This analytical cross-sectional study used nationally representative data from the 2004–05, 2010, and 2015–16 TDHS which collect data from women aged 15–49 who were either usual residents or visitors in the household on the night before the survey, men aged 15–49 who were either usual residents or visitors in the household on the night before the survey, and children aged 6 to 59 month that had a guardian or parent’s consent [21,22,23]. The study population was all married or cohabiting women of reproductive age (15–49 years).

The DHS uses a sample that is generally representative at the national, residence, and regional level and the sampling technique is of a stratified, two-stage cluster design, where first enumeration areas are drawn from a census file and from each of these enumeration areas a sample of households is drawn [23]. Across all three surveys—used in this study—women who reported to have never been in union, divorced, or separated were excluded from analysis and this resulted in sample sizes of 8189, 6786, and 6310 from the 2004–05, 2010 and 2015–16 TDHS, respectively.

Evidence of relevance from previous studies guided the selection of suitable variables from within the TDHS [2,3,4,5, 18, 19]. All levels of categorical variables were coded according to the direction of their influence on empowerment such that categories that reflected higher levels of empowerment had a higher ranking and those categories that were indicative of disempowerment had a lower ranking [18, 19]. The variables used in this study and how they were coded are described below.

The response for beating justified if the wife goes out without telling the husband/partner, beating justified if wife neglects the children, beating justified if wife burns the food, beating justified if wife argues with husband/partner, beating justified if the wife refuses to have sex with husband/partner were coded 0 for don’t know, 1 for yes, and 2 for no.

The response for the person who usually decides how to spend the woman’s earnings, the person who usually decides on the woman’s healthcare, the person who usually decides on large household purchases, the person who usually decides on what to do with husband/partner’s earnings, the person who usually decides on visits to family or relatives were coded 1 for husband/partner alone, 2 for woman alone, 3 for both woman and husband/partner.

The response for a woman’s educational attainment was coded 1 for no education, 2 for primary education, 3 for secondary education, and 4 for higher education.

The response for educational difference, which was measured in years, between a woman and their husband/partner was coded 1 for partner has more years of education, 2 for partner has the same years of education, and 3 for woman has more years of education.

The response for if the woman worked in the past year was coded 1 for no, 2 for worked in the year before, and 3 for currently working or on leave while those for earning from the woman’s work were coded 1 for payment in kind, 2 for cash and in kind, and 3 for cash only.

The response for earning ratio between a woman and her husband/partner was coded 1 for earning less than the husband/partner, 2 for about the same earning, and 3 for earning more than the husband/partner.

The responses for owning land alone or jointly with husband/partner and owning a house alone or jointly with husband/partner were coded 1 for does not own, 2 for lone ownership, 3 for joint ownership, and 4 for lone and joint ownership.

The responses for getting permission to go to health facility, getting money needed for treatment at health facility, distance to the health facility, and not wanting to go alone to the health facility were coded 1 for being a big problem and 2 for not being a big problem.

The variables Age at first birth, Age at first cohabitation, Frequency of watching television, and Frequency of reading newspaper or magazines were used as they were originally collected by the DHS.

Exploratory factor analysis (EFA) using variables presented in Table 1 was conducted to obtain the domains that represent women’s empowerment. In order to obtain meaningful domains, the suitability of the variables for exploratory factor analysis was tested using the Kaiser–Meyer–Olkin (KMO) test of sampling adequacy [24] where values greater than 0.70 were considered adequate. Domain retention was considered for those domains that together explained more than 50% of the total variance and variables were excluded if they had a loading of less than 30% or if they loaded on more than one domain [25]. Cronbach’s α was used to test the overall internal reliability of the index and that of individual domains [26, 27]. Here too, domains that had a Cronbach’s α value of less than 50% and variables that had a correlation that was not similar to the rest of the variables in the index were eliminated [28,29,30,31]. The final domain structure—the empowerment index—was obtained after performing an oblique Promax rotation.

The construct validity of these domains was checked through confirmatory factor analysis (CFA) [32] using the following fit statistics; the Root Mean Squared Error of approximation (RMSEA) and its respective 90% confidence intervals which represent parsimony of an index; the Comparative Fit Index (CFI) and Standardized Root Mean Squared Residual (SRMR) which represent relative and absolute fit of the index [31, 33]. The target for an index with good construct validity is a RMSEA with a value of less than 0.05, and where CFI and RMSR are close to 1 and 0 respectively [32].

A random half of the original data was obtained using the STATA command “random”, with respect to DHS clusters. All factor analyses were performed on one split-half sample and the other half was used to develop the index. All statistical analyses were performed using STATA version 14.2.