field of research is in. A more detailed study of publication bias will also reveal more details about the reliability of this literature. Different research communities may have different views on neighborhood and community effects, which may be reflected in published results. Clearly, all studies included in this review have some expectations from neighborhood or community effects, however, different disciplines may have different expectations about its importance as compared to other explanatory factors. In this review, I make the distinction between health science, social science, and human geography. These three disciplines overlap in many ways, but also have their own distinct discourses about neighborhood and community effects. Human geography, by its definition, deals with the relation between people and some spatial location, including neighborhoods and communities. Health science, on the other hand, although studying neighborhood effects on health, emphasizes behavioral change in health behavior over environmental interventions ( Sampson, 2003 ). In between human geography and health science is sociology, where most theory on the mechanisms behind neighborhood effects has been developed ( Galster, 2012 ). However, all disciplines also know critical accounts, which often argue that neighborhood effects are (partly) artefacts caused by individuals selecting themselves into neighborhoods based on individual characteristics such as income and preferences ( Galster & Sharkey, 2017 ). Furthermore, genetically informed designs have been used to discount the idea of neighborhood effects in health (e.g., Sariaslan et al., 2016 ).
When certain ideas are conventional in a research community, they are less likely to be contested ( Shwed & Bearman, 2010 ). Combined with the finding that positive results corroborating existing hypotheses are easier to publish ( Dwan et al., 2013 ), it is likely that in research communities where the conventional view is in favor of the existence neighborhood and community effects, journals in this field will be more likely to publish positive results for neighborhood and community effects. Human geography and sociology seem more attached to the idea of neighborhood and community effects than health research, therefore it is more likely that there is publication bias in the subset of studies from human geography and sociology than from health research. This leads to the following hypotheses about the sample of studies in this review: -In studies from the human geography and sociology communities, it is more likely than in the health community that there are significantly more results just below than just above the 5% significance threshold (hypothesis 2). JID: AJSS [m5GeSdc;August 1, 2022;22:41 ] -In studies from the human geography and sociology communities, there are more statistically significant results relative to nonsignificant results compared to in the health community (hypothesis 3).

ARTICLE IN PRESS
An alternative explanation for hypothesis 3 could be that in human geography and sociology there is more consensus about neighborhood effects because the "true " effect is larger/exists, while in health research there is no or less consensus, because the neighborhood in reality does not actually exert any influence on the outcomes studied. This problem is solved by using researchers' affiliations to define research community and not the dependent variables. Researchers from different disciplines have an overlap in study outcomes. Geographers and sociologists are also interested in health and health researchers also study happiness and quality of life. Therefore, when finding differences between disciplines, it is likely due to the community rather than the "true " neighborhood effect on a certain outcome.
Studying publication bias can also give insight into the life stage of the literature. Arguably, publishing positive or negative results follows a sort of bandwagon effect. When a new explanatory variable is first introduced in the literature and finding support for it is still considered novel, publication bias might be more often observed ( Gerber & Malhotra, 2008 ). Later in the life cycle it will be novel to show there is no effect, thus leading to a decline in publication bias ( Gerber & Malhotra, 2008 ). This reasoning leads to the following hypotheses: -In older studies, it is more likely than in more recent studies that there are significantly more results just below than just above the 5% significance threshold (hypothesis 4). -In older studies there are more statistically significant results relative to nonsignificant results compared to in more recent studies (hypothesis 5).
The idea of neighborhood and community effects in itself is not a very new idea, but within the East and Southeast Asian literature it was introduced later than in the Western literature (see Fig. 1 ). It is likely that the East and Southeast Asian research community is closer to the start of the life cycle and therefore more likely to exhibit publication bias, while the Western research community is in a later life stage and therefore less likely to have publication bias. This leads to the following hypotheses: -In studies by scholars with a local affiliation, in is more likely than in studies from scholars with a Western affiliation that there are significantly more results just below than just above the 5% significance threshold (hypothesis 6). -In studies by scholars with a local affiliation, there are more statistically significant results relative to nonsignificant results compared to in studies by scholars with a Western affiliation (hypothesis 7).
An alternative explanation for finding the above differences between scholars with a local or a Western affiliation could be a difference in the publication pressure culture between different regions. Even though "publish or perish " cultures between different countries have not been rigorously mapped yet, one study on demographers revealed that researchers experience publication pressure the most in English-speaking countries (United States, United Kingdom, Canada, Australia), and less so in Western Europe or the rest of the world (including Asia) ( van Dalen & Henkens, 2012 ). Others have described the strong pressure to publish English-language publications in China ( Li, 2015 ) and monetary incentives to publish in East Asian countries like China and South Korea ( Franzoni et al., 2011 ). Papers with corresponding authors from the United States were found to be more likely to publish positive results than papers with corresponding authors from Asia, but more than with corresponding authors from Europe, Australia and Canada ( Fanelli, 2011 ). No clear predictions can be made at this stage, however, an additional test distinguishing between (co-)authors from the United States vs. other Western countries is included to further explore the notion of publication pressure cultures.
Finally, the outlet in which studies are published likely makes a difference for publication bias. There are two competing ideas on how journals' prestige could exacerbate or alleviate publication bias. Researcher often judge journals' prestige on the impact factor and send their best work to journals with high impact factors. Where higher impact factor journals often have the first pick from high quality studies, they can potentially filter out studies that tried many model specifications in order to find a positive results ( Fanelli, 2010 ). This would lead to less publication bias in higher impact factor journals. On the other hand, when journals want to maintain their high impact factor, it is better to publish studies that have the potential to receive more citations. High impact factor journals might be more inclined to accept more sensational papers, or at least papers which can prove something novel, filtering out negative results. It was found that positive results are indeed more likely to be published in journals with higher impact factors ( Murtaugh, 2002 ), but are also more likely to be erroneous ( Szucs & Ioannidis, 2017 ). These two scenario's lead to opposing expectations about the relation between impact factors and publication bias. Therefore, I will not formulate a directed hypothesis, but rather just explore the relation.

The current study
In the current study, I review all the published literature on neighborhood and community effects in countries in East and Southeast Asia. This literature is relatively new, especially compared to the already existing literature on neighborhood and community effects in Europa and the United States. The growing interest in this topic and the lack of a coherent overview of what has been done suggests the need for this review. The distinction between these two literatures from Western and Eastern countries provides ground for an inquiry into the role of research communities and how they influence publication practices. By studying patterns of publication bias in the East and Southeast Asian literature, I will give new insight into the role of research communities as well as the reliability of this literature as a whole. Both study lines might reveal new pathways into positive directions for the field of study.

Data
Relevant studies for this review were selected through a systematic search of Scopus, conducted on October 1st, 2018. The search query included two sets of terms to search through articles' title, abstract, and keywords: (1) a set of search terms relating to neighborhood, community, or residential effects, and (2) a list of country names of East and Southeast Asian countries 1 . A filter was used to limit the results to social scientific studies. This initial search resulted in 1,258 papers (see Fig. 2 for a flow chart of the data collection process). Additionally, to identify more relevant studies, manual searches of the bibliographies from the articles of the initial search were conducted. This resulted in three additional studies. The search query did not include a language filter, but did use English search terms. Therefore, non-English studies were only included when they had an English title, abstract, or keywords. This resulted in the inclusion of three Chinese and one Japanese study. There was also no restriction on peer-reviewed papers, which resulted in the inclusion of one book chapter and one conference paper.
To be included in the review, studies needed to meet the following inclusion criteria: (1) there is an individual-level outcome variable; (2) there is at least one neighborhood or community characteristic used in the independent variables; (3) some form of multiple regression is used; (4) the neighborhood or community are defined as the place where people live, and not a place where people often come for other reasons (e.g., the neighborhood around the school or workplace); and (5)  Besides all bibliographic information from Scopus, from each article, the following elements were recorded: the country (or countries) the research sample was from; used methods; used data; cross-sectional or longitudinal design; delineation of the neighborhood or community; sample size; sample age; outcome variable(s); neighborhood and community variable(s); p -values; and coefficients and standard errors (or equivalent information) to calculate exact p -values. Other information was constructed from the bibliographic information, such as information about (international) collaborations and research discipline of authors. Finally, journal impact factors for the years in which articles were published were derived from the Journal Citation Reports. Where journals did not have an impact factor they were set to zero. For the two oldest articles (from 1992 and 1993), the journal's impact factor could not be determined.

Variables
There are two main dependent variables of interest in this study: the exact p -values of study results and a dichotomous variable distinguishing positive (significant) and negative (nonsignificant) results. p -Values are derived from the studies when an exact one is reported. When exact pvalues are not present, but enough information is provided to calculate one (e.g., coefficients or odds ratios and standard errors or confidence intervals), I calculated them by hand. Positive (551 results) vs. negative (715) results are based on whether results had a significance of < . 05 or ≥ . 05 , respectively.
Year of publication was divided in three categories : 1992-2012, 2013-2015, and 2016-2018. These three categories were used because it provided enough filling of cells when calculating the caliper tests. The first category is much wider, with 21 instead of three years, however, because the number of studies only started to take off around 2010, these three categories all represent a portion of the total sample comparable in size (specifically, 33.33%, 24.85%, and 41.82%, respectively). In the regression I use publication year as a continuous variable.
To study the effect of origin of the researchers, five categories were constructed: the research was conducted by only local scholars (i.e., researchers from the same country as the research sample) (61 studies); the research was conducted by only Western scholars (43); local scholars collaborated with Western scholars (53); local scholars collaborated with non-Western scholars (4); and other (4). An additional distinction was made between studies that included an author with an affiliation in the United States (72 studies) or with an affiliation in another Western country (i.e., from Europe, Australia, or Canada) (29).

Analyses
To test the hypotheses about publication bias, I use a series of caliper tests ( Gerber & Malhotra, 2008 ). Caliper tests are based on the arbitrary significance boundaries used in statistical reporting. I use the common 5% significance standard (or z = 1.96), and two equal intervals just above and just below this boundary. Under the assumption that published articles are a random draw from a sampling distribution, the chance of the significance value of a result falling in either of the two intervals should be equal. However, when the boundary of significance somehow has an effect on what is being published, there could be an asymmetrical distribution over the two intervals.
To test the hypotheses about positive vs. negative results, I use a three-level logistic multilevel model, with results nested in substudy, nested in study. Study denotes a specific paper included in the metaanalysis. Substudy denotes separate analyses within a study (e.g., one over a migrant subsample and one over a local subsample, or over two subsamples of men and women). Examining the variance on the study-level ( 2 = 1 . 44 ) and substudy-level ( 2 = . 14 ) in an empty model, shows that there is 9% variance on the substudy-level, suggesting the three-level structure is warranted. "True " neighborhood effects can confound the results of this analysis. If, for example, in reality, individual education is more sensitive to the neighborhood environment than individual health, studies looking at education are more likely to find a positive (i.e., significant) result. However, social scientists are more likely to study education than health researchers. So, not including subject areas could lead to spurious results for research disciplines. The same argument applies to the type of neighborhood and community characteristics studied: for example, accessibility characteristics are most likely to be used by human geographers. This analysis includes both subject area as well as type of neighborhood characteristic to control for this. Finally, to better position the East and Southeast Asian literature in a wider context, methodological issues pertaining to spatial dependency and causality are examined.

Descriptive results
Neighborhood and community effects studies from East and Southeast Asia started to take off early 2010s, with 29 studies published between 1992 and 2009; 52 between 2010 and 2014; and 84 studies between 2015 and October 2018. As can be seen in Fig. 3 , the studies mainly cluster in East Asia rather than Southeast Asia. China leads the  Note: Cells denote the number of studies with a positive results ( n ) by the total number of studies ( N ), presented as n / N . When studies examined more than one subject-neighborhood characteristic combination, they are present in each of the cells they studied. When studies examined one subject-neighborhood characteristic combination multiple times in varied ways, they are only present in the respective cell once.
charts with 85 studies conducted about the country. In order of frequency, China is followed by Japan (25) (61) were conducted by local scholars, followed by collaborations between local and Western scholars (53) and studies by Western scholars only (43). Most first authors (84) come from a broad social science discipline, which includes primarily scholars from sociology, economics, and demography. Forty authors come from human geography and 41 from health research. The unequivocally most studied subject is that of mental or physical health. Table 1 shows that 50% of the study effort concentrates on individual outcomes concerning health, followed by social cohesion or disorganization (32%), socio-economic status (12%), and 6% for an 'other' category that mostly comprises studies on transportation. Poverty or disadvantage and social cohesion are the two most studied neighborhood and community characteristics, irrespective of the subject. Logically, health related neighborhood and community characteristics are mainly used to study individual health outcomes, but not so much the other outcomes. Unlike in many Western studies , are characteristics pertaining to the migrant or ethnic population of neighborhoods hardly studied.

Publication bias
When studying the histogram plotting the distribution of z -values ( Fig. 4 ), the number of reported z -values just under the critical value of 1.96 (i.e., results not significant at the p = .05 threshold) is about twice as small as the number just over z = 1.96. To test how likely this distribution is, I perform a series of caliper tests ( Table 2 ). The results for the total sample show that the division between over and under caliper results is very unlikely to have happened by chance, independent from the width of the caliper intervals. A robustness check that omits all z -values derived from coefficient or odds ratios with low precision (i.e., less than four digits), show the same results, albeit with weaker evidence. Comparing results by publication year shows no evidence for publication bias for papers published between 2013 and 2015, while there is evidence for publication bias for publication from before and after these years. When examining authors' affiliations it shows evidence for publication bias for studies with only local scholars. No to very weak evidence is found for studies from collaborations or from only Western scholars. Studies that include at least one scholar from the United States are much more likely to exhibit publication bias than studies that include at least one scholar from another Western context (i.e., Europe, Australia, or Canada). Examining research disciplines shows most evidence for publication bias in health research, and only weak evidence in social science and human geography. Finally, dividing studies by impact factor shows evidence  a The robustness check is a subset from the total sample, which omits z -values that were calculated from coefficients or odds ratios with low precision (i.e., less than four digits). Note 1: Over caliper includes the results that are between 0% and X% greater than the 1.96 z -value. Under caliper results are between 0% and X% less than z = 1.96. Note 2: Calculations of the p -value are based on a probability distribution function.
for publication bias in both lower and higher impact factor journals, but strongest evidence in higher impact factor journals.
To test the hypotheses about differences in positive vs. negative results, I study significant vs. nonsignificant results. Fig. 5 shows a trend of more nonsignificant vs. significant results over time. Table 3 shows the regression results for significant vs. nonsignificant outcomes. The covariates in the model explain 18% of the study-level variance and 73% of the substudy-level variance. To control for potential "true " neighborhood or community effects, subject areas and neighborhood characteristics are included. Neighborhood or community characteristics did not affect the chance of a positive or negative result. However, stud-ies that examined social cohesion or disorganization were more likely to find positive results. Publication year and affiliation did not show differences in positive vs. negative results. Studies from human geography are more likely to find positive results than studies from social science. Higher impact factor journals are less likely to publish positive results.

Methodological and data issues
To be better able to compare the state of the East and Southeast Asian literature to the more established Western literature on neighborhood  Note 1: This model includes 163 studies and 325 substudies. Note 2: In logistic multilevel models, the lowest level always has a fixed variance of 3.29. and community effects, two methodological issues are examined: spatial dependency and causality. To take care of spatial clustering within neighborhoods and to avoid underestimating p -values, multilevel models are typically used. Seventy eight out of 165 studies used multilevel models. A simple logistic regression that also includes whether studies use causal methods and cross-sectional or longitudinal data, shows that studies using multilevel models are indeed less likely to find a neighborhood or community effect (b = − 0.32, s.e. = 0.12, p = .005). Fourteen studies did efforts towards causal claims by applying techniques ranging from fixed-effects models to instrumental variables and propensity score matching. Also these studies show fewer significant results (b = − 73, s.e. = 0.23, p = .002). Around 40% of the studies are based on one-off data collections of limited sample size in specific cities or neighborhoods. Other studies are based on nationally representative samples or based on data from larger projects. Datasets that are the most frequently used for studies in this review are the China Family Panel Study, the China Health and Nutrition Survey, the China Health and Retirement Longitudinal Study, the Indonesian Family Life Survey, the Korean Children and Youth Panel Survey, and the Taiwan Social Change Survey. Most studies (147 out of 165) used cross-sectional data. The use of longitudinal data by itself did not alter whether studies found a neighborhood effect (b = − 0.21, s.e. = 0.23, p = .348). Most studies are based on one (part of a) country, however, six of the 165 studies take a comparative perspective, mostly with the United States, but also with other nations within and outside of the region.

Discussion
This study set out to examine the reliability of the emerging literature of neighborhood and community effects in East and Southeast Asia. Because this literature has not been systematically reviewed before, it is crucial to understand patterns of bias in order to assess the current body of work. This section starts by discussing the patterns of publication bias, with a specific focus on different research communities (both location-and discipline-specific) and stage in the life-cycle of the literature. Furthermore, by examining methodological issues, this literature can be placed in the wider context of worldwide neighborhood and community effects.

Publication bias
Overall, publication bias is very likely to be present in the body of literature studying neighborhood and community effects in East and Southeast Asia (hypothesis 1). The results presented in this review may be overestimating the "true " neighborhood effect. The relatively large dataset allowed for an extensive review of publication bias by examining subsets of studies. Opposite to hypothesis 2, a study of different research communities' shows strongest evidence for publication bias in the field of health research, rather than in human geography and general social science. Unsupportive of hypothesis 3, human geography studies are more likely to find positive results than general social science studies. The results are not supportive of the idea that subfield human geography and social science are more prone to publication bias than health research, because the idea of neighborhood effects would be accepted more generally there ( Shwed & Bearman, 2010 ). The results rather seem to suggest that different academic fields have different levels of competition and publication pressures, although this is speculative.
Studying the life-cycle of neighborhood effects in East and Southeast Asia reveals more publication bias in earlier studies compared to later studies (hypothesis 4), although no age effect for positive vs. negative results (hypothesis 5). This is finding is in line with the idea that early in life-cycle of a research subject, publication bias is more likely, be-9 JID: AJSS [m5GeSdc;August 1, 2022;22:41 ] cause novel ideas are easier to publish in journals in search of citations ( Gerber & Malhotra, 2008 ). Later in the life cycle it may become novel to disprove an established idea ( Gerber & Malhotra, 2008 ). However, the findings also show that publication bias is more likely in the most recent studies compared to studies in the middle of the life-cycle. This calls to question where in the life-cycle the study of neighborhood effects in East and Southeast Asia stands. Whereas the 90s and 2000s may be early in the life-cycle for studies of Western samples, Asian-based researchers came much later to this topic, exemplified by that the earliest studies in this review are conducted by Western-based scholars. Supportive of hypothesis 6, publication bias is much more likely in studies published by only local scholars, compared to studies by Western scholars or by collaborative teams of locals and Western scholars. This difference was not found between positive and negative results (hypothesis 7). The results suggest that the idea of neighborhood and community effects is in an earlier life-stage in East and Southeast Asia than it is in the Western literature. Alternatively, the findings can mean that publication pressures are stronger in East and Southeast Asia than in Western academia. The finding that publication bias is more likely when a scholar with a US affiliation is present compared to an affiliation from another Western country, is also suggestive of a more competitive academic climate in the United States compared to Europe, Canada, and Australia. The latter is in line with higher levels of publication pressure experienced by U.S.-based researchers ( van Dalen & Henkens, 2012 ). Finally, the analysis of publication bias by journals' impact factor has mixed results. The p -value analysis shows more evidence for publication bias in higher impact factor journals, while the positive vs. negative results analysis shows more bias in the lowest impact factor journals. There are two competing predictions, one predicts less bias in higher impact factor journals, because these journals have first choice from all written papers and are therefore able to weed out the bad studies ( Fanelli, 2010 ). The other predicts more bias in higher impact factor journals, because these journals need to maintain their high status through novel and sensational findings, filtering out negative results. No clear conclusion can be drawn about these two competing predictions.

Methodological issues
Two methodological issues consistently show up in the neighborhood effects literature, that of spatial dependency and that of causality. First, spatial dependency refers to the issue that people who live in a certain neighborhood are not independent of each other. People cluster in neighborhoods based on many different aspects, such as income, neighborhood preferences, social networks, and education. Not taking into account this interdependency of neighborhood residents when modeling neighborhood effects and individual outcomes, p -values may be underestimated, leading to spurious neighborhood effects ( Hox, 2002 ). Almost half (78 out of 165) of the studies used multilevel models and these studies were indeed less likely to find a neighborhood or community effect. Even though multilevel adoption is high, effort towards causal claims are still lacking. The 14 studies that made this effort were again less likely to find a neighborhood effect. Because most (147 out of 165) studies were based on cross-sectional data, model design options are limited and studying change is often impossible. This number is much higher compared to, for example, the 70% of US-based studies linking neighborhoods and health that use cross-sectional data ( Arcaya et al., 2016 ). That the use of longitudinal data did not change the likelihood of finding neighborhood effects might be because the sample of studies using longitudinal data is too small to find any differences. A review of health studies ( Jivraj et al., 2020 ) and a study of educational outcomes  show that accumulated exposure to neighborhood deprivation is harmful and early-in-life interventions are beneficial. Stronger longitudinal data from East and Southeast Asia will allow for such studies, but especially methodologically there is also still a lot to gain in this literature.

Defining neighborhoods
Another widely discussed issue is neighborhood delineation ( Galster & Sharkey, 2017 ). How many neighbors do you study at when you want to capture neighborhood effects? How close do neighbors have to live from your house for them to have an influence over you? Such questions are not easily answered. The studies in this review often poorly describe what constitutes neighborhoods or communities exactly, that is, how many residents or what area size they comprise and what the variation is. This is a problem that occurs in this literature globally and is not limited to East and Southeast Asia. Besides, many studies let respondents decide for themselves what they perceive of as their neighborhood, resulting in different conceptions of neighborhood even within studies. Working with perceived neighborhoods is not unsound, however, it does contribute to the incomparability of neighborhood and community effects studies. The studies that do provide information about the spatial delineation are often a source for expanding your non-English neighborhood/community-related vocabulary. South Korean studies use gu (districts), eup (towns), and dong (neighborhoods); Chinese studies talk about juweihui (residents' committees), zirancun (villages), jiedao (subdistrict), and Tertiary Planning Units (specific to Hong Kong); Japan has chocho-aza (small area units) and kyuson (literaly, old village; a unit based on old municipal boundaries; Hanibuchi et al., 2015 ); Taiwanese studies use townships, villages, and zipcode areas; Thailand has census blocks; the Philippines have barangay (smallest administrative division; corresponds to neighborhoods); and Indonesia has kabupaten (regency; rather large sub-provincial units). What stands out from this list is that researchers' approach to studying neighborhood and community effects is highly pragmatic and they draw their contextual data from local statistical and administrative agencies. The approach is similar to the Western literature, in which we can read, for example, about census tracts (the US), Lower Layer Super Output Areas (LSOA; the UK), and buurten and wijken (both mean neighborhood, but are defined differently; the Netherlands). The overlap in approach results in an overlap of challenges as well. Using administrative data is often the only way to obtain reliable contextual data, but it also means researcher have no influence on how the data are defined. Questions about on which spatial scale neighborhood and community effects occur cannot be answered easily. Just pointing out the difficulties of comparing neighborhood studies, however, does not help us much further. The East and Southeast Asian context also provides the opportunity to reassess our thinking of what constitutes a neighborhood.
One thing that makes many East and Southeast Asian neighborhoods distinct from their Western counterparts is the relatively high population density and the accompanying urban form of high rise buildings, as compared to many American and European cities ( World Bank, 2015 ). Although high density has typically been associated with a range of social problems due to environmental stressors ( Regoeczi, 2003 ;Yu et al., 2016 ), it has also been associated with the possibility for better planning, facilities, and transportation ( Yeh & Yuen, 2010 ), which contrasts with the problems of public service provision that accompanies urban sprawl ( Carruthers & Ulffarson, 2003 ). When we consider the social mechanisms through which neighborhood and community effects are argued to occur ( Galster, 2012 ), the question arises what high density living does to social relationships. Although large scale societies have traditionally been associated with a decline in community ( Tönnies, 2001 ), there is no consensus about this association in work on urban life ( Fischer, 1972 ). The opportunities for neighbors to meet in public spaces will most likely drive social relationships ( Mollenhorst et al., 2008 ). For example, in Hong Kong, people do seem to have a strong perception of what their neighborhood is, based on their daily activity patterns and architectural features of the environment, identifying public parks and shopping centers as important neighborhood features ( Forrest et al., 2002 ). Social networks within high density East and Southeast Asian neighborhoods often consist of neighbors who are also work colleagues or meet at school ( Forrest et al., 2002 ), or meet through parks or sports JID: AJSS [m5GeSdc;August 1, 2022;22:41 ] facilities ( Yoo & Lee, 2016 ) or neighborhood associations ( Sujarwoto & Tampubolon, 2013 ). Thus, social contact seems to arise around shared activities, which are facilitated by shared spaces and facilities. Neighbors will still be able to interact and thus influence each other when there are ample spaces to encounter one another. The importance of high density and high rise buildings for neighborhood and community effects studies thus hinges on the meeting opportunities that exist in an area. Another specific example in which the concept of neighborhood is very different from Western concepts comes from the residents' committees in China. In the neighborhood and community effects literature, mechanisms through which these effects transpire relate to local (voluntary) organizations and local social networks ( Putnam, 1995 ). In many Western studies, the questions is asked under which circumstances these strong local social networks might emerge, for example, in safer and more homogeneous neighborhoods. However, in the case of China's residents' committees, neighborhood organization is institutionalized. Residents' committees have elections and have tasks such as the management of social safety, prevention work, and the mediation of disputes within a specific residential compound ( Ngeow, 2012 ). As the spatially smallest administrative unit, these residents' committees thus play an important role in residents' lives. In general, better neighborhood social organization is related to lower disadvantage, more collective efficacy, and more ties in social networks ( Sampson & Graif, 2009 ). Better organized neighborhoods have more tools to exert control over its residents and have more social cohesion and interdependencies between neighbors ( Völker et al., 2006 ). These factors all relate to neighborhood social capital, which is an neighborhood characteristic that is argued to be important for neighborhood effects to come about ( Galster, 2012 ). For example, higher levels of neighborhood social capital have been related to better health outcomes in both the Western and the Asian context ( Mohnen et al., 2011 ;Nieuwenhuis, 2020 ). In the case of China, residents' committees should thus cause much stronger neighborhood effects than neighborhoods that do not have this institutionalized social organization. The important role of residents' committees can be illustrated by their central role in combatting SARS ( Ngeow, 2012 ) and COVID-19 ( Miao et al., 2021 ). More supportive residents' committees during the COVID-19 lockdowns in Wuhan were related to higher levels of social cohesion and increased mental health and wellbeing ( Chen et al., 2021 ;Miao et al., 2021 ). The residents' committees are unique to China and are an example of how strong local organization makes the neighborhood more salient to its residents. In countries without residents' committees, the neighborhood as an entity may not be much on people's radar. Individually perceived neighborhoods, as used as measures in many studies, will be much more likely to measure the same area for everyone in the Chinese context, but this will not be the case for contexts without such a formalized institution.

Ethnic diversity and migration
Ethnic diversity and migration is one of the most studied topics in the Western neighborhood literature (e.g., Wacquant, 2008 ), but one of the least studied topics in the East and Southeast Asian literature. Many Western countries are classically migrant destinations, while many East and Southeast Asian countries are traditionally countries of origin for migrants. Therefore, the literature from the United States about African American ghettos and the European literature about ethnically diverse and deprived neighborhoods does not have an obvious equivalent in East and Southeast Asian cities. However, with changing economic conditions and increasing spatial inequality within Asian countries, internal migration is increasing. Especially China is a well-studied example of this process, with large migrant streams from China's inland provinces to the Eastern coastal cities, migrant enclaves are increasing in number and size ( Li & Wu, 2008 ). The increase in socio-spatial inequality in cities increases the need for studies on the effects of this new spatial distribution of people and poverty.
Ethnic heterogeneity is often framed as a burden to social cohesion, leading to less trust between people ( Putnam, 2007 ). However, as social support networks, higher shares of co-ethnics are also argued to be a beneficial form of social capital ( Zhou, 2009 ). Studies from South Korea shows that local ethnic integration is actually more of a strain, resulting in lower income for foreign wives ( Kim, 2018 ). Social capital is also less useful to attain good jobs for foreign wives when neighborhoods have higher rates of discrimination ( Kim, 2017 ). Shares of co-ethnics in the neighborhood is also related to foreign wives having children earlier and having more ( Kim & Song, 2015 ). The findings from South Korea show that co-ethnic social capital can also form a constraint for migrant communities (cf. Portes & Sensenbrenner, 1993 ). Studies from China found that, although poverty is related to lower trust between locals and migrants, higher proportions of migrants in neighborhoods are actually related to more trust between locals and migrants and under some circumstances to more positive attitudes ( Nieuwenhuis & Shen, 2021 ;Wang et al., 2017 ). That this finding contrasts with many of the findings from the Western literature is potentially caused by the higher cultural similarity between rural migrants and urban locals in China compared to the more ethnically diverse populations of some Western neighborhoods. Because people expect interactions with similar others to be more rewarding due to shared knowledge and better mutual understanding ( McPherson et al., 2001 ), it is possible that cultural similarities make it easier for locals to accept migrant residents in their neighborhood.

Limitations
Systematically comparing studies as in this review is not an easy task. Systematic reviews are generally the best available evidence for the existence for the relation between as risk factor and an outcome. In this case the risk factor is neighborhood or community disadvantage (or another neighborhood-or community-level characteristic) and the outcomes range from health to education and wellbeing. Whereas metaanalyses in the field of medicine often have the advantage of studying standardized measures on both sides of the equation as well as experimental research designs, the field of neighborhood effects is much more "sloppy ". That is not to say that researchers just wing it, however, there is no broad consensus on how to measure concepts and strict experiments are generally not possible (because mostly you cannot force people to move to different neighborhoods just for the sake of your study). Where it so that neighborhood and community effects studies were as standardized as the study of medicine, then the toolkit to study publication bias would be bigger, for example allowing for the use of funnel plots ( Egger et al., 1997 ). This could potentially identify small study effects. Standardization also gives the opportunity to calculate point estimates of the effect magnitudes, which can give insight in the importance of neighborhood and community characteristics for individual outcomes. However, this is not so much a limitation present in the East and Southeast Asian literature per se, but in the whole neighborhood and community effects field.

Conclusion
It is clear that there are a lot of studies being produced in the field of East and Southeast Asian neighborhood effects and that this literature is still expanding. This review is the first to systematically bring all the studies together. The main conclusion is that it is too early to make any strong generalizations because the literature is rife with publication bias and the reliance on mostly cross-sectional data makes causal claims difficult. Furthermore, within-country variability is likely nonnegligible, and because studies often use very specific case studies to study neighborhood and community effects, within-country generalization are also difficult to make. This field of study would benefit tremendously from stronger data collection with longitudinal and generalizable samples (with the disclaimer that it is not a desert out there and efforts are currently already being made). This could also lead to more JID: AJSS [m5GeSdc;August 1, 2022;22:41 ] informative comparative work within the East and Southeast Asian region, as well as beyond those borders. The current state of the literature shows that neighborhood and community effects are just as likely to exist in East and Southeast Asia as in the Western world. Promising avenues to be taken lay in the situations that are very specific to East and Southeast Asia: The prevalence of high density, high rise neighborhoods may shed new light on how social relationships are formed. Rural-tourban migration in many countries may find similarities as well as differences in the Western literature on the impact of ethnic residential segregation. China's residents' committees provide insight into a situation where neighborhoods are institutionalized, which is rather uncommon in other contexts. With the inevitable maturation of this field, future comparisons between the Asian and Western literature will likely yield more interesting conclusions from which we might learn about differences and similarities between the two contexts.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.