Ethics

Ethical approval for the research was provided by the Medical Research Council of South Africa; the College of Humanities, Beijing Forestry University; and the Faculty of Medicine at the University of Colombo, Sri Lanka [16]. We followed ethical and safety guidelines for research on violence against women [4]. Interviewees received an information sheet and provided written consent.

Sample

Data collection sites varied. In Sri Lanka, the city of Colombo and three contrasting districts were study sites. In China, research took place in a single county, with sampling split between a rural area and town. In PNG, there was a representative sample of the island of Bougainville. Further details of the research can be found elsewhere [16,17].

In each site, we selected census enumeration areas, with a probability proportionate to size, and systematically selected households within these areas. In households, we invited a woman aged 18–49 years (where necessary, randomly selected) for a face-to-face interview, with a trained female interviewer). In China, a household list of individuals in each cluster by age and sex was available and used for sampling within selected clusters, and the entire questionnaire was self-completed. Full details of the methods, sampling, are presented elsewhere [16,17]. The proportion of enumerated and eligible women interviewed per site varied, in Sri Lanka it was 73.9%, China 84.9%, and PNG 85.0% [16,17].

Measures

Intimate partner violence measures

The different forms of IPV women may have experienced were assessed through standardized measures in all three countries, based on questions first developed for the WHO multi-country survey on women’s health [3], and only asked to women who reported ever being partnered in their lifetime. These items have been refined in multiple studies globally and previously used in the Asia-Pacific context [3]. To assess emotional IPV, we asked women five questions about their lifetime experiences, for instance: “Has a current or previous husband or boyfriend ever insulted you or made you feel bad about yourself?” Responses were ‘never, once, few, or many times’. Similarly, four items asked about women’s lifetime experiences of economic IPV. An example question was: “Has a current or previous husband or boyfriend ever taken your earnings from you?” Responses were ‘never’, ‘once’, ‘few’, or ‘many times’.

Lifetime experience of physical IPV and sexual IPV were assessed through five and four, behaviourally specific items, respectively. An example of physical IPV was: “Has a current or previous husband or boyfriend ever hit you with a fist or with something else which could hurt you?”, while for sexual IPV: “Has a current or previous husband or boyfriend ever physically forced you to have sex when you did not want to?” Response options for both scales were “never, once, few, or many times”.

All three countries included a series of standardized questions about socio-demographics, including age, highest level of education, current relationship status, food-security and ability to mobilise resources (cash) in an emergency. We also asked about nine experiences of childhood abuse and neglect, using items adapted from the Childhood Trauma Questionnaire.

Analysis

The country data sets were combined into one. Data analysis were conducted using Stata IC/14.1 (StataCorp, College Station, TX, USA). Procedures accounted for the multistage structure of the data set, with stratification by data collection site within countries and enumeration areas as clusters. The data set was self-weighting.

First, we describe the sample by country, and as a pooled data set, using percentages and means as appropriate. We then examine the prevalence of each type of IPV assessed (emotional, economic, physical, sexual), through creating a three-level categorical variable to indicate lifetime experience of IPV: never, once, more than once. We then created an overall IPV experience variable, which parsed out all possible experiences of IPV, to look at overlaps and co-occurrence. This is also presented as a proportional Venn diagram. Descriptive analyses took into account the clustered nature of data, and 95% confidence intervals (CIs) were calculated using Taylor linearization [20], to account for data clustering.

Second, to describe prevalence and patterning of emotional IPV and economic IPV by item we: i) provide individual item responses as a percentage by country; and ii) estimate the item’s uniqueness, as the proportion of women who only reported that item in the scale. Essentially identifying the proportion that would not be categorised as experiencing that form of IPV if the item were not asked.

To assess associations between emotional IPV and economic IPV and health outcomes by individual items, and by country, we report the prevalence of each item by i) lifetime never/ever, ii) none or once, compared to few or more lifetime experiences, and iii) lifetime frequency of exposure (never, once, few, many). We present mean scores, and 95% CIs for depressive symptoms (China and PNG), and mean scores and 95% CIs for self-rated health (all countries). Where there is no overlap between 95% CIs, we consider this indication of the mean scores being different.

To assess whether emotional and economic IPV have independent associations with health outcomes, we pooled the data from all countries; country level analysis was infeasible because of sample size issues. We first assess this descriptively, through creating four-level categorical variables for multiple combinations of emotional and physical and/or sexual IPV, and economic and physical and/or sexual IPV. For instance, we created a four-level variable describing women who had i) no exposure to IPV ii) only exposure to economic IPV in their lifetime, iii) exposure to economic IPV and physical IPV in their lifetime, and iv) only exposure to physical IPV in their lifetime. We created a range of these four-level variables, based on frequency of exposure categories (never vs ever, and never or once, vs few or many) and types of exposure (sexual, physical, physical and/or sexual). For each permutation, we provide mean scores and 95% CI for depressive symptoms and self-reported health. We compare whether there is overlap between 95% CIs of each category and those reporting no IPV experience.

We then use unadjusted and adjusted multivariable regression models to assess whether IPV was independently associated with the health outcome in multiple ways. For depression, we used a negative binomial regression, and for self-rated health, ordered logistic regression. First, we separately model the impact of emotional, economic, and physical and/or sexual IPV, with no allowance for co-occurrence (Models 1-3), and then consider them together in a single model (Model 4). However, given that the constructs are largely overlapping, there was significant collinearity. We thus constructed four-level categorical variables comparing combinations of emotional IPV and physical and/or sexual IPV, and economic IPV and physical and/or sexual IPV experiences and health outcomes (Models 6-8). Finally, we created an eight level variable, which categorized women into all possible IPV combinations across the types of IPV its association with health (Models 9 and 10). All adjusted models included age, education, poverty and childhood traumas. All models included adjustment for the survey structure. We present the results of all of these models to explore the extent to which the measurement strategy impacts the point estimates for effect sizes and the conclusions drawn.

Description of the sample

In total 2438 ever partnered women were interviewed across the three countries (Sri Lanka n = 559; PNG n = 792; China n = 1087). In all countries (Table 1), the largest proportion of women were aged 35-49 years old (49.5% of the overall sample). In Sri Lanka and China over half the women interviewed (53.9% and 57.5% respectively) had some secondary education, but only 13.4% in PNG. Very few had no education whatsoever.

In all countries, lifetime emotional IPV was the most prevalent form of IPV women reported (Table 1), but with wide variation from 29.5% in Sri Lanka to 61.0% in PNG. Overwhelmingly, women who reported emotional IPV were likely to report either “few” or “many” episodes over their lifetime, rather than just “once”.

Lifetime economic IPV was reported by 35.7% of the overall sample, with variation by country (Sri Lanka 26.3%; China 24.8%; PNG 55.4% - Table 1). In Sri Lanka and PNG, economic IPV was as common as physical IPV, while in China it was less common than physical IPV, but more common than sexual IPV. Women who reported any exposure to economic IPV, mainly reported experiencing either “few” or “many” experiences, rather than just “once” in their lifetime.

When all forms of lifetime IPV were considered together and parsed out (Table 1) 35.8% reported never experiencing any IPV, 7.0% only emotional IPV, 6.0% only economic IPV, and 3.6% emotional and economic IPV only. A quarter (23.4%) reported lifetime experience of all forms of IPV. The Proportional Venn diagram (Figure 1) demonstrates these overlaps clearly.

Figure 1.  Proportional Venn diagram demonstrating overlaps between different forms of IPV women experienced in the three countries (pooled data).

Variation in reporting of items by country

Table 2 shows the prevalence of each individual item for emotional IPV. The most common emotional IPV item varied by country: in PNG it was ‘being insulted’ (55.4%), while in Sri Lanka and China it was ‘being scared or intimidated’ (22.1% and 23.8% respectively). In all countries, ‘hurt people you care about’ was the least reported item.

The extent to which women only reported that item for emotional IPV – uniqueness – varied by country. In all countries, the item “hurt others” was uniquely reported by <3% of participants, but there was no other item showing such consistency.

For economic IPV (Table 2) the most prevalent item varied across country; in PNG ‘earnings taken’ was reported by a third (35.0%) of women, while in Sri Lanka and China the most common form was “prohibited from getting a job” (16.7%; 14.4% respectively). In all three countries, the least common form of economic IPV reported was “thrown out of house” (2.5% Sri Lanka; 4.3% China; 21.9% PNG).

For economic IPV there was no clear patterning for uniqueness of responses by country. ‘Prohibited from working’ had high unique percentages in Sri Lanka and China (39.1% and 43.5% respectively), but not PNG (5.1%). While for a fifth of women in all three countries the only form of economic IPV they reported was ‘having earnings taken’ (21.0% Sri Lanka; 22.6% China; 22.5% PNG). Overall, the lowest unique item in all three countries was “thrown out of house” where uniqueness ranged from 0.7% (Sri Lanka) to 6.0% (PNG).

Descriptive association between emotional IPV items and health outcomes

Descriptive associations between each emotional IPV scale item and health symptom mean scores by country were assessed (Table S1 in the Online Supplementary Document). For individual items of emotional IPV, there was a consistent patterning for the never/ever binary, and never or once vs ‘few or many’ binary, and their association with health impacts, with women reporting experiencing these forms of emotional IPV having higher means scores of depressive symptoms and worse self-reported health compared to those reporting no experience. For depressive symptomology, there was no overlap between 95% CI for any item, and for self-rated health there was no overlap between 95% CIs in 14/30 instances.

By the frequency of reported lifetime experience (never, once, few, many) for individual emotional IPV items (Table S1 in the Online Supplementary Document), there was a consistent gradation, whereby women reporting more experiences of any item, reported higher mean depressive symptoms. For all items of the emotional IPV scale, women reporting ‘few’ or ‘many’ experiences had higher mean depressive symptoms, with no overlap of 95% CIs, compared to those reporting “never”.

For self-reported health, frequency of experience by emotional IPV item was associated with worse health, compared to those reporting no experience (Table S1 in the Online Supplementary Document). Only with one item “scare or intimidate you” was there consistency across countries, whereby in all three countries those reporting ‘many’ experiences had no overlap with “never” for 95% CIs. For “insulted you” there was no overlap between 95% CIs for China and PNG, among women reporting “many” experiences of this, compared to “never”.

Descriptive association between economic IPV items and health outcomes

Individual items of the economic IPV scale were assessed by health outcomes. For each item (Table S2 in the Online Supplementary Document), there was a similar patterning to emotional IPV, whereby higher mean scores were reported by those reporting more experiences of economic IPV. For depressive symptoms for never/ever, and never or once/two or more experiences, there was no overlap between 95% CIs for 13/16 combinations. For self-rated health, similarly the majority of binary combinations (17/24) had no overlap of 95% CIs.

For the frequency of experience of individual economic IPV items (Table S2 in the Online Supplementary Document), those reporting “once”, “few” or “many” experiences for each item had higher mean scores for depressive symptoms and self-rated health, compared to those reporting “never”. However, there was little consistency, whereby in 4/8 combinations for depressive symptoms women reporting ‘few’ had higher depressive scores than those reporting ‘many’ experiences.

Descriptive and adjusted associations between different combinations of IPV and health outcomes

We also assessed descriptive (Table S3 in the Online Supplementary Document), unadjusted and adjusted associations (Table 3, Table 4 and Table 5) between different IPV combinations and health outcomes. For emotional IPV there are relatively consistent patterns. In all different models (descriptive and regression) emotional IPV is associated with more depressive symptoms, whether emotional IPV was coded as one or more experience, or two or more experiences. However, there is less clear patterning between emotional IPV and self-rated health; descriptively (Table S3 in the Online Supplementary Document) in 2/6 cases there is overlap between 95% CIs for emotional IPV only, and no IPV experience. In adjusted regressions in 3/6 models (Models 6, 9 and 10 in Table 4 and Table 5) emotional IPV was not associated with self-rated health.

For economic IPV, independent associations with worse health outcomes are stronger for self-reported health, particularly when coded as two or more experiences. In descriptive models, and adjusted models, when economic IPV is coded as two or more there are consistent associations with worse self-rated health, with no overlap of 95% CIs in descriptive models (Table S3 in the Online Supplementary Document), and in all adjusted models (Table 4, Model 8; Table 5, Model 10). There are less consistent findings when economic IPV is treated as one or more experiences. Similarly, there are more mixed findings for economic IPV and its association with depressive symptoms, where the models show inconsistent findings.

Across all models women who experience physical and/or sexual IPV, when combined with emotional or economic IPV consistently strong associations with worse health outcomes, and the highest scores for depression and self-rated health. For instance in Table S3 in the Online Supplementary Document, all combinations of combined IPV show the highest mean scores for symptoms. Similarly, in all the adjusted regression models, combined experiences of IPV are show the largest coefficients for depressive symptoms and self-rated health, and are in all models these associations are highly significant (Table 4 and Table 5).