Spatial Distribution and Predictors of Domestic Violence Among Women Aged 15-49 in Ethiopia: Analysis of EDHS 2016 Dataset.

Introduction: Violence against women particularly that is commited by an intimate partner is becoming a social and public health problem across the world. Studies from different countries shows that the spatial variation in distribution of domestic violence was commonly attributed by neighborhood level predictors. Despite the importance of spatial techniques, studies that employ it in Ethiopia are limited. Therefore, the aim of this study is to determine the spatial distribution and predictors of domestic violence among women aged 15-49 in Ethiopia by using EDHS 2016 dataset. Methods: Secondary data from EDHS 2016 was used to determine the spatial distribution of domestic violence in Ethiopia. Spatial auto-correlation statistics (both Global and Local Moran’s I) was used to assess the spatial distribution of domestic violence cases in Ethiopia. Spatial locations of signicant clusters were identied by using Kuldorff’s Sat Scan version 9.4 software. Finally, binary logistic regression and generalized linear mixed model were tted to identify predictors of domestic violence. Result: The study found that spatial clustering of domestic violence cases in Ethiopia with Moran’s I value of 0.26, Z score of 8.26, and P-value < 0.01. The Sat Scan analysis nd out 24 signicant locations of domestic violence clusters. Among this, 10 are primary clusters with RR 2.18, LLR of 39.55, and P-value < 0.01. The output from regression analysis identies low economic status, husband/partner alcohol use, witnessing family violence as a child, marital controlling behaviors, and community acceptance of wife-beating as signicant predictors of domestic violence. Conclusion and Recommendation: There is spatial clustering of d domestic violence cases in Ethiopia. Areas with a high burden of the problem should get priority for intervention. Comprehensive and collaborative action should be taken by involving stakeholders at different levels. Specic activities may include Organizing media on awareness creation and continuous education on how to maintain a stable relationship between couples and employing long term and intensive effort for transforming culture and social norms that encourage violence against woman are among the major ones.


Introduction
The term domestic violence mainly refers to violence committed by an intimate partner but it can also encompass abuse by any member of a household. In recent years, violence against women particularly that is commited by an intimate partner is becoming a social and public health problem across the world.
According to the 2017 WHO report, 1 in 3 (35%) of women worldwide have experienced either physical or sexual intimate partner violence or non-partner sexual violence in their lifetime(1). Even though the burden of the problem varies across countries, existing studies show that the prevalence of the problem was higher in developing countries. A result from WHO global report on domestic violence shows that the prevalence of physical and/or sexual intimate partner violence is high in African and South-East Asian countries; where approximately 37% of ever-partnered women experienced physical and/or sexual intimate partner violence at some point in their lives (2). Similarly, individual studies from different countries support this nding and demonstrate that the burden of the problem was still high in many Asian and African countries. For example, a study conducted in Iran shows that 62% of women experienced domestic violence by their husbands (3). This was explained by cultural and social norms that made men have a high position in their family was responsible for it.
Likewise, a high prevalence of domestic violence was observed in studies conducted in various African countries. A study conducted in Tanzania shows that 65% of ever-married women experienced lifetime intimate partner violence and 7% of women had ever physically abused their husband (4). Although the study tries to show the prevalence of domestic violence against women, it could not identify the predictors of domestic violence. Another study conducted on Demographic Health Survey in Ghana shows a relatively lower prevalence of domestic violence where 33.6% of women experience it at some point in their life (5). This result is comparable with a study conducted on the Zambian Demographic Health Survey where 43% of women experienced domestic violence (6).
In the Ethiopian context, violence against women and girls continues to be a major challenge and threat to women's empowerment. Even though the government of Ethiopia has been doing so many jobs to reduce the prevalence of violence against women, still substantial levels of women and girls are facing physical, emotional, and sexual violence. According to EDHS 2016 report, 34% of ever-married women age 15-49 have experienced either physical, sexual, or emotional spousal violence (7). A systematic review of 15 articles on domestic violence from 2000-2014 also shows the high prevalence of domestic violence in different parts of the country. The lifetime prevalence of domestic violence against women by husband or intimate partner ranged from 20 to 78% (8).
Evidence from studies shows that domestic violence causes serious short and long term health problems.
Even though su cient studies that show the consequence of domestic violence has not been conducted, the existing few studies show that physical injuries such as fractures of bones, sight and hearing damage, head injury, back and neck injury (9) and mental problems such as depression, anxiety, and suicidal attempts were signi cantly higher among women who have ever experienced intimate partner violence(1, 10,11). Furthermore, intimate partner violence can lead to unintended pregnancies, induced abortions, gynecological problems, and sexually transmitted infections (12).
The social-ecological model is the most widely used model to determine factors associated with domestic violence. The model proposes that violence is the result of factors inter-relation at the individual, relationship, community, and societal levels (9). Previous studies use this model to identify factors associated with experiencing domestic violence. A study conducted in Brazil found individuallevel variables such as excessive alcohol use by husband, marital controlling behavior, and having multiple sexual partnerships, and community level predictors such as economic deprivation were signi cantly associated with domestic violence (13). Another study conducted in Nigeria also nds that permissive state-level social norms toward spousal violence and acceptance of spousal violence by women are signi cantly associated with experiencing domestic violence (14). Similarly, a multilevel study conducted in the same area also nds that community norms that justi ed IPV against women increase the Odds of experiencing domestic violence (15). In addition to this, previous studies identify that individual-level factors such as low level of education, witnessing family violence, harmful use of alcohol, having multiple partners, woman's attitude towards wife-beating, low level of wealth index and low level of women employment as factors for domestic violence (5,8,9,16,17). Spatial statistical analysis is becoming a well-known tool to determine the distribution and predictors of health problems across the world. Studies are employing this technique to demonstrate the distribution of domestic violence cases in different countries. Most of the studies were from developed countries and very few were from less developed countries. For example, a study conducted in the USA shows that signi cant clustering of domestic violence was observed in areas where black American women reside.
And this study also nds out that pregnant women living in this area were more likely to give birth to small gestational age infants as compared to women living in low prevalent areas (18). Another study in Canada identi es that clusters of gender-based violence were observed in the downtown and around entertainment areas (19). The clustering of cases in these areas was explained by high alcohol consumption and substance abuse around entertainment areas create favorable condition for the occurrence of violence cases. On the other hand, a study conducted in Spain indicates that Intimate partner violence risk was higher in neighborhoods with low educational and economic status, high levels of public disorder, and high concentrations of immigrants (20). Similarly, in a study conducted in Brazil, Spatial disparity in the distribution of intimate partner violence was observed. And socio-economical difference across the regions was responsible for this variation (21).
The spatial statistical technique gives importance for examining the geographical distribution of health problems speci cally, by identifying areas with a high burden of health problems. Despite this, studies that employ this technique for demonstrating the distribution of domestic violence are limited. The existing few studies are also con ned to developed countries. As far as my literature searching is concerned, no article was found that displays the distribution of domestic violence in Ethiopia using geospatial techniques. Therefore, the current study employs spatial analytic tools to demonstrate the spatial distribution and associated factors of domestic violence among women age 15-49 in Ethiopia by using the EDHS 2016 dataset. The results from this study could help to support the decision making of different stakeholders by identifying locations with a high burden of the problem.

Study design and setting
The study uses secondary data from EDHS 2016 dataset. In EDHS 2016, a community-based crosssectional study was conducted by the Central Statistical Agency (CSA) from January 18 to June 27, 2016, in Ethiopia. Ethiopia is the second populous country in Africa and located in the Horn of Africa from 3 0 to 14 0 N and 33 0 to 48 0 E. Administratively, Ethiopia is divided into nine geographical regions and two city administrations.
2.2 Sample size and sampling technique EDHS 2016 used two stages strati ed cluster sampling technique where each region was strati ed into urban and rural areas, yielding 21 sampling strata. In the rst stage, a total of 645 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 (7). In the second stage, a xed number of 28 households per cluster were selected and only one woman per household was randomly selected for interview. Finally, a total of 5860 women aged 15-49 were asked questions about domestic violence against women. All women aged 15-49 and who were either permanent residents of the selected households or visitors who stayed in the household the night before the survey were eligible to be interviewed(7).

Variables of the study
The dependent variable of the study is the experience of domestic violence by women aged 15-49. The independent variables are categorized into individual-level factors, household/relationship factors, and community-level factors. Individual-level factors include woman's age, educational status of woman and husband/partner, occupation, religion, perpetrator's alcohol use, experiencing family violence during childhood, and woman's attitude towards wife-beating. Household and Relationship level factors consist of wealth index, a number of live children, head of household, decision making power, marital controlling behaviors, and the educational difference between couples. Community-level factors include acceptance of wife-beating in community, level of female literacy in the community, early marriage in the community, place of residence, and region.

Data processing and analysis
Statistical analysis of the data was done on SPSS version 25. Cross-tabulation and summary statistics were performed to describe the populations according to their age, educational status, place of residence, and region. Binary logistic regression and a two-level generalized linear mixed model was employed to identify predictors of domestic violence. Finally, model comparison between the models was performed based upon the Log-likelihood ratio test to choose the best-tted model.

Spatial analysis of domestic violence
ArcGIS 10.7 software was used for spatial analysis of the data. Spatial autocorrelation (Global Moran's I) statistics and Anselin local cluster analysis was done to display the spatial distribution of domestic violence among woman aged 15- Anselin local Moran's I was used to identify local level clusters of domestic violence. A positive Local Moran's I indicate that the feature is surrounded by features with similar values and, such types of cases are called clusters. Whereas, a negative value for I indicates that the feature is surrounded by features with dissimilar values, and this was called an outlier(23).
Kuldorff's Sat Scan version 9.4 software was used to identify the geographical locations of statistically signi cant clusters of domestic violence. Scan statistics use a scanning window that moves across the study area. Bernoulli model was tted to identify statistically signi cant locations of domestic violence clusters. The Bernoulli model was selected because the structure of the data shows the binomial [0/1] distribution. Women who have experienced domestic violence were considered as case and labeled1 whereas, those who do not experience as control and labeled 0. The default 50% of the population was used as an upper limit for cluster size; because it allows the detection of both small and large clusters of domestic violence. Statistically signi cant clusters were identi ed by P-value and likelihood ratio tests.

Multi-level logistic regression analysis
A two-level generalized linear mixed model was tted by considering 4322 women aged 15-49 at level 1 nested within 645 clusters (communities) at level two. A multilevel analysis of the data takes three steps. The rst step was tting the null (intercept only) model without including predictor variables and the second step was a random intercept xed coe cient model by including individual and relationship level variables. The last was tting a random intercept and xed coe cient model by incorporating community-level predictors.

Model Comparison
Model comparison between the two nested (null model and random intercept xed coe cient model) and the logistic regression model was done in order to select the best-tted model. The commonly used parameter for evaluation of model tness is the Log-likelihood ratio test that compares the deviance (-log likelihood) of the models by subtracting the smaller deviance from the larger one. Deviance is an indicator that shows how well the model ts the data. A model with the lowest deviance is considered as the best-tted model than with large deviance. In addition to the log-likelihood ratio test, Akaike's information criterion (AIC) and Bayesian information criterion (BIC) were also used as measures of model tness to select the best one. Similar to the log-likelihood ratio test, the model with small AIC and BIC value is considered as the better model.

Socio-demographic characteristics of respondents
After removing 398 missing cases, a total of 4322 (weighted) women aged 15-49 were included for analysis. Majority 3375 (82%) of the respondents were from the rural part of the country and 1732 (40%) of them were from the Oromia region. The mean age of the respondents was 27.76 ± 9.1SD years and the majority, 2635 (61%) of the respondents do not attend formal education. Most of the respondents, 1832 (42.4%), were Orthodox by religion and 953 (22%) of them were from the richest family. Table 1 shows a cross-tabulation of the socio-demographic characteristics of the respondents with their experience of domestic violence.

Spatial distribution of domestic violence in Ethiopia
The result from this study shows that the spatial distribution of domestic violence among women aged 15-49 in Ethiopia was non-random with Global Moran's I 0.26 (P-value < 0.01). The z score value of 8.29 indicating less than 1% likelihood that the observed clustering of domestic violence among women in Ethiopia is the result of random chance. The result from Anselin Local Moran's I indicates the existence of hot spot, cold spot, and outlier clusters in the study area. Hot spot clusters are observed in Amhara regions (East Gojam and West Gojam zones), in the Oromia region (West Arsi, Guji, Bale, and Jimma zones) and in SNNP (Sidama, Gedio, Dawro and Gamo Gofa zones). Cold spot clusters were observed in Benishangul Gumuz, Tigray (Eastern, Central, and Southern zones) and eastern part of the Somali region. Figure 3 shows Output from Anselin Local cluster analysis of domestic violence in Ethiopia.

Sat scan analysis of domestic violence in Ethiopia
A total of 24 signi cant locations of clusters were identi ed. Among these, 10 considered as are most likely (primary) clusters and the rest as a secondary cluster. The primary cluster was located in Oromia, Somalia, and some parts of SNNP regional states. In the Oromia region, speci cally at (Guji, Borena, and Bale Zones), in Somalia region Liben and Afder zones and in SNNP Sidama zone were included. The secondary cluster was located in the Amhara region (in the Eastern Gojam zone) and in the Oromia region (Jimma zone). The primary cluster spatial window was centered at 5.203234 N, 40.019732 E with 187.83 Km radius with a relative risk (RR) of 2.18 and Log-Likelihood ratio of 39.55 at P-value <0.001. The spatial window of secondary cluster detected by Sat Scan analysis centered at 10.984556 N, 38.044450 E with 29.42 Km radius with RR of 2.96 and LLR of 28.56 with P-value of < 0.001. The bright red ring shown in gure 4 shows the primary signi cant cluster and the green ring shows the secondary cluster.

Result of Logistic regression
Binary logistic regression analysis was employed in order to see the association of predictor variables with domestic violence. According to the output from this model, age, education, religion, wealth index, husband/partner's education, husband/partner alcohol use, respondent's father ever beat mother, respondent afraid husband/partner, marital controlling behaviors, region and community acceptance of wife-beating shows signi cant association with domestic violence.
The experience of domestic violence was increased with increasing in woman's age. The Odds of experiencing domestic violence were 2 times higher for women aged 20-24 with AOR 2.09 and 95% CI (1.36, 3.17) and 3 times higher for women aged 45-49 with AOR 3 and 95% CI (1.84, 5.15) when compared with those women aged 15-19. Women from the richest family were 48% and those from richer families were 42% less likely to experience domestic violence when compared with those women from the poorest household with AOR and 95%CI 0.52 (0.36, 0.75) and 0.58 (0.45, 0.77), respectively.
Husband/partner education was also signi cantly associated with domestic violence. Women whose husband/partner's education is a secondary school were 42% and those with primary education were 19% less likely to experience domestic violence when compared to those with no education with AOR and 95% CI of 0.58(0.41, 0.82) and 0.81( 0.66, 0.97), respectively.
Women whose husband/partner drink alcohol were 2.6 times more likely to experience domestic violence when compared to those whose husband/partner does not drink alcohol with AOR 2.62 and 95%CI of (2.09, 3.29).
The Odds of experiencing domestic violence among women who witnessed family violence during childhood were 2.2 times higher than those who do not saw family violence with AOR 2.24 and 95%CI of (1.81, 2.58).
Women whose husband/partner exhibit at least one type of marital controlling behavior were 4.3 times more likely to experience domestic violence when compared to those whose husband/partner don't exhibit any kind of marital controlling behavior with AOR 4.26 and 95% CI (3.55, 5.11).
The Odds of domestic violence was 4.4 times higher among women who afraid their husband most of the time and 2.3 times higher among those who sometimes afraid their husband when compared to those who don't afraid their husband with AOR and 95% CI of 4 (3.45, 5.61) and 3.21 (1.83, 2.81), respectively.
Women who live in communities where wife beating is highly acceptable were 1.4 times more likely to experience domestic violence when compared to those who live in communities where wife beating is less acceptable with AOR 1.39 and 95% CI of (1.16, 1.66). Table 2 displays the output from the binary logistic regression analysis.

Result of multilevel logistic regression analysis
The null model is the rst model in multilevel regression analysis in which only the intercept randomly varies across level two units without adjusting for predictor variables. The intercept only model intends to verify the heterogeneity of communities for experiencing domestic violence. The result from the null model shows that the variance of random factor is .716 with its calculated Z statistics of 7.35 and pvalue of 0.000. This shows that experiencing domestic violence among women aged 15-49 randomly varies across clusters. The ICC value shows that 21.4% of the variation in the outcome variable was explained by the grouping variable and the rest was by predictor variables.
The second model is a random intercept model that has a random intercept component and a xed coe cient of individual and relationship level factors. The third model (full model) was developed by including community-level variables on model two. The output from this model shows that the experience of domestic violence was increased with increasing in women's age. The odds of experiencing domestic violence were 2.8 times higher among women whose age group was 30-34 and 4.2 times higher for those aged 45-49 when compared to women age 15-19 with AOR 2.8 and 95% CI of (1.05, 4.54) and 4.2 (1.82, 9.82), respectively.
Women from the richest family were 59% and those from richer families were 45% less likely to experience domestic violence when compared to those women from the poorest household with AOR and 95%CI of 0.41 (0.22, 0.77), 0.55 (0.36, 0.84), respectively.
Women whose husband/partner drink alcohol were 2.7 times more likely to experience domestic violence when compared to those whose husband/partner does not drink alcohol with 95%CI of (1.84, 4.01).
The Odds of experiencing domestic violence among women who witnessed family violence during childhood were 2.5 times higher than those who do not saw family violence with 95%CI of (1.86, 3.37).
Women whose husband/partner exhibit at least one type of marital controlling behavior were 4.2 times more likely to experience domestic violence when compared to those whose husband/partner don't exhibit any kind of marital controlling behavior with 95% CI (3.09, 5.63).
The Odds of domestic violence was 5.4 times higher among women who afraid their husband most of the time and 2.5 times higher among women who sometimes afraid their husband when compared to those who don't afraid their husband with 95% CI of (3.560, 8.132) and (1.652, 3.726) respectively. Table  3 shows the output from multilevel logistic regression.

Model Comparison
The logistic regression model and the two-level generalized mixed model were compared based upon their log-likelihood ratio and the two criterion measures (AIC and BIC). The model with small AIC and BIC measure was considered as the best-tted model.
The output from the analysis shows that employing a two-level generalized mixed model could not improve the model tness. Rather logistic regression analysis is considered as the best-tted model since it has signi cantly lower AIC and BIC values. Table 4 shows AIC and BIC values for logistic regression and generalized mixed model.

Discussion
This study uses national representative sample EDHS 2016 data to determine the spatial distribution and determinant factors of domestic violence in Ethiopia. Almost one-third (34%) of women aged15-49 experienced domestic violence in their lifetime. And 24%, 23.5% and 10.1% of women have experienced emotional, physical, and sexual violence by their husbands/partners, respectively. This nding is in line with WHO prevalence estimates of intimate partner violence for African countries (2), a study conducted in Ghana (5), and almost similar to the 2016 DHS national report (7). This high prevalence indicates that domestic violence remains as the major social and public health problem in the country.
The study also nds out that the spatial distribution of domestic violence cases was non-random in Ethiopia. The Global Moran's I value of 0.26 and Z score of 8.29 with p-value < 0.0001 indicates that there was signi cant clustering of domestic violence throughout the country. This means the distribution of domestic violence cases was more prevalent in some communities than the others. Community-level factors such as community norms towards wife-beating may have a signi cant role in the distribution of violent cases. The spatial clustering of domestic violence cases was also reported from a study conducted in Brazil (24), a spatial epidemiologic study conducted in Spain (20) and from a study conducted in Rwanda (22).
The result from Sat Scan analysis of the data identi es primary and secondary most likely clusters. The primary signi cant cluster was located in Oromia (Guji and Borena zones), Somali (Liben and Afder zones), and SNNP (Sidama zone) regions of the country. The secondary cluster was located in the Amhara region east Gojam Zone and in Oromia in the Jimma zone. Women who live in these clusters have a high risk of experiencing domestic violence when compared to those women who reside outside these clusters. The observed high clustering of domestic violence cases in these areas may be attributed to community and societal-level characteristics such as societal norms that encourage wife-beating, socio-economic status of communities, or weak legal and community sanction on aggressors (13)(14)(15)20). The outcome from regression analysis supports this re ection since community-level factors speci cally, community acceptance of wife-beating increases the likelihood of experiencing domestic violence. The results from previous studies conducted in foreign countries also nd that spatial variation in the distribution of intimate partner violence clusters was mainly attributed to neighborhood-level characteristics (18,19,25). High risk of intimate partner violence was observed among socioeconomically disadvantaged communities, high immigrant concentration, and a high level of public disorder (20). In the current study, su cient community-level variables (neighborhood level) were not included. Therefore, future studies need to incorporate su cient community-level predictors when conducting a similar study.
The result from logistic regression analysis shows that woman's age is signi cantly associated with domestic violence. As a woman's age increases, the likelihood of experiencing domestic violence was also increased. The reason why older age women have a high risk of experiencing domestic violence when compared to younger ones may be because younger women may hide their real status because of different cultural in uence or older women are more likely to be in a union for a longer time and this gives them more chance to encounter violent situation. This result is consistent with an ecological study conducted in Brazil (26) and in Nigeria (15). The socio-economic status of women shows a signi cant association with domestic violence. Women from the richest family were 48% and those from richer 42% lower risk of domestic violence than those from the poorest households. This nding suggests that living in poverty is a signi cant factor in experiencing domestic violence. The nding from this study is supported by studies from Brazil (13), Zambia (6), Rwanda (22), and Ethiopia (8). The relationship between low economic status and domestic violence may be explained by a husband/partner with low income might not be able to support the household expense properly and this might also be one cause for disputes.
This study also nds out that the respondent's husband/partner alcohol use is signi cantly associated with experiencing domestic violence. A woman whose husband/partner drink alcohol was 2.6 times more likely to experience domestic violence when compared to those whose husband/partner does not drink alcohol. This nding is in line with a previous study conducted on 14 sub-Saharan countries (27), with a study conducted in Ghana (5), Nigeria (15), Zambia (6), a systematic review of 15 articles in Ethiopia (8) and a study conducted in Robe Hospital, southeast Ethiopia (28). The result of this study is lower than a study conducted in southeast Oromia (29) and in northwest Ethiopia (16). This difference may be due to differences in the study population (because the previous studies are conducted mainly among pregnant women) and sample size differences (the current study was employed on a large sample size). Despite this difference, harmful alcohol consumption by husband/partner is considered as the main risk factor for domestic violence against women. The main reason why women whose husband/partner drink alcohol have a higher risk of domestic violence could be because excessive alcohol drinking may affect the cognitive function of mind, reducing self-control and makes individuals incapable of peaceful resolution to con icts (30).
Respondent witnessing family violence as a child was also show signi cant association with experiencing domestic violence. A woman who saw family violence as childhood was 2.2 times more likely to experience domestic violence. This nding was consistent with a previous study conducted in Nigeria (15), southeast Oromia (29), and North West Ethiopia (31) but, lower than a study conducted in Ghana (5). The difference of results between the current study and the study in Ghana ay be due to population differences. The relationship between observing family violence and experience of domestic violence may be explained as a child who witnesses family violence may develop a behavioral or emotional problem in letter life and this could make him/her incapable to form stable relationship.
Marital controlling behaviors by husband/partner also show signi cant association with experiencing domestic violence. Women whose husband/partner exhibit at least one type of marital controlling behaviors were 4 times more likely to experience domestic violence when compared to those whose husband/partner do not exhibit any type of marital controlling behaviors. The relationship between marital controlling behaviors and experience of domestic violence can be explained as if a husband/partner exhibits repetitive marital controlling behaviors, good feeling and effective communication with his wife will disappear and this will lead to disputes and occurrence of violent circumstances. The result of this study is in line with studies conducted in Brazil (32), Nigeria (33), and southwest Ethiopia (34).
Fearing of husband/partner by a woman also strongly associated with experiencing domestic violence. Women who sometimes afraid of their husband/partner were 2 times and those who most of the time afraid were 6 times more likely to experience domestic violence when compared to those who do not afraid of their husband/partner. This nding is consistent with a study from Nepal (35) and Uganda (36). Fear of husband/partner was considered as the consequence of many hostile behaviors and it is associated with many violent activities (36). This is explained as the husband's/partner's repetitive aggressive behaviors could make a woman fear him and obligated to stay in a relationship with a violent partner. Social norms are community-level factors identi ed by previous studies to have a strong association with domestic violence. The analysis of social factors by different scholars shows that social norms can be manifested in two ways the rst one is through gender norms and the second one is through gender norm perpetuating violence against women. Gender norms are informal social rules and expectations that distinguish males from females whereas gender norms perpetuating violence against women are norms that normalize violence within a speci ed community (37). The current study focus on the second type of social norm; community acceptance of wife-beating that shows a strong association with a woman's experience of domestic violence. Women who live in a community where wife-beating for husband is highly acceptable were 3.6 times more likely to experience domestic violence when compared to those who do not accept the beating of a wife. This result shows that the existence of a permissive social norm in the community plays a signi cant role in facing domestic violence by a woman. This nding is consistent with two previous studies from Nigeria (14,15) and a study from Ethiopia (38).
This study has some strengths and weaknesses. The rst strength of the study is it uses nationally representative data to show the magnitude and determinant factors of domestic violence in Ethiopia.
Second, it demonstrates the spatial distribution pattern and displays signi cant locations of domestic violence clusters across the country. Finally, it employs both conventional and multi-level logistic regression analyses and compares model tness parameters to choose the best-tted model. Despite this, the study also has some limitations. The rst one is due to the secondary nature of the data, the study could not incorporate su cient community-level variables. Therefore, future studies should consider more community-level variables when conducting similar studies. The other limitation is in order to maintain the con dentiality of respondents, location data in EDHS 2016 was displaced by 2 KM for urban areas, and 5 KM for rural. Therefore, the study may not display the actual locations of domestic violence clusters.

Conclusion
This study nds out that nearly one-third of women have experienced domestic violence by their husbands/partners. The output from the spatial statistical analysis shows that the spatial distribution pattern of domestic violence cases was non-random in Ethiopia. The Global Moran's I statistics shows that there is signi cant clustering of domestic violence cases in Ethiopia. And the output from Sat Scan analysis identi es primary and secondary clusters of domestic violence. Primary clusters were observed in southern Oromia, Somali, and some parts of SNNP whereas, secondary clusters were observed in Amhara and Oromia regional states.
In this study, a strong association of domestic violence with an individual, relationship, and community factors were observed. The output from logistic regression shows that husband/partner's alcohol use, witnessing family violence as a child, marital controlling behaviors, being afraid of husband/partner and community acceptance of wife-beating were predictors of domestic violence. The strategy for prevention and control of intimate partner violence should follow a multi-sectorial approach. Comprehensive and collaborative action should be taken by involving stakeholders at all levels including health professionals, government o cials, non-government organizations, community and religious leaders. Based upon this, speci c activities may include Organizing media on awareness creation and continuous education on how to maintain a stable relationship between couples, taking strong action on controlling alcohol and other drugs, preparing standard limit of alcohol use across the country and Employing long term and intensive effort for transforming culture and social norms that encourage violence against woman is among the major ones.

Declarations Ethical consideration
Ethical clearance letter that explains the appropriateness of the study was obtained from the University of Gondar Ethical Review Board. Written consent was obtained from DHS Program International Inc. to access the dataset. To ensure the con dentiality of respondents, in EDHS 2016 dataset, any personal information was well coded and location address was displaced by 2 KM for urban and 10 KM for a rural resident.

Consent for publication
Not applicable for this section.

Availability of data and materials
The datasets used for analysis are available from the corresponding author on reasonable request.

Competing of interests
"The authors declare that they have no competing interests" in this section.

Funding
The study was funded by Jimma University Medical Center, Jimma Ethiopia. The funding body has no any role in the design of the study, collection, analysis, and interpretation of data.
Authors' contributions ES designed the study, develop the proposal, worked in data extraction, performed analysis and interpretation of the results and prepared the manuscript. TM assisted the analysis and interpretation of result, approved the proposal, revised the manuscript. KA provided technical support on geospatial analysis of the data, assist in writing of proposal and approve the proposal. All authors read and approved the nal manuscript.  Tigray 1 *** P-value < 0.01 ** P value < 0.05