Introduction

Women’s empowerment is a multidimensional and multilevel concept that encompasses claims on new human, social, and economic resources. These claims potentiate transformations in individual, interpersonal, and collective agency, or the ability to imagine, act on, and realize personal and shared aspirations [1]. Thus, women’s empowerment is a process of actualization—of oneself and of collectives—that arises in an evolving structural and normative context characterized by historical constraints and expanding choices. The process of empowerment is multidirectional, involving reciprocal influences of investments in resources, changes in agency, and realizing aspirations over time (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Women’s multidimensional, multilevel empowerment in social context.

https://doi.org/10.1371/journal.pone.0345742.g001

Resource investments in women are central to global health and sustainable development. Food-system interventions with women’s empowerment components have had favorable effects on measures of food security, affordability, availability, diet quality and adequacy, and BMI for age [2]. Women’s engagement in self-help groups, an investment in social capital, has had positive effects on their economic empowerment, mobility, and political empowerment; mixed or null effects on their interpersonal agency (family-size decision making); and some effects on their individual agency (e.g., self-confidence and gender-equitable attitudes) [3,4]. While not ensuring positive outcomes, economic transfers to women have improved children’s well-being, health, and education [5], and economic interventions with ‘gender-sensitivity’ social-norms components may reduce partner violence [6,7] and increase women’s antenatal visits [8]. The effects of women’s empowerment interventions on multidimensional measures of their agency show promise [9,10]; however, synthesizing this evidence is limited by the heterogeneous conceptualization and measurement of women’s agency [11–13].

Approaches to measuring women’s empowerment have strengths and challenges. First, long-form measures of women’s empowerment capture more dimensions with more questions, potentially offering a fuller picture of women’s lives [14]. However, lengthy surveys increase respondent burden [15], potentially compromising data quality. Moreover, enumerators may lack the skills to clarify many nuanced questions on women’s empowerment, resulting in redundant responses to questions intended to capture different constructs [16]. These challenges may make long-form women’s empowerment surveys burdensome, costly, and of variable quality for routine data collection [15]. Second, abbreviated surveys of women’s empowerment have been sector-specific [17] or focused on selected dimensions of women’s empowerment [18], which provide an incomplete picture of women’s intersectoral, multifaceted lives [19]. Finally, women’s empowerment surveys often focus on sub-populations, such as adolescents [20] or women of reproductive age [21]. This approach produces measures of women’s empowerment tailored to specific periods of the life course but precludes tracking change in women’s empowerment across the life course and comparing age-specific and overall estimates of women’s empowerment for adolescent and adult populations across countries.

As a result, only 42% of the data needed to monitor Sustainable Development Goal (SDG) 5 and other gender-related SDGs is routinely produced [22]. At this pace, over two decades would be needed to fill gender-related data gaps. A concise generalized measure is needed for routine administration by national statistical offices (NSOs) and global survey platforms [23] that covers concepts of empowerment aligned with Sustainable Development Goal (SDG) 5 to “achieve gender equality and empower all women and girls,” and other gender-specific SDGs, such as SDG 1 (no poverty) and SDG 10 (reduced inequalities).

Relevant survey platforms include the Demographic and Health Surveys (DHS), United Nations Children’s Fund Multiple Indicator Cluster Surveys (UNICEF MICS), World Values Surveys (WVS), World Bank Living Standards Measurement Study (WB LSMS), World Bank Living Standards Measurement Study–Integrated Surveys on Agriculture (WB LSMS-ISA), and the 50x2030 Initiative to Close the Agricultural Data Gap (50x2030 Initiative) (Table 1). These platforms differ in their analytical objectives, and as such, their sampling designs and interview targets also differ. A concise empowerment module needs to consider these differences across survey platforms and the unique benefits of integrating a common module into each platform (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Benefits of integrating WEMNS into global survey platforms.

https://doi.org/10.1371/journal.pone.0345742.t001

To meet this need, our global partnership developed the Women’s Empowerment Metric for National Statistical Systems (WEMNS) survey module. We worked with the 50x2030 Initiative to develop, test, and refine WEMNS as a concise, theoretically informed, psychometrically valid, and scalable module that NSOs and survey organizations in LMICs could adopt wholly or in part to measure women’s and men’s empowerment in national and sub-national household surveys. Users then could calculate specific WEMNS indicators and a counting-based WEMNS index [24,25]. Here, we report findings from Phase III of this project, contributing evidence on the feasibility, acceptability, and suitability of administering a concise, conceptually comprehensive face-to-face survey module to compare empowerment measures for adult women and men across three LMIC settings.

Materials and methods

Development of the WEMNS module

Development of the WEMNS module was implemented in three phases (Fig 2). Phases I and II are summarized in Fig 2 and are described briefly here, as these phases are described in detail elsewhere [26]. In Phase I, we identified theoretically relevant constructs and aligned them with gender-related SDG targets and indicators. We reviewed existing women’s empowerment questionnaires to operationalize identified constructs. We conducted key-informant interviews with 59 experts from governmental, civil-society, academic, and bilateral or multilateral organizations across Africa, Asia, and Latin America. Finally, we conducted two virtual, group-based stakeholder consultations [19]. All of these steps informed the development and refinement of an initial WEMNS module.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Three-phase approach to WEMNS module development and testing.

https://doi.org/10.1371/journal.pone.0345742.g002

Phase II involved phone-based cognitive interviews (N = 128) and pilot surveys (N = 100 in Bangladesh, N = 82 in Malawi, N = 50 in Nepal) to refine further the initial WEMNS module. An intermediate WEMNS module then was implemented in a single phone survey in Guatemala (N = 2,031) and two-wave phone surveys in Bangladesh (N = 1807), Malawi (N = 1,657), and Nepal (N = 2,461). Psychometric assessment of data from the two-wave phone surveys (N = 5,455) revealed that WEMNS item sets captured common constructs of empowerment; however, few item sets achieved scalar invariance, indicating that the remaining item sets may not measure their underlying empowerment construct in equivalent ways across genders, country settings, or time. All of the steps undertaken in Phases I and II informed revisions to the WEMNS module that was implemented in Phase III. Phase III involved cross-sectional face-to-face surveys in the same country settings plus Guatemala and a psychometric assessment of those data. Guatemala was excluded from the psychometric assessment because it was not part of the original, three-country protocol, and was implemented later. Phase III is detailed here.

Phase III. Survey and psychometric assessment of WEMNS module

Study sites and samples.

In all three country settings, a multi-stage probability sample design was implemented to identify households with at least one woman and one man 18–64 years. A target sample size of approximately 800 households per country was adequate for psychometric analysis [27] and indicator development [28]. Within each household, one eligible woman and one eligible man were randomly selected from the household roster. The sampling design differed across country settings because sampling frames differed (S1 File). These variations are of less concern in psychometric assessments since arriving at nationally representative estimates was not the analytical aim. Sample recruitment and data collection took place in 2022 from November 2 to November 27 in Bangladesh, from October 11 to November 17 in Malawi, and from December 4 to December 22 in Nepal.

Piloted WEMNS module.

Table 2 presents the dimensions in the face-to-face WEMNS module and psychometrically assessed here, number of items and response options that operationalize each dimension, and SDG targets/indicators and theoretical constructs of empowerment that each WEMNS dimension was designed to capture. These item sets are updated from those administered in the Phase II phone-survey [26] and align more closely with those selected for the Alkire-Foster-based index [24].

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. WEMNS dimensions,^a Gender-related SDGs, and theoretical domains of empowerment.

https://doi.org/10.1371/journal.pone.0345742.t002

The WEMNS module for face-to-face surveys included item sets designed to measure 13 latent variables of empowerment that conceptually captured claims on critical resources and major domains of agency (Table 2). Economic resources was captured with two item sets measuring access to and use of financial resources (8 items) and access to credit (5 items). Technological resources was captured with one item set measuring the use of information and communication technologies (ICT, 6 items). Intrinsic economic, reproductive, bodily, and time-use agency was captured with four item sets measuring the endorsement of women’s freedom to choose livelihoods (5 items), endorsement of women’s freedom to choose family formation (4 items), rejection of women’s subjection to sexual harassment (5 items), and rejection of gender inequality in time spent on paid and unpaid activities (4 items). Instrumental time-use and economic agency was captured with two item sets measuring the extent of influence over own time

allocation (6 items) and decision-making and control over income (4 items). Collective agency in organizations was captured with two item sets measuring confidence to participate in community organizations (5 items) and participation and leadership in community organizations (5 items). Collective agency in communities was captured with two item sets measuring perceptions of women’s and men’s community engagement, each with 9 items. Two measures—ownership and tenure security of land and other property and ownership of a mobile phone—were included in the WEMNS module but were not psychometrically assessed, as they were considered observed (non-latent) variables.

Validation module.

A validation module also was developed to assess the construct validity of final, maximum a posteriori factor scores [29] for all 13 dimensions of women’s empowerment that were derived from final models (see Analysis, below). The questions included in this validation module were extracted from the World Values Survey and were designed to measure basic needs, resources, agency, and subjective well-being. Basic needs was measured as the self-reported frequency of going without enough food in the last 12 months (often = 1; never = 4). Resources was measured as the self-reported frequency of obtaining information from a mobile phone, and separately, the internet (never = 1; daily = 5). Intrinsic agency was assessed as the self-reported extent of justifying divorce, and separately, not justifying wife beating (never = 1; always = 4). Instrumental agency was assessed as the self-reported extent of: choice and control over one’s life (none = 1; great deal = 4), input into decisions about household income and spending (none = 1; most = 3), and making one’s own decisions about household income and spending (none = 1; high = 4). Subjective well-being was measured as self-rated health (very poor = 1; very good = 5), happiness (not at all = 1; very happy = 4), life satisfaction (completely dissatisfied = 1; completely satisfied = 4), and satisfaction with household financial situation (completely dissatisfied = 1; completed satisfied = 4).

Survey platform.

Team members developed a customized data system using World Bank Survey Solutions software for computer-assisted personal interviews (CAPI) (https://mysurvey.solutions). Interviewers administered a household-level module to the primary respondent, preferably the household head. Survey Solutions then was programmed to select randomly one woman and one man 18–64 years from the roster for face-to-face administration of the WEMNS module and a validation module (S1 File).

Analysis.

All analyses were conducted using STATA Version 17 (StataCorp, College Station, Texas, USA) and MPlus version 8.6 (Muthén & Muthén, Los Angeles, California, USA). In step 1, we performed univariate analysis to assess the distributions and item-level missingness of demographic variables and all items in the WEMNS module that were included in the psychometric analysis. We also estimated tetrachoric and polychoric correlations between all binary or ordinal WEMNS items, within and between theoretical empowerment constructs.

In step 2, we performed exploratory factor analysis (EFA) in a random split-half sample of the pooled (three-country) dataset (n₁ = 2,401). We analyzed each WEMNS item set separately (Table 2), resulting in 13 EFAs. Our decision to examine each item set separately during this developmental stage was guided by analytical goals and practical empirical considerations based on initial EFAs with all 75 items (S1 File). Specifically, we aimed for a nuanced individual analysis and validation of all 13 dimensions of women’s empowerment. This approach made the various analyses and interpretations more feasible and ensured that the results were more interpretable, accessible, and useful for researchers, policymakers, and stakeholders. We performed the separate EFAs with means and variance adjusted weighted least squares estimators, which is suitable for binary and ordinal variables [30]. EFA models were fit using geomin (an oblique) rotation to simplify interpretation of the factor structure [27]. Criteria to remove an item from an EFA included low factor loadings (<|0.350|), significant negative factor loadings, and theory (e.g., the remaining items could adequately reflect the construct) [31].

In step 3, for each item set, we performed confirmatory factor analysis (CFA) on the next random split-half sample (n₂ = 2,402) to test the factor structure identified in the EFA. For all EFA and CFA models, we interpreted overall model fit as ‘good’ based on the following indices and thresholds: upper limit of the confidence interval for the root mean squared error of approximation (RMSEA) equal to.07 or.08, comparative fit index (CFI) close to 0.95 or higher, and Tucker–Lewis index (TLI) close to 0.95 or higher [32].

In step 4, for each empowerment factor identified in the EFA and confirmed in the CFA, we conducted two separate multi-group CFAs (MGCFAs) using the CFA sample (N = 2,402) to test for measurement invariance, one MGCFA for genders and one for country settings. In MGCFA, nested CFA models are estimated, in which equality constraints on specific model parameters are added sequentially to test whether the models with and without those constraints are equivalent across groups [33,34]. In the MGCFA models, one loading per latent variable is fixed to 1 to identify the latent variable and establish its scale for cross-group comparisons. In a baseline model, configural invariance is observed when the same factor structure (e.g., the same pattern of factor-item relationships) holds [34], providing evidence that the factor structure is invariant and that a similar construct of empowerment is observed across groups [35]. For the empowerment factor with ordinal responses, metric invariance is tested by constraining all factor loadings (relating items to the underlying construct) to be equal across groups [34,35]. Scalar invariance is tested by constraining all factor loadings and item thresholds to be equal across groups [34]. In exact invariance testing, scalar invariance is required to validly compare and to interpret differences in scores for a latent empowerment construct across groups [35]. For the dimensions of access to/use of financial services and access to credit with binary indicators, we tested configural versus scalar invariance directly, bypassing metric invariance. For the dimension of use of ICT, initial MGCFA models by country settings failed to converge, possibly due to skewed distributions of the responses, so we dichotomized these items and similarly bypassed metric invariance. This decision was due to the use of the weighted least squares mean and variance adjusted (WLSMV) method, which does not permit testing metric invariance with scale factors or residual variances allowed to vary across groups [29].

The first MGCFA series assessed measurement invariance of each item set across genders, and the second across country settings (Bangladesh, Malawi, Nepal). In each case, we assessed the fit of the configural model using the same criteria, as described for the EFAs and CFAs. To assess fit for the nested metric and scalar models, we considered changes in model-fit statistics (chi-square difference test Δχ², ∆CFI, ∆RMSEA) with sequentially added equality constraints on model parameters [33,36,37]. We inferred whether each item set was invariant based on a non-significant chi-square difference at p = 0.05 [34], and guidelines for changes in other model fit statistics with increasing equality constraints (∆CFI ≤ 0.002; ∆RMSEA ≤0.007) [36,37].

In step 5, for item sets in which metric or scalar invariance across groups was not observed, we reviewed results from the MGCFAs, including the modification indices for loadings and thresholds, to assess whether MGCFA with partial invariance testing was neccesary. This review allowed us to identify specific model parameters that were non-invariant across groups and to assess invariance among the remaining parameters [33]. For measures demonstrating at least configural invariance but not scalar invariance, we tested partial scalar invariance only for those measures with few modification indices >10 (the default value in Mplus 8 [28]) for the current scales with a small number of items, under the argument that small-scale noninvariance likely has a negligible impact on the estimation of construct-related parameters [38]. In those cases, we released equality constraints across groups one parameter at a time, starting with the parameter having the largest modification index, indicating the expected improvement in the overall model χ² if the equality constraint for the parameter was released [33]. We then tested the partial invariance of models across groups, based on the Δχ², ∆CFI and ∆RMSEA.

In step six, we assessed the construct validity of all 13 WEMNS item sets in the pooled sample. We estimated Spearman pairwise associations between the weighted factor score for each WEMNS item set and variables representing basic needs, resources, agency, and subjective well-being from the World Values Survey (see Validation Module, above). Factor scores were calculated using the factor loadings from the final CFA models. Expected bivariate relationships are summarized in the supplemental information (S1 File). Conventions to interpret correlation coefficients were used (<0.30 = weak; ≥ 0.30 and <0.50 = moderate; ≥ 0.50 = large [39]. Finally, we calculated Spearman pairwise correlations between scored item sets to confirm that they were distinct and non-redundant.

Results

Step 1. Sample characteristics and distributions of WEMNS items

Across settings, participation rates were 85.5%−99.0% for the sample of women and 85.5%−99.8% for the sample of men (Table 3). The mean age was 31.9–35.6 years for women and 33.5–38.5 years for men. The percentages with no schooling were 4.3%−40.1% among women and 3.9%−23.1% among men.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. WEMNS face-to-face survey participant characteristics, by gender and country setting.

https://doi.org/10.1371/journal.pone.0345742.t003

Table 4 provides the distributions of each of the 75 WEMNS items included in the psychometric analysis. Missingness was extremely rare (<1% of observations) or rare (<7%) for most items (67 of 75), and responses were treated, conditional on observed variables, to be missing at random in the analysis. A majority of the sample reported that they did not own or have a mobile money account, bank account, bank card, or credit card (58% to 99%), and similar percentages of the sample reported not to have used any of these financial services in the prior 12 months (62% to 98%). A majority of participants reported that they could take out a loan from any of the financial institutions mentioned (66% to 74%), except from an ‘other’ NGO program (43%). Use of ICT varied widely by type, with 97% of the sample reporting never use of a landline and 75% reporting daily use of a mobile phone. A majority of participants “fully disagreed” with most items measuring women’s freedom to choose their own livelihoods (54% to 89% across all items) and freedom to choose family formation (59% to 68% across three of four items). In addition, around half or more participants reported that various forms of sexual harassment against women were “usually acceptable” (48% to 67% across all items). Similarly, a majority of participants “fully agreed” with statements acknowledging the existence of gender inequality (62% to 66% across all items). That said, two-fifths to a majority of the sample reported having “a lot of influence” over their own time allocation (42% to 66% across all items). About half or more participants reported having “some” or “a lot” of influence over financial or personal healthcare decisions (48% to 75% across all items). While almost half to a majority of participants reported being “very confident” to participate in community organizations (45% to 71% across all items), a majority did not participate in community organizations (55% to 91% across all items). In general, higher percentages of the sample “fully agreed” with statements regarding men’s community engagement than with statements regarding women’s community engagement (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Distributions of WEMNS, by dimension of the WEMNS module, pooled sample (N = 4,803).

https://doi.org/10.1371/journal.pone.0345742.t004

Steps 2–3. Factor analyses in pooled sample

For a majority of the 13 item sets intended to reflect single dimensions of empowerment, model fit statistics were adequate, with the upper bound of the confidence interval for the RMSEA being ≈0.08 or lower and the CFI and TLI being ≈0.95 or higher. Moreover, the item-specific factor loadings in all cases except two equaled or exceeded 0.35 (Table 5). Item sets for intrinsic economic, reproductive, and time-use agency and for intrinsic collective agency in organizations were the only ones for which one or more model fit statistic in the EFA and/or CFA were not close to the recommended thresholds.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Results from Exploratory Factor Analyses (EFA) and Confirmatory Factor Analyses (CFA) for each WEMNS item set administered in face-to-face surveys, random split-half samples from all participants across Bangladesh, Malawi, and Nepal.

https://doi.org/10.1371/journal.pone.0345742.t005

Step 4. Measurement invariance testing across gender and country setting

In MGCFA of the same 13 item sets, configural invariance (Yes or Close to) across gender was observed for nine item sets, such that all instrumental, and collective agency constructs appear to have been understood similarly across gender in this sample (Table 6). For all three measures of resources—access to and use of financial services, access to credit, and use of ICT—configural invariance was not observed. Similarly, for two measures of intrinsic agency—endorsement of women’s freedom to choose livelihoods and endorsement of women’s freedom to choose family formation—configural invariance was not observed. Metric invariance was observed for seven measures—two for intrinsic agency, one for instrumental agency, and four for collective agency. Scalar invariance was observed for only one measure, rejection of women’s subjection to sexual harassment (Table 6).

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Results of measurement invariance testing across gender for each WEMNS item set administered in the face-to-face survey, CFA sample of participants from Bangladesh, Malawi, and Nepal (n₂ = 2402).

https://doi.org/10.1371/journal.pone.0345742.t006

In MGCFA to assess measurement invariance across country settings, configural invariance (Yes or Close to) was observed for 9 of the 10 item sets for which MGCFA models converged (Table 7), suggesting that most WEMNS constructs for resources and agency appear to be understood in similar ways across country settings in this sample. As noted above, response options for the item set on use of ICT were dichotomized for the MGCFA across country settings, and metric invariance was not assessed. Metric invariance was observed for four item sets—endorsement of women’s freedom to choose livelihoods, rejection of gender inequality in time allocated to paid/unpaid activities, extent of influence over own time allocation, and decision-making control over income. Scalar invariance was not observed for any item set.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Results of measurement invariance testing across country setting for each WEMNS item set administered in face-to-face surveys, CFA sample of participants from Bangladesh, Malawi, and Nepal (n₂ = 2402).

https://doi.org/10.1371/journal.pone.0345742.t007

Step 5. Partial invariance testing across gender and country setting

To consider testing for partial scalar invariance across gender, seven measures were eligible, as the results of MGCFA across gender indicated at least configural but not scalar invariance (Table 6). For six of these measures, the presence of many large modification indices suggested many potential sources of mis-fit; therefore, we determined that performing partial invariance testing was not appropriate for these measures. We assessed partial scalar invariance for one measure, decision-making and control over income; however, the data did not support partial invariance (results available upon request). Given that the data did not support scalar invariance of any of the WEMNS item sets across country settings, and in view of the large changes in CFI and RMSEA indicating pervasive structural differences (Table 7), we determined that it was not appropriate to pursue partial invariance testing across country settings.

Table 8 summarizes the results of invariance testing of WEMNS item sets.

Download:

• PNG
  larger image
• TIFF
  original image

Table 8. Results of partial invariance testing of WEMNS item sets.

https://doi.org/10.1371/journal.pone.0345742.t008

Step 6. Construct validity of derived factor scores for WEMNS item sets

Overall, Spearman correlations among the weighted factor scores for each item set were weak (Table 9): 58 (87.9%) of 66 correlation coefficients were <0.30, seven (10.6%) were ≥0.30 and <0.50, and one (1.5%) was ≥ 0.50, with the maximum correlation being 0.52 between endorsement of women’s freedom to choose her preferences in livelihoods and to choose her preferences regarding family formation. In other words, all item sets were distinct and most were weakly correlated, further justifying the decision to analyze them separately.

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. Spearman pairwise correlations between WEMNS factor scores derived from CFA models.

https://doi.org/10.1371/journal.pone.0345742.t009

Spearman correlations between the weighted factor score for each WEMNS item set and variables for basic needs, resources, agency, and subjective well-being also were weak (Table 10): 139 (96.5%) of 144 correlation coefficients were <0.30, four (2.8%) were ≥0.30 and <0.50, and one (0.01%) was ≥ 0.50. The five moderate-to-high correlations conceptually aligned. Two were between the score for decision-making and control over income and variables for input and decision-making over income. Another two were between the score for access to/ use of ICT and variables for the frequency of obtaining information from mobile phones or the internet. The fifth was between the score for use of financial services (including mobile money) and the variable for the frequency of obtaining information from the internet.

Download:

• PNG
  larger image
• TIFF
  original image

Table 10. Spearman’s pairwise correlations of WEMNS factor scores derived from CFA models and constructs from the world values survey.

https://doi.org/10.1371/journal.pone.0345742.t010

Discussion

This paper presents findings from the psychometric assessment of the WEMNS module, a concise yet comprehensive survey module measuring domains of women’s empowerment that align with theoretical constructs of resources and agency as well as key gender-related SDG targets and indicators. No such module exists for use across NSOs and global survey platforms to compare national changes in a common measure of women’s empowerment. Such a module would fill a major global gap in national-level data on women’s empowerment [22] and would enable researchers to harmonize analyses of the relationships of women’s empowerment with other SDGs collected by different survey platforms (Table 1). Such data would provide the evidence needed for governments and donors to better target investments to advance women’s empowerment for the benefit of women, families, and societies.

Results presented here suggest cautious optimism regarding the WEMNS module’s capacity to fill critical data gaps. In the pooled sample, most WEMNS item sets had clear factor structures that aligned with theoretical dimensions of empowerment and SDG targets and indicators. Configural invariance (e.g., items in a set load on a single latent variable in the same pattern across groups) was observed for nine of 13 item sets across gender and nine of 13 item sets across country setting. Thus, most WEMNS item sets capture common constructs of empowerment across gender and diverse country settings. However, scalar invariance was observed for only one item set across gender and no item sets across country setting, suggesting that the item sets did not measure the same empowerment constructs in equivalent ways across gender or across country setting. Most WEMNS constructs would benefit from further refinement and testing to be suitable for face-to-face data collection and cross-cultural comparative analysis.

Notably, we assessed the psychometric properties of item sets in the face-to-face WEMNS module separately. This decision was guided by the many item sets, evidence from the phone-survey analyses [26], initial EFA findings in which multidimensional models had poor fit or failed to converge (S1 File), and findings here that the factors represented by single item sets were weakly correlated (Table 9). These practical and empirical considerations favored assessing psychometrically WEMNS item sets separately and deriving separate scores for each item set. None of the psychometric analyses, therefore, provided empirical support for creating a combined scale, based on the WEMNS empowerment factor scores. Moreover, the separate analyses of the item sets here do not provide guidance about the empirical performance of the counting-based WEMNS index, which combines resource and agency indicators, treating them as directly observed (non-latent) variables [24].

Notably, the samples for our analyses were not drawn in identical ways (S1 File). For developing and testing scales, however, representative samples are not initially required, provided that the study samples are large and diverse enough for cross-group comparisons. Also, the long-term goal is to recommend a scale for cross-gender, cross-cultural, and cross-time comparison by testing and confirming its equivalence across many diverse samples. Still, any point estimates derived from these study samples may not be representative of the geographic areas from where they were drawn.

These considerations notwithstanding, development of the WEMNS module was informed by a multi-disciplinary, multi-institutional, multi-cultural team. The WEMNS module benefitted from a review of questionnaires across multiple surveys platforms, inclusion of previously validated questions [40,41] , knowledge from experts in the field, local knowledge from multiple rounds of testing and piloting, and a large face-to-face survey in urban/rural samples in three diverse settings. The team also followed best practices for the development and psychometric assessment of new scales [42]. This level of rigor is rare in the development of measures for women’s empowerment. Finally, the WEMNS module covers major domains of resources and agency that align with theories of empowerment and three gender-related SDGs, covering 13 distinct empowerment-related constructs with only 75 survey items here and 64 items at most for the counting-based index, detailed elsewhere [24]. Its concise, yet comprehensive structure and cross-sectoral applicability among adult women and men make it unique among measures of women’s empowerment.

The results of this analysis suggest the benefit of further refinement of item sets in the WEMNS module, particularly if their intended use is the comparison of estimates across men and women or across country settings. As a next step, the WEMNS module has been integrated into household surveys as part of the 50x2030 Initiative, with suggested refinements from country partners. We encourage further investments in repeated, nationally representative surveys to ensure that measures of women’s empowerment can be used validly to compare derived estimates across countries and over calendar time to monitor SDG5 and gender-related SDGs globally. We also recommend investments to include a refined WEMNS module in cross-cultural panel studies of adult women and men to understand the comparability of measures of empowerment over the life course, trajectories of empowerment over age, and reciprocal influences with indicators of demographics; household needs, socioeconomic status, and welfare; individual beliefs and community social norms; as well as women’s, children’s, and family health and well-being.

Conclusion

The WEMNS module offers a concise, comprehensive measure of women’s and men’s empowerment that shows potential for use to monitor SDG5 and other gender-related SDGs. Further refinement of item sets is advised, especially if the item sets are to be used for comparative purposes. The field of women’s empowerment would benefit from similarly rigorous studies to develop and to assess measures for women’s empowerment to ensure that cross-cultural, cross-gender, cross-group, cross-program, and cross-time comparisons are valid and meaningfully inform equitable national policies and investments.

Supporting information

Acknowledgments

The WEMNS team comprises a team of researchers from (in alphabetical order): Data Analysis and Technical Assistance (DATA) in Bangladesh, Emory University in the United States (U.S.), Interdisciplinary Analysts (IDA) in Nepal, the International Food Policy Research Institute in the U.S., the National Statistical Office in Malawi, Tufts University in the U.S., Vox Latina in Guatemala, and the World Bank Living Standards Measurement Study in the U.S. WEMNS has been developed as an empowerment module for the 50x2030 Initiative to Close the Agricultural Data Gap (50x2030 Initiative), which aims to build capacity and close the agricultural data gap in 50 countries by 2030 and is implemented through a partnership of the International Fund for Agricultural Development (IFAD), the UN Food and Agriculture Organization (FAO), and the World Bank. The authors thank Erin Johnson for assistance with coding for the prior phone-survey analysis referenced in this paper. We also thank colleagues who provided feedback on the WEMNS survey module at various stages of its development. Finally, we thank the participants, without whom this work would not have been possible.