Evaluating Community Factors Associated With Individually Held Intimate Partner Violence Beliefs Across 51 Countries

Abstract

Globally, one in three women will experience intimate partner violence (IPV) within her lifetime. IPV attitudes are highly predictive of IPV. While a wealth of literature on risk factors related to IPV exist, an overarching critique in the field is the lack of studies examining risk factors across the socioecological framework. Using data from multiple administrative and individual surveys, this study fills a gap in the literature by evaluating the effect of meso-influences on physical IPV attitudes (i.e., permissibility of a man beating his wife) while accounting for known micro- and macro-risk factors in 64,466 individuals across 51 low-, middle- and high-income countries. Mixed-effects modeling was used to evaluate factors and identify comparative contributions of each factor representing the socio-ecological levels. We tested five multivariate logistic models. The final model indicated that greater perceived neighborhood disorder and less perceived neighborhood security were associated with physical IPV attitudes, while individual endorsement of interpersonal violence, belief in corporal punishment of children, holding greater patriarchal beliefs, being male, being separated from a significant partner, reporting greater household hunger and nationally lower levels of female literacy were associated with beliefs that IPV is acceptable. Overall, the findings of this study support that IPV is a complex behavior, influenced by factors across socio-ecological domains. However, data on neighborhood structural factors (i.e., exosystem) would help unpack the mechanisms between macro-, meso- and micro-level factors and may be important for protecting women from violence.

Overview

Intimate partner violence (IPV) is a global problem with one in three women reporting physical and/or sexual violence lifetime victimization within an intimate partner relationship (García-Moreno et al., 2013). As IPV is a complex phenomenon, the factors that influence this type of violence are multifold spanning the socioecological model (Heise, 1998). A growing area of interest is understanding attitudes involving the acceptability of IPV across contexts (Kovacs, 2018; Sardinha & Catalán, 2018; Serrano-Montilla et al., 2020; Tausch, 2019). Believing IPV is acceptable under a given set of circumstances is thought to be highly predictive of IPV behavior, and understanding what influences attitudes is critical for informing prevention strategies (Izugbara et al., 2020; Nayak et al., 2003; Kury et al., 2004). While studies on IPV attitudes exist, the majority of studies that consider multiple levels of the socio-ecological model evaluate macro-level factors and individual-level factors but fail to account for the influence of community-level (meso- or exo-) factors on attitudes (Kovacs, 2018; Sardinha & Catalán, 2018; Serrano-Montilla et al., 2020). To help fill gaps in the literature, we coalesced data from several sources to evaluate the effect of perceived community influences on IPV attitudes while accounting for known micro- (individual, interpersonal), and macro-level (national indicators) risk factors across 51 low-, middle-, and high-income countries.

Utilizing a Socioecological Framework

Given its complexity, many researchers have called for IPV to be studied using the socio-ecological model (Capaldi et al., 2012; De Koker et al., 2014; Hardesty & Ogolsky, 2020). The socio-ecological model emphasizes that individual behaviors, including IPV, are subject to multiple levels of influence including micro- (individual), meso- (interactions with socially constructed neighborhood, community), exo- (services and structures in the neighborhood, community, and/or workplace), and macro- (widely shared cultural beliefs, laws) levels (Bronfenbrenner, 1977). The benefit of using an ecological model framework is that it allows researchers to expand conceptualizing behaviors beyond one-level of influence (Sallis & Owen, 2015) and helps to promote the idea of multiple level of influences on behavior. As such, the socio-ecological model can serve as a wider framework to which other theories are applied.

Micro-level Factors and IPV Attitudes

While researchers generally agree that more studies that apply a socioecological framework to IPV research are needed (Dardis et al., 2015; Hardesty & Ogolsky, 2020; Heise, 1998; Lawson, 2012), most empirical studies focus on factors at the micro-level (individual) of the socioecological model (Capaldi et al., 2012; Dardis et al., 2015). At the individual-level, researchers have found that rates of IPV perpetrated against women is greatest in contexts where the use of violence is normative within the society (Jewkes, 2002); this includes endorsement of corporal punishment against children (also known as use of physical punishment, such as spanking or hitting) (Button, 2008; Lansford et al., 2020) and greater attitudes on the permissibility of physical violence in certain social situations (e.g., to hit back if one is hit) (Lysova & Straus, 2019). These associations may be in part explained by Cultural Spillover Theory, whereby the permissibility of violence in one context may increase non-permissible forms of violence in other circumstances or contexts (Baron et al., 1988). For example, individuals who believe it is permissible to use violence against children or against another human may result in individuals believing partner violence is permissible, even in contexts where it is not legal. Thus, when evaluating IPV attitudes, it seems important to consider how beliefs around allowable forms of violence may inform IPV attitudes.

Another major construct in IPV work is gender inequality. At the individual level, researchers have evidence that holding patriarchal beliefs is associated with IPV attitudes (Hayes & Boyd, 2017; Hossain et al., 2020; Uthman et al., 2009), and that IPV attitudes are differentially held by men and women (Uthman et al., 2010b). Feminist theory can conceptualize the occurrence of violence against women at the hands of men relative to power structures across the socio-ecological framework. Starting in the earliest conceptualizations of Feminist Theory in the late 1970s and early 1980s, inequality between men and women led to men having more power than women across settings including the household, workplace, and education, and thus men used power over women in various ways including enacting violence against women (Dobash, 1979; Tierney, 1982; Walker, 1979). Though Feminist Theory has had multiple iterations, the central tenement of one socially constructed identity (such as male gender) having more power within a given society leads to violence against other identities is fundamental to understanding why IPV is pervasive, and why so many individuals may find IPV to be permissible.

Finally, several demographic variables are important, including: age, given that risk of IPVexposure changes over time (Abramsky et al., 2011; Peterman et al., 2015); relationship type, with those who are married or separated being at greater risk (Abramsky et al., 2011; Tran et al., 2016); and education level and household socioeconomic status, as lower levels of education and socioeconomic status are associated with greater risk of IPV behaviors and attitudes (Abramsky et al., 2011; Peterman et al., 2015; Tran et al., 2016; Wang, 2016).

Macro-level Influences on IPV Attitudes

As individual behavior has many complex factors that influence it, several researchers have begun to evaluate the influence of macro-level factors on individual attitudes across contexts (Bucheli & Rossi, 2019; Lysova & Straus, 2019; Uthman et al., 2009). Macro social determinants are distal determinants that influence outcomes, and can include factors such as culture or macroeconomics (Galea, 2007). One area of focus has been on national ability to provide services, including education systems, social service provision, and access to health care (Lysova & Straus, 2019; Uthman et al., 2010a). Wealth of a country, evaluated with the gross domestic product (GDP), has been found to be negatively associated with IPV (Kovacs, 2018). Another area of interest is in gender equity norms which can reflect prioritize and legitimize of the privilege of men relative to women, which might lead to greater endorsement of the use of violence by a man against his wife. In one study examining country-level gender norms, researchers found evidence that the gender development index (GDI) which is based on the gendered difference of human capabilities based on life expectancy, education level, and a decent standard of living (i.e., Human Development Index), explained heterogeneity in endorsement of wife beating across contexts (Uthman et al., 2010b). Another measure that has been studied—female literacy rates—may reflect both gender norms and a nation’s ability to provide educational resources, as higher female literacy may indicate a national approach with greater emphasis and normativity around the education of all children compared to countries with lower female literacy. As such, researchers have found that female literacy rates are protective against IPV attitudes (Sardinha & Catalán, 2018). Thus, macro level gender norms are necessary to consider because they may both influence gender equity issues associated with IPV and a general index of human capital development.

Meso- and Exo-level Factors and IPV

In recent years, researchers have called for researchers in the field of IPV to better understand how neighborhood and community factors influence IPV (Capaldi et al., 2012; Dardis et al., 2015). Community and neighborhood are represented in the meso- and exo-levels of the socioecological model, with exo-level factors reflecting structures in communities, such as the existence of domestic violence shelters or rate of crime in a given community, while meso-level factors are how individuals interact with their communities, including factors like perceived community support and perceived community violence. Researchers who have examined neighborhood and community factors and IPV outcomes have found social disorder, witnessing community violence, and neighborhood poverty level to correlate with greater risk for victimization (Benson et al., 2003; Cunradi et al., 2000; Dekeseredy et al., 2003; Raghavan et al., 2006; 2009; Stueve & O’Donnell, 2008; Thulin et al., 2020; Thulin et al., 2021). Researchers have found an inverse association between positive perceptions of neighborhood and IPV, although promotive factors of neighborhood and IPV have primarily been studied in cohort studies and less is known about trends across contexts (Browning, 2002; Thulin et al., 2020). While research on neighborhood-level factors expands the field of IPV, few researchers have examined how community-level perceived norms influence individuallevel IPV attitudes, fewer used cross-contextual samples, and we note a paucity of work examining meso-level factors with macro-level factors (Voith, 2019).

Current Study

The current study will evaluate the association between physical IPV attitudes and micro- (individual, interpersonal), meso- (community) and macro- (national indicators) factors across 51 low-, middle-, and high-income countries. We coalesce data from multiple sources to test our hypothesis that factors corresponding to micro-, meso-, and macro-level influences will all predict physical IPV attitudes, and then evaluate relative contribution of a given risk factor in relation to other risk factors to better evaluate the effect size of distal and proximal risk factors and physical IPV attitudes.

Methods

Overview

The current study draws from several sources of data to construct a dataset with micro-, meso- and macro-level data. Data are drawn from the World Values Survey 6, the World Bank, and the United Nations. We first provide an overview of each data source, organized by socioecological levels that each data source corresponds to. We then provide detailed information on the measures selected from each data source.

Micro- and meso-level data.

We used the World Values Survey (WVS; http://www.worldvaluessurvey.org/wvs.jsp), a publicly available multi-national survey that examines values that effect social and political life for micro- and meso-level factors. Started in 1981, WVS has collected 6 waves of data across 100 countries, representing 90% of the global population. Data are purposively sampled to be representative of all people age 16 and above; see website for additional detail on sampling. Data are collected primarily through face-to-face interviews or phone interviews. The WVS has rigorous internal checks and cleans the data before it is released for public use. All micro- and meso-level data used in our analysis are from Wave 6 of the WVS which were collected from 2010 to 2014.

Macro-level data.

Data on country-level factors is also publicly available, and were drawn from the World Bank (WB: https://data.worldbank.org/) and the United Nations (UNICEF: data.unicef.org and UNDP: http://hdr.undp.org/en/data). For one variable of interest, female literacy, data for 11 countries (Australia, Germany, Japan, Kyrgyzstan, The Netherlands, Palestine, South Korea, Sweden, Taiwan, United States, and Yemen) were obtained from Index Mundi (https://www.indexmundi.com/).

Data from the WVS, WB, and UN were downloaded as.csv files, and then collated into a single dataset; all analyses were conducted using Stata 15.1 (StataCorp, 2017).

Measures

Outcome variable

Physical IPV attitudes.

Attitude about intimate partner violence was measured with a single item asking participants to indicate how justifiable it was “for a man to beat his wife.” On a 10-point Likert scale, respondents could indicate that it was “Never justifiable” (1) to “Always Justifiable.” This variable was rescaled to 0 for those who indicated it was never justifiable, or one if the respondent indicated any level of acceptability.

Micro-level factors

Belief in Corporal Punishment Against Children.

Respondents indicated how justifiable it was for parents to “beat” their children on a 10-point Likert scale from “never justifiable” (0) to “always justifiable (9).

Belief in Use of Violence Against Another Human.

Respondents indicated how justifiable it was for violence to be used against others on a 10-point Likert scale from “never justifiable” (0) to “always justifiable (9).

Patriarchal Beliefs.

This measures was a mean of five items rated on a 4-point Likert scale with high scores indicating stronger patriarchal beliefs (α=0.71). Sample items include “On the whole, men make better political leaders than women do” and “A university education is more important for a boy than a girl.”

Demographic Variables.

The demographic variables included in this analysis were individual’s education level ranging from “no formal education” (1) to university-level education with degree (8), age in years, sex (male/female), relationship status of either separated (1) or not (0), household income status ranging from dissatisfied (1) to satisfied (10), and household hunger measured as frequency of hunger over past 12 months from never (1) to often (4).

Meso-level factors

Perceived Neighborhood Disorder.

Perceived neighborhood disorder was evaluated with five items (α=0.81) asking about the frequency of crime (robberies), racism, over policing, and substance use (alcohol and drugs) within the neighborhood (α=0.81). Respondents indicated frequency on a 4-point Likert scale from “Very frequent” (3) to “Not at all frequent” (0) that each event happened in their neighborhood, and a mean across items was calculated.

Perceived Risk of Crime to Household.

Participants indicated perceived risk of crime to their household with a single item that asked them to report on how frequently they or their family “felt unsafe from crime in [their] home” over the past 12 months from Often (3) to Never (0).

Perceived Neighborhood Security.

Participants indicated how secure they felt in their neighborhood with a single item from “Very secure” (1; recoded to 3) to “Not at all secure” (4; recoded to 0).

Perceived Neighborhood Trust.

Participants reported on how much they trusted people they encountered in their lives, including friends, neighbors, and strangers. Participants responded to the five-items using a 4-point scale (α=0.75): “Trust completely” (3) to “Do not trust at all” (0). A mean across items was calculated.

Country-level covariates

Gini Index.

The Gini Index is a distribution of income across income percentiles. Data were drawn for each country from the World Bank, and an average of each country-level covariate was calculated between 2010 and 2014. In 3 cases, Gini Index between 2010 and 2014 was not an available; as such, the most recent Gini Index was used for Azerbaijan (2005), Nigeria (2009), and Uzbekistan (2003).

Gross Domestic Product (GDP).

The GDP per capita is a statistic representing a given country’s economic output relative to population size. Data were extracted by country for each year from 2010 to 2014 from the World Bank, and an average was calculated.

Female Literacy Rate.

The female literacy rate represents the percent of women in the country who are literate. Data were drawn from UNICEF, except for 11 countries where the data were not available through UNICEF and instead was drawn from Index Mundi. Most countries had literacy rates reported between 2010 and 2013, but for countries that did not have a formal literacy rate for that range, the rate was reported from 2005 to 2010. For data drawn from UNICEF, an average literacy rate was calculated between 2010 and 2013 or 2005 and 2010 in the case that no data between 2010 and 2013 were available. For data drawn from Index Mundi reports, the single literacy rate (accurate as of early 2019) was used.

Gender Development Index (GDI).

The Gender Development Index (GDI) was drawn from UNDP. GDI is based off the Human Development Index (HDI), which is a measure of human capabilities based on life expectancy, education level, and a decent standard of living. GDI is calculated based off the ratio of HDI for females to HDI for males. Data were extracted by country for each year from 2010 to 2014, and a mean was calculated.

Further details on measures can be provided upon request.

Sample

Of the 86,276 observations across 60 countries in the World Values Survey 6, nine countries (Bahrain, Taiwan, Hong Kong, Kuwait, Libya, Trinidad and Tobago, Singapore, Qatar, and New Zealand) do not have values for the Gini Index (n = 11,744) and thus the entire country was excluded. Of the 74,528 observations representing 51 countries remaining, 13.5% of observations had one or more missing variables, with 1201 individuals missing the variable for IPV attitudes, 6255 missing data on at least one individual-level independent variable and 2606 were missing at least one community-level independent variable. A total of 64,466 observations across 51 countries were included in analyses.

Analysis

First, we examined descriptive statistics of physical IPV attitudes by country and all independent variables. Levels of missingness of each variable were evaluated by country.

Second, we utilized mixed-effects logistic models to evaluate which factors were associated with physical IPV attitudes at the micro-, meso-, and macro-level of the socio-ecological model. For interpretability, results are presented in adjusted odds ratios; thus, co-efficient values less than one are interpreted as a negative association while co-efficient values greater than one are interpreted as a positive association. The base model examines physical IPV attitudes by country, to confirm the use of clustering. Due to prior research indicating sex differences in IPV attitude reporting (Capaldi et al., 2012; Uthman et al., 2010b), we tested fixed and random effects of gender. Upon confirmation that sex needed to be modeled as a fixed and random effect (i.e., the effect of sex varies across clustering units), both were included in all subsequent models. Model 3, 4, and 5 iteratively add individual-, country- and community-level variables.

In our final analysis, we standardized all independent variables except for the categorical micro-level variables of sex and relationships status. We then utilized the standardized variables and unstandardized categorical variables to examine the relationship between variables at each level and physical IPV using a mixed-effects logistic model.

Given the intention of the WVS to be representative of each country, all regression models included weights for slight deviations from the intended sampling frame within country (see http://www.worldvaluessurvey.org for details on weight construction).

Missingness

Those missing were more likely to be female (χ(1)=117.804, p<0.001) and on average 3.6 years older (F=300.85, p<0.001) than individuals from the countries that were included in our analysis. We also found differences in missingness by country, with some countries such as Yemen having a large percent of individuals missing some data (43.6%) while other countries such as Georgia had a very small percent (0.6%) of missingness. As missingness does not seem to be at random, we did not use multiple imputation (as this likely would have enhanced the bias), and instead re-ran analyses restricting to countries with less than one third (i.e., 35%) missing data to evaluate result sensitivity. The restricted analysis eliminated Morocco, Algeria, Australia, China, India, Egypt, and Yemen.

Results

Descriptive Statistics

Table 1 reports on the proportion of respondents reporting any level of physical IPV attitude endorsement, by country. Rwanda, India, and Iraq had the greatest endorsement of the acceptability of a man using violence against his wife. The countries of Australia, Georgia, and Brazil had the lowest endorsement of acceptability of physical IPV.

Table 1.

Weighted prevalence of endorsement of physical IPV attitudes by country.

	Prevalence of IPV Attitudes		Total Number Surveyed
Country/Region	n	%	N
Rwanda	1193	96%	1242
India	778	74%	1052
Iraq	683	66%	1042
Egypt	631	63%	1001
South Africa	2057	63%	3278
Algeria	507	59%	865
Nigeria	854	55%	1551
Zimbabwe	759	51%	1477
Philippines	576	50%	1159
Lebanon	491	46%	1075
Uzbekistan	612	46%	1341
China	736	45%	1626
Palestine	410	44%	932
Malaysia	507	41%	1223
Kazakhstan	482	37%	1316
Yemen	203	36%	558
Ghana	514	36%	1427
Kyrgyzstan	423	36%	1190
Morocco	152	33%	460
Tunisia	301	33%	922
Peru	361	32%	1115
Azerbaijan	308	31%	984
Russia	645	31%	2092
Thailand	316	30%	1040
Pakistan	265	30%	890
Ukraine	332	26%	1256
Ecuador	311	26%	1190
Belarus	325	26%	1269
South Korea	258	25%	1019
Jordan	284	25%	1139
Estonia	310	21%	1442
Germany	401	21%	1940
Slovenia	198	20%	999
Turkey	267	18%	1445
Mexico	351	18%	1920
Armenia	168	16%	1051
Japan	303	16%	1911
Cyprus	119	14%	861
Colombia	195	14%	1440
Argentina	126	13%	983
United States	244	11%	2125
Chile	103	11%	901
Poland	105	11%	919
Spain	126	11%	1125
The Netherlands	187	11%	1722
Sweden	119	11%	1117
Uruguay	96	10%	917
Romania	142	10%	1365
Brazil	142	10%	1444
Georgia	103	9%	1111
Australia	54	5%	997
Total	19,059	30%	64,466

Descriptive statistics on independent variables are presented in Table 2. The mean endorsement of corporal punishment was higher than participant’s average belief in use of violence against others. On average, participants held some patriarchal beliefs. Overall, neighborhood disorder and perceived risk of crime was lower than average report of trust in neighbors and perceived security in the neighborhood. Average female literacy was 89.9%, but ranged from 41.0% to 100%. GDP per capita ranged from USD625.56 to USD62,537.18, and on average was USD14,709. The gender development index (GDI) was on average 0.95, indicating that on average female HDI per country was slightly lower than the average male HDI for that country.

Table 2.

Descriptive statistics and source of independent and covariate variables.

	Source	Mean	Standard Deviation	Min	Max
Endorsement of physical IPV attitudes	WVS	0.31	0.46	0	1
Individual-level covariates
Belief in corporal punishment	WVS	2.75	2.51	1	10
Belief in violence against others	WVS	1.96	1.90	1	10
Patriarchal beliefs	WVS	1.43	0.64	0	3
Sex	WVS	0.48	0.50	0	1
Age	WVS	41.63	16.55	16	99
Education level	WVS	4.95	2.16	1	8
Relationship status	WVS	0.02	0.14	0	1
Household income	WVS	5.85	2.45	1	10
Household hunger	WVS	1.56	0.86	1	4
Community-level factors
Perceived neighborhood disorder	WVS	0.74	0.69	0	3
Perceived risk of crime to household	WVS	0.59	0.89	0	3
Perceived trust in others	WVS	1.42	0.59	0	3
Perceived neighborhood security	WVS	2.06	0.81	0	3
Country-level covariates
Female literacy	UNESCO	89.90	14.78	41.00	100.00
GDP per capita	WB	14,709.39	16,558.67	652.56	62,537.18
GDI	WB	0.95	0.07	0.63	1.03
Gini-Index	UNDP	38.00	9.03	24.54	63.20

Mixed-effects logistic Model Building: Unstandardized Factors

Model 0: The base model indicated that accounting for country clustering was appropriate. Almost a quarter of IPV attitude variation (intraclass correlation=0.250) was accounted for by clustering by country. Fit statistics of Model 0 were AIC=68,788.57, BIC=68,806.72.

Model 1/2: We compared modeling sex as a fixed- (Model 1) and random-effect (Model 2), and found Model 2 to be superior (LRT: χ=11,540.65, p<0.001). As such, sex is included as a fixed- and random-effect in all models. As compared with the Model 0, the fit of Model 2 was superior (AIC=68,070.81, BIC=68,107.11).

Model 3: Model 3 incorporated individual covariates. Greater belief in corporal punishment (aOR=1.34), violence against others (aOR=1.88), higher patriarchal beliefs (aOR=1.24), being male (aOR=1.57), less education (aOR=0.97), being separated from a partner (aOR=1.23), and greater household hunger (aOR=1.15) were all significantly associated with endorsement of IPV attitudes. Addition of individual factors lowered the country intercept from 1.10 to 0.68. Fit statistics of Model 3 (AIC=50,816.14, BIC=50,925.03) were substantially better than Model 2.

Model 4: Model 4 included country-level factors. Lower female literacy (aOR=0.97) is inversely associated with IPV attitudes. GDP (aOR=1.00), GDI (aOR=25.51) and the Gini-Index (aOR=0.98) were not significantly associated with IPV attitudes. The trends of micro-level factors did not substantially change. The random effect intercept co-efficient for country decreased to 0.64. Fit statistics of Model 4 (AIC=49,110.27, BIC=49,245.94) were better as compared with Model 3.

Model 5: Model 5 incorporated perceived meso-covariates. Perceived neighborhood disorder was positively associated with IPV attitudes (aOR=1.05) while perceived neighborhood security was negatively associated (aOR=0.96). Perceived trust in others (aOR=0.96, p=0.069) and perceived neighborhood crime (aOR=1.02, p=0.291) were not significant at p<0.05. The micro- and macro-level trends did not substantially change. The fit statistics for Model 5 were better for AIC (49,093.12) but slightly worse for BIC (49,264.97) as compared with Model 4.

Results from model 3, 4, and 5 are presented in Table 3.

Table 3.

Meso-level influences on physical intimate partner violence attitudes while accounting for micro- and macro-level factors across 51 countries, presented in adjusted odds-ratios and 95% confidence intervals.

	Model 0	Model 1	Model 2	Model 3	Model 4	Model 5
Fixed-effects
Micro-level covariates
Belief in corporal punishment				l.34**[1.33, 1.35]	l.33**[1.32, 1.35]	l.33**[1.32, 1.35]
Belief in violence against others				1.88**[1.85, 1.92]	1.88**[1.85, 1.91]	1.88**[1.84, 1.91]
Patriarchal beliefs				1.24**[1.19, 1.29]	1.24**[1.20, 1.30]	1.25**[1.20, 1.30]
Sex		0.45 [0.41, 0.48]	0.54 [0.45, 0.62]	1.57**[1.43, 1.72]	1.56**[1.4l, 1.71]	1.56**[1.42, 1.72]
Age				1.00 [0.997, 1.00]	1,00*[0.997, 0.999]	1.00 [0.997, 1.00]
Education level				0.97**[0.96, 0.99]	0.97**[0.96, 0.98]	0.97**[0.96, 0.98]
Relationship status				1.23*[1.04, 1.46]	1.25*[1.05, 1.48]	1.24*[1.04, 1.47]
Household income				0.99 [0.98, 1.00]	0.99 [0.98, 1.00]	1.00 [0.99, 1.01]
Household hunger				1.15**[1.12, 1.18]	1.15**[1.12, 1.18]	1.13**[1.10, 1.17]
Macro-level covariates
Female literacy					0.97*[0.94, 0.99]	0.97*[0.94, 0.99]
GDP per capita					1.00 [1.00, 1.00]	1.00 [1.00, 1.00]
GDI					25.51 [0.14, 4615.83]	25.78 [0.14, 4701.04]
Gini-index					0.98 [0.95, 1.01]	0.98 [0.95, 1.01]
Meso-level factors
Perceived neighborhood disorder						1.05*[1.01, 1.09]
Perceived neighborhood crime						1.02 [0.99, 1.05]
Perceived trust in others						0.96 [0.92, 1.00]
Perceived neighborhood security						0.96*[0.93, 0.99]
Random effects
Individual sex			0.12 [0.06, 0.22]	0.11 [0.06, 0.21]	0.09 [0.05, 0.17]	0.09 [0.05, 0.16]
Country	1.10 [0.74, 1.62]	1.11 [0.75, 1.64]	1.00 [0.69, 1.46]	0.68 [0.46, 1.00]	0.64 [0.43, 0.96]	0.64 [0.43, 0.97]
Model fit statistics
AIC	68,788.57	68,225.99	68,070.81	50,816.14	49,110.27	49,093.12
BIC	68,806.72	68,253.22	68,107.11	50,925.03	49,245.94	49,264.97

^*p < 0.05

^**p < 0.001.

Reanalysis excluding 35% missingness.

We re-ran the analyses excluding the seven countries with more than 35% missing data, and found similar trends. In the model without the seven countries, the micro-variable of age was significance but coefficient size and direction was the same as the full model (b=1.00, p=0.012). The meso-variable of perceived neighborhood trust became significant (0.93, p<0.001) but perceived neighborhood disorder was no longer significant at p<0.001 (aOR=1.04, p=0.084). As was the case for the micro variable changes, the coefficients of the meso-factors were the same direction and roughly the size as in the full model. The macro-variable trends were not statistically different, but the co-efficient size for the GDI was smaller (aOR=3.78, p=0.694). The fit statistics of the model excluding the seven countries with missingness of greater than 35% were superior as compared with the model that included them (i.e., Model 5) (AIC: 42,656.11, BIC: 42,825.79). Further details on the restricted analysis are available upon request.

Relative Effect: Standardized Factors

The final analysis involved standardizing all micro-, macro- and meso-factors to evaluate relative size of influence on physical IPV attitudes. The categorical variables of sex and relationship status were not standardized to retain interpretability. The most predictive factors were the micro-level factors of belief in the use of violence against others (aOR=3.36) and belief in the use of physical discipline against children (i.e., corporal punishment) (aOR=2.10). Other significant factors in the standardized model included holding patriarchal beliefs (aOR=1.15), reporting greater household hunger (aOR=1.11), and reporting lower levels of education (aOR=0.94). The macro-level factors of GDP per capita (aOR=0.67) was the most protective factor, followed by higher female literacy rates (aOR=0.70) and Gini-Index (aOR=0.71). GDI was not significant (aOR=1.24). The meso-level factors of perceived neighborhood crime (aOR=1.04) and perceived neighborhood security (aOR=0.97). The full model is reported in Table 4.

Table 4.

Physical intimate partner violence attitudes and standardized micro-, meso-and macro-factors across 51 countries.

Fixed-Effects
Individual-level covariates	Adjusted odds ratio [95% confidence interval]
Belief in corporal punishment	2.10**[2.06, 2.17]
Belief in violence against others	3.36**[3.23, 3.46]
Patriarchal beliefs	1.15**[1.12, 1.18]
Age	0.98 (0.01)
Education level	0.94**[0.92, 0.97]
Household income	0.99 [0.97, 1.01]
Household hunger	1.11**[1.08, 1.14]
Country-level covariates
Female literacy	0.70*[0.53, 0.97]
GDP per capita	0.67**[0.51,0.80]
GDI	1.24 [0.91, 1.65]
Gini-index	0.71*[0.56, 0.89]
Community-level factors
Perceived neighborhood disorder	1.04*[1.01, 1.06]
Perceived neighborhood crime	1.02 [0.99, 1.05]
Perceived trust in others	0.98 [0.96, 1.00]
Perceived neighborhood security	0.97*[0.95, 0.99]
Random effects
Individual sex	0.093**[0.05, 0.16]
Country	0.472**[0.33, 0.72]
Model fit statistics
AIC	50,777.89
BIC	50,959.38

^*p < 0.05

^**p < 0.001.

Model controls for the unstandardized categorical variables of sex and relationship status.

Discussion

In the present study, we were able to evaluate individual, perceived meso- and macro-variables across multiple countries. In the unstandardized models, we identified multiple individual and perceived meso-variables, and one significant macro-variable while accounting for differences due to country and gender. By using standardized forms of each micro-, meso- and macro-factor, we were able to compare relative effect within and across the socioecological model. Our finding of greatest effect at the individual variables shows the importance of individual influence and an opportunity for targeting proximal factors for change. Although neighborhood and national factors were smaller than individual factors, accounting for variation by country accounted for a quarter of the variance of IPV attitudes. Further, finding that meso- and macro-factors remained significant even when accounting for individual influences indicates that they are also critical, albeit more distal. It may be that neighborhood and national factors have a longer-term influence on individual beliefs, and that change of those factors would have a widespread influence on reducing physical IPV-attitudes across individuals, as compared with individual-level interventions that may only influence those who participate.

Contextualizing Findings within the Field of IPV Research

Though our findings supports prior investigators’ advocacy for the use of the socio-ecological model in IPV research, the lack of research addressing multiple levels stems from several overarching challenges in the field. First, few datasets exist with relevant data at each level of the socio-ecological model. For our study, we coalesced data from multiple sources to build a dataset to investigate relative effects at each level. Unfortunately, the data did not allow us to evaluate structures within neighborhood or communities (i.e., exosystem) as this data was not collected and is not readily available for many locations. Additionally, unlike publicly available macro-level data which can be merged in with few if any ethical considerations, even if community-level data were available, merging it in may require additional identifying details which could create opportunities for privacy violations. While the present study was able to evaluate the perceived neighborhood characteristics which correspond with the meso-level, expanding multi-national surveys that examine IPV such as the frequently used Demographic and Health Survey, or funding researchers to build novel datasets to include more levels of data while preserving individual respondent privacy may help fill the broad gap in the literature.

A second but potentially larger challenge is a lack of theoretical application and development within and across levels. Application of theory is one of biggest challenges facing the field of IPV research, noted by multiple researchers who have conducted systematic reviews on IPV-related research (Capaldi et al., 2012; Voith, 2019). This challenge also influences data being collected as well as the present study, and will attempted to be addressed below.

Theoretical Application and IPV Research

Within IPV literature, Feminist Theory is often used to conceptualize violence as an outcome of the power imbalance between men and women, whereby men have more power within relationships and systems and use this to enact violence against women (Dobash, 1979; Tierney, 1982; Walker, 1979). Feminist Theory originated in the late 1970s and early 1980s, focusing on social and structural power imbalances that prioritized men over women; though multiple iterations of Feminist Theory have expanded the theory to include other socially constructed identities and novel contemporary challenges (Snyder, 2008), binary male/female gender inequalities which prioritize men continue to exist. Influencing potentially all levels of the socioecological model, the access and legitimacy of inequitable power dynamics between men and women entitle men to use coercion and force to gain what they desire, even if their female partner disagrees. The findings from this study are consistent with the patriarchal view of influence on IPV attitudes (Hayes & Boyd, 2017; Hossain et al., 2020; Uthman et al., 2009). Yet, Feminist Theory alone does not explain all of the findings at the micro-level of this study, nor does it address the meso-level findings nor the macro-level findings. Rather, other theories need to be examined and explored.

Our findings also support the application of the theory of Cultural Spillover to IPV research. Cultural Spillover Theory postulates that legitimizing the use of violence in any circumstance or context will increase violence in other circumstances or contexts such as rape and IPV (Baron et al., 1988). Though predominately applied within the United States legal systems to explain the use of violence in intimate partner relationships between those formally employed to maintain order and safety (i.e., police officers, military, etc.) (Bradley, 2007; Klostermann et al., 2012; Rosen, 2007), Lansford et al. (2020) applied Cultural Spillover Theory to understand the association of endorsement of corporal punishment of children and IPV attitudes across 21 low- and middle-income countries (Lansford et al., 2020). In another study, researchers found that in a population of university students across 32 nations, greater attitudes on the permissibility of physical violence in certain social situations (e.g., to hit back if one is hit) were found to be associated with greater endorsement of IPV attitudes (Lysova & Straus, 2019). Our study is consistent with prior research and extends it by identifying that micro-attitudes of permissibility of physical violence against another are the largest belief-risk factor, followed by attitudes toward corporal punishment of children, followed by patriarchal beliefs across 51 contexts. Believing that violence is at times justifiable likely overlaps at least in part with the belief of using corporal punishment against a child, and both spillover to IPV. Additionally, Feminist Theory and Cultural Spillover Theory may co-exist, whereby the effects of Spillover might be greater in societies with greater gender inequalities. Although we were not able to test the intersection of Feminist Theory and Cultural Spillover Theory in this study, future studies that delineate justifiable corporal punishment of girl versus boy children and justifiable violence against male versus females (and what situations are justifiable) would help to unpack these theories and beliefs further.

Researchers have a growing interest in understanding community-level factors that may be predictive of IPV because it extends our knowledge beyond individual and interpersonal ecological levels, and provides information that may reveal new foci for prevention (Capaldi et al., 2012; Dardis et al., 2015). Although the co-efficient size is relatively smaller than other significant factors representing the micro- and macro-levels, our findings that less perceived community-risk and greater perceived neighborhood safety were associated with lower IPV attitudes across 51 countries are a novel contribution to the field. These findings are consistent with other researchers who studied the associations between qualities of communities and forms of interpersonal violence in cohort studies in a single country (Pickover et al., 2018; Thulin et al., 2020; Voith, 2019). Our finding that neighborhood disorder was associated with more positive attitudes supports Social Disorganization Theory which suggests that physical disorder leads to increased risk of violent behavior (Sampson et al., 1997).

While the findings of the present study provide novel information about community-level factors in cross-contextual analyses, future research must examine the influence of micro- and macro-factors that might increase or decrease disorder and security. For example, macro-policies that decrease access to social services including schooling for certain geographic areas may over time decrease community earning potential and wealth, and in turn may increase the risk of community blight and disorder. Similar relationships may exist between macro- and meso-factors; for example, greater gender equality may translate into women’s having greater access to formal roles within community development and leadership and enhance community trust and feelings of security. One way to explore these ideas further is to examine the cross-level relationship between macro- and meso-factors and the individual outcome of IPV beliefs over time. Additionally, adding promotive measures of individual, community, and country variables would also expand the field. Though the purpose of the present study was to incorporate meso-variables to better understand the role of neighborhoods and communities, we utilize available measures which are largely deficit focused. Future work evaluating promotive community factors that may be related to IPV, such as feeling respected by and able to trust police and judicial services, availability and provision by social services, and feeling physically, emotionally, and spiritually safe within one’s community may help expand ways to prevent and protect against IPV.

At the macro-level, female literacy, GDP, and Gini-index were protective factors against physical IPV attitude endorsement in the standardized model. The distal influence of country wealth may translate into ability to invest in social services. While our findings that GDP is an important factor for physical IPV attitudes (Uthman et al., 2009), GDP assumes equity in income distribution. As distribution is often inequitable, with various communities or identities experiencing marginalization within systems, it is important to account for inequality (such as the Gini co-efficient and female literacy rates) in these analyses. This is because the use of violence against another person is a form of exerting power and influence, which implies perceived inequalities between humans. Just as Link and Phelan described fundamental causes of disease (Link & Phelan, 1996), exploring the mechanisms through which macro social determinants of IPV, including the factors that influence how subsets of humans are devalued based on identity (i.e., gender inequality), may be important to further our understanding of where to intervene. Future research examining which indices are most useful for predicting IPV attitudes, including the gender social norms index recently launched by the United Nations Development Programme (Conceição, 2019; Conceição et al., 2020), may be useful.

Limitations

Our results support that utilizing the socio-ecological model in multinational IPV attitudes research is possible, but challenging due to overarching challenges present in the field of IPV research. First, few single data sources contain data pertaining to multiple levels of the socio-ecological level. While macro-level data can at times be added, such as was done in this study, connecting micro-level data with meso- and exo-level data is more difficult. The second challenge is the lack of conceptual development to inform how the micro-, meso-, and macro-levels inform and interact with each other; this particularly problematic for the incorporation of the meso-level. However, as is shown in this study, there are important risk factors for physical IPV attitudes at each level of the socio-ecological model.

Although we used a uniquely constructed dataset, representing high-, middle- and low-income countries, and provided novel insight to factors predictive of physical IPV attitudes, we note several limitations present in our study. First, physical IPV attitudes are measured by a single item measure. While this does not capture different aspects of this attitude, the item is face valid and does provide initial evidence that more research is needed to characterize these relationships in more detail. A more detailed measure may provide more nuanced findings, but the fact that we found effects with a single item that may have limited psychometric properties actually suggests our findings are quite robust. An expansion of the measurement of attitudes could also allow researchers to evaluate attitudes involving forms of IPV beyond physical to include sexual, emotional, and electronic forms. Second, we assessed neighborhood factors based on individual perceptions versus more objective measures that reflect the exosystem (services and structures in the neighborhood, community, and/or workplace) of the socio-ecological model. Individual lived experience represented in the meso-level, however, is important to understand because perceptions often drive behavior. Nevertheless, more objective assessments of neighborhood factors would provide important information and allow for the investigation of meso- and exo-levels across contexts. Third, this study does not provide information on potential promotive factors across the socioecological model. Though we utilized available measures, an association between a risk factor and a violence attitude outcome does not provide information on factors that may promote attitudes that exemplify universal human respect and thus reduce attitudes on the permissibility of violence against others, including IPV attitudes. Fourth, we are not able to speak to cultural diversity and variations present in countries and regions in this paper, but rather focused on identifying which micro-, meso-, and macro-factors are consistent predictors across heterogenous contexts. Finally, our study is cross-sectional in nature so our results are only correlational in nature. It would be useful to examine how change in factors at different ecological levels may influence IPV beliefs over time. This kind of longitudinal analysis may also help inform prevention strategies that help break the cycle of risk to IPV victimization.

Implications

We note several implications of the current study. First, our findings support the importance of community on physical IPV attitudes. This suggests that change in community safety may be important for reducing the acceptability of physical IPV attitudes. Second, despite substantial challenges, we find that it is both feasible and important to account for ecological levels of influences beyond the individual to understand IPV attitudes in multi-national research. Third, the present study highlights the challenge of incorporating meso level variables (i.e., neighborhood, community) as we were not able to model geographic community within country in a multi-level model. Rather, we utilized individual perceptions of community on individual perceptions of the justification of physical IPV attitudes. While perceptions are important to capture, the findings of the present study demonstrate the need to collect data on observed exposure to risk or access to promotive factors in the community setting in future administrative datasets.

Conclusion

These limitations notwithstanding, our results contribute to understanding IPV attitudes by examining predictors across socioecological levels and across nations. Our study suggests that norms at micro-, meso-, and macro-level influence individual beliefs the acceptability of IPV against women. Thus, modeling of multiple factors across the socio-ecological model is important because it expands the existing empirical work on IPV attitudes in several ways. First, we employed a socio-ecological framework to physical IPV attitudes to better understand the complex web of factors associated with IPV attidues. To do so, we combined multiple sources of data to create a unique dataset that had factors corresponding to micro-, meso-, and macro-levels of the framework. Next, by standardizing, we were able to compare relative contributions of factors within and across levels of the socioecological model. Finally, we explored theoretical underpinnings that may be driving our findings, and suggested future steps for testing theory and empirical findings further. This study provides a useful model for others interested in multi-national that examines variables across the socio-ecological model.

Biographies

Author Biographies

Elyse Joan Thulin, MS, is a doctoral student at the University of Michigan School of Public Health, a scholar at the University of Michigan Prevention Research Center, and a Rackham Merit Fellow. Her research interests include intimate partner violence and gender-based violence in US and global populations.

Justin E. Heinze, PhD, is currently an assistant professor in the Department of Health Behavior and Health Education in the School of Public Health at the University of Michigan. Dr. Heinze’s research interests include developmental transitions, social exclusion/ostracism, school safety and longitudinal data methodology.

Marc A. Zimmerman, PhD, is the Marshall H. Becker Collegiate Professor (and former Chair) in the Department of Health Behavior and Health Education in the School of Public Health, and a Professor of Psychology and the Combined Program in Education and Psychology all at the University of Michigan. He received his Ph.D. in Psychology from University of Illinois. Dr. Zimmerman is the Director of the CDC-funded Michigan Youth Violence Prevention (yvpc.sph.umich.edu) and Prevention Research Centers (prc.sph.umich.edu). He led the development of Youth Empowerment Solutions program (yes.sph.umich.edu) and public health applications of place-based change for community improvement. He is Co-Principal Investigator (PI) of the NICHD-funded initiative on Firearm Safety among Children and Adolescents (FACTS). He is also the Co-Director of the Bureau of Justice Assistance funded National Center for School Safety (nc2s.org). Dr. Zimmerman is the Editor of Youth and Society and editor emeritus of Health Education and Behavior. He has published over 300 articles and book chapters, and co-edited two books on adolescent development including topics on violence, mental health, substance abuse, evaluation methods, and empowerment. His research focuses on adolescent health and resiliency and the application of empowerment theory.