Socio-demographic status: Age in years, current living arrangements (joint, partly joint, not joint), marital status (not married, engaged to be married/married/separated or divorced), and fertility (parity, interval between marriage and first child, fertility problems yes/no, number of abortions).
Psychiatric morbidity: The Self Rating Questionnaire (SRQ-20) it is a 20 item self-report scale based measure of psychiatric morbidity developed and validated cross-culturally as a psychiatric screening tool for primary health care settings in developing countries . It has been translated into Urdu and used, and validated in four studies in Pakistan [4, 20–22]. Case criteria for the study was set at ≥ 8 cutpoint on SRQ-20 . When validated against the Present State Examination (PSE) as a second stage assessment of psychiatric morbidity, the sensitivity varied between 73%-83%, and specificity between 72%-85% . When validated against the Psychiatric Assessment Schedule; the best threshold for women was 7/8 with a sensitivity of 78% and specificity of 81% .
1) Perceived gender preference of parents from birth. A parental preference for boys over girls, and a preferential allocation of care and family resources towards boys are the two most commonly cited elements of gender disadvantage in countries where the bias is culturally embedded. We asked
i) Did you feel that your parents would have preferred you to have been a boy? (Yes/no)
ii) Did you feel that your parents favoured your brothers or other male relatives over you? (Yes/no)
iii) Did you have the same access to health care and education as your brother or other male relatives? (Yes/no)
These questions were easily comprehended by Pakistani women and deemed appropriate to elicit experiences of disadvantage.
2) The Parental Bonding Instrument (PBI)  assesses the adequacy of a child's bonding with parents in early life in two dimensions; care and overprotection. Scale scores ascertained in adulthood reflect childhood experience , remain stable over long periods  and are relatively little coloured by current affective status [25, 26]. We used the shorter 16 item version of the PBI , which we had previously validated in the same research setting . We hypothesized that women disadvantaged by their gender would report lower levels of care and higher overprotection.
3) Level of education, collapsed into none, primary (up to 5 years), middle (up to 8 years), matriculation (up to 10 years), intermediate (up to 12 years), Bachelor's, Masters and Doctoral. We hypothesized that women disadvantaged by their gender would have attained a lower educational level.
4) For married women only
i) Age at marriage (in years)
ii) Marital satisfaction: The shorter 14 item Marital Satisfaction Scale  with a modified three point item response scale, as validated in an earlier pilot study in the same setting . We hypothesized that women disadvantaged by their gender would have been married earlier and would report lower levels of marital satisfaction.
5) Autonomy. We included an ad hoc measure to assess the participants' degree of liberty and autonomy in adulthood in their parents' or husband's household. This was based on yes/no answers to the following four items (Cronbach's alpha, 0.76):
i) do your parents/husband restrict you going out with your friends?
ii) would your parents/husband allow you to work outside of home?
iii) would your parents/husband allow you to work in a male environment?
iv) if you like a man would you tell your parents? (for married women this was asked in the past tense)
We hypothesized that women disadvantaged by their gender would report lower levels of autonomy.
Socioeconomic status (SES): To ensure adequate variance in SES, sampling was stratified into two sets of catchment areas characterized by differing economic circumstances. For each participant, we also assessed current SES and SES of the family of origin as follows: household income, husband's income, father's income and a household assets index based on possession of the following utilities and items: television, video recorder, computer, air conditioning, car, domestic help, visits abroad, home owner, number of bedrooms, number of bathrooms. Cronbach's alpha was 0.89.
Life Events: The List of Threatening Events  identifies 12 event categories most likely to be rated as providing significant contextual threat. We ascertained events occurring in the last one year.
Social Support/Social Network: Close Persons Questionnaire (CPQ)  is a structured questionnaire assessing several dimensions of social support, including emotional/confiding, practical and negative aspects of support from up to four identified sources. The Close Persons Questionnaire has been used on Asian population including Pakistanis living in the United Kingdom  however it has not been previously used in Pakistan.
Woman's gender preference for her own child: assessed using the question 'How important is it for you that your first child should be a son?' with the responses 'not so important', 'important' and 'very important'.
We report the principal characteristics of the sample by catchment area. Next we conducted a pathway analysis, to test the model that gender disadvantage is a life course phenomenon with pathways from parental preference for a boy to low care and high overprotection, to low educational level, to early age at marriage and low marital satisfaction to low levels of autonomy in adulthood. Each component is represented by an observed variable. The full model could be run on married women only. A model omitting age at marriage and marital satisfaction was run in a secondary analysis on the whole sample. Next we developed a structural equation model to test the hypothesis that gender disadvantage is independently associated with psychological morbidity. The measurement model initially consisted of five measurement sub-models for the following unobserved variables (latent constructs)
1) psychological morbidity, for which the indicators were the 20 SRQ items
2) negative marital satisfaction (the seven negatively orientated marital satisfaction items)
3) positive marital satisfaction (the seven positively orientated marital satisfaction items)
4) socio-economic status (high versus low SES catchment area, household assets index, husband's income, father's income, household income and level of education)
5) gender disadvantage (the three gender disadvantage perception questions, PBI care, PBI overprotection, level of education, age at marriage and autonomy).
Level of education was therefore an indicator for socioeconomic status and gender disadvantage. Positive marital satisfaction, negative marital satisfaction, socioeconomic status and gender disadvantage were all assumed to be correlated with each other. We tested for pathways from socioeconomic status and gender disadvantage to psychological morbidity. We also tested for direct pathways from each of the gender disadvantage indicators to psychological morbidity (i.e. effects of the observed variable not mediated through the latent construct of gender disadvantage), and for direct effects of age, number of children, interval between marriage and first child, fertility problems, number of abortions, number of life events and emotional/confiding social support. Non-significant pathways and variables were removed from the model. Again, the full model could only be run on married women. A partial model, excluding age at marriage and marital satisfaction, was run on the whole sample as a secondary sensitivity analysis. Pathway associations for the prediction of psychological morbidity are expressed as crude and standardized regression weights. Correlations are reported for associations between unobserved variables. A squared multiple correlation indicates the proportion of the variance in the latent construct psychiatric morbidity accounted for by the variables in structural model. Model fit was assessed using the Tucker-Lewis index (TLI), Root Mean Square Error of Approximation (RMSEA) and Akaike's Information Criterion (AIC). Akaike's Information Criterion (AIC)  adjusts the model chi-square to penalize for model complexity. The lower the AIC value, the better the fit of the model . The Tucker-Lewis index (TLI) indicates the proportion of co-variation among indicators explained by the model relative to a null model of independence, and is independent of sample size. Values near 1.0 indicate good fit; those greater than 0.90 are considered satisfactory [37, 38]. The Root Mean Square Error of Approximation (RMSEA) assesses badness of fit per degree of freedom in the model and is zero if the model fits perfectly; RMSEA values of less than 0.05 indicate close fit and 0.05 to 0.08 reasonable fit of a model .
Finally, across the whole sample, we used ordinal regression to identify independent associations between socio-demographic, socioeconomic and gender disadvantage variables and the extent of the woman's preference that her first child should be a boy.