With the high population growth rate of Afghanistan and the security situation, providing reasonable jobs to people becomes challenging because lower security condition makes it difficult to conduct businesses and decreases the incentive to start a new business and to purchase equipment. According to Afghanistan Living Conditions Survey 2016–2017 ( Islamic Republic of Afghanistan Central Statistics Organization, 2018), unemployment rate, the ratio of people who are seeking a job but do not have a job in the labor force, is 25 %. In particular, the male's unemployment rate is 18 %, whereas the female's unemployment rate is 41 %. Even among those who have a job, 19.7 % of males stated that they want additional work, and 23.7 % of females stated that they want additional work. Among those employed, the salaried worker is 17.2 % (including public sector). Specifically, unpaid male family workers are 24.5 %, and daily labor is 15.7 %. The ratio of people who are working for their own business is 40.1 %.

The security condition in Afghanistan is getting worse. According to the Special Inspector General for Afghanistan Reconstruction, in 2017, the middle year in the experiment, approximately 59.7% of the country's 407 districts were under Afghanistan government control or influence; 11.1% of total districts were under insurgent control or influence; 29.2% of all districts were contested districts. ( Special Inspector General for Afghanistan Reconstruction, 2017). According to the United Nations Assistance Mission to Afghanistan, the number of civilian death due to conflicts and explosives was 3701, 3565, 3510, and 3428, from 2014, 2015, 2016, and 2017, respectively ( United Nations Assistance Mission in Afghanistan, 2018). Meanwhile, the number of injuries caused by these activities for the same years was 6384, 7419, 7924, and 7015, respectively.

In terms of education, there is a substantial disparity between males and females, depending on the area. For example, in urban areas, 66.9% of males and 40.8% of females are literate, whereas in rural areas these values decrease to 45.6% and 13.1% for males and females, respectively. The poverty rate is also severe in Afghanistan. In 2016, it was reported that 54% of households had expenditure less than the poverty threshold.

Gender roles in both intra-household and business in Afghanistan are influenced by socio-political aspects ( Abirafeh, 2009). Moreover, Afghanistan has suffered from a power vacuum driven by the absence of a decisive central government, which prepared the ground for the influence of alternative socio-political and religious elements. These circumstances primarily applied to the rural areas, which comprise the majority of the country's social class ( Abirafeh, 2009; Barfield, 2010). In the following, two influential social and religious factors affecting Afghan women's access and control over resources within the household and social spheres that are widely practiced are presented: (i) the principle of the Purda (derived from Pashtunwali principles) and the (ii) Rigid interpretation of the Salafi sect.

The "Purda" principle restricts women's mobility outside the house. This principle is derived from the Pashtunwali; a quasi-legal system practiced among the Pashto ethnic in Afghanistan and Pakistan (particularly in the so-called Tribal Areas of the Northwest Frontier Province). ( Abirafeh, 2009; Barfield, 2010; Beath et al., 2013; Ginsburg, 2011; Hawkins, 2009). According to Ginsburg (2011) and Hawkins (2009), although women are among the fundamental elements of the Pashtunwali system (Purda Principle),

women are inferior beings who are subordinate to men and must be strictly protected for males' honor ( Ginsburg, 2011; Hawkins, 2009).

The implementation of the Salafi (a branch of Sunni Islam) religious sect has negatively impacted women's access and control over resources at both domestic and community levels. Therefore, this interpretation emphasizes restoring the lifestyle and tradition of the Prophet Muhammad (from the 7th century AD) and fighting against any religious innovation (so-called bidah) in terms of interpretation ( Barfield, 2010) ( Abirafeh, 2009), including women's engagement in various socio-economic activities that requires presence in the mixed environments with other men.

Due to those two aspects, there were no changes in individuals' attitudes about women's role in the intra-household decision-making and the general status of women in society in the past two decades. Meanwhile, increasing insecurity exacerbated restrictions on women's mobility. Furthermore, no positive effect was found concerning the positive attitude toward women's participation in the village governance in Pashto-majority areas, and this attitude difference was significant compared with other ethnolinguistic districts. ( Beath et al., 2013)

Randomized controlled trial experiments have been conducted to examine the effect of receiving a non-trivial amount of cash grant to relieve the capital constraint. In the Ultra Poor (TUP) project in Bangladesh from 2007 to 2011, Bandiera et al. (2013) examined the effect of giving a non-trivial amount of cash grant. The recipient group were females living in chronic poverty and either chosen to receive a one-time cash grant of 140 USD (= 9500 TK) or, the equivalent, two cows (or sometimes one cow and five goats). The authors found that the recipient group's profit increased by 38 percent compared to the control group and was sustained even after the support was stopped.

In a randomized trial control conducted in Tanzania, Berge et al. (2015) examined the effect of an experiment combining two programs: the support to access financial resources and giving business training for small-scale male entrepreneurs. The authors compared the effect of combining two programs with the effect of implementing one of the above two programs. They find that the effect of combining two programs is stronger than the effect of implementing only one program. However, this stronger effect is not observed among female microentrepreneurs. As a potential reason, the authors suggest the importance of non-cognitive abilities and family constraints.

In connection to alleviating capital constraints, Banerjee et al. (2015) conducted a similar field experiment to an experiment conducted by Bandiera et al. (2013) in six countries on three continents (Ethiopia, Ghana, Honduras, India, Pakistan, and Peru), enabling access to capital among 10,495 male and female living in chronic poverty. The study showed the positive effect of access to capital on self-employment activities, particularly in terms of increasing productive assets, income, and revenue. The results were consistent in nearly all countries.

In an experiment conducted by Blattman et al. (2014) in Uganda from 2007–2012, the authors examined the effect of cash grants on youth employment among illiterate young males and females aged 13–35. The target groups had to form new groups at the time of application, 265 groups were formed to receive the grant, and 270 groups were also formed as control groups. The treatment groups were given an amount of 382 USD per group member in a lump sum. The authors observed that the workload, income, and working hours of the treatment group significantly increased; however, there was no evidence of changing social cohesion in the study.

Haushofer and Shapiro (2016) studied the effect of unconditional cash transfer among impoverished households in rural areas of western Kenya. They found a positive effect of the unconditional cash transfer on self-employment, asset holdings, consumption, food security, and psychological well-being. However, they also found that the monthly and lump-sum transfers yielded different effects. The author found that smaller transfers positively affected food security more than larger transfers. Regarding the gender differences in the initiative's impact, the author observed an increased spillover effect among female recipient groups over the control group in treatment villages, suggesting that alternative methods such as in-kind assistance or skills transfer are superior to cash transfer as a female empowerment tool. In a nutshell, studies demonstrate the positive impact of unconditional cash transfers on self-employment and promoting small-scale entrepreneurship in developing countries. However, the positive effects alter depending on additional elements such as gender, frequency and amount of cash transfers, and the inclusion of other initiatives such as business training. Furthermore, based on the obtained results, despite all the conditions being the same, the outcome among female recipients was more diminutive than male recipients', suggesting the more significant influence of gender differences impacting the return of unconditional cash transfers.

In summary, those studies demonstrate the positive impact of unconditional cash transfers on self-employment and promoting small-scale entrepreneurship in developing countries. However, the positive effects depends on additional elements such as gender, frequency and amount of cash transfers, and the inclusion of other initiatives such as business training. Furthermore, based on the obtained results, despite all the conditions being the same, the outcome among female recipients was less clear than male recipients, suggesting the more significant influence of gender differences affecting the return of unconditional cash transfers.

The program "Promoting Entrepreneurship among the Youth in Afghanistan" (PEYA) was an RCT-based project funded by the Afghan Reconstruction Trust Fund. It was launched in August 2015 and closed in July 2018. The PEYA program was implemented by the Ministry of Labor and Social Affairs of Afghanistan in Balkh, Kabul, and Nangarhar provinces. The program's objective was to help establish new micro-enterprises or expand existing businesses by providing a cash grant of USD 500 (approximately 33,500 Afghan (Afs)), which is equivalent to 2017 GDP per capita in the country to zero or low education individuals. In this study, 3490 eligible applicants were chosen depending on their business proposals. Based on a lottery process, half of the study participants (1745) were assigned to the treatment group where they received a cash grant, whilst the other half who were assigned to the control group did not receive a cash grant. The lottery and the baseline interview were conducted in August 2016. Due to the potential lack of sufficient numbers of mentors who could give mentorship on businesses, mentorship on businesses was randomly given to the half of individuals who received the cash grant and lived in Kabul ( Figure 1).

As presented in Table 1, the key eligibility criteria for the applicants were male and female aged between 18 – 35 years, who were illiterate or had a low level of school education, and who resided in one of three provinces: Balkh, Kabul, and Nangarhar. The selected applicants had to have a national ID (Tazkira) and a bank account. This project was implemented through the collaboration of four teams: i) the Non-formal Approach to Training, Education, and Jobs in Afghanistan team, ii) the Provincial Directories of the Ministry of Labor and Social Affairs, iii) the Employment Service Centers (ESCs) (set up for better coordination of the intervention), and iv) the internal implementing team of the Ministry of Labor and Social Affairs.

The selection process of the grantees started by providing information about the intervention and eligibility criteria. Four districts in each province were specified. The operation teams were assigned to mobilize the communities in these areas. This process lasted for 30 days, and each community was visited twice. During the initial briefing (week 1) and refresher briefing (week 2), information about the objective and selection criteria was also shared with local authorities, such as the governor's office staff, provincial councils, and community leaders.

The application form was a three-page questioner containing 19 questions that covered six sections (skills, support, and investment; product/services; price; place; promotion; customers) (Underlying data) ( Kashefi & Naito, 2021a). The answers to most questions were closed-ended; however, there were several questions that had open-ended options to provide space for detailed answers. Each question and option had equivalent marks. Completion of the questioner on average took 15–20 minutes. Quality assurance was made by the training and moderating committee at the local level. Collectedly, the project had 8238 applications from the three provinces.

Considering the low level of literacy within the communities, application forms printed on large banners were also provided in addition to conventional brochures and posters. Subsequently, the applicants were selected based on an evaluation of their initial submitted business plan, which included information, such as education, age, demographic characteristics, income, work experience, business type, and their plans on how they would spend the grant. In many cases, the applicants were assisted by other family members, neighbors, etc., who had a higher level of education and were able to convey the information in the application form. The applications with a higher score than the threshold level of the score become qualified for the lottery process. After the qualified applicants were selected, scheduled meetings for baseline interviews and lottery were announced to be held in each province. At the meetings, professionally trained interviewers conducted the baseline interviews. After which each qualified applicant draws a lottery number that determines the receipt of the cash grant.

Given the amount of the cash grant and concern about corruption in the process of choosing the recipients, it was important that all applicants understood that the recipients of cash grants were chosen fairly. Thus, the process of lottery took place in the presence of representatives from provincial councils, provincial governors, the Implementing Partner Agency (IPA), ESCs, and representatives of provincial directorates of labor and social affairs. After determining the recipients of the grant, a three-member panel verified the information provided, and the grant agreement was signed. Finally, the full grant was transferred in one transaction to the grantees' submitted bank account.

Regarding the endline, the evaluation teams started to interview individuals among those who participated at the baseline in June 2018 and closed in July 2018. The teams interviewed 3005 individuals. However, due to miscommunication between the Ministry of Labor and Social Affairs and the external teams, the external team did not record the ID numbers for some surveyed individuals in the endline which could match individuals of the endline with those in the baseline. To ensure that correct matching was implemented, we included only individuals whose ID number and national identity number were matched in the baseline and the endline data in the final analysis sample. As a result, the final sample size was 2177 individuals (n=2177) in this study, with the number of the individuals in the treatment group (1146) slightly higher than that of the control group (1031) at the endline ( Table 2), implying 38 % attrition rate.

As the PEYA program is a Randomized Control Trial (RCT)- Experiment, we estimate the effect of the policy by estimating the intent to treat (ITT) with the following analysis of covariates specification:

where i is the index of an individual, Y _i,A is the outcome variable in the endline period (e.g., monthly earnings in the last month) for individual i, Y _i,B is the outcome variable at the baseline of individual i, and Grant _i is the dummy variable which is equal to one if individual i received the grant and zero if not.

As discussed in the Experiment's setting, the grant is randomized at each individual level. Thus, to calculate the standard error, we use the robust standard error. β ₁ captures the impact of the cash grant; X _i,B is a set of control variables at baseline, which includes age group dummy, province dummy, years of schooling, and zero education dummy; and ε _i refers to the individual error term. We estimate the above equation using ordinary least squares (OLS). In order to address the potential bias that can occur due to the high attrition rate, several robust bounding methods are applied in this study ( Lee, 2009; Manski, 1990). Stata, a statistical software, is used for our analysis.

Table 3 shows the summary statistics of baseline variables and the balancing test of the sample used in the study analysis. Panel A shows that 72% of the individuals are male and approximately half of the sample are between age 18 and age 23. Of the 2177 participants, 42% were in 18-23 age group. Moreover, 54% of the participants were married at baseline. Approximately 47% of them were heads of the household at baseline.

Panel B shows the geographic distribution of the three provinces and their native language. The geographic distribution is evenly apportioned between the treatment and control groups. This panel shows that approximately 51 % of the participants have stated Dari language as their mother tongue, 46 % of them were Pashto native speakers, and 3 % were minority language speakers (i.e., Turkemni, Uzbeki, and Pashai).

Panel C shows income-related information at the baseline. The results indicate that 92 % of the individuals in our study were new to business, implying that they had not started their businesses yet at the baseline. Specifically, the average income in the last month at the baseline was approximately 4,000 Afs. The P-value of the joint significant test, which is 0.412, shows that our analysis sample was balanced.

Figure 2 shows the histogram of monthly income for those who had received cash grants and those who had not at baseline. According to estimates by the National Statistical Office in a 2016–2017 study, the poverty line threshold was 2,064 Afs per person per month ( Islamic Republic of Afghanistan Central Statistics Organization, 2018). Thus, the average income in our study sample was two times higher than the income at the poverty line. However, as shown in Figure 2, a substantial percentage of the population had zero income. Figure 2 also shows that the distribution of income was quite similar between those who received cash grants and those who did not.

Figure 3 (a) and 3(b) display the histogram of monthly earnings for the male and the female groups, respectively. Figure 3(a) and 3 (b) show that the distributions of earnings for the male sample and the female sample are quite different. For example, for males, less than 20% of the male sample have zero earnings. However, for females, more than 70 % of the female sample have zero earnings.

Table 4 shows the mean and balancing test of the key variables in the male group. About 42 % of this group were without education, and 57 % were married. The average monthly income for this group was 5130 Afs, although 17 % were without income.

Table 5 shows the mean and balancing test of the key variables in the female group. About 47 % of this group were without education, and 48 % were married. The average monthly earnings of this group was 5130 Afs; however, 76 % were without income.

Table 6 displays the balancing test regarding the mentorship to be given among those who received cash grants and who lived in Kabul. The p-value of the joint significant test shows that the data is balanced between the group who received the mentorship and who did not.

Table 7 displays the index of different types of assets between the control and treatment groups. These assets were divided into Group 1: productive durable, including animal ownership, and local transportation equipment (e.g., a cart, three-wheel/truck, and bike), and Group 2: consumer durable, including household goods (e.g., home appliances and mobile phones. To construct household asset-related indices, the values of each variable were standardized to ensure that the value of each variable represents its difference from the mean. Then, the variables are compiled into indices using factor analysis.

In our analysis sample, according to Table 7, the p-value of the joint significant test is 0.49, and it shows that the mean was balanced between the control and treatment groups.

Despite the high attrition rate, overall, the balancing tests of Table 3– Table 7 show that the balance between the treatment and control groups was maintained in the analysis sample.

In Table 8, column 1 shows the effectiveness of receiving cash grants on monthly earnings on the entire group. These results show that receiving cash grants increased the treatment group's monthly earnings by approximately 4000 Afs. As such, the increased annual income will be 48,000 Afs, which is more than the amount of the grant that an individual in the treatment group received. Considering the 33,500 Afs cash grant, the percentage of the increased annual income to the amount of the cash grant would therefore be:

Column (2) displays the effect of receiving a cash grant on monthly earnings in the male group. It shows that receiving the cash grant increased the monthly earnings by about 4839 Afs, and therefore the percentage of the increased annual income to the amount of the cash grant for this group was 173 %.

Column 3 displays the effect of receiving the grant on monthly income on the female group. The column shows that receiving the cash grant increased the monthly earnings by about 1900 Afs, which is less than half of the estimated effect on the male group. It implies that the percentage of the increased annual income to the amount of the cash grant for this group was 69 %.

Thus, in either of three cases, if the state of increased income continues for the next few years, the sum of increased income will be more than the sum of the administration cost, the interest cost and the amount of cash grant given. It implies that the project passes the standard criteria to determine the appropriateness of government program.

In columns 4 – 6, we added several important covariates and checked whether our estimated coefficients were sensitive to the inclusion of those control variables. These columns indicate that the estimated coefficients of the grant variable reported in columns 1 – 3 are not sensitive to the inclusion of various control variables (Extended data) ( Kashefi & Naito, 2021b). Additionally, the results in Table 8 show that providing mentorship did not have an effect on increasing the monthly earnings.

As mentioned previously, the endline study sample size was much smaller than that of the baseline due to the administrative error during the data collection. One concern was whether such attrition of the sample affects the estimated coefficients. Therefore, we controlled for the attrition by applying the inverse probability weighted (IPW) estimation ( Wooldridge, 2002; Wooldridge, 2010) ( Table 9, Panel A). To calculate the standard error correctly, we use the Generalized Moment Method estimation (GMM) with IPW.

In Table 9, column 1 of Panel A shows the estimation results of applying IPW to the full sample. It shows that the effect of receiving the cash grant on monthly earnings was 3984 Afs, which was quite similar to the estimated coefficient in column 4 of Table 8 (i.e., 4036 Afs). Column 2 displays the estimated coefficient using IPW for the male group. The estimated coefficient was 4876 Afs, similar to the estimated coefficient displayed in column 5 of Table 8 (i.e., 4984 Afs). Column 3 of Panel A presents the estimated coefficient for the female group by using IPW. The estimated coefficient is 2062, which is close to the estimated coefficient in column 6 of Table 8 (i.e., 2070 Afs). Panel A shows that, when attrition is controlled by IPW estimation, the estimated coefficients of the cash grant were similar to the estimated coefficients obtained in the OLS estimation presented in Table 8.

In Table 9, Panel B, we estimated bounding treatment effects. As demonstrated by Manski (1990), Karlan and Valdivia (2011), and Blattman et al. (2014), we first imputed the outcome variable for the unfound sample by assuming that the monthly earnings of unfound treatment group members were equal to the average monthly earning of the found treatment group minus 0.25 standard deviation of the monthly earnings of the found members. We also assumed that the monthly earnings of the unfound control group was the average monthly earnings of the found control group plus 0.25 standard deviation of the found members. After this imputation, we estimated the effects of the cash grant and other covariates using both the found and unfound samples. With this assumption, we found that receiving the cash grant increased the monthly income by 2000 Afs and 1200 Afs for the male and female groups, respectively. Thus, the cash grant is likely to have increased the monthly earnings.

To further correct for attrition bias, Lee's treatment effect bounds for receiving the cash grant on monthly earnings was estimated ( Table 9, Panel C) ( Lee, 2009). The lower bound of the effect of receiving the cash grant was 2328 Afs and 1637 Afs for the male and female groups, respectively. The OLS estimates were between the lower and upper bounds for each specification and the lower bounds, 2328 Afs and 1637 Afs for the male and female groups were statistically different from zero. Thus, even at the conservative estimates, the percentage of the increased annual income to the amount of the cash grant would be 83 % for the male group and 58 % for the female group.

Overall, analysis based on several estimation methods demonstrates that the effect of receiving the cash grant on monthly earnings was economically and statistically significant.

Table 10 shows the effect of the cash grant on having an officially registered business, labor supply and asset levels. In this table, Panel A displays the effect of the cash grant on the probability of having an officially registered business. Column 1 of Panel A shows that an individual who received the cash grant had an 8 percentage point higher probability of having an officially registered business on average. Column 2 and column 3 show that a male individual and a female individual who received the cash grant have a 7 percentage point and a 9 percentage point higher probability of having an officially registered business, respectively. As shown in columns 4–6, we added several important covariates as control variables. The columns also show that the estimated coefficients of the treatment dummy were similar to the estimated coefficient in columns 1–3.

Panel B shows the effect of the cash grant on the number of household members working for a micro-enterprise. Panel B shows that an individual who received the cash grant increased the number of household members working for their microenterprise by 0.2 persons on average. Column 2 of Panel A, a male individual who received the cash grant increased the number of family members working for their microenterprise by 0.18 persons on average. Column 3 shows that a female individual who received the cash grant increased the number of family members working for their micro-enterprise by 0.254 persons on average.

Panel C shows the effect of the cash grant on the number of non-household members working for a micro-enterprise. Column 1 shows an individual who received the cash grant increased the number of non-family members who work for their microenterprise by 0.23 persons. Column 2 shows that, for the male group who had received the cash grant, the employment number of non-family members increased by 0.232 persons. The results in column 3 show that, for the female group who had received the cash grant, the number of non-family members who were employed for their microenterprise increased by 0.229 persons.

When starting or extending a business, additional equipment, such as a phone, a truck, and a cart, might be needed. Panel D shows the effect of receiving cash grants on the general asset index for production. To calculate the general asset index for production, we first standardized the level of each asset for production. Then, through the principal component analysis, we determined the weight for each asset and calculated the weighted index ( Jolliffe, 2005). As shown in columns 2 and 3, we find that, for the male group, receiving the cash grant increased standard deviations of the index of assets for production by 0.176, and for the female group, by 0.062. When the control variables were included, the cash grant increased the index of assets for production by 0.167 standard deviation for the male group and by 0.074 standard deviation for the female group (columns 5 and 6).

Results in Panel E display the index of assets for consumption. Columns 2 and 3 show that receiving the cash grant increased the index of assets for consumption by 0.1 standard deviation for the male group, and decreased it by 0.02 standard deviation for the female group. When the control variables were included in the estimation, the estimated coefficient and statistical significance did not change. Columns 5 and 6 show that the cash grant increased the index of assets for consumption for the male group by 0.094 standard deviation and decreased it by 0.073 standard deviation for the female group. Thus, for the male group, receiving the cash grant increased the index of assets for consumption. However, our results do not reflect the same for the female group.

Table 11 shows the effect of receiving the cash grant on the levels of different types of assets. The results in this table show that, for the male group, receiving the cash grant increased the level of assets, e.g., bicycles, carts, three-wheel trucks, mobile phones, and cows. Conversely, the number of cows was the only asset that increased after receiving the cash grant in the female group.