Spatial Distribution and Determinant Factors of Intimate Partner Violence Among Reproductive age Group Women in Ethiopia; Generalized Structural Equation Modeling

Background: Intimate Partner Violence (IPV) is the most serious and pervasive yet under recognized human rights violation in the world as well as in Ethiopia. The objective of this study was to nd the spatial distribution of IPV and its determinant factors in Ethiopia. Methods: Secondary data analysis was conducted among 2,687 reproductive age group women (15–49 years). The distribution of IPV across the country was observed by ArcGIS software. In SaTScan software, the Bernoulli model was tted by Kulldorff methods to identify the purely spatial clusters of IPV. Generalized Structural Equation Model (GSEM) was used to determine factors associated with each domain of IPV (physical, emotional & sexual violence). Result: The spatial distribution of IPV was found to be clustered in Ethiopia with Global Moran’s I 0.09 (p < 0.001) and the highest IPV cluster was observed in Oromia (p < 0.001), Somali (p < 0.001) and Southern Nation and Nationality and Peoples (SNNP) (p < 0.001) regions . Watching television and not having attitudes toward wife beating were negatively associated with physical violence. Being richest and nonsmoker were inversely associated with emotional violence. The odds of experiencing sexual violence were increased among pregnant and wife of uneducated husband/partner. Women’s decision-making autonomy and husband/partner drinking alcohol have positive and negative association with all domains of IPV respectively. Conclusions: There was a signicant clustering of IPV in Ethiopia and the highest IPV cluster was observed in Oromia, Somali and SNNP regions. Being richest, watching television, not having attitudes toward wife beating, women’s decision autonomy and husband’s/partner’s high education and non-alcohol drinker status were negatively associated with IPV. The likelihood of experiencing IPV was also increased among smoker and pregnant women. We recommend that improving the economic status of the household and power of women's decision autonomy, increasing community awareness about the consequences of IPV with particular emphasis in Oromia, Somali and SNNP regions.

Intimate partner violence isn't a little issue that as it were happening in a few pockets of society, but rather is a wide-ranging public health problem of epidemic proportions, requiring urgent action. Globally, nearly one third (30%) of all women who have been in a relationship have experienced physical and/or sexual violence by their intimate partner. This prevalence was highest in the WHO Africa, Eastern Mediterranean and South-East Asia regions (37%) and lowest in the high-income region (23%) (12) In Ethiopia, the prevalence of lifetime IPV ranges from 20-78% (13). Even though the country rati ed many of international and continental agreement that promote and protect against women right, around 63% of women and 28% men agree that a husband is justi ed in beating his wife (14). This makes IPV not only a deep-rooted problem but also somehow acknowledged rather than challenged. In addition, there is an imbalance between men and women in institutionalized gender roles and structural power (15). Different studies were conducted at different part of Ethiopia using a univariate analysis to determine factors associated with IPV (16)(17)(18)(19)(20)(21). In this analysis only one dependent variable is allowed. However, IPV have three domains (physical, emotional and sexual violence) which need to be considered as independent variables. Furthermore the prevalence of IPV is not geographically homogenous (13,14). Therefore, the present study applies spatial analysis to identify of geographic distribution IPV and generalized structural equation model (GSEM) to determine factor associated with three domains of IPV.

Study design and setting
Secondary data analysis was conducted using the Ethiopian Demography and Health Surveys (EDHS) 2016. Ethiopia is composed of 9 National Regional states: namely Tigray, Afar, Amhara, Oromia, Somali, Benishangul-Gumuz, SNNP, Gambella and Harari, and two Administrative states (Addis Ababa City administration and Dire Dawa city council). It has 68 zones, 817 districts, and 16,253 kebeles (smallest administrative units of a country) (14). The current population of country is 114,708,673 as of Tuesday, June 2, 2020, based on Worldometer elaboration of the latest United Nations data (22).

Data source and study period
The data source for this study is secondary data, which was retrieved from the DHS program o cial database www.measuredhs.com, after permission was granted through online request by explaining the objective of our study. The 2016 Ethiopia Demographic and Health Survey (EDHS) is the fourth Demographic and Health Survey conducted in Ethiopia. The study period for the EDHS 2016 was from January 18, 2016, to June 27, 2016(14).
Sampling procedure, study population and sample size The 2016 EDHS sample was strati ed and selected in two stages. In the rst stage, a total of 645 Enumeration Areas (EAs) (202 in urban areas and 443 in rural areas) were selected with probability proportional to EA size and with independent selection in each sampling stratum.
In the second stage of selection, a xed number of 28 households per cluster were selected with an equal probability systematic selection from the newly created household listing (14). In total, 15,683 women aged 15-49 who reported ever being married participated in the survey. For the domestic violence module, only one married woman per household was selected and 2,687 woman were selected and interviewed .The current study included women who reported ever being married and completed the IPV questionnaire (weighted sample = 2,734). Latitude and longitude coordinates were also taken from selected enumeration areas (clusters). The detailed sampling procedure was presented in the full EDHS report (14).

Measurements of outcome variable and operational de nition
The outcome variables with important predictors were extracted from Ethiopia Demographic and Health Surveys individual data set. In the 2016 EDHS, IPV was assessed using women's self-reported responses to questions depending on the modi ed Con ict Tactic Scales of Straus (23). Speci cally, violence committed by the current husband/partner for currently married women and by the most recent husband/partner for formerly married women was measured by asking all ever-married women if their husband/partner ever did the following:  (14).

Measurements of explanatory variable and operational de nition
Depend on different literature review (24-37); variables related to women, husband/ partner and family were included in this analysis Fig. 1.
1. Attitudes toward wife beating: It was measured based on the following ve questions that women were asked about whether situations of hitting or beating a wife is justi able: if she goes out without telling him; neglects their children; argues with him; refuses to have sex with him; and burns the food.
If they said 'yes' to any one of the above questions, they were classi ed as having attitude towards wife beating.
2. Women's decision-making autonomy: Categorized as 'yes' if she was involved in all decisions regarding her own health care, major household purchases and visits to her family or relatives Data processing The analysis was done using STATA 14, ArcGIS 10.3 and SaTScan 9.6 software's. Before any statistical analysis, the data was weighted using sampling weight (weight for domestic violence), primary sampling unit, and strata to restore the representativeness of the survey and to tell the STATA to consider the sampling design when calculating standard errors to get reliable statistical estimates. Descriptive statistics and summary statistics were showed using text and tables.

Spatial autocorrelation analysis
To check whether IPV were spread or cluster or randomly distributed, spatial autocorrelation (Global Moran's I) statistic measure was used. Moran's I values close to-1 indicated disease/event dispersed, whereas Moran's I values close to + 1 indicated disease/event clustered, and disease/event distributed randomly if Moran's I value was zero. A statistically signi cant Moran's I (p < 0.05) led to the rejection of the null hypothesis (IPV is randomly distributed) and indicated the presence of spatial autocorrelation. This analysis was done using ArcGIS software.

Spatial scan statistical analysis
Statistically signi cant spatial clusters of IPV among reproductive age group women were identi ed by spatial scan statistical analysis using Kuldorff's SaTScan software. The maximum cluster size was set at 50% of the population at risk. Bernoulli model was tted by considering women who did not experience life time IPV taken as controls and those women who experience life time IPV were taken as cases represented by a 0/1 variable. The number of cases in each location had Bernoulli distribution and the model required data with or without life time IPV. A Likelihood ratio test statistic was used to determine whether the number of observed life time IPV cases within the potential cluster was signi cantly higher than the expected or not. Primary and secondary clusters were identi ed using p-values and likelihood ratio tests based on the 999 Monte Carlo replications.

Model building for Generalized Structural Equation model
Generalized Structural Equation Model (GSEM) was used to determine factors associated with each domain of IPV (physical, emotional and social violence). Each domain of IPV were binary variable that was analyzed with Bernoulli family and a logit link function.
The analysis was started with a hypothesized model in Fig. 1. Modi cations were taken iteratively by adding a path link. At the end, an over identi ed model with minimum information criteria was retained. A nal model was selected based on statistical signi cance of path coe cient, the theoretical meaningfulness of the relationship and minimum information criteria. Statistically signi cant effects were considered for P < 0.05 at Con dence interval of 95%.

Result
Characteristics of study population A data of 2,687 reproductive age group women were included in the nal analysis. Among these study participants, more than one third of them (39.62%) were from Oromia region, more than two third (68.87%) of them were not currently working and 1,107 (41.20%) are Muslim. More than third quarters (78.5%) of the respondents were not watching television at all. Around 11.1% of the respondents were pregnant. More than two third of the women had decision making autonomy (69.00%) and have attitudes towards wife beating (68.63). Regarding husband /partner of respondents, 44.80% of them did not have any formal education and 30.23% of them drunk alcohol. The spatial distribution of IPV was found to be clustered in Ethiopia with Global Moran's I 0.09 (p < 0.001). Given the z-score of 5.45, there is less than 1% likelihood that this clustered pattern could be the result of random chance. The bright red and blue color to the end tails shows an increased signi cance level (Fig. 2). Spatial clustering of IPV was found at regional levels. The highest IPV was spatially clustered in Oromia, SNNPR,Somalia and Amhara (Fig. 3).

Spatial SaTScan analysis of intimate partner violence (Bernoulli based model)
Spatial scan statistics detected a total of high and modest preforming spatial cluster of IPV. Among these, 10 clusters were high performing cluster (LLR = 40.67, RR = 2.46, P-value < 0.001) and 4 clusters were lowest performing cluster (LLR = 13.19, RR = 1.98, P-value < 0.001).The highest performing clusters of IPV were identi ed in Oromia, Somali and SNNPR Table 2. The bright red colors (rings) indicate that the most statistically signi cant spatial windows of IPV. There was high IPV within the cluster than outside the cluster (Fig. 4). The likelihood of experiencing physical violence were decreased by 36% and 42% among women who were watching television less than once a week and at least once a week as compared with those were not watching television at all respectively. Smoker women were 2.59 times more likely to be emotionally violated than their counterpart. Pregnant women were 1.62 times more probable to be sexually violated than non-pregnant women.
Women who have attitudes toward wife beating were 1.32 times more likely to be physically violated than their counterpart. The chances of physical, emotional and sexual violence were decreased by 28%, 30% and 35% among women who had decision making autonomy than their counterparts respectively.

Discussion
The purpose of this study was to nd the spatial distribution of IPV and its determinant factors in Ethiopia. According the present study, there was a signi cant clustering of IPV in the study area and the highest IPV cluster was observed in Oromia, Somali and SNNP regions. This clustering of IPV in this area might be due to culture of a given society that did not recognize IPV as violation of human right rather accept and expect that wife beating is part of a normal union. This also might be low awareness and wrong attitude of husbands/partners toward negative consequent of women violence.
The present study documented that women age as a determinate factor of IPV. Relatively younger women were less likely to be physically abused as compared to older women. This nding is lined with household surveys in eight southern African countries (25) and Rwanda (26). This similar nding might be related to duration into the partnership of women's rst IPV victimization. Most of time relationships of longer durations are more likely to involve a history of male-perpetrated violence (38).
This study revealed that women who were watching television were at low risk of physical violence as compared with their counterpart. This result might be due to positive effect of mass media like television on women's attitudes towards violence against women (39) and those women who such attitudes are less likely to be abused.
Cigarettes Smoking status is signi cantly associated with emotional violence. According to the present study, smoker women were more likely to be emotionally violated than their counterpart. Since the current study use secondary date that was conducted through community based cross-sectional study, it is di cult to sort out the causal ordering between cigarettes Smoking status and emotional violence. Some studies also documented that experiencing intimate partner violence (IPV) increases women's risk for cigarette smoking (40)(41)(42)(43).
The results of this study suggested that pregnant women were more probable to be sexually violated than non-pregnant women. Such violence during pregnancy does not only affect the women's reproductive health, but also leads to adverse outcomes of maternal and child health (44)(45)(46).
In line with a study conducted in Uganda (47), the present study reported that women who have attitudes toward wife beating were more likely to be physically violated than their counterpart. The possible explanation for these results, women with attitudes supportive of IPV is more likely to be victims of IPV.
Because of such attitudes, women may accept and expect that wife beating is part of a normal union.
Our nding demonstrated that women's decision-making autonomy were signi cantly associated with physical, emotional and sexual violence. The chances of physical, emotional and sexual violence were decreased among women who had decision making autonomy than their counterparts. This nding is in line with other studies that was conducted in Uganda (37), in Ghana (33) and in Peru (32). This consistency could be supported by a male-dominated marital power structure has been documented to be highly related with marital con ict and husband-to-wife violence (48). Women's decision-making autonomy is not only reduced the risk IPV but also increased the utilization of maternal service (31,49).
In line with previous studies (29,30), the current study reported that the greater wealth were protective against emotional violence. The possible explanation for these consistent results might be low household economic status is the reason for con ict between couples.
The current study also identi ed husband /partner related factors that associated with IPV such as husband /partner's age, educational level and alcohol drinking status. The present study reported that the likelihoods of emotional violence were decreased among young husband/partner's as compared with old husband/partners. A similar result has been found in Haiti (26). In concurrent with other studies (27,28), the present study reported that the likelihoods of sexual violence were decreased among higher educated husband/partner than uneducated.
This study revealed that Husband/partner drinks alcohol as a predictor of IPV. According to current study drinker husband /partner were more likely to physically and sexually violate his wife/ partner than their counterparts. This result is in line with other studies that was conducted in Angola (36), in Uganda (34) and India (35). The possible explanation for these results alcohol uses directly disturbs mental and physical function, decreasing self-control and leaving individuals less capable of negotiating a nonviolent resolution to con icts in relationships. Excessive drinking by one partner can worsen nancial problems, child care di culties, in delity or other family stressors. This can lead to marital tension and con ict, increasing the risk of violence occurring between partners.
The primary strength of the current study was using large population-based data with a large sample size, which is representative at national and regional levels, so it can be generalized to all women in reproductive age group in Ethiopia. The joint use ArcGIS and Sat Scan statistical tests facilitated to identify similar and statistically signi cant area with high IPV (hot spot area). Furthermore, it used multivariate analysis (GSEM) to determine factors associated with three domains of IVP simultaneously.
However, the nding of this study interpreted with some limitation. First, the location of data values was displaced up to 2 kilometers for urban and up to 5 kilometers for rural areas to ensure respondent con dentiality, thus, this was the challenge to know the exact cases' location. Since EDHS was conducted using cross-sectional study design, it is di cult to sort out the causal ordering. Recall bias may be the other impediment for this study as EDHS was a questionnaire-based survey and relied on the memory of the respondents.

Conclusion
There was a signi cant clustering of IPV in the study area and the highest IPV cluster was observed in Oromia, Somali and SNNP regions. Being richest, watching television, not having attitudes toward wife beating, women's decision autonomy and husband's/partner's high education and non-alcohol drinker status were negatively associated with IPV. The likelihood of IPV was also increased among smoker and pregnant women. We recommend that improving the economic status of the household and power of women's decision autonomy, increasing community awareness about the consequences of IPV with particular emphasis in Oromia, Somali and SNNP regions.   Primary and secondary clusters of health insurance coverage among women across regions in Ethiopia,