Urbanization, and Child Mental Health and Life Functioning in Vietnam: Implications for Global Health Disparities

Abstract

Purpose:

Urbanization is linked to increased health risks, including mental health. However, the large majority of this research has been conducted in high-income countries, and little is known about effects in low-and-middle-income countries (LMIC) where urbanization is occurring most frequently and most rapidly. Within global mental health, children and adolescents are a critical but understudied population. The present study assessed relations between urbanization factors, and child mental health in Vietnam, a Southeast Asian LMIC.

Methods:

Most studies investigating urbanization and mental health have used geographically-based dichotomous urban-vs.-rural variables. Because of significant limitations with this approach, the present study assessed parent-reported urbanization factors (e.g., pollution, crime). In Sub-study #1 (cross-sectional), 1,314 parents from 10 Vietnam provinces completed the Urbanization Factors Questionnaire, Child Behavior Checklist (mental health), and Brief Impairment Scale (life functioning). In Sub-study #2 (longitudinal), 256 parents from one highly urban and one highly rural province completed the same measures, at three timepoints across 12 months.

Results:

Cross-sectional canonical correlations identified relatively small (e.g., R²=.08) but significant relations between urbanization factors, and child functioning. Parallel analyses using a geographically-defined urban-vs.-rural variable did not produce significant results. The large majority of longitudinal relations between different urbanization factors and child functioning were non-significant.

Conclusions:

This study, among the first to assess urbanization as a multi-dimensional continuous construct in relation to child psychopathology, highlights the value of use of urbanization factors approach. A new “urbanization factors differentials” theory is proposed to suggest how effects of urbanization factors might result in global health disparities.

BACKGROUND

The goal of the global health movement is to reduce health disparities between high income countries (HIC), and low and middle income countries (LMIC)[1]. Within this field, over the past decade there has been increasing attention focused on mental health, as it is recognized that mental health problems are a leading cause of morbidity and excess mortality worldwide while at the same have one of the largest HIC vs. LMIC disparities[2]. And within global mental health, child mental health (for conciseness the term “child” is used to refer to children and adolescents) has been identified as a priority area. Children represent one-third of the world’s population, many mental health conditions begin prior to adulthood, and early intervention provides for increased cost effectiveness[3]. However, in LMIC children’s mental health needs receive relatively little attention, even in comparison to the needs of LMIC adults, with global child mental health development lagging even more than global mental health development more generally[3].

One factor contributing to impaired health in LMIC may be the minimally controlled urban development that many LMIC experience, which can increase stress on the physical environment (e.g., via pollution) as well as social systems (e.g., via crime), with consequent increased risk for a wide variety of health disorders[4]. There is some evidence that urbanization is associated with increased risk for mental health disorders. In their meta-analysis, Peen and colleagues[5] found that living in an urban area was associated with a 38% increase in risk for a mental health disorder. However, the large majority of research in this area has been conducted in HIC, and its applicability to LMIC is unclear. In Peen and colleagues’ meta-analysis[5], all 20 studies were from high income countries, and in Solmi and colleagues’ recent comprehensive review of urban vs. rural differences in mental illness[6], less than 20% of datasets were from LMIC. Both reviews concluded that the lack of research in LMIC was an important gap in the research literature.

An additional complexity in regards to drawing strong conclusions is that results across studies have not been highly consistent[6], suggesting effects of unassessed moderators may be influencing results. This may be due in part to how “urbanization” has been operationalized. Urbanization often has been discussed and assessed as geographically-based differences (e.g., in health) between “urban” vs. “rural” areas[5], sometimes at neighborhood levels but typically more broadly[6]. There are several limitations with this approach[7]. First, urbanization and its effects are not unidimensional but are much broader than can be accurately captured in a single “urban vs. rural”, geographically-defined dichotomous variable[8,9]. Urban and rural areas may differ in regards to the physical environment on multiple factors (e.g., pollution, noise, crowding) that vary across sites, and they may differ in regards to multiple social environment factors (e.g., crime, the presence of unwanted but legal businesses such as short-term hotels used for sex) that vary across sites[7]. They may differ in regards to access to social and health services, and the availability of social support from neighbors. The rich complexity of these differences cannot be captured by a single “urban” vs. “rural” variable[10]. A second limitation with the use of a dichotomous urban vs. rural variable is that the level of these factors (e.g., crime; low social support) also will differ within urban areas and within rural areas, and use of a dichotomous variable does not capture this variability. That is, two geographical areas, both defined as “urban” (or as “rural”) may be quite different in terms of crime, social support, pollution, etc., and placing the two units of observation in the same category adds significant error to models[11]. Finally, an urban vs. rural variable is more distal from factors underlying health-related effects than assessment of urbanization factors (e.g., exposure to crime), making it more difficult to understand processes underlying effects of urbanization[12]. Knowing that exposure to high levels of neighborhood crime is associated with increased risk for adolescent conduct problems is more informative in regards to developing supportive interventions than simply knowing that living in an urban (vs. rural) setting is associated with increased risk for adolescent conduct problems. In the present study therefore, we used a continuous parent-reported “urbanization factors” approach as our independent variables, rather than geographically-defined, urban vs. rural areas of the country.

The study focused on the Southeast Asian LMIC of Vietnam. Vietnam has one of the most rapidly expanding economies in the world, in 2017 in the top 10% worldwide in terms of GDP growth[13], recently moving from lower income to lower-middle income status[14]. As part of this development, it has experienced significant urbanization with, for instance, the portion of its population living in urban areas increasing by 50% from 2000 to 2018[15], with accompanying stress on the social and physical environments[16,17]. In regards to child mental health, a child epidemiological survey that was nationally representative, including in regards levels of urban development, found that between 19% to 20% of children and adolescents had significant emotional (anxiety, depression) mental health problems, with 5% to 6% showing significant conduct / behavioral mental health problems[18]. Although there has been increasing interest in mental health in Vietnam, its mental health infrastructure remains under-developed, particularly in regards to children and adolescents[19,20].

METHODS

The present paper reports on two sub-studies, both of which were approved by the U.S. FWA IRB (#00018223) at Vietnam National University – Hanoi, with all participants providing voluntary informed consent.. This first sub-study, reported immediately below, was cross-sectional, utilizing a large national sample of families in Vietnam, to test the value of a multi-dimensional conceptualization of urbanization factors in relation to child mental health and life functioning

Participants, Sampling Frame, and Recruitment

The goal of the study sampling frame was to capture the range and variability in urbanization factors across the country, for a sample of 6-16 year children and their parents, with 10 of Vietnam’s 63 provinces selected based on: (a) level and rate of urbanization, (b) economic status, and (c) geographical character (e.g., mountain vs. coastal). Provinces included: (a) Binh Thuan (rural, under-developed), (b) Da Nang (regional capital, urbanizing); (c) Hanoi (national capital, rapidly urbanizing); (d) Hai Phong (major port city, urbanizing); (e) Ha Tinli (rural); (f) Hau Giang (rural, under-developed); (g) Ho Chi Minh City (industrial capital of the country, rapidly urbanizing); (h) Hoa Binh (rural, under-developed); (i) Phu Yen (rural); and (j) Thai Nguyen (semi-rural). Within each province three districts were chosen representative of the province, from (relative to the province) urban, semi-rural, and rural areas. Within each district, two neighborhoods were randomly selected. In Vietnam, all citizens must register with local authorities, and these population lists are public record. Residents are registered by household, with basic information including family members, and their age and gender. Using these lists, within each neighborhood 22 families were randomly selected for participation, stratified on gender and age of the child, from 6 to 16 years of age, producing 60 data collection sites and 1,320 potential participant families. Data were provided by parents, primarily the mother (75%). A total of 1,320 families were selected for recruitment, six of whom declined to participate, for a final sample consisting of 1,314 parents or guardians reporting on their child. Median annual parent income was $1,227, as compared to the 2010 Vietnam per capita annual GDI of $1,250[21]. Sample demographics and descriptive statistics for the various outcome variables are reported in Table 1.

Table 1.

Sub-study #1 sample descriptive statistics

Variable	Mean (SD)1,2
Informant (proportion mothers)	.75
Age of parent (years)	39.91 (6.74)
Child gender (proportion female)	.50
Age of child (years)	11.15 (3.17)
UFQ – Crime	2.22 (0.69)
UFQ – Neighbor Support	2.69 (0.51)
UFQ – Sleaze	1.48 (0.62)
UFQ – Unclean	1.85 (0.66)
CBCL – AnxDep	0.22 (0.21)
CBCL – WithDep	0.18 (0.23)
CBCL – Somatic	0.16 (0.21)
CBCL – SocialProb	0.23 (0.21)
CBCL – ADHD	0.33 (0.28)
CBCL – RuleBreak	0.08 (0.11)
CBCL – AggBeh	0.18 (0.20)
CBCL – Internalizing	0.19 (0.18)
CBCL – Externalizing	0.13 (0.14)
BIS – Interpersonal	0.13 (0.23)
BIS – School	0.20 (0.25)
BIS – Self	0.60 (0.35)

Notes:

^1.Unless otherwise indicated, descriptive statistics are the Mean and Standard Deviation.

^2.In order to increase interpretability, means and SD for the UFQ, CBCL, and BIS are presented on the item response scale: For the UFQ the scale is 1 to 3; for the CBCL 0 to 2; and for the BIS 0 to 3.

Procedures

The project was conducted by Vietnam National University-Hanoi (VNU). For each participating region, VNU officials contacted the provincial Department of Education to request project support; all agencies chose to participate. Provincial agencies identified local staff from each locale who accompanied the VNU project interviewer to the family’s house, introduced the interviewer and project to the family and then left. Project interviewers described the project in detail, obtained consent from those interested, and scheduled a time for the interview. Parents were interviewed in a private area of the home. Families were paid based on the economic level of their locale, ranging from about $4 to $10. Data were collected between April 2014 and September 2014.

Measures

All measures were scored following the guidelines provided in the measures’ manual or reports. Missing data was treated as missing.

Demographics.

In this study, all measures were completed by the parent, typically the mother. The demographics questionnaire assessed basic information including child gender and age, family structure, and parent education and income.

Child Behavior Checklist-VN.

The parent completed the Vietnamese version of the Child Behavior Checklist[22], which assesses children’s emotional and behavioral problems. It contains 118 items (e.g., “Fears going to school”) rated on a 0=Not True, 1=Somewhat or Sometimes True, 2=Very True or Often True scale. In the present study, the CBCL narrowband psychopathology scales Anxious-Depressed, Withdrawn-Depressed , Somatic Complaints, Social Problems, Attention Problems, Rule-Breaking Behavior, and Aggressive Behavior were used in one set of analyses, and the two broadband scales, Internalizing Problems (emotional problems) and Externalizing Problems (behavioral problems) in a second set of analyses. The CBCL is widely used and validated internationally[23], including in Vietnam[24]. Because this study was not focused on dichotomous clinical levels of symptoms, raw scores rather than t-scores were used.

Brief Impairment Scale.

Parents completed the Vietnamese version of the Brief Impairment Scale (BIS)[25] to assess functional impairment in three domains: (a) Interpersonal functioning (relationships with family members, friends, etc.); (b) School functioning (academic performance, and behavior at school); (c) Self-care (e.g., self-hygiene). BIS items (e.g., How much of a problem has your child had getting along with his/her father?) are rated on a 0 (no problem) to 3 (serious problem) scale. BIS scales are scored as the sum of the items. The BIS has been used and validated in Vietnam[25].

Urbanization Factors Questionnaire.

The Urbanization Factors Questionnaire (UFQ) was developed in Vietnamese for use in Vietnam and similar countries[25], based on reviews of conceptual urbanization papers[8,11,26] and existing urbanization factors questionnaires[27,28]. The reviews identified eight items that were translated and tailored for Vietnamese culture. Four additional items were generated to more fully cover the domains identified by the conceptual reviews. The UFQ’s 12 items assess four factors, with each factor containing three items, related to urbanization: (a) Crime (e.g., theft from homes or stores); (b) Unclean (physical environment; e.g., pollution; noise); (c) Social Evils (e.g., presence of karaoke bars), with this factor’s name taken directly from the Vietnamese term “tệ nạn xã hội” used to describe the content of the subscale; and (d) Neighbor Support (e.g., a neighbor with whom one could discuss a personal problem), reverse coded since it was hypothesized that neighbor support would be lower in urban areas. The UFQ is answered on a 1=No, 2=Not certain, 3=Yes scale for the presence of the item, with each scale representing the sum of the items for that scale.

RESULTS

1. Validation analyses.

Because the UFQ was developed as part of the present study, two sets of analyses were conducted to support its validity. First, using SAS (9.4) Proc GLM, means for the four UFQ factors were compared via an ANOVA across the two most highly urbanizing provinces (Hanoi, HCMC) and the three least developed provinces (Binh Thuan, Hau Giang, Hoa Binh), focusing on the most urban districts in the former, and the most rural districts in the latter. ANOVA for all four of the UFQ factors were significantly different as a function of the urban vs. rural factor, with Crime F(1,217)=8.81, p<.005, Neighbor Support F(1,217)=9.49, p<.005, Social Evils F(1,217)=31.92, p<.0001, and Unclean F(1,217)=5.76, p<.05. Means (SD) were in anticipated directions, with Crime (urban) = 2.34 (.66) greater than Crime (rural) = 2.05 (.77); Social Evils (urban) greater than 1.69 (.64) vs. Social Evils (rural) = 1.26 (.50); Unclean (urban) = 2.00 (.71) greater than Unclean (rural) = 1.80 (.57); and Neighbor Support (urban) = 2.46 (.62) lower than Neighbor Support (rural) = 2.70 (.53). A confirmatory factor analysis using SAS Proc Calis was conducted to support the validity of the UFQ’s hypothesized factor structure. Each of the four factors had the three UFQ items hypothesized to load on the factor, with the four factors allowed to correlate. Using maximum likelihood estimation, the model showed good fit, with RMSEA=.04, NFI=.93, and NNFI=.94.

2. Overall relations between urbanization factors, and mental health and life functioning.

Overall relations between the (a) urbanization factors and (b) CBCL narrowband scales were assessed with canonical correlation analyses using SAS Proc Cancorr, with a .40 loading cutoff. Canonical correlation analysis identifies relations between two sets of variables (in the present case, the UFQ and CBCL), by finding linear combinations within each set of variables that are maximally correlated with each other across the two sets of variables[29]. These linear combinations produce “canonical variates” that are similar to the factors produced in exploratory factor analysis. The first canonical correlation (between the two canonical variates) represents the linear combination of the variables within each set that have maximum correlation between sets, using a statistical process similar to that used in multiple regression. Subsequent canonical correlations are based on the residual covariance and represent unique relations relative to preceding canonical relations. The loadings of each variable (e.g., the UFQ Crime factor) on the canonical variate represent, as in factor analysis, the relation (often presented as a correlation) between the item and the canonical variate. In our analysis, the overall canonical relation between the UFQ and CBCL was significant, with Wilk’s Lambda=.92, F(28,4710)=3.82, p<.0001, R²=.08. Three (of four) individual canonical relations between the UFQ factors and the CBCL narrowband scales were significant. As Table 2 (Analysis A) reports, overall there was a positive correlation between urbanization and child mental health problems. In the first canonical relation, all of the CBCL scales loaded positively on their canonical variate, and three of the four of the UFQ factors loaded on their canonical variates (neighbor Support had a negative loading, reflecting the fact that lower levels of neighbor support were associated with higher levels of urbanization). Thus, results indicate that higher levels of urbanization were associated with higher levels of child mental health problems. To test whether these relations might be due in part to confounding with demographic variables, we included family income, and mother’s and father’s education as covariates in the model; the overall model remained significant at p < .0001.

Table 2.

Cross-sectional canonical correlations

Analysis A – UFQ and CBCL
Overall analysis: F(28,4710)=3.82, p<.0001, R2=.08
	Canonical variate
	#1	#2	#3
Canonical Correlation	.20****	.15****	.12*
	Loadings1
UFQ – Unclean	0.51	0.65	0.19
UFQ – Crime	0.35	0.22	0.87
UFQ – Neighbor Support	−.79	0.59	0.09
UFQ – Sleaze	0.59	0.50	−.26
CBCL – AnxDep	0.72	0.24	0.23
CBCL – WithDep	0.82	−.44	0.28
CBCL – Somatic	0.62	0.22	−.32
CBCL – SocialProb	0.60	0.20	0.46
CBCL – ADHD	0.78	0.48	0.15
CBCL – RuleBreak	0.51	0.18	0.44
CBCL – AggBeh	0.76	0.13	−.01
Analysis B – UFQ and BIS
Overall analysis: F(12,3469)=4.36, p<.0001, R2=04
	Canonical variate
	#1	#2
Canonical Correlation	.17****	.10*
	Loadings1
UFQ – Unclean	0.26	0.62
UFQ – Crime	0.44	0.64
UFQ – Neighbor Support	−.90	0.41
UFQ – Sleaze	0.37	0.60
BIS – Interpersonal	0.72	0.66
BIS – School	0.41	−.16
BIS – Self	0.83	−.55

Notes:

^*= p < 05;

^**= p < 01;

^***= p <.001;

^****= p <0001 for canonical correlations. Bolded coefficients are above the .40 loading cutoff.

¹= Loadings are for loadings of the UFQ, CBCL, and BIS on their canonical variate, in the metric of a correlation.

We conducted a second canonical correlation analysis, between the (a) four urbanization factors and (b) three BIS scales. The overall canonical relation was significant for relations between the UFQ and the BIS, with Wilk’s Lambda=.96, F(12,3469)=4.36, p<.0001, R²=.04; two (of three) individual canonical relations between the UFQ factors and the BIS were significant (see Table 2). As with the CBCL, the loadings (Table 2, Analysis B) in this canonical correlation analysis indicated there was a positive relation between urbanization and life functioning impairment (i.e., higher levels of urbanization were associated with higher levels of child functional impairment). As with the analyses with the CBCL, we tested whether these relations might be due to confounding with demographic variables by including family income, and mother’s and father’s education as covariates in the model; the overall model remained significant at p < .0001.

3. Relations between dichotomous geographic variable, and mental health and life functioning.

To assess the extent to which these relations between urbanization factors, and child mental health and life functioning might (or might not) be captured by a dichotomous, geographically-based urban vs. rural variable, we conducted two MANOVA, using the multivariate option in SAS Proc GLM. In both MANOVA, the dichotomous independent variable was the same variable in the Validation Analyses (above) that compared the two most highly urbanized provinces versus the three least developed provinces. In the first MANOVA, the dependent variables were the CBCL narrowband factors, and in the second MANOVA the dependent variables were the three BIS variables. These MANOVA are statistically equivalent to the canonical correlations in the analyses that assessed overall relations between the urbanization factors, and mental health and life functioning. However, both MANOVA were non-significant, F(7,213)=1.27, and F(3,216)=2.15, respectively.

4. Moderating effects of urbanization factors on relations between child psychopathology and functional impairment.

The last set of analyses focused on the extent to which urbanization factors moderated relations between child psychopathology and child functional impairment. Psychopathology is defined as “pathology” based on its relation to life functional impairment[30]; i.e., it is relations of symptoms such as anxiety and depression or conduct problems to life functioning that make these conditions “pathology”. One possible effect of urbanization factors might be to make relations of mental health symptoms more or less closely linked to life functional impairment. For instance, it is possible that under high levels of urbanization factors, relations between psychopathology and functional impairment will be stronger, as the urbanization factors overload the child and family’s ability to cope with the psychopathology, with the psychopathology consequently impacting more on the child’s life functioning. On the other hand, it is possible that at high levels of urbanization factors relations between psychopathology and functional impairment might be weaker, as the urbanization factors themselves impact as overlapping variance on life functioning. To evaluate these possibilities, a series of general linear model analyses were conducted in SAS Proc GLM, with life functioning (the BIS scales) as the dependent variables, and the interaction between the urbanization factors and CBCL factors (as well as their main effects) the independent variables. To interpret significant interactions, we computed standardized beta parameter estimates for the relation between the CBCL and the BIS, at −1 standard deviation below the mean and at +1 standard deviation above the mean urbanization factor[31,32]. To minimize the number of individual analyses, these analyses focused on the two CBCL broadband subscales. Of the 24 analyses (2 CBCL scales X 3 BIS scales X 4 UFQ scales), 10 of the interaction effects were significant (see Table 3). To determine the probability that this overall result reflected capitalizing on chance rather than true population differences, we conducted a likelihood chi-square test assessing the probability that 10/24 of the tests would be significant by chance, given a population null hypothesis (that .05 of the tests would be significant, due to chance). The test was significant, with χ₍₁₎=10.02, p<0.002, indicating that our findings deviated significantly from the overall null hypothesis; we therefore proceeded to the interpret the interactions. Results were consistent across models, with the relation between the CBCL and the BIS larger at lower levels (−1 SD from the mean) of the urbanization factors than at higher levels (+1 SD from the mean) (see Table 3).

Table 3.

Cross-sectional moderator analyses

Dependent Variable	CBCL	UFQ	F(1,1313)	β @ −1 SD	β @ +1 SD
BIS InterPersonal	Ext	Crime	13.24***	0.47	0.28
BIS InterPersonal	Ext	Social evils	6.23*	0.44	0.31
BIS InterPersonal	Ext	Neighbor Support	2.57
BIS InterPersonal	Ext	Unclean	12.13***	0.47	0.29
BIS InterPersonal	Int	Crime	7.54**	0.35	0.21
BIS InterPersonal	Int	Social evils	0.02
BIS InterPersonal	Int	Neighbor Support	1.51
BIS InterPersonal	Int	Unclean	0.90
BIS School	Ext	Crime	32.76****	0.51	0.20
BIS School	Ext	Social evils	0.37
BIS School	Ext	Neighbor Support	1.96
BIS School	Ext	Unclean	14.11***	0.46	0.26
BIS School	Int	Crime	0.85
BIS School	Int	Social evils	0.47
BIS School	Int	Neighbor Support	3.13
BIS School	Int	Unclean	6.87**	0.36	0.21
BIS Self	Ext	Crime	9.62**	0.40	0.23
BIS Self	Ext	Social evils	0.02
BIS Self	Ext	Neighbor Support	0.19
BIS Self	Ext	Unclean	6.26*	0.39	0.25
BIS Self	Int	Crime	1.37
BIS Self	Int	Social evils	0.68
BIS Self	Int	Neighbor Support	5.85*	0.21	0.10
BIS Self	Int	Unclean	0.20

Notes. F value is for CBCL x UFQ interaction term. Main effects of the CBCL and UFQ were included in models but are not included in this table because they are not directly relevant to the research question, β = beta for relation between CBCL and BIS, at −1 / + 1 SD from the mean of the urbanization factor.

^*= p <.05

^**= p <.01

^***= p <.001

^****= p <.0001

METHODS

The second sub-study was a short-term (12 month) longitudinal study, designed to identify potential causal effects of urbanization on child functioning. Because the study was one of the first to assess relations between urbanization factors and child mental health in an LMIC, analyses were considered exploratory, and no specific hypotheses were made.

Participants, Sampling Frame, Recruitment, and Procedures

The longitudinal sub-study involved a separate sample of participants recruited from one highly urban area of Vietnam (Hanoi city) and one highly rural area in Vietnam (Hai Duong province). Following the focus of this project on use of an urbanization factors approach, the goal of this sampling frame was to not to create urban vs. rural groups but rather to provide a range of conditions related to urbanization (as assessed by the UFQ). The same recruitment procedures used for the cross-sectional sub-study were used for the longitudinal sub-study, with one exception. Because of increased rates of youth psychopathology among adolescents[33], the longitudinal sub-study focused on adolescents, aged 11 to 16 years of age. Data were collected at three time points, with six months between timepoints. At Time 1, 260 families were invited to participate in the study, with 256 enrolling in the study and providing T1 data, 253 at T2 (6 months), and 250 at T3 (12 months), with data collected from March 2015 to April 2016. Data were provided by parents, primarily the mother (76%). Table 4 provides sample demographics as well as Time 1 scores on the UFQ, CBCL, and BIS.

Table 4.

Sub-study #2 sample descriptive statistics

Variable	Mean (SD)1,2
Informant (proportion mothers)	.76
Mean (SD) age of parent (years)	39.91 (6.74)
Gender of child (female)	.51
Mean (SD) age of child (years)	11.15 (3.17)
UFQ – Crime	2.07 (0.70)
UFQ – Neighbor Support	2.61 (0.56)
UFQ – Sleaze	1.50 (0.63)
UFQ – Unclean	1.84 (0.67)
CBCL – Internalizing problems	0.20 (0.18)
CBCL – Externalizing problems	0.13 (0.15)
BIS – InterPersonal	0.15 (0.24)
BIS – School	0.21 (0.25)
BIS – Self	0.87 (0.39)

Notes:

^1.Unless otherwise indicated, descriptive statistics are the Mean and Standard Deviation.

^2.In order to increase interpretability, means and SD for the UFQ, CBCL, and BIS are presented on the item response scale: For the UFQ the scale is 1 to 3; for the CBCL 0 to 2; and for the BIS 0 to 3.

Measures

The same measures used in the cross-sectional sub-study were used in the longitudinal sub-study.

Analyses

The primary purpose of the longitudinal analyses was to determine the extent to which change in child functioning in mental health (CBCL) and life impairment (BIS) across the three timepoints varied as a function of the level of urbanization factors as reported by the parent at T1. Towards this end, we used a mixed models analysis[34], with Time (the three timepoints) as a random factor nested within Subject, the two broadband CBCL and three BIS subscales as the dependent variables (across the three timepoints), and Time 1 levels of the four Urbanization Factors Questionnaire as fixed effect predictor variables. The effect of Time represented the slope for change in the dependent variables across the three timepoints, taking into account the cross-timepoint (i.e., lagged) relations in the dependent variable. A positive Time parameter estimate indicates an increasing level of the dependent variable across time whereas a negative Time parameter estimate indicates decreasing levels. The Time X UFQ interaction was the effect of interest, assessing the extent to which these slope for the dependent variable varied as a function of Time 1 levels of the UFQ factors.

RESULTS

Following the above analytic framework, across the 20 analyses (5 dependent variables X 4 UFQ factors), one Time X UFQ interaction was significant, the Time X Unclean interaction on BIS School Functioning, F(1,500)=8.34, p<.005. The likelihood chi-square test for the extent to which this deviated from the overall null hypothesis (that .05 of the tests would be significant, by chance, given a population null hypothesis) was non-significant, with χ(1)=0.00, p=1.00. Consequently, this interaction was not interpreted. In order to help interpret these null findings themselves, we tested the extent to which child functioning was stable, and thus unlikely to have predictive relations. The statistical model involved the CBCL and BIS variables as dependent variables, with Time as a random factor assessing the slope of change. Of the five tests, one was significant, for CBCL Internalizing Problems subscale which decreased slightly over time, F(1,255)=18.64, p<.0001, β_Time = −0.18. We also tested the stability of the urbanization factors, using similar analyses. The one significant effect was for Social Evils which increased slightly over time, with F(1,255)=5.56, p<.05, β_Time = 0.09; none of the other urbanization factors showed significant change across the one year.

GENERAL DISCUSSION

Results of the validation analyses provide support for the validity the Urbanization Factors Questionnaire, with all four of the UFQ subscales showing significant differences in the expected directions between highly urban and highly rural areas of Vietnam, and the confirmatory factor analysis supporting the four factor structure of the measure. In the substantive cross-sectional analyses, child psychopathology and life impairment were positively correlated with urbanization (i.e., higher levels of urbanization factors were associated with higher levels of child psychopathology and functional impairment), suggesting that urbanization may be a risk factor for increased child impairment, and highlighting the utility of the urbanization factors approach. However, in the longitudinal analyses, overall urbanization was not a significant predictor of change in child functioning.

More specifically, in the cross-sectional substantive analyses, three of the four UFQ – CBCL canonical correlations, and two of the three UFQ – BIS canonical correlations were statistically significant. In regards to the UFQ – CBCL relations, the first canonical correlation represented a general relation between urbanization factors and child psychopathology. All of the CBCL factors were above the .40 loading cutoff, as were three of the four UFQ factors (Unclean, Low Neighbor Support, and Social Evils). The first canonical relation thus indicates a broad relation with higher levels of urbanization factors associated with higher levels of child psychopathology, with the child psychopathology and urbanization factors perhaps linked through general stress rather than more specific processes, given the broad nature of both canonical variates. In the second canonical relation, the CBCL canonical variate was defined by Attention Problems, and Withdrawn – Depressed with a negative loading. The child psychopathology variate thus appears to represent poor attention and high / excessive energy levels (i.e., hyperactivity), reflected in the CBCL Attention Problems factor (which includes hyperactivity items such as Can’t Sit Still) but also in the negative loading for Withdrawn – Depressed which includes items assessing low energy and activity levels in the child (e.g., Underactive, slow moving, or lacks energy)[35]. On the UFQ side, the canonical variate again represented a general urbanization factor, with three of the four UFQ factors (excluding Crime) above the loading cutoff, although for this canonical variate the loading for the UFQ Neighbor Support factor was positive. This canonical relation suggests that attention problems / hyperactivity have a distinct relation with urbanization factors, beyond the broad relation between child psychopathology and urbanization found in the first canonical relation, with higher levels of urbanization associated with higher levels of attention problems / hyperactivity. In considering the positive loading for Neighbor Support on its canonical variate, it is important to consider that canonical relations represents residual covariance in relation to the prior canonical relations, in this case the general relation between child psychopathology and urbanization factors in the first canonical correlation. The reason Neighbor Support had a positive loading may be that a parent may receive and be more aware of potential neighbor support if they are coping with a child with easy to identify, public problems such as attention and hyperactivity problems[36].

The third canonical relation reflects a specific relation between the Crime urbanization factor, and child rule-breaking behavior (stealing, lying, etc.), with higher levels of crime urbanization associated with higher levels of child rule-breaking. This finding is similar to the significant literature linking exposure to neighborhood crime, and child and adolescent conduct problems via social modeling, peer pressure, etc.[37], although most of this literature has not focused on crime in the context of urbanization per se. The positive loading on the CBCL variate of the Social Problems narrowband factor likely reflects the fact that children with significant conduct problems often also have significant social problems, as they tend to have conflicts with the broader peer group, and are rejected in response to their anti-social behavior by the broader peer group[38]. It is interesting that this was the only canonical relation involving the UFQ Crime factor, which suggests that there are independent effects between urbanization-related exposure to crime and child functioning, relative to the broader relations in the first and second canonical relations.

Similarly, the overall canonical relation between BIS functional impairment and the urbanization factors was significant, with R²=.04, and two of the three canonical relations significant. The first canonical relation involved all three BIS subscales, and UFQ Crime, and UFQ Neighbor Support (with a negative loading), indicating that higher levels of urbanization were associated with higher levels of child life impairment . Given that this is the first canonical correlation, this suggests that exposure to crime and low neighbor support are particularly closely linked to impaired child functioning, perhaps because low neighbor support for the parent and family, and exposure to or threat of crime may have a more direct impact on a child’s life functioning than the other UFQ factors such as pollution[39]. In the second canonical relation involving the BIS, the urbanization canonical variate represented a general factor with all four of the UFQ subscales above the loading cutoff, with the BIS canonical variate including the interpersonal and self-impairment domains but not the school domain. The fact that the interpersonal domain (which focuses on the family) and the self domains loaded on the variate but the school domain did not load on the variate suggests that the effects represented by this second canonical relation may be more directly connected to the local neighborhood where urbanization processes are occurring. The structure of this second canonical relation was relatively complex. On the BIS canonical variate, the BIS Self domain had a negative loading whereas the BIS Interpersonal domain had a positive loading. One possible explanation is that urbanization likely impacts on family members and friends as well as on the study-targeted study participant him or herself, putting more stress on interpersonal relationships. Under such circumstances, people may attempt to cope by taking better care of themselves and participating in enjoyable life activities, two main areas targeted by the BIS Self scale, resulting in the Self domain moving opposite to the Interpersonal domain. The positive loading for the UFQ Neighbor Support factor may be a function of awareness of social support being higher when there is a greater need, related to one’s child having difficulty in life functioning.

A central study finding was that although there were five significant canonical correlations between child functioning and urbanization factors, none of the relations assessing urbanization as a dichotomous urban vs. rural variable were significant. In these analyses, MANOVA were used, which are statistically equivalent to a canonical correlation with a dummy-coded dichotomous variable on one side of the equation. In contrast to our study, some but not all prior research using geographically-defined dichotomous variables has found urban vs. rural differences in mental health functioning[5,6]. One possible explanation for why cross-study results for geographically-defined dichotomous variables are at least somewhat inconsistent may be that when these dichotomous units are heterogeneous (i.e., different “urban” or “rural” areas within a study are particularly heterogeneous on the underlying urbanization factors), results may be less likely to be statistically significant. These results highlight the value of an urbanization factors approach, that it can capture within-location variations on participants’ experiences of urbanization. However, geographically defined urban vs. rural variables do have the advantage that they are and can use pre-existing population data, and one disadvantage to the urbanization factors approach, at least as implemented in the present study, is that the level of the urbanization factors is subjectively rated by the participants. A recent study by Generaal et al. (2018)[7] directly compared use of objectively obtained urbanization factors data (e.g., neighborhood air pollution from a national database) vs. a dichotomous urban / rural variable. As with the present study, that study found that a number of these urbanization factors were related to mental health problems, but not the dichotomous geographic variable.

The cross-sectional moderator analyses found that for 10 out of 24 relations between the BIS and the CBCL, the relation was smaller at higher levels of urbanization factors. One possible explanation for this apparent protective effect of the urbanization factors is that areas with higher levels of crime and pollution, etc. resulting from urban development may also have higher levels of mental health related support in clinics or other facilities. Thus, this finding might have resulted from a confounding between these urbanization factors and access to health care, which might mitigate the effects of psychopathology on life functioning. This is of course speculative, and unfortunately access to health care was not assessed in the present study, but this does suggest an important area for future research. It is important to note that 10 of 10 significant effects in these analyses showed the same pattern (i.e., that higher levels of the urbanization factor were associated with a smaller relation between the CBCL and BIS). The overall significance of the 10 of 24 significant analyses was less than .005, which suggests that this finding was not simply due to random variability.

Finally, only 1 of 20 longitudinal predictive relations was significant, which is the expected rate of Type 1 errors if the null hypothesis were true in the population. One possible explanation for the lack of significant results is that relations between urbanization factors and child functioning identified in the cross-sectional analyses may be non-causal, due to various confounding variables. However, it is also possible that the one year time frame of the longitudinal sub-study was too short to identify change in child functioning. Although certainly not stable, urbanization factors by their nature tend to be chronic rather than acute in their onset (as opposed to stressors such as death of a family member, loss of parent employment, etc.), and thus their effects may be relatively difficult to identify with short- to medium-term longitudinal studies.

Urbanization factors differentials, and global health disparities

By their nature, high-income / developed countries are more urbanized than low-and-middle-income / lesser developed countries; e.g., in 2018, approximately 83% of people living in the UK were living in urban areas whereas 36% of people in Vietnam were in urban areas[40]. Thus, it might seem counterintuitive that urbanization could be related to global health disparities, since HIC will generally have higher levels of urbanization. However, there are both positive and negative aspects to urbanization, and urbanization does not occur in a uniform manner. Positive aspects of urbanization include increased economic opportunities, greater access to health care and education, and other related attributes[10]. Negative aspects include increased crime, reduced neighbor support, etc. factors assessed by the UFQ. These different attributes do not develop equally rapidly. For instance, almost by definition economic opportunities related to industrial development (e.g., factories) will develop early in the urbanization process and at least initially, typically consequent degradation of the physical (e.g., pollution; noise) and social environments (e.g., crime) although not uniformly[10]. Conversely, increased access and quality of health care and education likely will typically develop more slowly, as the government and business respond to increased need, and develop the infrastructure and capacity over time. Thus, in urban areas in LMIC, positive aspects of urbanization such as increased access to health care and education may be less developed than in HIC. As a consequence of this “urbanization factors differentials”, urbanization in LMIC may be more associated with increased risk for health problems, including mental health problems, increasing HIC vs. LMIC health disparities. To test this theory would require assessment of both positive and negative urbanization factors, in less and more developed areas of a country, in LMIC vs. HIC.

Limitations

One limitation indicated by the above discussion is that the Urbanization Factors Questionnaire focused only on negative aspects of urbanization, and did not include positive aspects of urbanization such as increased access to health care or education. Inclusion of such positive urbanization factors potentially could have provided some support for the Urbanization Factors Differentials (UFD) theory. For instance, the UFD theory suggests that there would be increased access to health care in areas that had been urbanizing for longer periods of time (i.e., in HIC), and that increased access to health care would be negatively correlated with mental health problems and life impairment in general Thus, a potential area of future research may be further development of the Urbanization Factors Questionnaire, to include positive urbanization factors (e.g., access to health care and higher quality education). Another study limitation was that the longitudinal study had a relatively short time-frame, one year, which may have reduced the study’s ability to identify predictive relations. Finally, data were provided by a single informant, the parent. Thus, the extent to which relations were influenced by informant bias is unclear.

CONCLUSIONS

Most centrally, the results of this study suggest that when considering effects of urbanization, urbanization factors may be a more powerful method than geographically defined urban vs. rural variables. It also will be important for future research to assess positive as well as negative aspects of urbanization, such as access to health care. Such an approach should provide more information regarding potential intervention targets, and support more effective policy planning at the public health level. Results also suggest the attention problems / hyperactivity, and conduct / behavior problems may be two areas of child psychopathology with specific processes linking to urbanization, that thus may warrant special attention. Finally, although there has been a moderate amount of research investigating effects of urbanization in HIC and to a lesser extent in LMIC, to date there has been little consideration of how urbanization might play a role in global health disparities, in particular global mental health disparities. The Urbanization Factors Differentials theory suggests a testable hypothesis for how this might occur, and may be a useful focus for future research in this area.