Agricultural Information Institute, Chinese Academy of Agricultural Sciences, Beijing 100081, P.R.China

Abstract Ending poverty is a top priority of the international development agenda,and governments worldwide have attached great importance to poverty alleviation measures. However,poverty reduction policies have mostly focused on men,which has widened the gap in productivity and income between men and women and increased gender inequality. This paper aims to determine the impacts of a multi-component program on women’s empowerment and poverty reduction,and explore the role empowered women play in poverty reduction. The dataset used in this study was collected in nine poor counties of Ulanqab City in the Inner Mongolia Autonomous Region of China at the end of 2014,yielding a sample of 900 households.Recall questions were used to reconstruct the baseline data and build a panel dataset. Smaller groups of rural households were further identified to better target the women in the beneficiary group. To control the selection bias,propensity score matching,inverse probability weighting,and the difference-in-differences matching method were used to analyze the effect of the program and undertake robust checks. The results show that the program has positive effects on women’s empowerment and poverty reduction simultaneously. Empowering women also has positive effects on poverty reduction,and the women who were the beneficiaries have contributed to increasing the incomes and living standards of households.Training,microfinance,and associations are common means or strategies to empower women to address poverty. This paper provides new empirical evidence that women can benefit from a gender-focus program through portfolio intervention such as training,cooperatives,and credit. Empowered women further improve the livelihoods of poor households and help lift them out of poverty. The results suggest that researchers and policymakers need to pay more attention to poverty issues from the perspective of gender.

Keywords:women,empowerment,poverty reduction,Inner Mongolia Autonomous Region

1.lntroduction

Since the Beijing Declaration and Platform for Action adopted at the United Nations Fourth World Conference on Women,held in Beijing in 1995,empowering women has been recognized as essential to realizing sustainable development. It underpins the United Nations 2030 Agenda.According to data from the World Bank in 2018,for every 100 men there were 104 women in poor households,and it has been found that girls and women of reproductive age are more likely to live in poor households than boys and men (Boudetet al.2018). Women are more vulnerable to poverty than men due to cultural norms and values,the gendered division of assets,and the power dynamics between men and women (Atozouet al.2017). Women usually bear a disproportionate burden of unpaid household responsibilities. If they have a job,it is most likely to be casual,informal work,which is easily overlooked by policymakers. Empowerment is the process whereby women learn knowledge and skills,overcome difficulties,and benefit from helpful resources along the way (Cornwall 2016). It is not only an outcome but also an intermediate variable to further observe other development outcomes(Moyoet al.2012),especially poverty. To achieve the final goal of poverty reduction,greater emphasis must be placed on gender equality and removing the barriers that affect women.

Empowering women is a vital instrument in the fight against poverty (Sharmaet al.2012; Faborode and Alao 2016). In the short run,increasing women’s productivity(Seymour 2017; Diiroet al.2018),employment (Khumalo and Freimund 2014; Maligaliget al.2019),and earnings can contribute to poverty reduction and economic growth.And in the long run,gender equality is an important factor in preventing the intergenerational transmission of poverty(Morrisonet al.2007),because women often bear the most important responsibility for raising the education levels of children and mothers are directly related to the improvement of children’s health and education (Malapit and Quisumbing 2015; Sraboni and Quisumbing 2018; Holland and Rammohan 2019; Joneset al.2019). However,poverty reduction policies have mostly focused on men,which has widened the gap in productivity and income between men and women and greater gender inequality. Meanwhile,gender differences have also not been fully accounted for in the traditional poverty analysis,design,and monitoring system,which has dramatically reduced the effectiveness of poverty reduction. Therefore,it is necessary to analyze and address poverty issues from the perspective of gender.

Despite the theoretical potential of various development strategies to simultaneously promote women’s empowerment and poverty reduction,there are mixed messages in the existing literature. For example,microfinance is wellknown to be an important means to empower women and address poverty,usually combined with other thematic areas (Taylor and Pereznieto 2014). However,there is no consistent conclusion about microfinance’s effect on women’s empowerment (van Rooyenet al.2012; Ganleet al.2015; Hossainet al.2016). While some researchers found microcredit programs had positive effects on women’s empowerment (Lyngdoh and Pati 2013; Raphael and Mrema 2017),others found they had no effect or negative effects.One possible reason for these discrepancies is the usage of different measures for empowerment (Supriya 2013).In addition,the fact that women have little control over productive assets may account for their disempowerment(Supriya 2008). The seemingly paradoxical findings from previous studies suggest that the impacts of microfinance programs depend on the context in which they are implemented and how they are implemented.

Another example is agricultural cooperatives,which are also increasingly promoted as part of rural development strategies to empower women. Many studies have examined the potential positive role of cooperatives in women’s empowerment (Ferguson and Kepe 2011; Ataei and Miandashti 2012; Biru 2014),but the conclusions are also mixed (Mayoux 1992,1995; Paryabet al.2014).Moreover,membership of a cooperative has heterogeneous impacts,implying rural financial markets can maximize the potential positive impacts of cooperative services on farmers’ productivity and welfare (Wossenet al.2017),which suggests that the intervention portfolio might generate opportunities to effectively empower women and further contribute to poverty reduction. In addition,training is another strategy,which can improve women’s empowerment on aspects of increased control beliefs and intra-household decision-making power (Huiset al.2019). Overall,access to resources (including credit,training,and social networks)enable women to generate income and fight against poverty,but it is not sufficient (Kabeer 1999; Cornwall 2016). The relationship between women’s economic participation and empowerment is complicated; increasing women’s incomes does not necessarily balance gender relations (Bailur and Masiero 2017).

This paper uses a dataset from one of the programs of the International Fund for Agricultural Development (IFAD)in China to detect the simultaneous impact of a multicomponent program on women’s empowerment and poverty reduction. The Inner Mongolia Autonomous Region Rural Advancement Programme (IMARRAP) became effective in 2008 and ended in 2014. Its objective was to achieve poverty reduction in the program area in a sustainable and gender-equitable way. Generally,women in the program area faced many constraints on their participation in economic activities. They had limited access to resources such as land,credit,technical advice,and the market,which restricted their capacity to make a living and made them more vulnerable to various risks. In this context,the program aimed to reduce poverty through a “gender lens”and provided training and credit to support women.

After the initial sample design using village-level propensity score matching (PSM),project villages and nonproject villages were chosen based on a series of indicators such as the distribution of investment and beneficiaries.Then direct beneficiaries and non-beneficiaries were randomly selected in two different groups of villages,respectively. In addition,indirect beneficiaries were also identified and selected to detect possible spillover effects.They were defined as the households that did not participate in the program,which were adjacent to direct beneficiary households. The data were collected in nine poor counties of Ulanqab City at the end of 2014,where the program was implemented,yielding a sample of 900 households.To better target women beneficiaries,households in which women reported their participation in the program were further identified as being in the direct beneficiary group.To better control the selection bias,household-level PSM,inverse probability weighting (IPW),and the difference-indifferences (DID)-PSM method were used to analyze the effect of the program as well as provide robust checks.Recall questions were used for the baseline data as neither the beneficiary groups had a comparison group nor provided the necessary outcome indicators. The respective indicators of women’s empowerment were measured using intra-household decision-making scored in the dimensions of livelihood. Different poverty incidences calculated based on income,consumption expenditure,and the wealth index were used as the objective measurements of poverty and households’ own judgments of their economic status as the subjective measurement.

The results show women are empowered due to participation in the program,and poor households potentially move out of poverty due to empowered women. This paper employs an empirical strategy to analyze the effect of a package of interventions,including microcredit,training,and association on women’s empowerment and poverty reduction. With women assuming an ever more important role in agricultural production in China,the results will also provide evidence for researchers and policymakers to better understand the relationship between women’s empowerment and poverty reduction.

The paper is organized as follows. Section 2 provides a brief introduction of program context and raises the hypothesis that needed to be examined under the empowerment framework. Section 3 presents the methodology to detect the relationship among program participation,women’s empowerment,and poverty reduction,including the data and sampling strategy,empirical strategy,indicator definition,and descriptive statistics. Sections 4 and 5 report the results of the empirical models and discuss the main findings.Section 6 summarizes and concludes the findings.

2.Context and framework

2.1.Program area

Ulanqab City is located in the middle of the Inner Mongolia Autonomous Region of China,with its longitude between 40°10´N and 43°28´N and latitude between 110°26´E and 114°49´E. By the end of 2006,the city was administratively divided into Jining District,Fengzhen City (county-level),and nine rural banners or counties. All nine banners or counties involved in this study were Chayouqian Banner,Chayouzhong Banner,Chayouhou Banner,Siziwang Banner,Shandu County,Huade County,Xinghe County,Zhouzi County,and Liangcheng County. Ulanqab belongs to the mid-temperate zone and experiences a continental climate. The annual average temperature is around 3°C,and there are just 125 frost-free days. In 2008,the total area of land in Ulanqab was around 5.45 million hectares,of which 11% was arable land,8.2% was forest,and 63%was grassland (Jin 2009). Major crops included potatoes,maize,vegetables,oilseeds,and sugar beet. The per capita net income per year for the rural population (including herdsmen and farmers) increased from 2 003 CNY1100 USD=686.91 CNY (2020).(291.6 USD) in 2000 to around 4 061 CNY (591.2 USD) in 2008(Jin 2016),and income for herdsmen was traditionally much higher than for farmers engaged in crop growing.

In the program area,women had difficulties in accessing resources such as new technology,extension services,market information,and credit. More and more farmers had been migrating from rural to urban areas before the program implementation,of which most of the migrant workers were male. Under these circumstances,women were therefore rapidly becoming the main labor force for farming activities. But women’s access to technical services such as extension and skills training was very limited. Few women were able to access such services provided through the formal agricultural extension channels. On the other hand,most rural women were highly motivated to improve the livelihoods of their households. But women suffered from relatively limited access to credit. Most loans provided by formal financial credit institutions were to men. Some poor women even had to pay high interest through informal loan channels.

In response,the program attached importance to gender focus at all levels of implementation and management.First,there was a good proportion of women (about 35%)in the program management team. Second,women had equal opportunities or priority to access technologies such as greenhouses for vegetable production and net sheds for potato production,extension services including crop planting and animal feeding,and the market. Third,specific modules were designed to provide strategic support for women; for example,the establishment of village-level micro-credit groups and the provision of training to assist women to establish income-generating activities.

2.2.Conceptualizing empowerment

From the previous studies,empowerment can be seen as an approach to address poverty and gender inequality.In the context of poverty reduction,empowerment refers to the expansion of assets and capabilities so that poor people can participate in,negotiate with,influence,control,and hold accountable institutions that affect their lives(Narayan 2002). Compared with conventional anti-poverty approaches,an empowering approach is grounded in the conviction that poor people themselves are invaluable partners for development (Narayan 2005),with the emphasis on reducing inequality through improving human capabilities and the distribution of tangible assets (Narayan 2002).

As one of society’s empowerment subsets,the empowerment of women has its unique concerns such as cross-cutting category and intra-household relations(Malhotra and Schuler 2005). In the context of gender consideration,women’s empowerment is about the process by which those who have been denied the ability to make strategic life choices acquire such an ability (Kabeer 1999). Agency is generally viewed as the essence of women’s empowerment (Malhotra and Schuler 2005),which includes processes of decision-making and lessmeasurable manifestations such as negotiation,deception,and manipulation (Kabeer 1999),though actors may be constrained by their opportunity structure,which may influence their ability to transform agency into action (Alsopet al.2006). Women always experience development programs differently from men and are affected differently.The existing evidence suggests that financial services and training programs are not gender-neutral and that genderspecific design can yield more positive economic outcomes for women (Buvinić 2019).

The empowerment of women and poverty reduction are intertwined with each other under the framework of empowerment. A growing amount of evidence shows that empowerment is instrumental in reducing income and consumption poverty (Alsopet al.2006). As a member of a poor household,empowering a woman has the potential to empower the whole household.

This paper aims to examine whether empowering women through interventions (training,cooperatives,financial services,etc.) could help achieve the poverty reduction goal. Empirically,we examine the effect of the program on women’s empowerment and poverty reduction simultaneously. In detail,there are several steps as follows(Fig.1):

The first step is to examine whether the program had positive effects on women’s empowerment and poverty reduction simultaneously at the household level. In this step,the participants might be both female and male members in a poor household. Therefore,the possible positive results may imply that empowered women have contributed to poverty reduction,but the effect on poverty may not only come from female beneficiaries but also from male beneficiaries.

The second step is to examine whether the program has positive effects on women’s empowerment and poverty reduction simultaneously at the individual level. In this step,the participants are just the female members of a poor household. Therefore,the possible positive results may imply that empowered women have contributed to poverty reduction.

Fig.1 The roadmap of this study.

The third step is to discuss the empowerment process under different gender-specific interventions such as training,cooperatives,and micro-credit services. In this step,the possible reasons for women’s empowerment through interventions and how they contribute to poverty reduction are explored.

3.Data and methods

3.1.Data and sampling strategy

The data were collected in nine poor counties of Ulanqab City in the Inner Mongolia Autonomous Region of China at the end of 2014,yielding a sample of 900 households.Prior to conducting the household survey,a series of steps were taken to facilitate an evaluation of the program. First,treatment villages (project villages) were identified based on the distribution of investment and beneficiary households as well as the local coordinator’s opinion. Second,control villages (non-project villages) that were most comparable to treatment villages were identified using PSM with the baseline data of 2008,conditional on a set of observable characteristics such as per capita income,per capita arable land,landform,and main production type. Third,after discussion with key local organizations,a final village list comprising 60 villages (30 treatment villages and 30 control villages). Fourth,within each treated village,there were households who participated in the program(participants) and others who did not (non-participants).Since non-participants live near the participants,they may obtain indirect benefits from the program. A useful group of households were identified to detect the potential spillover effects as well as avoid potential bias in the estimates of impact,otherwise the estimates may reflect fundamental differences between participants and non-participants rather than the impact of the program. The final sample,therefore,included three sets of households:(1) the direct beneficiary households of the program,with 10 households randomly chosen from selected treatment villages based on lists of direct beneficiary households provided by village leaders; (2) indirect beneficiary households in the treatment village,with neighbor households of the direct beneficiaries chosen from the treatment village; (3) non-beneficiary households in the control villages,with 10 households selected randomly from each selected control village.Finally,the sample included a total of 900 households of which 600 were in the treatment villages (300 were direct beneficiaries and 300 were indirect beneficiaries) and 300 in the control villages.

For projects which lack valid baseline data,efforts should be made to construct baseline data through respondent recall (Bamberger 2009). Recall was used in this study as the baseline data of 2008 did not have a comparison group; in addition,inadequate baseline data were collected at the household level on outcomes. The recall period was indexed to a landmark event,the Beijing Olympic Games coinciding with the baseline year,which ensured some variables such as household asset ownership could be reliably remembered. Again,specific training and pilot interviews were performed to validate the recall data.

To avoid biases due to the effects of household structure on decision-making outcomes,we restricted our analysis to dual-adult households. In all,816 dual-adult households(those with male and female adults) were selected from all 900 households,excluding single-male and single-female households.

Among the direct beneficiaries,rural households in which only women were participants of the program were identified,whatever they access to training,credit,and other resources from the program. While in the indirect beneficiary group and non-beneficiary group,no women participated in the program directly. As a result,there were 177 households in the direct beneficiary group,237 households in the indirect beneficiary group,and 251 households in the nonbeneficiary group.

3.2.Empirical strategy

The estimate of the treatment effect may be biased due to possible non-random treatment assignment in observational studies. PSM is a statistical matching technique that minimizes bias and identifies every possible observation under the treatment group and constructs a comparison group with similar observable characteristics. The implementation of PSM requires two steps. In the first step,the propensity scoresp(Xi) of households has been estimated with a logit model containing the explanatory variablesXiof the probability of enrolling in the program(Ti=1 if a household is enrolled in the program andTi=0 otherwise). The estimated propensity scores are then used to construct the comparison groups.

In the second step,the average treatment effect on the treated (ATT) has been estimated. The final estimator for this average treatment effect is obtained as the average of the differences in the situation of households in the treatment group and their counterfactuals. The mean difference of the two groups should be statistically significant to show an effect on the surveyed households.

whereYi1is the potential outcome whenTi=1;Yi0represents the counterfactual outcome whenTi=0.

There are many different methods to find the “similar households” in the control group with propensity scores close to those treated households. Among them,the nearest neighbor method chooses counterfactual households for each treated household who is closest in terms of the propensity score. Nearest neighbors are not determined by comparing treated observations to every single control,but rather by first sorting all records with the estimated propensity score and then searching neighbors forward and backward for the closest control unit. In addition,balancing tests for PSM should be used to confirm the validity of matching.

Inverse probability weighting (IPW) is an alternative to PSM,which is also used to reduce the bias caused by nonrandom treatment assignment. It tends to perform better than even the most effective matching estimators in finite samples (Bussoet al.2014). IPW refers to weighting the outcome measures by the inverse of the probability of the individual with a given set of covariates being assigned to their treatment. The weights are constructed based on propensity scorep(Xi). They are 1/p(Xi) for the treated participants and 1/[1–p(Xi)] for the untreated participants.

Since the systematic differences in unobservable variables may still bias PSM estimators,several extensions have been proposed as a response. One of the extensions is the difference-in-difference PSM estimators,which is an attractive estimator because it permits the selection to be based on potential program outcomes and allows for selection on unobservable variables (Heckmanet al.1997).Here,t1represents the time period to when the program was completed andt0denotes a time period from when the program started. The conditional difference-in-differences estimator compares the conditional before–after outcomes of program participants with those of non-participants.It extends the conventional difference-in-differences estimator by defining outcomes conditional onXand using semiparametric methods to construct the differences.

The key assumption for standard DID is that the outcome in the treatment group and the control group would follow the same time trend in the absence of the treatment. First,a placebo test and DID estimation using different comparison groups can be used to check the assumption of equal trends. Second,for the PSM-DID strategy,a standard DID is done using propensity score-matched groups. The point of matching is to provide robustness against the potential violation of the parallel trends assumption.

3.3.Key indicators

There are two main outcome focuses:poverty reduction and women’s empowerment (Table 1). With regard to the empowerment of women,its measurement is still being debated. Some of the key methods that researchers have used to measure empowerment include Gender Development Index (GDI),Gender Empowerment Measure(GEM),and Women’s Empowerment in Agriculture Index(WEAI). Except for these indexes,the extent to which women participate in intra-household decision-makingprocesses is often used as a measurement of women’s agency (Donaldet al.2017) and women’s empowerment(Alkireet al.2013; Seymour and Peterman 2018). The Decision Power Index introduced in 1960 asked questions about “who has the final say” in respect of eight family decisions and weighted the answers from 5 (husband always) to 1 (wife always) (Donaldet al.2017). Similar to this idea,this paper scored on women’s decision-making role within the household in six domains,including (1)agricultural production,(2) access to and control over productive resources,(3) access to and control over household assets,(4) access to and control over food and high-frequency non-food items,(5) access to credit,and(6) social connection in the community. For each domain,score was assigned in terms of the following rules (Hwanget al.2011):male has the final say alone=1,male has more final say than woman=2,male has equal final say with woman=3,male has less final say than woman=4,woman has the final say alone=5. The total score was calculated with equal weights to measure the women’s decision-making power. Another method to measure women’s empowerment in this paper is to calculate the disempowered headcount following the main idea of WEAI (Alkireet al.2013) but only with the decision-making module. All adequacy indicators in the above six domains were first coded,assuming that the value 1 if the individual lacked adequate achievements in that indicator and zero otherwise. An inadequacy score was then computed for each person,according to his or her inadequacies across all indicators. Assuming equal weighs for simplicity,a second identification cut-off was set to identify who is disempowered. After exploring the sensitivity of the empowerment classification for different cut-offs,we selected a disempowerment cut-off of 20% (Alkireet al.2013). For those whose inadequacy score was less than the disempowerment cut-off,their score was replaced with zero. Then the disempowered headcount ratio was the total population divided by the number of individuals who were disempowered.

Table 1 Brief description of key outcome variables

With regard to poverty,it can be measured using objective and subjective indicators. The objective measures focus on people’s access to different kinds of resources,and the subjective measures emphasize the standard of living people actually enjoy. From the objective aspect,there are three principal indicators of economic status:household income,household consumption expenditures,and household wealth(Rutstein and Johnson 2004). The use of asset indices as proxies for welfare,wealth,economic status,and living standards has rapidly become very popular in social studies.The method has been introduced in the analysis of poverty(Sahn and Stifel 2000; Filmer and Pritchett 2001). This paper used income-based,consumption expenditure-based,and assets-based poverty incidences to measure poverty status respectively and objectively. First,we used per capita rural household net income per year of 2 300 CNY (334.83 USD) in 2010 constant price as the income poverty line to calculate the income-based poverty incidence. Second,the expenditure-based poverty incidence was calculated using the World Bank’s household consumption poverty line. Two poverty lines were selected in accordance with the program period. In 2008,the World Bank updated the international poverty line to 1.25 USD per capita per day at 2005 PPPs.The international poverty line was adjusted at the Chinese 2008 constant price using the PPP conversion factor and the official Chinese consumer price index. In 2015,the World Bank again updated the international poverty line to 1.90 USD per capita per day at 2011 PPPs. Third,the wealth index was composed of key asset ownership variables such as the household ownership of a number of consumer items and dwelling characteristics. Then five wealth quintiles were divided equally,and the lowest quintile was defined as poor. From the subjective aspect,poverty can be defined by examining which people consider themselves poor or defined by collecting peoples’ beliefs about their position in a system of inequalities (Nándori 2010). The economic status of rural households in the program area was classified into four categories:better-off (A),ordinary (B1),poor (B2),and very poor (C). The categories of very poor and poor were considered as poor households. Likewise,sampled households gave us their subjective answers about which categories they belonged to. We viewed these answers as the subjective measurement of poverty. Also,we measured the very poor category separately.

We used a number of covariates that were believed to be associated with program participation,covering a set of household characteristics such as household size,age of household head,education of household head,number of children less than 5 years old,number of children aged between 6 and 14 years old,number of elderly people who were more than 64 years old,number of female workers,number of agricultural laborers,size of arable land,and access to credit.

3.4.Descriptive statistics

Tables 2 and 3 present descriptive statistics of rural household characteristics in the full sample and various subgroups (G1,direct beneficiaries; G2,indirect beneficiaries;G3,non-beneficiaries; G4,indirect beneficiaries and nonbeneficiaries),and detect the differences between the direct beneficiaries and its different counterpart groups in 2014 and 2008. From the data for the full sample in 2014,shown in the first column,we can see nearly three persons in a household on average. The average age of the household head was about 57 years old. The formal schooling ofhousehold heads was,on average,slightly more than six years. Few households had young children who were less than 14 years old,but most households had old people who were more than 64 years old. More than half of the household members could work as agricultural laborers.The average per capita arable land was around 0.51 ha.There were around 26% of households who had credit in 2014. The per capita expenditure was 14.50 CNY per day

on average,and 45% of women were disempowered. The poverty status varied according to different poverty lines and methods.

Table 2 Descriptive statistics in 20141)

Table 3 Descriptive statistics in 20081)

Compared with 2008,household livelihoods had improved in 2014. From the subjective measurement of poverty,we can see that poverty incidence (S1) decreased from 67% in 2008 to 30% in 2014,and poverty incidence(S2) decreased from 10% in 2008 to 4% in 2014. In addition,credit access doubled from 2008 to 2014 due to rural finance intervention.

From the results of thet-test of differences between the direct beneficiary group and various counterpart groups,we can see that there were many significant differences among groups,which suggests no group could provide a good counterfactual. The factors which affected program participation may potentially influence the outcomes we wanted to evaluate. If so,the difference in outcome was not the true impact of the program. For example,there were obvious different distributions in the wealth quintiles between the direct beneficiary group and the non-beneficiary group. The economic status in the direct beneficiary group seemed to be better than that of the non-beneficiary group. The selection bias might be a possible reason for the significant difference in economic status between the direct beneficiary and non-beneficiary groups. The program aimed to provide technologies,extension services,and credit for poverty reduction,which targeted households who were poor but with potential production capacity,so some absolutely poor households targeted for assistance by the government might not be included in the program.That is to say,these households were potentially excluded from the direct beneficiary group,but households in the non-beneficiary group were randomly selected,they could be contained in the non-beneficiary group.

4.Results

4.1.lnfluencing factors of participation in the program

As shown in Table 4,we use a logit estimator with an odd ratio to explain the determinants of participation in the program in 2008. The selected variables can predict 74.14% of outcomes in the full sample of 665,in which non-participants came from the indirect beneficiary group and the non-beneficiary group (G4). Consistent with descriptive statistics above,the estimation results show that the education of household head,number of elderly people more than 64 years old,number of agricultural laborers,and access to credit have significant positive effects on program participation. The results indicate that participants are likely to have higher levels of education and access to credit. And participant households have more agricultural laborers and old people than non-participant households. In addition,the age of the head of the household is significant and negative,which means that younger heads of households are more likely to participate in the program. Moreover,other variables,such as the number of children aged between 6 and 14,are significant and positive when non-participants only came from the indirect beneficiary group (G2). Using these variables,propensity scores can be calculated for the different groups,and more estimates based on the propensity scores are further considered to ensure an unbiased estimate of impacts.

4.2.Program intervention,empowerment of women,and poverty reduction

Different model specifications,including PSM,IPW,and DID,are used to evaluate at the household level the impact of theprogram on women’s empowerment and poverty reduction simultaneously. Table 5 presents the average treatment effects of the program on the treated households for a selection of indicators such as intra-household decisionmaking,women’s disempowerment headcount ratio,and various poverty incidences based on expenditure,wealth,and self-reporting,considering direct beneficiaries (G1)as the treatment group and non-beneficiaries (G3) as the control group. As expected,the size of effects and the degree of significance vary,but signs are generally similar across specifications. According to the results on women’s empowerment,all estimates of the decision-making score are significant and positive,which implies the program has contributed to increasing women’s decision-making power.The PSM and IPW estimates of the disempowerment headcount ratio are significant and negative,suggesting that women are empowered due to their participation in the program. With regard to poverty reduction,the program has a significant and positive effect on expenditure using PSM,IPW,and DID approaches. All IPW estimates except the self-reported very poor category are significant and negative,which indicates that the number of poor households declines due to the program. Viewed from the asset-based poverty incidence (W),the program changes the distribution of poor households between the direct beneficiary group and the non-beneficiary group. All estimates denote that more poor households in the direct beneficiary group have been lifted out of poverty,and accordingly the number of poor households increases in the non-beneficiary group. From the subjective measurement of poverty,all estimates show that poor households have moved out of poverty because of the program.

Table 4 Influencing factors of participation in the program1)

A number of sensitivity analyses were also carried out to check the results and explore the potential spillover effect.Given the significance of most IPW estimates in Table 5,Table 6 only reports the results using the IPW approach among different pairs of comparison groups. First,in the column of “G1vs.G2”,we replace the control group with the indirect beneficiary group (G2). The results show that almost all estimates are lower than what we have seen in Table 5,a consistent sign indicating potential spillover effects in the G2. Second,in the column of “G1vs.G4”,we change the control group into the indirect beneficiary and non-beneficiary groups (G4). The effect’s size is located between the result of the first column in Table 6 and the third column in Table 5 for nearly all indicators,as expected. Third,in the column “G2vs.G3”,the results are very different from others in Table 6 because there are no direct beneficiaries in this pair of groups. The estimates are nearly insignificant,which provides information that no other interventions except the program may differ for all indicators in the program area. The results also imply that the spillover effects haven’t turned into substantial effects on the outcome indicators. Fourth,in the column “G5vs.G3”,it is reasonable to put direct and indirect beneficiary groups together (G5) to detect the program’s total effect due to the existence of spillover effects.

4.3.Empowered women and poverty reduction

We can see from the results in Tables 5 and 6 that the program has simultaneously positive effects on women’s empowerment and poverty reduction,which to some extent provides evidence that empowering women is one of the mechanisms to fight against poverty,but the effect on poverty has not come exclusively from female beneficiaries as it may also come from male beneficiaries. To further examine the relationship between women’s empowerment and poverty reduction,we select a smaller group of households in the sample in which women were the onlyparticipants. Table 7 reports the robust results on women’s empowerment with direct beneficiaries (G1) in the treatment group and non-beneficiaries (G3) in the control group. Due to participation in the program,women’s decision-making power increases across different model specifications,and the ratio of the disempowerment headcount decreases under the estimation of PSM and IPW. Moreover,the estimates of per capita income (except the DID-PSM estimate) and income-based poverty incidence are significant,which indicates that empowered women have contributed to increasing household income and reducing income poverty accordingly. Intuitively,access to technology,extension services,market,and credit can help women involve in income-generating activities,which improves their ability to earn income and contribute to the household income.In addition,significant estimates of asset-based poverty incidence suggest that empowered women are conductive to improve living standards compared to other households in the sample. Again,the self-reported poverty incidence denotes that empowered women have a positive effect on poverty reduction. On the contrary,the estimates of consumption expenditure and expenditure-based poverty incidence are almost insignificant. One possible reason for this is that except income,consumption is also related to consumers’ preferences and habits,which are not easy to change in the short term.

Table 5 Effects of program participation on women’s empowerment and poverty reduction1)

5.Discussion

5.1.Training is a common means to empower women in the agricultural sector

Traditionally,women are good partners to their husbands with regard to agricultural production in China. Among the direct beneficiary households from the program modules,such as the technical envoy system,greenhouses,potatonet sheds,and strengthening village livestock service stations,91% of women engaged in income-generatingactivities along with their husbands. But as their husbands migrate to urban areas to find work,a growing proportion of women play an increasingly important role in agricultural production. However,a lack of formal education and training opportunities is a key factor to limit women’s access to new technologies in crop agriculture and the livestock sector.Field survey data showed that women’s educational level was nearly three years lower than men’s in the same household on average. Therefore,the training component was well-designed in the program to provide equal or more opportunities for women to gain new knowledge and skills.The number of women participants was systematically monitored as well. As a result,an increasing number of women were involved in technical activities due to project implementation. Data shows that more women were trained than men in most modules of the program.Women empowered through the training component have the potential to increase their share of family incomes and further strengthen their bargaining power in the household.Meanwhile,participating in stable income-generating activities can also increase rural women’s confidence and promote their participation in community activities.

Table 6 Effects of program participation among different pairs of groups based on inverse probability weighting (IPW)1)

Table 7 Effects of empowered women on poverty reduction1)

5.2.Microfinance support for women is an effective gender strategy to increase women’s role in the production

In the 1990s,microfinance programs have increasingly targeted women because access to credit can enable women to play a more active role in household decisionmaking and increase investment in family welfare. Some evidence shows that access to microfinance can initiate the virtuous spiral of economic empowerment,increased well-being,and women’s social and political empowerment(Cheston and Kuhn 2002). Therefore,the formation of Women’s Group Micro-credit (WGMC) became one of the important modules in the program,which aimed to raise poor women’s income and status by improving their access to credit and knowledge. Previous studies showed that resources and incomes controlled by women are more likely to be used to improve family food consumption and welfare,reduce child malnutrition,and increase the overall well-being of the family (García 2006). WGMC was more inclined to create opportunities for women to increase income and reduce poverty. Compared with household credit,gender-specific credit from WGMC showed a different result (Fig.2),which illustrates that gender-specific credit was most used to buy animals or productive materials.There are three possible reasons for this:First,raising animals is profitable,because the demands for meat and milk have been growing with the increase of population and improvement of living standards; second,livestock is one of competitive industries of the Inner Mongolia Autonomous Region and rural households there have the habit of raising animals; and third,small animals are always good choices for poor women to generate income and improve food and nutrition security.

5.3.Associations or cooperatives for women are grassroots hubs where productive resources can be channeled to enhance women’s capacity for selfimprovement

Fig.2 Different purposes of gender-specific credit and household credit.

Compared with men,women face a number of disadvantages as farmers,including less mobility,less access to training,less access to market information,and less access to productive resources. Some evidence suggests that women tend to lose income and control as a product moves from the farm to the market (Gurung 2006). Women’s association support was another module of the program designed to provide organizational,technical and market support to women’s credit groups. Members of the women’s association could gain training about association management,contract farming,processing,and marketing.They also had opportunities to build human and social capital and further increase their capacity to participate in village and township activities. In addition,the women’s association has the potential to raise women’s awareness to become members of other cooperatives. From the field visits,we can see that the number of women participating in producer cooperatives in the direct beneficiary group was far more than the numbers of the two other groups(the indirect beneficiary and non-beneficiary groups).For women,associations or cooperatives’ organizational empowerment impact may be more important than the direct economic benefits. The program has especially enhanced the skills,opportunities,and prestige of women who were active leaders in these organizations. Being leaders in producer cooperatives has improved women’s self-confidence in working collectively to realize positive changes to their communities.

6.Conclusion

In recent decades,more and more attention has been paid to gender differences in poverty analysis. Among them,microfinance,training,and cooperatives are the common means to empower women for poverty reduction. However,the results of existing studies have varied depending on how the researchers designed the program,identified the target group,measured the key indicators and chose model specifications. Based on a gender-focus program implemented in the Inner Mongolia Autonomous Region of China,this study has looked at women’s participation in the program for dual-sex households to detect the program’s effect on women’s empowerment and poverty reduction.The role of empowered women in poverty reduction has been examined by further identifying rural households in which women were the only participants.

In the initial sample design,project villages and nonproject villages were chosen based on village-level PSM.After the villages were determined,both direct beneficiaries and non-beneficiaries were randomly selected in the two kinds of villages. Especially,indirect beneficiaries were identified to detect possible spillover effects. They were defined as households who were adjacent to direct beneficiaries,but which did not directly benefit from the program. As a result,there are substantial differences between the direct beneficiary group and various counterpart groups. The logistic model’s estimation indicates that the education level of the head of the household,number of elderly people above the age of 64,number of agricultural laborers,and access to credit have significant positive effects on program participation. Other variables such as the age of household head and the number of children aged between 6 and 14 may also have influenced participation in the program. In order to control the selection bias,household-level PSM,IPW,and DID method were used to analyze the effect of the program as well as do robust checks,of which recall questions were used because the baseline data did not have a comparison group and outcome indicators.

The measurement of women’s empowerment and poverty were also key contents of this study. On the one hand,we calculated both women’s intra-household decisionmaking score and the disempowerment headcount ratio as the respective indicators of women’s empowerment.On the other hand,income,consumption expenditure,and wealth index were used as the objective measurement of a household’s poverty,and households’ judgments of their economic status were used as the subject measurement of their poverty. These indicators were considered together because we can observe and explain the results from different perspectives. Under the empowerment framework,this paper examined whether empowering women through interventions could achieve poverty reduction. The results show that empowered women positively affect poverty reduction,and training,microfinance,and women’s associations are important mechanisms for women to fight against poverty. Especially,gender-specific credit in the program area is most effective when used to purchase animals or productive materials for income-generating activities. Given gender differences have not been fully accounted for in the traditional poverty analysis,which has greatly reduced the effectiveness of poverty reduction,these results suggest researchers and policymakers need to analyze and discuss poverty issues from the perspective of gender.

Acknowledgements

This work was supported by the People’s Republic of China Inner Mongolia Autonomous Region Rural Advancement Programme led by IFAD,the National Natural Sciences Foundation of China (71661147001) and the Special Fund for Basic Scientific Research in Agricultural Information Institute of Chinese Academy of Agricultural Sciences(2020JKY040). This research would not have been possible without the support from IFAD that permit our team to conduct collaborative research. We acknowledge the farmers and the local extension workers in the study area for their help during the pilot testing survey and main fieldwork. We are grateful to Dr.Bi Jieying,Ms.Huang Jiaqi,Dr.Sun Mengyao,Mr.Zhu Cong,Dr.Zhang Xuebiao,and Dr.Samesh Adhikari from Agricultural Information Institute,Chinese Academy of Agricultural Sciences for their constructive contributions to evaluation design and data collection.

Declaration of competing interest

The authors declare that they have no conflict of interest.