Unconditional cash transfer programs and women’s empowerment: evidence from Pakistan

Abstract

This study explores the impact of unconditional cash transfers on the multiple dimensions of women’s empowerment in Pakistan. Emphasizing the importance of cultural and religious norms, empowerment is considered as a latent construct manifested through three distinct choice dimensions, viz., “self,” “familial,” and “economic.” For the empirical estimation, a structural equation model is used on the country-representative Impact Evaluation Survey data of 2015–16 for the Benazir Income Support Program. The measurement model identifies various indicators for the dimensions considered. The results confirm the importance of providing cash transfers to the country's poorest women in all three dimensions, while the impact on self-choices is almost 40% that of the impact on the other two aspects of empowerment. Our findings provide recommendations for the successful implementation of social assistance programs.

1 Introduction

Over the last few decades, women’s empowerment has emerged as a primary concern of the sustainable development agenda. It is also viewed as a significant measure of social development. Both policymakers and academics are searching for ways to enhance women’s empowerment, particularly in developing countries where gender imbalances persist.¹

Many studies across various disciplines discuss the measurement of women’s empowerment and analyze the impacts of policies and programs² to expand empowerment as an outcome of sustainable development (Adato et al. 2000; Narayan et al. 2009; Schuler and Hashemi 1994). Most of these studies measure empowerment using proxy indicators³ while not encompassing multiple dimensions and ignoring cultural norms and traditions, in defining empowerment and measuring impacts of policy interventions. We choose Pakistan as a case study to fill these gaps—a country for which relatively little research has been published on the impact of cash transfer programs on women’s empowerment.⁴ Our study evaluates the impacts of unconditional cash transfers provided by the Benazir Income Support Program (BISP) in expanding empowerment for women, the major beneficiaries. We estimate the changes in various domains of women’s empowerment using a structural equation model (SEM).

Conceptually, women’s empowerment is a complex multidimensional process rooted in Sen’s capability approach (Chakrabarti and Biswas 2012; Samman and Santos 2009) and is context-specific (Kabeer 1999; Malhotra and Mather 1997; Mason and Smith 2003). While it is evident that agency enables women to define self-interest, make choices. and pursue capabilities (Kabeer 2001; Nussbaum 2000; Sen 1989, 2001), the measurement of agency must also focus on access to and control over assets, including physical, financial, human and social capital, and the ability to make decisions. Much of the literature uses information related to decision-making to infer empowerment (see, for example, Ashraf 2009; Braaten and Martinsson 2015; Carlsson et al. 2012; De Brauw et al. 2014; Peterman et al. 2015; Baranov et al. 2020). This is particularly useful on two accounts. First, this approach disentangles the puzzle of empowerment and brings us to the concept of agency. And secondly, it provides us with a measurement solution for such a complex and seemingly difficult concept by using self-reported decision-making in various domains of one’s life. For conceptual foundations, we follow Kabeer’s (1999) notion of empowerment and use decision-making and agency interchangeably as the crux or spirit of empowerment.

Several studies on women’s empowerment support the notion that the provision of social assistance to women improves welfare at the household level (Acharya and Bennett 1982; Attanasio and Lechene 2002; Deere and Doss 2006; Duflo 2003; Hoddinott and Haddad 1995; Schady and Rosero 2008; Yoong 2012; Ganle et al. 2015; Muralidharan and Nishith 2017). It is indeed rational to think that targeting cash grants directly and entirely to mothers would strengthen these women’s autonomy and decision-making power, strengthen their engagement in activities that lift their standing in the community, and bring them out of social and political isolation by cultivating links to other grantees and to the state itself. The existing empirical literature on various countries, however, does not provide ample and clear evidence of the effectiveness of cash transfers in expanding women’s empowerment. For example, micro-credit programs were found to increase women’s participation in household decision-making in Mexico (Angelucci et al. 2015), but no effect is found in India (Banerjee et al. 2015) and Pakistan (Said et al. 2019). However, Bandiera et al. (2020), Haushofer et al. (2019), and Kinyondo and Ntegwa (2023), among others, found improvement in the economic empowerment of women due to cash transfers,⁵ despite more limited impacts on their decision-making power. These mixed effects could be explained by measurement issues related to decision-making, program implementation and delivery details, context-specific gender relations, and cultural norms. The apparent insignificance of cash transfers in improving empowerment is intriguing and suggests that the assumed mechanism through which a shift in women’s access to resources translates into her decision outcomes is not working out as first conceived by policymakers and program implementers.

This paper contributes to the growing literature in two main ways. First, it enriches the area of the literature that uses latent variables analysis to inform various domains of empowerment.⁶ We define empowerment as a woman’s decision-making ability regarding her life choices, encompassing her “self” (the ability to act as an individual), “familial” (the ability to make choices about children), and “economic” (the ability to participate in the labor force and make decisions about household purchases and investments) choices.⁷ Building on this definition and extending the latent variable model of women’s empowerment, this study presents a structural equation model to inform the construction of empowerment indices for women from the poorest households in Pakistan. Ballon (2018)⁸ and Abreha et al. (2020) used multiple indicators, multiple causes (MIMIC) models to analyze women’s empowerment, although their models differ in structure, reflecting their different objectives. In the case of Ballon (2018), the objective was to estimate the different aspects of women’s empowerment as latent variables and to then apply stochastic dominance analysis to assess the relative importance of the different aspects of empowerment. In contrast, Abreha et al. (2020) used their model to estimate the different aspects of women’s empowerment to measure their association with children's health status. In our paper, we wish to assess the impact of unconditional cash grants on women’s empowerment, so our model differs from those of both Ballon (2018) and Abreha et al. (2020).⁹

Second, our paper presents a study of women’s empowerment in the context of a country that, with its intricate gender relations and strong patriarchal norms and religious characteristics, with huge gender imbalances in almost every field of life, including access to basic education, healthcare delivery, employment, and income, is very different from the Latin American countries that are often considered in the literature on women’s empowerment.¹⁰ While international empirical research is inconclusive on the effect of cash transfer on women’s empowerment, the available handful of studies found a positive impact of BISP on women’s mobility, tolerance of domestic violence, and sharing of household work (see Ambler and De Brauw, 2017; Waqas and Awan 2019a, b).¹¹ These studies are either based on samples from only one part of the country or evaluated the effect of BISP on women empowerment by a few selected outcome-related/decision-making variables like voting rights, women mobility, or labor market activities. On the contrary, our paper develops a comprehensive measure of women empowerment as a multidimensional construct incorporating gender roles and norms specific for the religious and cultural setting of Pakistan. This approach adds value in understanding the impact of cash transfer on various domains of women’s empowerment. We used the latest 2016 BISP Impact Evaluation data covering the whole country. The use of national representative data for evaluating the effectiveness of the BISP on women empowerment in the aforementioned three dimensions is advantageous compared to the qualitative focus group-based approach (see Ganle et al. 2015) or systematic survey method (Broady et al. 2016), as well as the popular but highly costly randomized controlled trial (RCT) approach (see for example, Baranov et al. 2020; Dhar et al. 2022) available in the literature.¹²

Thus, our findings offer important considerations regarding social assistance programs’ gender-based targeting design and the consequent effects of those programs on women’s empowerment. This study complements a growing literature that indicates that unconditional cash transfer programs may be conducive to alleviating poverty and improving households’ well-being.¹³

The paper is structured as follows: Sect. 2 introduces the definition and conceptual framework of empowerment used in the study. Section 3 describes the main elements of the program; Sect. 4 presents the structural equation model; Sect. 5 describes the empirical results; and Sect. 6 concludes.

2 Conceptual framework

The question of what constitutes women’s empowerment is a matter of continuing debate. Nonetheless, it is widely regarded as a multidimensional concept. For example, Anderson and Eswaran (2009), Saleem and Bobak (2005), and Sathar and Kazi (2000) use the term “autonomy” when measuring women’s empowerment in Bangladesh and Pakistan, and Adato et al. (2000) use the term “status” when measuring the impact of Progresa (a cash transfer program in Mexico) on women’s empowerment. Similarly, Ashraf (2009), Braaten and Martinsson (2015), Carlsson et al. (2012), De Brauw et al. (2014), and Peterman et al. (2015) use “decision-making” to analyze women’s empowerment, while “gender equality” is used by the World Bank. There is no clear demarcation between these terms, and they are often used interchangeably.

One of the most comprehensive theoretical foundations of women’s empowerment emerges from Sen’s capability approach. Sen (1985) puts forth a theoretical debate on empowerment being the enhancement of one’s agency, conceived as the enlargement of effective opportunities for women to live the life they want. Kabeer (1999), on the other hand, extends the definition and connects power, symbolizing the ability to make choices. She explains this process of empowerment in terms of three inter-related elements: resources, agency, and achievements. For our study, we operationalize empowerment as defined by Kabeer and approach it as a process comprising access to resources leading to agency and achievements as empowerment outcomes.

The literature suggests that “resources” and “agency” are the two most stated domains of empowerment (Acharya and Bennett 1982; Chen 1992). Some studies present resources as a pre-condition for empowerment, not a source; for example, Kishor (2000) analyzes resources such as education or employment as enabling factors of the empowerment process. But in societies like Pakistan, it is harder for a woman to exercise choice in accessing both material and human resources. Pakistani women are easily oppressed by their mothers-in-law, often accept domestic violence to keep the family intact, prefer sons over daughters, and relinquish their right to inheritance in favor of their male siblings (Chung and Gupta 2007; Winkvist and Akhtar 2000).

Alderman and Gertler (1997) analyze the differences in rates of human capital investment by gender due to differences in household resources. They find that the difference in price elasticities of educational attainment falls as family resources rise. Poorer families invest less in daughters relative to sons, and the difference in the level of discrimination between wealthier and poorer families grows as the price of human capital rises. In a study on gender difference in household expenditure on education in Pakistan, Aslam and Kingdon (2008) find significant biases toward male children in education-related expenditure in both the enrollment decision and the decision on private expenditure for schooling. Their results suggest that the observed strong gender difference in education expenditure is a within rather than an across household phenomenon. It is evident that equal access to resources is hard to practice in societies like Pakistan. Therefore, we assert that an increase in female-oriented “resources” such as a gender-targeted cash transfer program (e.g., BISP in Pakistan) provides opportunities that improve a woman’s ability to make decisions. We also suggest policies designed to alleviate poverty and raise incomes will reduce gender discrimination among the poor.

Early studies considered women with control over material resources and labor income as a robust measure of agency or empowerment.¹⁴ Another branch of the literature uses bargaining models to measure female agency. In this context, a woman’s degree of empowerment is evaluated by the strength of her bargaining power within the household, and therefore, improvement in her status will enhance her decision-making ability or empowerment.¹⁵ Shibata et al. (2020) and Agarwal (1994), among others, suggest that female bargaining power in rural areas depends on factors such as a woman’s ownership of land, her sources of income, her educational status, her access to family and tribal support systems, and support received from NGOs or the state. It is therefore by means of these parameters or factors that policy interventions may improve a woman’s bargaining position in the household and consequently her empowerment. Doss (2013) explains intra-household allocations via the impact of endowments, preferences, human resource investment prices, household resource levels, labor market opportunities, and marriage markets (see Behrman 1997 also). Women’s participation in household decision-making is considered an outcome or achievement or a manifestation of the agency.

Thus, while resources are critical for women’s empowerment, they are not always enough. Without women’s individual or collective ability to recognize and utilize resources in their own interests, resources cannot empower them. Having claimed that “agency” should be treated as the soul of empowerment, and resources and achievements as enabling conditions or sources of empowerment, correspondingly, another domain is necessary. Social settings, norms, and religious practices shape women’s ability to exercise agency. Most empirical models used to measure women’s empowerment ignore the complexities of gender relations beyond the household and the important role played by social norms, culture, values, and perceptions in the bargaining process. Ballon (2018) proposed a comprehensive approach to measure women’s empowerment, building on Sen’s capability approach and exploring intra-household gender dynamics. In her model, women’s empowerment is a latent variable reflected in the observed decisions made by women but shaped through available “resources” and determined by exogenous values and traditions. Jayachandran (2021a, b) argued that cultural norms help explain the large differences in female employment among countries at the same level of development. Thus, gender gaps in the labor market may be narrowed through policies attuned to cultural norms. Using a structural equation modeling approach, Bahadir-Yilmaz and Öz (2018) found a significant and negative effect of violence against female spouses on their level of autonomy, measured by participation in household-related decisions. They reported that the incidence of intimate partner violence (IPV) also diminishes with higher female autonomy. While looking at the impact of social norms on IPV, the authors find that IPV increases along with the husband’s commitment to social norms, upholding traditional gender roles characterized by women’s subordination and restricted agency. Chaudhuri and Yalonetzky (2018) demonstrated that social comparisons are possible using stochastic dominance techniques suited for multiple ordinal and dichotomous variables. Their empirical findings in India highlight that whenever these dominance conditions hold for a pair-wise comparison, the multidimensional autonomy distribution in one state is more desirable than in another one in terms of a broad range of criteria for the individual and social welfare evaluation of autonomy.

Based on the literature reviewed, we propose a framework of women’s empowerment that interlinks the following three aspects/dimensions.

2.1 Access to enabling resources as sources of empowerment: “economic” choice

A reliable source of income—like cash transfers—offers regularity and predictability of income and flows of resources that provide women with the security to plan and act. It also emphasizes just how significant social resources are to women’s empowerment. Education, health and fertility decisions, socio-demographic characteristics (including age, family size, family structure, and region), and religious and social norms each play a decisive role in determining women’s agency.¹⁶ Haddad and Hoddinott (1994) find robust results suggesting that increasing women’s share of income increases spending on food and reduces the budget shares of alcohol and cigarettes for men. Similarly, Eswaran (2002) argues that expanding a woman’s autonomy within the household is positively linked with her bargaining power and is shown to reduce fertility rates and child mortality. Schuler and Hashemi (1994) find that participation in credit programs empowers women, mostly through enhancing their economic roles, and is positively correlated with the use of contraceptives. Allendorf (2007) argues that women's land rights also positively impact young children’s health since children of mothers with land ownership are significantly less likely to be severely underweight. She also finds that men receiving pensions have almost no effect on children’s nutritional status. Pre-marital enabling resources can ensure post-marital agency (Yount et al. 2016, 2018) and allow women to negotiate rights and physical safety within marriage (Miedema et al. 2016). In this study, we call this the “economic” dimension of empowerment.

2.2 An ability to exercise choice in household decision-making: “familial” choice

In our framework, agency refers to making decisions for oneself and the family (Gammage et al. 2016; Ghuman 2003; Malhotra and Mather 1997; Yount et al. 2016). The decision-making indicators are self-reported by the respondents and measured the capability of respondents to make decisions alone or jointly with their spouses. Several studies propose and use household decision-making as a measure of agency.¹⁷ For the purpose of our study, we use decision-making in various domains of a woman’s life as a manifestation of agency and name this dimension “familial.”

2.3 The expression of equitable gender beliefs and attitudes (intrinsic agency or power within ¹⁸): “self”-choice

It can be argued that participation in poverty reduction programs may not impact social norms because such values and norms are deeply rooted in religion and culture. On the other hand, the normative structures that limit women’s decision-making may be altered through processes arising from individual actions in response to exogenous changes like gender-targeted cash transfer programs.¹⁹ For example, restrictions on women’s freedom of movement and interactions with men who are not closely related to the family may relax as a result of cash transfer programs or other similar opportunities, and women’s access to social networks outside the family may increase along with their presence in the public sphere. Irrespective of whether they have been intentionally fostered through development policies (as opposed to being merely by-products of policies with other aims), such changes may eventually lead to a shift in women’s positions within family systems, or markets, and may represent a transformation in gender norms. This is the “self”-aspect of empowerment.

The key features of our conceptual framework explained above suggest that empowerment is a “process.” Malhotra et al. (2002) list four major hurdles when measuring empowerment, including: “the use of direct measures as opposed to proxy indicators, the lack of availability and use of data across time, the subjectivity inherent in assessing processes, and the shifts in the relevance of indicators over time” (p. 20). In response, we make efforts to measure empowerment using direct indicators (i.e., based on women's responses to specific survey questions about their decisions and mobility²⁰) as these provide the closest measures in specific cultural contexts. Secondly, we attempt to measure empowerment at the individual level as a function of the individual’s decision-making and community norms. As discussed above, in our model, “empowerment” can be manifested through the multidimensional elements representing three distinct choices/dimensions of women’s empowerment: “economic” (participation in the labor force as well as deciding on investment and household purchase); “familial” (choices made particularly for the children’s well-being); and “self” (that is the ability to act as an individual or self). These choices are determined through several decision outcomes/indicators (observed self-reported indicators of decision-making in various domains such as minor household purchases, labor force participation, child education, family planning, having another child, mobility in social participation, and marketplaces, etc.). The norm/attitude is that men only make the decisions that influence these outcomes. In our estimation model, “choices” interact into a system of structural equations where a latent or unobserved variable is specified to measure empowerment. This latent variable represents a woman's decision-making ability that is (at least partially) measured by her decision outcomes and is formed by her access to resources, “cash transfers,” and the prevailing values/traditions in the society, observed as exogenous factors.

Table 1 Balance between treatment and control groups

3 Overview of BISP

Almost one-third of the population of Pakistan lives in abject poverty. In 2008, the country witnessed a major shift in its poverty reduction policy and announced the Benazir Income Support Program (BISP) as its main social safety-net program. The BISP provides unconditional cash payments and is piloting conditional cash transfers. The unconditional cash transfer provides money directly to the female heads of poor households. The transfer is paid every quarter using automated payment mechanisms like debit card and mobile phone transfers. The program has three main policy goals: (1) eradicating extreme and chronic poverty; (2) empowering women; and (3) achieving universal primary education. The intention is to achieve the first goal through regular cash transfers and the second goal by giving the transfers to women. The government launched a separate conditional cash transfer program in 2013 to increase primary school enrollments.

Getting BISP benefits is contingent upon recipients having a computerized national identity card. While establishing the BISP scheme, an effort was focused on creating a modern, well-managed, large-scale, efficiently targeted cash transfer program. This flagship operation has successfully scaled up to its current coverage of almost 5.6 million families across all provinces, representing about 18 percent of the population of Pakistan. The selection of BISP beneficiaries is made through a poverty scorecard survey based on household demographics, assets, and other measurable characteristics that, in principle, cannot be manipulated by beneficiaries and the survey firms. During the period under analysis for this paper (2015–2016), the benefit level was PKR 1500 per month (~ US $10), paid in quarterly installments. This is enough to provide a monthly supply of wheat flour for a family of six. BISP is also piloting conditions-based cash transfers and is providing assistance to disabled and calamity-stricken communities across the country.

4 Empirical strategy

4.1 The dataset

The BISP has been targeted using a proxy means test (PMT), which is used to attempt to reach 15 percent of households nationwide with regular, unconditional cash transfers. The national targeting mechanism based on PMT was developed in 2010–11. Weights for the PMT were developed using the 2007/8 Pakistan Living Standards Measurement Survey, and the PMT uses 23 variables to compile a poverty score. To target the BISP, a poverty scorecard survey was initiated in 2010/11, collecting information on those 23 variables across Pakistan. Upon completion of the data collection, a PMT score was generated for every household. A PMT threshold (cut-off score) of 16.17 was established based on budget availability to reach at least poorest 15 percent of the population of Pakistan.

Our dataset is derived from the BISP Impact Evaluation Survey 2015–16. This is a nationally and provincially representative survey that collects data on BISP-eligible households in the four target provinces (Punjab, Sindh, Khyber Pakhtunkhwa or KPK, and Baluchistan). The sample was chosen to be representative of households in the four provinces that are close to the pre-determined poverty threshold of 16.17. The evaluation sample was created with the intention of exploiting the eligibility criterion to provide a sample including “control” and “treatment” groups. Thus, the evaluation sample has beneficiaries as well as comparable non-beneficiaries. In addition, for ever-married women, the survey gathers information concerning the spouse’s education and age, decision-making outcomes, attitudes toward gender roles, children education, and healthcare, as well as the woman’s labor force participation, control of earnings, and expenditures. As our study is concerned with the appraisal of women’s empowerment regarding herself, familial life choices, and economic choices, the sample we consider includes all ever-married women.

Treatment households are defined as those households with PMT scores equal to and below the cut-off score of 16.17 who received the BISP payment. Control households are defined as those households with a PMT score greater than the cut-off score of 16.17 and within a pre-determined range up to a score of 21.17. Respective samples of treatment and control households were chosen from within each primary sampling unit (PSU) using simple random sampling. This leads to a final sample size of 6447 women, where 3,939 are BISP beneficiaries (~ 61%) and 2,472 (~ 39%) are in the control group.

By limiting the control households to have PMT scores close to but above the threshold of 16.17, it is hoped that factors that affect women’s empowerment, other than the BISP payment, will have low variability between the control and treatment groups. In this sense, the sample may behave similarly to one in which the treatment was randomly allocated. Table 1 presents evidence that this has not been entirely successful.

The second and third columns of Table 1 present the mean values of the observable characteristics for the control and treatment groups, respectively. The fourth column of Table 1 presents p values corresponding to the null hypotheses that each of the variables has identical expected values in the control and treatment groups. These were computed by regressing each of the characteristics on a treatment dummy variable and computing t-statistics using the heteroscedasticity-consistent covariance matrix of White (1980). Since there are 30 different hypotheses, standard hypothesis testing procedures that control the probability of false rejection for a single hypothesis are likely to result in a number of spurious rejections. To avoid this problem, we adjust the p values of the tests using the step-wise multiple testing procedure of Holm (1979). Holm’s procedure provides strong control of the family-wise error rate. Thus, for example, if we reject the null hypothesis in each case for which the adjusted p value is less than 0.05, then the probability of rejecting at least one true null hypothesis is less than 0.05, for any combination of true and false null hypotheses.

Note that, even though the sample was drawn from households with a narrow range of PMT scores, there exists evidence in Table 1 of systematic differences between the treatment and control groups. However, while these differences are statistically significant, they are not economically significant. For example, there exists strong evidence that the mean age of the head of the household is higher in the treatment group than the control group. While it is possible that an age difference that was measured in decades would have important implications for women’s empowerment, the measured difference in Table 1 is slightly less than one year, and it is implausible that such a small difference could have a meaningful impact on women’s empowerment. Similar arguments exist for the other statistically significant variables in Table 1. Consequently, it is reasonable to conclude that differences in women’s empowerment between the treatment and control groups are due to the treatment effect, rather than the small differences in the measurable characteristics of control and treatment groups. Nonetheless, in our empirical work, we include some of the statistically significant variables from Table 1 as control variables in order to ensure that any measured differences in empowerment between the control and treatment groups are not actually due to those variables rather than the BISP payment.

4.2 Variable selection

Our empirical model is designed to measure the impact of the receipt of the BISP payments on the three dimensions of empowerment discussed in Sect. 2—the ability of a woman to act as an individual (“self”), the ability to make choices about children (“familial”), and the ability to participate in the labor force and make decisions about household purchases and investments (“economic”). This task is complicated by the fact that these aspects of empowerment are not directly observable but are instead reflected in the living circumstances of women, as revealed in the Impact Evaluation Survey. In particular, the “economic” aspect of empowerment is reflected by the ability of a woman to make joint decisions with her husband about minor household purchases, lending and borrowing money, making small investments, and the type of work that she does. The “familial” aspect of empowerment is reflected in the woman's ability to make joint decisions with her husband about methods of contraception, having more children, and choices about the education and marriages of their children. The “self” aspect of empowerment is reflected in the ability of a woman to go to the market, to visit friends and neighbors, religious places, and health facilities, and to make joint decisions with her husband about social participation, voting, and serious health problems. We show below that the possession of data on these variables is sufficient to statistically identify the three aspects of empowerment, their mutual correlation, and the impact that receipt of the BISP payment has on each of them individually.

The names of the variables used in the model are listed in Table 2, along with a short functional description and an indication of whether the variable is latent and/or binary. In Appendix Table 6, we provide the empowerment-related questions from the BISP Impact Evaluation Survey Questionnaire.

Table 2 Variables used in the model

4.3 The model

In this paper, we propose a latent variable model that reflects two characteristics of empowerment. Firstly, empowerment is not directly observable. Instead, it may be inferred from several observable indicators (Ballon 2018). Secondly, empowerment is a multidimensional process, and a single variable cannot explain all the underlying concepts (Malhotra et al. 2002; Samman and Santos 2009). Nonetheless, the dimensions of empowerment are likely to be mutually correlated.

Let $f_{i} = \left( {\begin{array}{*{20}c} {self_{i} } & {familial_{i} } & {economic_{i} } \\ \end{array} } \right)^{\prime}$ be a 3 × 1 vector that measures the three domains of women’s empowerment for person i. Self denotes the individual or self-choice, while familial and economic refer to familial and economic choices (See Sect. 2 for the details). Without loss of generality, we will assume that these variables have means of zero. We would like to estimate the following regression equations.

$$ f_{i} = \beta \,treat_{i} + \Gamma x_{i} + \varepsilon_{i} $$

(1)

where $treat_{i}$ is a dummy variable that indicates whether individual i is part of the treatment group (i.e., is a BISP recipient), $x_{i}$ is a qx1 vector of control variables, $\varepsilon_{i}$ is a 3 × 1 vector of potentially mutually correlated unobservable disturbances, $\beta$ is a 3 × 1 vector of regression coefficients, and ${\Gamma }$ is a 3xq matrix of regression parameters. In Table 1 in Sect. 4.1, we present evidence of small systematic differences between the treatment and control groups. In order to ensure that these do not result in spurious findings about the impact of BISP payments on empowerment, we specify

$$ x_{i} = \left( {\begin{array}{*{20}c} {head\_age} & {femage_{i} } & {num\_dependent_{i} } & {hhsize_{i} } & {wall_{i} } & {floor_{i} } & {toilet_{i} } \\ \end{array} } \right)^{\prime} . $$

We stress that our inclusion of these variables is motivated by a desire to minimize any potential correlation between $treat_{i}$ and $\varepsilon_{i}$, rather than a direct interest in their relationship with $f_{i}$.

If $f_{i}$ was directly observable, then the estimation of, and tests of hypotheses about $\beta$ would be a standard regression problem. The fact that $f_{i}$ is latent complicates matters. To resolve this issue, we specify the following confirmatory factor analysis (CFA) model for $f_{i}$:

$$ y_{i}^{*} = \Lambda f_{i} + \xi_{i} $$

(2)

where $\xi_{i}$ is a px1 vector of random disturbances, $\Lambda$ is a px3 full column-rank matrix of coefficients, and $y_{i}^{*}$ is a px1 vector of latent variables. Denote the elements of $y_{i}^{*}$ as $y_{ij}^{*}$, j = 1,…,p. For each $y_{ij}^{*}$, there exists a latent threshold $\tau_{j}$ and an observable binary variable $y_{ij}$ such that

$$ y_{ij} = \left\{ {\begin{array}{*{20}l} {1{\text{ if }}y_{i}^{*} \ge \tau_{j} } \hfill \\ {0{\text{ if }}y_{i}^{*} < \tau_{j} } \hfill \\ \end{array} } \right.. $$

The elements $y_{ij}$, j = 1,…,p indicate the observable binary responses of individual i to a range of questions intended to measure women’s empowerment. Specifically, we set:

y_i = (mobility_market_i mobility_health_i mobility_friends_i mobility_religous_i health_problems_i social_participation_i voting_i another_child_i childrens_educ_i childrens_marriage_i family_planning_i minor_purchases_i job_choice_i lending_borrowing_i investment_i)^/.

Equation (2) may be rewritten as

$$ f_{i} = Ky_{i}^{*} + \eta_{i} $$

(3)

where $K = \left( {\Lambda^{\prime}\Lambda } \right)^{ - 1} \Lambda^{\prime}$ and $\eta_{i} = - \left( {\Lambda^{\prime}\Lambda } \right)^{ - 1} \Lambda^{\prime}\xi$. Together, Eqs. (1) and (3) constitute a Structural Equation Model (SEM).

Note that, for any 3 × 3 matrix $M$, we may write

$$ \tilde{f}_{i} = Mf_{i} = \tilde{K}y_{i}^{*} + \tilde{\eta }_{i} , $$

where $\tilde{K} = MK$, $\tilde{\eta }_{i} = K\eta_{i}$, and $\tilde{f}_{i}$ is an observationally equivalent representation of the latent women’s empowerment variables. In this sense, in general, the SEM is not identified. Fortunately, our application provides restrictions that are sufficient to resolve this issue. Note that each of the elements of $y_{i}$ loads onto only a single factor. In particular, mobility_market, mobility_health, mobility_friends, mobility_religous, health_problems, social_participation, and voting are related to only the “self” component of women’s empowerment; another_child, childrens_educ, childrens_marriage, and family_planning are related to only the “familial” component; and minor_purchases, job_choice, lending_borrowing, and investment are related to only the “economic” component. Thus, for an appropriately ordered vector $y_{i}^{*}$, $\Lambda$ is block diagonal, consisting of three blocks, each of which is a column vector. It follows that $K$ is a block diagonal matrix, consisting of three blocks, each of which is a row vector. The only values of $M$ that respect this structure are diagonal matrices. Thus, $f_{i}$ is identified up to a rescaling of the elements. To resolve the indeterminate scaling, we (arbitrarily) set the variances of the elements of $f_{i}$ to 1. Note that there exist other variables that might be included in $y_{i}$. Examples include attitudes to IPV and the ability of the woman to express an opinion. However, the addition of such variables complicates arguments about identification since, unlike the variables that we have included in $y_{i}$, they are not associated with only a single element of $f_{i}$.

The model is represented diagrammatically in Fig. 1.

Fig. 1

Diagram of Eqs. (1) and (3)

5 Results and discussion

The model described in the previous section was estimated using the diagonally weighted least squares (DWLS) procedure implemented in the Lavaan package for the R programming language.²¹

Table 3 shows the estimated covariance matrix for the three aspects of women’s empowerment. Also shown are the t-statistics corresponding to the null hypotheses that each of the elements of the covariance matrix is equal to zero. Recall that the variances of the empowerment variables are unidentified and have been arbitrarily set to unity.

Table 3 Estimated covariance matrix of empowerment variables

As might be expected, the aspects of women’s empowerment are positively correlated with each other. The large t-statistics provide strong evidence against the hypotheses that the correlations are zero. Note that the correlation between the “economic” and “familial” aspects of women’s empowerment is stronger than that between the “self” aspect and either of the other two aspects. Nonetheless, the fact that none of the correlations are particularly close to 1 supports the proposition that women’s empowerment needs to be understood as a multidimensional concept.

Table 4 shows the estimates of the elements of $K$. As discussed above, $K$ is block diagonal due to the restrictions implied by the elements of $y_{i}$. Note that, for the “self” aspect of empowerment, the largest loadings are on the mobility variables, with joint decisions about social participation, serious health problems, and voting having a less strong relationship with “self.”

Table 4 Estimates of K

For the “familial” aspect of empowerment, each of the variables is approximately equally important. For the “economic” aspect of empowerment, joint decisions about lending and borrowing and small investments are the most important variables. As might be expected, all the elements of $K$ are positive. They are also all highly statistically significantly different from zero. For completeness, Table 4 also includes the values of the estimated thresholds, $\tau_{j}$, j = 1,…,p.

Finally, in Table 5, we present the estimates of $\beta$ and $\Gamma$.

Table 5 Estimates of $\beta$ and $\Gamma$

Note that there exists evidence that some of the control variables have a statistically significant relationship with some of the aspects of women’s empowerment. Thus, the fact that there are some differences in the characteristics of the treatment and control groups beyond the receipt of the BISP payment (as illustrated in Table 1) is a relevant consideration. However, the estimated coefficients for the control variables are generally quite small. This, combined with the fact that these variables differ by only a small amount between the treatment and control groups, as established in Table 1, supports our conjecture that these variables account for little variation in women’s empowerment between those households in the sample that receive the BISP payment and those that do not. Controlling for these variables, we find that the relationship between the receipt of the BISP payment and each of the three aspects of women’s empowerment is positive and statistically significantly different from zero. A Wald test of the null hypothesis that each of the elements of $\beta$ is jointly equal to zero, so that the BISP payment has no impact on women’s empowerment, returns a test statistic of 21.335. Since the statistic has a $\chi_{3}^{2}$-distribution, the null hypothesis is strongly rejected at any conventional significance level. Note that the estimated impact of the BISP payment on the “self” aspect of empowerment is ~ 37% higher than that of the impact on the “familial” and “economic” aspects. A Wald test of the null hypothesis that the BISP payment has an equal impact on each of the three aspects of empowerment generates a test statistic of 18.158. Under the null hypothesis, such statistics have a $\chi_{1}^{2}$-distribution. Consequently, the differences in the estimates of the impact are highly statistically significant. This extends the result of previous studies measuring women’s empowerment in Pakistan including Akram (2018), Chaudhry and Nosheen (2009), and Ghuman et al. (2006), which found that women have greater decision outcomes concerning their self-domain compared to the familial or the economic choice domain, specifically in the treatment group.

The results of Table 5 report evidence that the BISP-targeted cash transfer program in Pakistan has wide-ranging impacts on the lives of beneficiaries beyond the intended scope as a poverty reduction intervention. Our results demonstrate that the transfer was successful in improving multidimensional agency and decision-making of women and that these effects were robust to various measures of women’s agency in various domains of their lives. While the existing literature hypothesizes that economic development/poverty alleviation by itself cannot ensure significant achievements in women’s empowerment (see Duflo 2012 in this respect), our results provide reasonably strong evidence that the BISP transfer has indeed had successful impacts on women’s empowerment. Giving cash, even small but regular amounts, to women appears to have significantly impacted all three aspects of empowerment we studied in this research.

Our results in Tables 4 and 5 indicate that BISP is a crucial determinant of self-choices that lead to improvement in female mobility, healthcare, social participation, and voting. This means participation in the program relaxes some of the norms like “freedom of mobility” and “meeting with non-relatives of the family” for the female participants. Mobility remains the most important and widely used indicator of socio-cultural empowerment in the literature. Jayachandran (2021a, b) noted that the practice of “purdah” in Islam (which is also the religious faith in Pakistan) enhances the low employment rate in the Middle East and North Africa [see also Koomason’s (2017) study on Ghana]. Several researchers declared that advancement in women’s liberty of mobility is essential for increasing their capacity to make personal choices, adjust their behaviors, expand their public linkages, attain better employment, and lessen their state of poverty (D’Acci 2011; Gram et al. 2019).

Women’s awareness of their and their partner’s roles is an important component of women’s empowerment. Our findings show that the husband’s support in joint decision-making is a significant determinant of women’s empowerment in the domains of familial, economic, and some aspects of self-choices. Our results are in line with those of Khan and Awan (2011) and Batool et al. (2019) in that husbands’ support and co-operation have statistically significant effects on women’s empowerment. When a husband provides support to his wife by setting a cordial relationship and allowing her the freedom to express her state of mind and helps her resolve her problems, she may feel protected and consequently gain confidence.

As seen in Table 4, the indicator “job_choice” has a statistically significant relationship with the economic aspect of women’s empowerment. When women are allowed to work outside the family, they gain greater freedom in making choices which in turn affects their own health and mobility, important economic matters, and the education of their children. The women that BISP targets are the poorest of the poor in Pakistan. These women are not in paid employment but work as farm helpers for their landlords or husbands—most of these women work for food and clothing and are unable to make even small purchases. This also has huge implications for the success of cash transfer programs in impacting empowerment. Women who have capacity to make decisions may keep control of the cash transfer and use the money to improve the health and education status of their family. Similarly, if a woman has a say in minor household purchases, it is more likely that she will spend the money on improving her child's nutritional outcomes. Our economic choice outcome also includes the investment and borrowing choices of women. In the Indian state of Madhya Pradesh, Field et al. (2020) noted that when women have a bank account, their labor force participation rate increases. Schaner's (2011) case study in Kenya makes the same conclusion. Jayachandran (2021a, b) considers this as the change in social norms toward better women’s empowerment.

Note that BISP beneficiary status has a positive impact on the “familial” aspect of women’s empowerment, indicating that BISP transfers influence such decisions as family planning, increasing family size, and, most importantly, children’s education. This is a critical finding that could have important implications for Pakistan with the world’s second-highest number of out-of-school children (OOSC). An estimated 22.8 million children in Pakistan aged 5–16 do not attend school, which represents 44 percent of the total population in this age group. Therefore, BISP can play a critical role in promoting education through its unconditional as well as conditional cash transfer programs. Ashraf et al. (2010) noted in the case of Zambia that women try to hide their family planning decision from their husbands. Our results show improvement of women’s empowerment of the BISP beneficiaries’ joint decision-making in family planning despite the presence of a strong cultural norm against this.

Our analysis of empowerment and the role of cash transfers in improving empowerment also suggests that it is the cash that influences the wider behavioral changes reported, without any conditions imposed on women, contradicting the idea that only cash that comes with conditions brings changes within households leading to the supposed empowerment of women (Adato et al. 2000).

6 Conclusion

In this paper, we present a model of women’s empowerment for the BISP target population in Pakistan. We define empowerment as the ability of a woman to make decisions related to self, familial, and economic choices, and we use this definition in the specific context of women as beneficiaries of poverty alleviation programs. We distinguish three key elements that contribute to a woman’s ability to make such decisions, as theorized by Kabeer (1999): her enabling resources; decision outcomes as a measure of her agency; and values/traditions as a measure of the settings or pre-conditions. Based on this definition, this paper presents a structural equation model, to measure and explain the causes of women’s empowerment. Our measurement model confirms that women’s empowerment is indeed a multidimensional construct. The results of the structural model reveal the importance of providing cash transfers to the woman of the house at all three dimensions of empowerment. Specifically, the treatment group communities are more likely to report that they can visit friends, market, health facilities, and religious places without permission. Furthermore, they can make joint decisions on serious health-related issues, social participation, and participation in voting in elections. However, their voice is less pronounced compared to the above indicators in familial (children marriage and education, as well as family planning) and economic (household purchase, investment, etc.) dimensions.

Based on these results, various policy implications can be put forward. The first is that cash transfers without conditions may help to alter behavioral patterns, but choices related to various socio-economic norms are much harder to change. BISP can initiate small Community Leaders forums, at the level of village and at the Tehsil,²² to generate discussion and peer learning practices for a better understanding of the rights and obligations of women. Materials providing education about women’s rights can also be distributed to promote awareness. Further participatory research is needed to explore this phenomenon.

Secondly, there is clear evidence that BISP has the potential to assist women by addressing their needs and enhancing their capacity for economic, social, and personal development. But to fully capitalize on this opportunity, it is important for BISP and similar interventions to mainstream gender practices that uplift the status of women not only within households but also in society at large. For instance, women beneficiary representatives should be included in important policy formulation processes at various administrative and operations levels of BISP Boards to provide recommendations for promoting gender equity. Therefore, more gender-sensitive design features in BISP or any other cash transfer programs in the developing world in general are required to reduce poverty and help governments advance their goals of achieving greater gender equality.

Appendix

See Appendix Table 

Table 6 Questions used in this study from the BISP Follow-up Questionnaire: women decision-making and empowerment (all ever-married women)

6.