Key independent variable.

Rural Women Composite Empowerment Index (RWCEI). In order to measure and capture a multidimensional picture of women empowerment in rural sectors of Pakistan, we developed the “Rural Women Composite Empowerment Index (RWCEI).” It is based on the WEAI developed by Alkire et al. and adapted by Nuzhat Ahmad & Huma Khan [21, 23] but includes a wider range of empowerment dimensions such economic empowerment, autonomy, decision making, qualification, time allocation, mobility empowerment, political empowerment, awareness and violence. The RWCEI therefore measures women’s empowerment through nine domains, fourteen sub-domains and eighty-nine indicators (S2 Table).

Many assessments of women’s empowerment have relied on a power typology founded in the seminal writings of Freire (1968) on autonomy and Lukes (1974) on power, and expressed with regard to gender and women’s empowerment by Rowlands (1995, 1997) [26–29]. This typology contrasts dominating or exerting "power over" others with creative types of empowerments such as "power within" (encompassing self-respect, identity, and a knowledge of rights), "power to" (implementing individual goals), and "power with" (acting collectively toward shared interests). Different studies also categorize types of empowerments in different dimensions of power [30–33].

In regard to power, empowerment can be framed in both positive and negative ways (as per Rowlands 1997). In a positive sense, agency occurs when people recognise their own worth and the purpose they bring to their acts (individual’s internal agency, or "power within") and are able to act to achieve their objectives (instrumental agency, or "power to"), even when others or social norms oppose them. In a negative sense, it refers to actors exceeding others’ agency and exerting control or "power over" their lives and resources [34]. Empowerment, then, is about changes in these various forms of power, which interact and reinforce one other to produce unequal outcomes.

With regard to the Rowlands (1997), Malapit et al. (2019), Meinzen-Dick et al. (2019), and Elias et al. (2021), theory of power, it is significant to substitute nine domains of RWCEI (in addition to those in WEAI) under the classifications of power [29–32]. The theory categorizes power (empowerment) into four dimensions, and every dimension is important while constructing the index. Rowlands (1997) states that empowerment is a power and power may be manifested in four ways: power within, power to, power with and power over. Power within relates to self-confidence and personal strength to make decisions. In this dimension, qualification and awareness domains are included. By getting an education and awareness a woman can gain self-confidence and can make better decisions in her life [30, 31, 33]. Power to (instrumental agency) means individual agency, and the capability to choose and carry out activities. So, in this category, domains of economic empowerment, autonomy, and decision-making lies. Moreover, the domain of time allocation and mobility empowerment also fits under this category. Time allocation domain is about woman having enough personal time for herself, and it is possible only if work is divided among family members [30–33]. Mobility empowerment means that women can move freely within and outside her settlement. It simply reflects the power to use one’s time for different activities and the power to go to different (important) places. Women’s capacity for decision making is enhanced if she is financially empowered and earning money. Moreover, autonomy is also important for decisions regarding spending money to buy her own personal things or not. Similarly, capability or capacity for performing actions is also reflected in whether she has power to make decisions about agricultural activities, household decisions and financial decisions [30–33].

Power with (power to participate in collective process) perceives that empowerment is a collective process that involves the support and interaction of families, societies, organizations, and communities. The domain of political empowerment fits under this category. Politically empowerment means that is she can vote and participate in political or other organisations and is not dependent on whether her family allows her to do so or not. Power over measures the degree of strength in relationships between woman and other members of family or community. Therefore, for this category violence domain is included. Violence refers to whether the woman is facing domestic violence in her family or not. If relationship with her family members is strong, then she is considered empowered in this category of power [30–33].

RWCEI was formulated in four steps. Firstly, we made binary responses of all indicators by assigning a value ‘1’ if woman had achievement/ adequacy in that indicator; otherwise ‘0’. The transformation of the variables to a binary format was necessary to create an “empowerment” variable. This was needed prior to the factor analysis.

Secondly, we summed all these binary scores under each sub-domain; a cut-off 30% was used to identify woman’s empowerment in each sub-domain. We made the judgement that if women had 30% adequacy in their responses, then she was considered empowered in that sub-domain. As there are no set standards/guidelines for deciding on such a threshold for a given index [35] we made this judgement based on a number of factors. Measurement or discussion of women’s empowerment often ignores the fact that women are tied together by strong emotional and structural bonds and their interests are strongly vested in their families which influences individual autonomy. Previous studies focus on individual empowerment irrespective of social status of the family but in Pakistan women rarely make any independent decisions. Joint family system is the bedrock of Pakistani societies [36]. Therefore, for devising the cut-offs for index we focused on norms of society, and as such a 30% cut-off level is more realistic. In absence of this support structure, the threshold may be higher. This does mean that our index has a relatively low threshold for women to be classified as ‘empowered’, but it is a starting point for measuring these important domains.

Thirdly, the same procedure was followed to identify woman empowered in each domain. Lastly, women’s empowerment index was calculated by using factor analysis with tetrachoric correlations. Weights were assigned to each domain based on their factor loadings and index was generated in STATA. A factor analysis of a matrix of tetrachoric correlations is more appropriate when input variables are dichotomous [37, 38]. Studies investigated that factor analysis is a useful technique for the construction of index [39].

Outcome variables.

a) Household food security. To construct the household food security (FS_h) variable, calorie intake per capita was calculated for each selected household in each region through the dietary intake assessment (DIA) approach. Quantities of food items were converted into grams from kg’s or litres and then into calories with the guidance of the food consumption table (FCT) for Pakistan [25]. FCT provides the calories amount of every food item in 100 grams of edible portion. Therefore, for each food item, calories per gram were computed. The data of food items was based on the 7-days recall period; therefore, household daily calorie intake was obtained by dividing the weekly data by seven and converted into a single unit i.e., kcal per day to obtain household daily calorie availability. Calculated calories were then converted into per adult equivalent (AE) by adjusting according to gender, age, and physical condition (pregnancy and lactation) of household members by using caloric adjustment table and expressed in adult equivalent units. Energy requirement depends on the sex and age of the person and therefore it varies across household members. To account for these changes adult equivalence scale was used in the development of this variable. Details about calculating calorie availability per adult equivalent (AE) are given in S1 Appendix.

Total calorie availability per day per adult equivalent TC (tkcal)_h at the household level “h” was measured by dividing the total calories of every household from the AE size of the respective household “h”.

(1)

To estimate the household’s food insecurity status, TC (tkcal)_h was compared with minimum dietary energy requirement (MDER). MDER was the threshold level defined by FAO i.e., 1745 kcal per capita per day. It is the minimum amount of calorie requirement that is necessary to meet the adequate energy needs of a person. It was assumed that when the household was not capable enough to meet the MDER, it was unable to sustain its health and therefore it lied in the food insecurity group.

(2)

Where; FS_h is the food security status of h^th household, TC (tkcal)_h is total calorie intake of h^th household per day per adult equivalent and MDER is recommended threshold level of food security for the Pakistani population. A household unit was declared food secure if FS_h was greater than and equal to “0”. Furthermore, FS_h converted in to three categorical variables i.e., food insecurity, mild food security and high food security.

b) Nutrition security. Healthy food and good quality of diet are usually associated with a diversified diet and adequate amount of nutrient intake. Different indices were developed in the past to measure dietary diversity and quality e.g. Healthy Eating Index [40] and Chinese Diet Quality Index [41].

To construct the nutrition security variable, we created household dietary diversity scores (HDDS) and household dietary quality scores (HDQS). They serve as qualitative measures that exhibit household’s diversity among food items and food groups [42] and are simple tools that exhibit the dietary quality of households [43].

Household Dietary Diversity Score (HDDS).

Food consumption score (FCS) developed by the World Food Program (WFP) to measure the HDDS [44] was used in the present study. It measures the scores by assigning weights to food groups consumed by a household during the seven days. Eight food groups were deduced from twelve food groups, used by the present study on the grounds of nutritional attributes and proposed by FAO to measure the nutritional security at the household level [42]. There were standard weights designed for each food group that constituted the food consumption score and a continuous variable was generated with a possible range of 0 to 112. The food consumption groups include starches, pulses, vegetables, fruit, meat, dairy, fats, and sugar. The formula, based on these groups, with the standard weights, is: (3)

The reason for assigning unequal weights was that every food group did not have equal importance in calculating the total scores because of different nutritional attributes. For example, fats and sugar have less significance in a healthy diet as compared to milk, meat, and pulses. Therefore, accounting for these differences was crucial in estimating the true picture.

Household Dietary Quality Score (HDQS).

Household Dietary Quality Score (HDQS) is considered as an essential measure when assessing food and nutrition security. Different studies argue that people are consuming enough calories but still their diet is missing in essential nutrients. For example, if they are only taking cereals in their diets and do not include fruits and vegetables, milk and milk products, so their diet remains deficient in essential micro and macronutrients [6, 7]. Therefore, the required amount of nutrients are essential for a healthy and active human body. To measure whether a household is consuming the right amount of nutrients the study estimated macronutrients (proteins) and micronutrients including vitamins (A, B, and C) and minerals (calcium, iron, iodine, and zinc) in their diet plans, and utilised the Pakistan dietary guidelines with FAO standards for daily recommended allowances of these nutrients according to different age groups [45]. Each type of nutrient was converted into standard milligrams equivalents according to the food composition table (FCT) for Pakistan [25]. The daily intake of all household members was summed up and compared with the recommended daily nutrient intake (RNI) by age, sex, and physical condition (pregnancy and lactation) of household members.

Those households who fell below the recommended threshold level were considered nutrient-deficient households as they are consuming poor quality of diet. And those households who score above the threshold level were considered nutrient-rich households, as it was likely they were meeting their nutrient requirements and consuming good quality of diet.

(4)

(5)

HDQS is the score of households ‘h’, and it equals one when its score is equal to or greater than recommended nutrient Intake (RNI_h). After creating the nine macro and micronutrient variables, we then developed a household dietary quality index through principal component analysis (PCA) and used this index as a dependent variable in the analysis [46]. PCA is used as it provides a better approximation of the required factors [47]. It uses Pearson correlations for identifying important factors that contains most of the information [48–50]. It is also most appropriate technique for the construction of indices [39]. Different studies also used PCA for construction of FNS indices in history [50, 51].

Modelling.

Multi-level mixed-effect logistic regression methodology. Different control variables were used in the analysis that are crucial to incorporate according to previous studies. For example, household size and its occupation [4, 52–55], unemployment to employment ratio [56], employment status of household head [57], expenditures on daily food items [58], household wealth index [21, 59], educated and health communities [58, 60, 61] etc. Description and measurement of these control variables are given in S3 Table. As described previously, we had three dependent variables in the analysis and the type (numerical, categorical etc.) of all variables was different. We used ordered logistic, linear, and binary logistic regression for FS, HDDS, and HDQS respectively. To incorporate the hierarchical composition of the data set (i.e. individuals were nested in households and households were nested in the 19 communities) multi-level modelling was adopted as suggested by Kamanda, Madise, and Schnepf [60]. For this purpose, we added the role of health facilities as a proxy for identifying food insecure households in the community and role of educated women in the community. A community with educated women was included, because a higher level of women education is associated with better health outcomes and wellbeing of their families [57, 60].

Three models were used for comparison. Model 1 is the food security (FS) model and it focused on calorie intake by the households. Model 2 is the dietary diversity model (HDDS) and it focused on a number of food groups consumed by the households. Model 3 referred to the dietary quality (HDQS) and it described nutrient intake by the households.

The two-level model is written as follows: (6)

Where π_ik is the probability of household FNS for the i_th household in the k_th community. RWCEI is the key independent variable and X_ik is the vector of covariates corresponding to the i_th household in the k_th community, β_o is the vector of parameters and μ_ok is the random effect at the community level. The intercept (average probability) of being food and nutrition secure was assumed to vary randomly across communities. The fixed effects (measures of association) were expressed in odds ratios (OR) and 95% confidence interval (95% CI). The random effects (measures of variation) are expressed as variance coefficients. We used the post estimation command to determine the p-value of variance.