Study setting
This study used the cross-sectional data from the 2017-18 Pakistan Demographic Health Survey (PDHS 2017-18). PDHS 2017-18 is the latest nationally representative survey conducted by the Pakistan National Institute of Population Studies and the ICF International funded by the United States Agency for International Development (USAID).
Study design and population
PDHS 2017-18 collected data for women aged 15-49 years and their children under 5 years old through a stratified two-stage cluster sampling to estimate the key indicators at the national level, in urban and rural areas. The survey was conducted across four provinces (Punjab, Sindh, Khyber Pakhtunkhwa, Balochistan); Azad Jammu & Kashmir (AJK) and Gilgit Baltistan (GB); Islamabad Capital Territory (ICT); and the former Federally Administrated Tribal Areas (FATA) in Pakistan. In the first stage, 580 clusters (enumeration areas from the previous national census consist 200-250 households) were selected. In the second stage, through an equal probability systematic selection process, 16,240 households were selected within 580 clusters. To obtain representative estimates at the national level, sampling weights were calculated and applied. A sample of the women aged 15-49 years (n=15930) who were either permanent residents of the selected households or visitors who stayed in the households the night before the survey were recruited after informed consent, out of them 15,068 women were successfully interviewed in PDHS, and their response rate was 94.6%. For the purpose of the present analysis, we restricted our analysis to married women aged 15-49 years old because for some variables the data were only collected for married women.
Study variables
Variable selected for the explanatory (EFA) and confirmatory factor analysis (CFA)
A total of 26 important variables related to women empowerment (15, 16, 20, 24-27) including those from our previous study (4) that were available in 2017-18 PDHS, were selected for EFA CFA. All categorical variables were either recoded or used in their original format based on their suggested direction and influence on women empowerment so that the categories with higher ranks represent higher levels of empowerment and those with lower ranks indicate low empowerment (16, 26). A summary of 26 variables, seven domains, and four dimensions that were conceptualized in this study along with the details of recoded variables are in appendix 1.
a) Economic dimension
This dimension included two domains; namely, labor force participation and property-owning. Labor force participation included the following indicators: respondent's occupation, type of earning from respondent's work, seasonality of respondent's occupation, and income ratio (women/men). Property owning was represented by legally owning a house or land variables.
b) Socio-cultural dimension
This dimension included three domains; decision-making, attitude toward violence, and age at critical life events. Participation in decision-making was assessed by three items, namely: (1) person who decides respondent's healthcare; (2) person who decides on large household purchases; and (3) person who decides whether the respondent can visit her family or relatives. Attitudes toward violence were assessed using five variables describing whether beating was justified if the wife: goes out without telling her husband; neglects the children; argues with her husband; refuses sex with her husband; burns food. Age at critical life events domain was measured by two indicators including age at first birth and age at first cohabitation.
c) Education dimension
This dimension included two domains; literacy which was measured by the ability of the participants to read and the frequency of reading a newspaper and education level which indicate the highest educational level of participant.
d) Health Dimension
This dimension includes negotiating sex and access to healthcare domains. Women's ability to negotiate sex was measured by indicators describing if they could refuse sex or ask their partner to use a condom. Access to healthcare was classified by four indicators examining the difficulty in getting medical help (not a big problem=1, big problem=0), namely: (1) receiving permission before getting medical help; (2) having money for healthcare; (3) distance to health facility; (4) not wanting to go healthcare facility alone.
Variables related to the access to reproductive and maternity care
Four indicators related to access to reproductive and maternity care were selected as outcome variables including; 1) unmet needs for family planning; 2) adequate ANC; 3) institutional delivery; and 4) skilled birth attendance.
a) Unmet needs for family planning: Unmet need was defined as the unmet need for limiting (i.e. women whose most recent pregnancy was not wanted at all, fecund women who did not use contraception despite their desire to have no more children, women who were postpartum amenorrheic for 2 years following an unwanted birth and were not using contraception) and spacing (i.e. women whose most recent pregnancy was not wanted initially but wanted later, fecund women not using contraception who were undecided when/if they wanted a to have a child or who wanted a child 2+ years later, and women who were postpartum amenorrheic for 2 years following a mistimed birth and were not using contraception) (15). The relevant questions had dichotomous response alternatives (i.e., ‘yes’ or ‘no’ responses) and unmet needs for family planning were coded as “yes=1” and “no=0”.
b) Adequate ANC: Based on the World Health Organization (WHO) recommendation, having at least four ANC visits is necessary for optimal maternal and child outcomes [23]. Therefore, adequate ANC was coded as 'yes=1' for women with at least four ANC visits before their most recent (4+ ANC visits) in the last five years and coded as 'no=0' if there were fewer than four visits.
c) Institutional delivery: This variable is coded into “yes=1” indicating delivery at health facilities and “no=0” indicating delivery at home/elsewhere.
d) Skilled birth attendance: Defined as and coded “1” if birth is delivered with the assistance of a doctor, nurse, midwife, lady health visitor, or community midwife.
Data analysis
Data analysis was conducted in four steps following the procedure from our previous paper (4). First, the variables were extracted from PDHS 2017-18 dataset and either recoded or retained in their original forms for factor analysis (Table 1), then the dataset was randomly split into half using the STATA command “splitsample”. Assuming that homogenous samples of married women aged 15-49 years are being generated, the first half was used for EFA and the second half was later used for CFA as recommended in previous literature (28, 29). The suitability of data for EFA was tested using the Kaiser–Meyer–Olkin (KMO) test of sampling adequacy and Bartlett test of sphericity (30) in which, respectively, values greater than 0.70 and p-value <0.05 are considered favorable. In the second step, the first half of the sample was used to identify the latent constructs that reflect women’s empowerment using EFA. The decision on which domains to be retained was made based on the eigenvalue (>1), scree plot (Figure 1), and the amount of explained variability by each individual domain. The variables with a loading factor <0.3 and those loaded on more than one domain were dropped in the further analysis as recommended by Stevens 2009 (31). To construct the final model and obtain the structural domains-empowerment indices- oblique rotation was adopted over orthogonal rotation to account for the potential correlation between factors (27). In the third step, the internal reliability of the overall index and individual domain was examined by Cronbach’s α test (Table 2) (32, 33) and domains with a Cronbach’s α value <50% as well as the variables that removing them significantly improve the Cronbach’s α coefficients, were dropped (34, 35). In the last step, the construct validity of the index was assessed by confirmatory factor analysis (CFA) in the second half of the sample to estimate how well the measured variables represent the number of emerged constructs. The CFA produces the fit statistics based on the covariate structure of observed data (Table 3) to determine the appropriateness of the model such as the Root Mean Squared Error of approximation (RMSEA) which represent the parsimony of an index; the Comparative Fit Index (CFI), Tucker-Lewis index (TLI), and Standardized Root Mean Squared Residual (SRMR) which represent relative and absolute fit of the index (36). An index with good construct validity has RMSR and RMSEA <0.05 and CFI and TLI more than 0.95 (37). In addition to CFA, to evaluate the convergent validity of the final model, the association between the emerged domains and four indicators of access to reproductive and maternity care; namely unmet needs for family planning, adequate ANC, institutional delivery, and skilled birth attendance were measured. Higher access to reproductive and maternity care services has been observed among more empowered women (9, 12, 19, 38, 39). These associations were estimated using Poisson regression as recommended by Barros et al. (40) and adjusted for household wealth to assess the association of empowerment with the four outcomes of interest independent from the household’s wealth (20). The categories (low, medium, high) for women empowerment domains were obtained by pooling the individual indicators’ scores and approximating the terciles as the cutoff points (20, 41). All the analyses were performed in STATA software version 16 and the p-value<0.05 was considered a significant statistical level.