This study analyses the cross-section data of the Tanzania section of the WHO Multi-Country Study on Women's Health and Domestic Violence, a large population based survey that gathered data from more than 24,000 women from a representatively selected samples of households in fifteen study sites in ten countries . Conducted from 2001 to 2002, the Tanzanian section of the WHO multi-country study conducted representative surveys of women of reproductive age in the capital Dar es Salaam and Mbeya region. Dar es Salaam is made up of three municipals which are further subdivided into a total of ten divisions. These ten divisions are comprised of a total of 73 wards. Each ward in urban Dar es Salaam is further subdivided into streets, which itself are further subdivided into ten cell units. In rural Dar es Salaam wards are sub-divided into about 370 villages, which are also further subdivided into ten cell units. The later will have about 10-50 household. Mbeya region has six districts, each of which is further divided into divisions, which are then subdivided into wards. In the divisions Mbeya Urban and Mbeya Rural there are 53 wards, which further subdivide themselves. This administrative structure was used for the two stage cluster-sampling scheme. In Dar es Salaam all districts were included from which 22 wards from the list of 73 were selected, but for Mbeya, only two districts, urban and rural Mbeya, were included where 22 wards were selected from a list of 53. Using probability proportional to size of the wards (in each site) ten cell units were then selected randomly and their households were enumerated and mapped to allow random selection. Only one woman aged 15 to 49 was randomly selected per household and interviewed in complete privacy, except for infants younger than two years, to protect confidentiality and ensure safety . After participation, all respondents received information about available local women's services, which also dealt with domestic violence. Women reporting thoughts on or attempts of suicide were seen by the supervisor responsible for the group and in a few cases referred to the psychiatrist responsible for the mental wellbeing of the team. No compensation was offered to participants.
The survey used female interviewers who were trained using a standardised three-week training course. The response rate at the household level was 100% in both sites and 96% in Dar es Salaam and 97% in Mbeya at the individual level .
The study adhered to the WHO ethical and safety recommendations for research on domestic violence against women  and received ethical approval from WHO Secretariat Committee for Research in Human Subjects and the research and publications committee of the then Muhimbili University College of Health and Allied Sciences. Regional and district directors and ward executive offices of the surveyed communities also gave their assent.
Definitions of key measures used in analysis
The study adopted a number of working definitions. 'Ever partnered' included all women who report having ever been married to, ever lived with or currently have a steady, regular male sexual partner, regardless of the actual length of the relationship. 'Ever pregnant' included all women who reported having ever given birth to a child or having ever been pregnant.
In the survey women were asked directly whether they had experienced specific acts of violence. These include if her partner had slapped or had thrown something at her that could have hurt her, pushed or shoved her, hit her with a fist or with something else that could have hurt her, kicked, dragged, or beaten her up, choked or burnt her on purpose, threatened to use or actually used a gun, knife, or other weapon against her, physically forced her to have sexual intercourse when she did not want to, had sexual intercourse with her when she did not want to because she was afraid of what her partner might do, or forced her to do something sexual that she found degrading or humiliating. Additional questions about the frequency and timing of acts (ever, past year) were also asked. All women who answered positively to at least one of the questions about specific acts of physical or sexual violence committed by a partner towards her at any point in her life were considered to have experienced intimate partner violence.
Lifetime history of miscarriage, induced abortion, and stillbirth was assessed for all women through a single item asking if they ever had a pregnancy that ended in miscarriage, stillbirth or induced abortion. Recent research from Tanzania reported that surveys investigating 'miscarriage' and 'stillbirth' separately may not measure the intended outcome since these are Western concepts which do not directly translate into local categories . Positive answers to these questions are therefore combined under the term pregnancy loss.
The additional explanatory factors in this analysis, as derived from prior studies on pregnancy loss and induced abortion [4, 7, 11], include location (urban/rural), age (continuous), education (no schooling/primary/secondary), partner's education (no schooling/primary/secondary), socio-economic status (low/medium/high), marital status (married/not married), partner having other wives or affairs (Yes/No) and number of live born children (continuous).
The analysis was restricted to ever partnered, ever pregnant women. To analyze the data, descriptive statistics were calculated and possible associations between factors were explored by conducting cross tabulations. Crude odds ratios (OR) between different forms of intimate partner violence and possible associations with induced abortion and pregnancy loss were estimated using bivariate logistic regression. Multivariate logistic regressions were used to obtain adjusted odds ratios (AORs), controlling for the more commonly recognized explanatory factors, such as women's and partner's age, education, socio-economic status, marital status and their partner having other wives or affairs, women's number of live born children and living in a rural versus urban area.
An odds ratio larger than one represents a greater likelihood of the outcome than for the reference category, and an odds ratio smaller than one represents a smaller likelihood compared with the reference category. A p-value below 0.05 was considered statistically significant.
Data were missing for less than five percent of respondents for most variables, and women with missing data were excluded from analyses with that variable. All analyses were conducted using STATA version 11, taking the cluster sampling design into account.