Our study indicates that in India of 2005/06 and Bangladesh of 2007, education had a protective effect on DV irrespectively of its distribution among spouses. Couples who were equally low-educated (primary education level or less) were the most adverse groups in terms of DV. In contrast, equally high-educated couples (secondary educational level or higher) were less likely to experience DV. The likelihood for experiencing DV in couples with educational gap lies in between couples with equally high and equally low educational level. Wives with higher education than their husbands were less likely to be exposed to DV than wives whose husbands are higher educated. Furthermore, our results point towards a reverse association of the severity of violence and educational level.
Our results are inconsistent with the literature [7
]. An adverse effect in wives with higher education than their spouses could not be verified. However, it has to be considered that different classifications were used in different studies. A crucial limitation of earlier studies is that they did not divide the category of couples with no education gap into two subgroups (no gap-low education and no gap-high education). According to our results, these two subgroups differ remarkably with respect to violence. Assuming a general protective effect of education [7
], merging these subgroups into one group is inappropriate.
To check the consistency of our results, we employed an additional regression model where the outcome variable was dichotomized and equally educated couples were merged into one category. The results show that equally educated spouses were the least likely to experience DV. In couples with SEG and higher educated husbands, the chance of DV was higher compared to couples with higher educated wives. These findings are also not consistent with the literature [7
]. Several factors may explain our findings. One may be the difference in data basis. Ackerson et al. (2008) [7
] used data from Indian 1998/99 NFHS. Part of the disparity of the results may come from the seven-year period between those results in comparison to the ones that were found in the present study. Burazeri et al. (2005) [11
] based their research on Albanian women, which constrains comparison with data from Indian subcontinent because of culture, ethnicities and belief. A second plausible cause may be the differences in coding of the variables. As there is a high number of equally educated couples with no
or only primary education
and low education is associated with high odds for DV, merging the groups with no SEG should lead to an overestimation of the prevalence of violence. Thus, violence should be highly present in the merged group of the second regression model. In reality, the opposite outcome occurred which makes the matched group the least violent (Table ). This fact can be due to a very low prevalence of DV in equally high-educated couples.
Stratified analyses were made to compare the two countries. Few differences were found between the two countries. The decreasing chance of DV in higher educated couples can be regarded irrespectively of countries. The prevalence of DV is higher in Bangladesh compared to India. Except from Bangladeshi less severe violence group, wives who were higher educated than their husbands were less likely to experience DV. Perhaps several results were not statistically significant for Bangladesh due to lower numbers.
The prevalence of overall DV observed in our study sample was 31.4%. In Bangladesh of 2007 the prevalence was higher (47.9%) than in India of 2005/2006 (30.3%). The prevalence in Bangladesh differed from the results of surveyed Bangladeshi men in 2004 which showed that 68% of the interviewees’ wives experienced DV [6
]. This might be due to underreporting of the women because of fear and shame. The results for India are more comparable to earlier results (26% in India of 2007) [23
Our study has some limitations. Firstly, the presented models had a low goodness of fit. Hence, the validity of the statistical models is low and other predictors of DV that were not included have to be considered.
Secondly, there might be selective dropouts in the selection of women for the questionnaire on DV. It can be assumed that violent settings are less likely to provide the privacy which was required to perform the survey. This fact may lead to an underestimation of violence in the current sample. Some characteristics of our final sample were significantly different from the sample that was not interviewed on DV. This narrows the generalizability of our results. Furthermore, estimating DV might be biased by the exclusion of never married, divorced and separated women as DV could be a reason for separation. Our results are affected if the chance of being separated is associated with the educational level of spouses. Nevertheless, we included only currently married women because they were supposed to have a more stable commitment and separation due to DV is less likely. Moreover, recall bias might be lower in currently married than in separated or divorced women as they are reporting on their ongoing partnership.
Another limiting factor is the validity of the outcome variable as it is categorized by DHS [12
]. Violent acts such as kicking or punching with fist or something harmful are classified as less severe violence
. These cases may also be considered to be classified as severe
violence. If so, these cases are missing in the severe violence
category, leading to an underestimation of the sample’s prevalence of severe violence.
Furthermore, as part of the construction of the outcome variable, cases that reported both types of violence, were assigned to severe violence which might have lead to an underestimation in the less severe violence-group.
We used multinomial logistic regression rather than ordinal regression since an ordinal scale could not be assumed. As a result, information of the ordinal-scaled outcome variable was lost and interpretation of the severity of DV has to be done with caution.
Furthermore, there might be a social preference bias that led to underreporting of DV, although the general set-up during the survey was arranged as supportive as possible by DHS. Women who were selected for the additional questionnaire concerning DV had to provide the necessary privacy and were excluded if the arrangement was not appropriate. Nevertheless, it has to be considered that the true unknown prevalence is higher than the surveyed due to reasons of anxiety, shame, or underestimation of the respondent. Also recall-bias can be assumed because the ever-experienced violence by the current husband had to be reported.
Finally, all considered factors are not capable of explaining a satisfying amount of the variance of DV. Some works show further factors that might have an impact but were not surveyed in 2007 BDHS or 2005/06 NFHS. Those factors may be alcohol or drug abuse [18
], poor life satisfaction and well-being [21
], psychiatric and psychological dysfunctions [21
] as well as possible genetic dispositions.