Endogenous Longevity and Optimal Tax Progressivity

We study the impact of endogenous longevity on optimal tax progressivity and inequality in an overlapping generations model with skill heterogeneity. Higher tax progressivity decreases both the longevity gap and net income inequality, but at the expense of lower average lifetime and lower aggregate labor supply and income. We find that the welfare-maximizing income tax is less progressive than in the case of exogenous longevity and that the present US income tax should redistribute less. Our result is robust to the empirically observed range of labor supply elasticity and the assumptions of both missing annuity markets and tax deductibility of private health expenditures.


Policy Research Working Paper 8691
Despite its growing popularity, evidence that volunteering enhances civic values and social cohesion among different communities remains limited in developing countries. This study presents novel evidence from Lebanon on the impact of offering a volunteering program that consisted of inter-community volunteering activities and soft skills training on self-reported social cohesion values in the short term. The results show that youth who were selected to participate in the program were more likely to report higher tolerance values as well as a stronger sense of belonging to the Lebanese community roughly one year after the completion of activities. The results show that selection into the program had no impact on improving volunteers' soft skills that were thought to contribute to social cohesion. This finding implies that the mechanism for improved social cohesion values most likely came from the program's innovative feature, which required 20 percent of selected youth to come from communities outside where the area where the project was implemented. Selection into the program had no impact on other secondary measures, namely, employability and employment outcomes. The results should be interpreted with caution, given study design limitations that relate to the nonrandom assignment of youth into the treatment and comparison groups, as well as the presence of nonrandom attrition between the two tracked time periods. This paper is a product of the Social Protection and Jobs Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at jalaref@worldbank.org and rleonsolano@worldbank.org.

INTRODUCTION
Volunteerism has been examined in research in broader terms as part of civic engagement and falls within the existing literature on social capital, with the latter defined as a structure of relations between actors, connections, and networks among individuals and organizations (Coleman, 1988;Kapucu, 2008;Kapucu, 2011). Putnam (2000) emphasizes the importance of trust and reciprocity within networks and between individuals as one of the possible ways to enhance social capital in terms of collaboration and engagement. Uslaner and Brown (2005) argue that community participation is often influenced by citizens' perceptions of equality and trust in others. Those involved in associational activities are more likely to become civic participants when they hold trusting attitudes towards others. Such activities are also reported to have positive effects on the participating individual in terms of better physical and mental health, longer life, and reduced antisocial behavior (Wilson and Musick, 2000;Kwak, Shah, and Holbert, 2004).
The evidence of positive effects of volunteering, as one form of civic engagement, on enhancing civic values and social cohesion remains rather limited and tentative in developing countries. The notion that volunteering may affect positively youth's sense of social cohesion has gained policy relevance since the publication of the World Development Report 2013: On Jobs, which stresses that in countries affected by conflict situations, creating the types of productive opportunities that strengthen social cohesion can help reduce the volatility of economic growth and achieve international development goals by defusing tensions and building trust among the different communities involved.
This paper provides novel empirical evidence on the impact of volunteering on enhancing social cohesion values in Lebanon, a country with a fragile and highly complex political, religious and social landscape, as well as high degrees of social and economic exclusion among its young population. To our knowledge, this is the first impact evaluation that rigorously addresses this research question in Lebanon and in the Middle East and North Africa (MENA) region. The main results show that youth who were selected to participate in a volunteering program that consisted of 80 hours of inter-community volunteering activities and 20 hours of soft skills training were more likely to report higher and improved values of social cohesion in the short term. In specific, they were more likely to report higher tolerance values as well as a stronger sense of belonging to 3 the Lebanese community. The results show that the selection to the program had no impacts on improving volunteers' soft skills that were thought to further contribute to social cohesion, implying that the mechanism for improved social cohesion values most likely came from the program's innovative feature that required 20 percent of selected youth to come from communities outside where the project is implemented. Finally, the selection into the program had no impact on other secondary measures, namely employability and employment outcomes.
The remainder of the paper is organized as follows: Section 1 provides context and an overview of the intervention. Section 2 provides a description of study timeline and data. Section 3 details the empirical identification strategy. Section 4 provides the main results of the intervention on social cohesion values as well as employment outcomes. Section 5 presents results on other outcomes, namely attitudes towards volunteering. Section 6 concludes.

SECTION 1: CONTEXT & INTERVENTION
Lebanon's political development system since Independence has been heavily influenced by its confessional system. While originally established to balance the competing interests of Lebanon's diverse religious communities, it is seen as an impediment to inclusive growth and effective governance (World Bank, 2016), and has been closely tied to the economic and social inclusion challenges facing Lebanese youth today.
The confessional system of governance has heavily impeded the equitable and efficient distribution of investments and public services. Provision and targeting of public services tend to be guided by considerations of confessional quotas and electoral geography rather than needsbased service delivery that favors the poor. In the absence of effective state institutions, sectarian organizations have played a key role in the provision of social services such as education, health, and welfare support to the most vulnerable groups linked to their electorates, thus deepening a sense of discriminatory and inequitable system (World Bank, 2016;Kraft et al., 2008). Regional disparities are stark, with the bulk of the poor living in peripheral areas (particularly the North and the South), with visible inequality in access to and quality of social services. According to the 2012 Institutional Profiles report, the quality of public services and its territorial coverage, which was weak to being with, have significantly deteriorated since 2006. 1 A combination of rising poverty, rising insecurity, and deteriorating public services have further strained inter-communal relations and contributed to deteriorations in social cohesion. Many Lebanese youth do not trust their state and become disillusioned as they are not able to affect their own life or contribute productively to society at large. 2 Political and civic engagement is reported to be low (Status of Women in the Middle East and North Africa Survey Project, 2010). 3 In an already fragile context with a highly complex political, religious and social landscape consisting of 18 religious sects, numerous political parties, and large numbers of refugees, many Lebanese youth today remain unexposed to other parts of their country before they enter the world of education or work.
The confessional system and the inequalities it breeds have also played a key role in perpetuating inequality of economic opportunity among youth, placing further strains on social cohesion. Youth in areas where services and infrastructure are inadequate are more likely to have limited access to economic opportunities and improvements in quality of life. Evidence indicates that area of residence and parents' education at birth 4 largely contribute to youth exclusion and limit their socio-economic mobility (World Bank, 2016). The poor quality of public school education generates large inequality of opportunity among youth with regards to human capital accumulation and obtaining high-quality jobs. Additionally, for many young Lebanese, finding job opportunities is more a function of confessional policy and 'wasta' (connections) rather than meritocracy. 5 Youth economic exclusion is manifested in worsening labor market outcomes. In 2010, youth (15-24) unemployment reached 34 percent, and the percentage of youth not in education, employment, or training (NEET) reached 21 percent in 2007. These poor labor market outcomes have likely worsened as a result of the sluggish economic growth of the past few years, which can be partly explained by the impact of the Syrian conflict on Lebanon's economy. Challenges around skill mismatch exacerbate exclusion, as many young Lebanese lack the skills and competencies demanded by private sector employers, particularly 'soft skills'.
To address some of these challenges, the Lebanese government (GOL) identified volunteerism as a mechanism to enable diverse youth to work together for improved community assets and service delivery as well as increased employability. In September 2012, the GOL issued a Decree (Number 8924/2012) that created a new extra curriculum program that requires secondary school students to complete 60 hours of civil work. In addition, the Ministry of Social Affairs (MOSA), through its Volunteering Department, launched annual action plans for the implementation of youth volunteer summer camps across Lebanon.

6
Against this backdrop and as part of the GOL efforts to promote volunteerism in Lebanon, the World Bank, through a US$2 million grant from the State and Peace Building Fund, supported the MOSA to conceive the National Volunteer Service Program (NVSP). The NVSP, launched in 2013 for a period of 3 years, sought to promote social cohesion among Lebanese youth aged 15-24 through the financing of volunteering activities and the provision of soft skills. 6 The benefits of volunteering as a promising pathway to social integration for youth are well documented in the literature. Besides the intrinsic rewards obtained from the act of helping others, studies show that volunteering has positive effects on the individuals who volunteer in terms of better physical and mental health, longer life, reduced anti-social behavior, and increased civic participation and social trust (Wilson and Musick, 2000;Perry and Katula, 2001;Laurence and Heath, 2008;UN Volunteers, 2011).
The project's theory of change relied on the idea that volunteering has the potential to defuse social tensions by bringing youth from communities other than their own together to work around shared goals and to find solutions to the most pressing problems affecting their communities. One of the most innovative features of NVSP is that at least 20 percent of youth who participate in an NVSPfinanced project must come from communities outside where the project is implemented.
Volunteering outside one's own community is considered a best practice by development psychologists who claim that a change in context is a prerequisite for improved trust, respect, and cooperation among people from different backgrounds. On the other hand, the provision of soft skills is thought to further contribute to social cohesion by fostering among participating youth qualities such as teamwork, leadership, communication, and conflict resolution skills.
A secondary objective of the NVSP was to enhance the employability of participating youth, by strengthening their soft skills as well as by providing them hands-on experience through volunteerism, which they can add to their resumes to signal to employers that they have become more employable. However, rigorous evidence on the impact of volunteering on employability and successful youth transition to jobs in developing countries is scarce. There is some evidence in developed countries (Powel and Bratovic, 2007;Spera et al, 2013;Paine, McKay and Moro, 7 2013), but it focuses on case studies describing programs and their outcomes (Graham and Perold, 2013) rather than causally measuring their impacts.
The NVSP launched its volunteering activities through the Small Grants Program (SPG) in 2015.
After a Call for Proposals (CfP), 38 different proposals were received. 22 proposals were selected for financing, which benefited 1,293 youths (of which 54% females and 46% males) across five different regions in the country (North, Mt. Lebanon, Beirut, Bekaa, and South). Thirty-eight percent of these volunteers came from communities outside the ones where projects were implemented. 7 All projects selected for funding received a $30,000 grant, which was used to buy project materials and provide incentives (transportation and meals) as well as soft-skills training to participating volunteers. On average, selected projects lasted about 3 months, and participating youths were required to complete 20 hours of employability/soft skills training and 80 hours of volunteering. With regards to the soft-skills training delivered, the NVSP partnered with the International Youth Foundation (IYF) to develop and deliver a 30h soft skills training to participating youth. The delivery of the soft skills training relied on a series of Training of Trainers (ToT) sessions for staff from MOSA and participating NGOs, to enable them to then deliver the soft skills training to youth during the implementation of volunteering activities. IYF used its world renowned "Passport to Success" (PTS) curriculum, which focuses on a core group of widely endorsed life skills, such as self-confidence, responsibility, and respect. Additionally, the curriculum includes workplace readiness skills, such as interviewing and time management.

SECTION 2: STUDY TIMELINE & DATA
A quasi experimental impact evaluation design was embedded into the NVSP. As mentioned before, the NVSP received 38 applications from eligible NGOs. Per well-developed selection criteria, 8 the highest 22 ranked proposals were selected to receive funding. Each of the 38 proposals included a list of 50 youth (the minimum number of youth set by the NVSP) who would benefit from the project if selected for funding. However, as mentioned before, the 22 selected projects benefited a total of 1,296 youth, exceeding the set target of 1,100 volunteers.
Of the 50 volunteers included in each of the 38 proposals, 22 youth per proposal were randomly selected to participate in the impact evaluation study. Therefore, the initial sample size of the study comprised a total of 825 youth: 473 youth who served as the treatment group (representing the 22 selected NGOs that received NVSP funding) and 352 youth who served as the comparison group (representing the 16 non-selected NGOs). However, two NGOs refused to participate in the study once informed that their proposals had not been selected for funding. Therefore, the final sample size consisted of 759 youth, of which 473 treatment and 286 comparison.
Detailed baseline data were collected through face-to-face interviews from July to September 2015 prior to implementation. The actual implementation varied between projects and ranged between the second half of August and end of December 2015. Sampled youth from both selected and nonselected NGOs were invited to fill out a questionnaire with detailed information on volunteers' socio-economic backgrounds, education levels, interests and attitudes towards volunteering, employment, soft skills, as well as social cohesion values.
Follow-up data were collected between November 2016 and March 2017, approximately one year following the start of implementation, through phone and face-to-face interviews. The questionnaire contained the same modules asked and collected at baseline. Despite the high mobility of sampled volunteers, thorough tracking procedures led to relatively low non-response rates at follow-up: 88 percent of the 759 youth were tracked and completed the follow-up survey.

SECTION 3: EMPIRICAL IDENTIFICATION STRATEGY
Identification of program impacts relies on a difference-in-difference (DiD) estimation, where outcomes are observed for the treatment and comparison groups for two time periods: at baseline, prior to the launching of the selected projects, and at follow up, one year following the implementation of the selected projects.
The DiD estimator compares the changes in outcomes over time between selected and non-selected youth. The first difference is the difference in before-and-after outcomes for selected youth to control for factors that are constant over time in that group. The second difference is the difference in before-and-after change in outcomes for non-selected youth who did not enroll in the program, but were exposed to the same set of environmental conditions. Subtracting both differences averages out any constant (time-invariant) observable & unobservable differences between both groups to attribute the impact estimate to program participation.
The report focuses on intent-to-treat (ITT) estimates, measuring the impact of offering volunteering opportunities and soft skills training independently of actual take-up. 9 We estimate the following individual-level intent-to-treat regression: where is the outcome of interest for respondent i in period t, is a post-treatment year binary variable, is a binary variable for being assigned to the treatment, and is a fixed effect for NGOs. represents the baseline average for the outcome of interest for non-selected youth. is the difference in after-and-before intervention in outcomes for non-selected youth. is the difference in after-and-before intervention in outcomes for selected youth. is the difference in outcomes between selected and non-selected youth at baseline. is the DiD estimator. is a mean-zero error term. Standard errors are robust and allow for intra-cluster correlation at the NGO level. 10 The DiD estimator can be derived from the above regression as follows: The DiD estimator relies on the "Equal Trend Assumption" that does not require both selected and non-selected youth to be on average balanced at baseline on key observable & unobservable characteristics. Table 1 shows that both groups differ on some key characteristics. Non-selected youth are more likely to be older, more educated (hold more academic degrees), come from Beqaa and Nabatiye, and have parents with intermediate education (grade 7 to 9). Selected youth are more likely to be males, younger, students, come from Mount Lebanon and the North, and have mothers with university education. Both groups appear balanced on key outcomes related to soft skills, tolerance values, and labor market outcomes. The exception is that non-selected youth exhibited a better sense of belonging to the Lebanese community and selected youth were more likely to have been unpaid employees (interns) at the time of baseline data collection.
In addition to comparing means of observable characteristics, the study also tested for the differences in the statistical distributions of key outcomes using two sample Kolmogorov-Smirnov tests of the equality of distributions. Results indicate that the only key outcome for which there is a statistically significant difference in its distribution between the treatment and comparison groups at baseline is the sense of belonging to the Lebanese community. The largest difference between the distribution functions in the direction that the comparison group contains larger values than the treatment for this key outcome is 0.1043. The approximate p-value for this small difference is 0.021.
What the DiD estimator requires, however, is that both selected and non-selected youth are subject to the same time trends and that differencing removes any confounding factors. The study was not able to obtain pre-intervention data from previous years on both groups to check for this assumption. The implication of this is that if there are other factors that affect the difference in trends over time between the two groups, then the estimation will be biased. This may be particularly the case due to time varying non-response (attrition), which means that even if one is to assume that the "Equal Trend Assumption" holds on the original full sample for the previous years prior to the implementation of the NVSP program, non-random attrition at follow-up biases this assumption.
In this regard, attrition was not balanced between both groups with the comparison group more likely to drop out of the sample at follow-up than treatment. 11 It appears that there are few statistical differences between the treatment and comparison groups with regards to some key characteristics of volunteers who were more likely to drop out in each group. Evidence of some asymmetric selective attrition by treatment status biases the OLS estimates.
Given the concern of a correlation between entity's error term (in this case, volunteers in our sample) and the predictor variable (in this case, the treatment variable), individual fixed effects (FE) are utilized. FE remove the effect of individual-specific time-invariant characteristics so as the net effect of the predictor variable on outcome variables can be assessed, per the following equation: where is entity n (i.e. the individual volunteer). Since they are binary (dummies), there are n-1 included in the model (i.e. 758 individual volunteers). is the coefficient for the binary regressors (the 758 volunteers).
Additionally, we propose dealing with attrition in two ways. First, we utilize the standard "Manski Bounds" approach (Horowitz and Manski, 2000) by imputing upper and lower bound estimates for missing data on estimated outcomes of interest at follow-up, where lower bound estimates take the lowest possible value and upper bound estimates take the highest possible value for individuals who could not be tracked over time. This allows us to provide the two extreme possible scenarios for estimated impacts had data been successfully collected for attritors. Second, we use the Inverse Probability Weighting (IPW) procedure to establish narrower bounds that might provide a better sense of whether there is a robust treatment effect. This entails first estimating a probit model that predicts the probability of data being observed (i.e. not attrition) using a set of covariates at 13 baseline that were found to be uncorrelated with the treatment in Table 1. 12 Observations are then weighted by the inverse of their probability of having data observed. Therefore, those who had a small chance of being observed are given increased weight, to compensate for those similar observations who are missing. The pseudo R-squared from the probit model suggests that those baseline covariates explain about 8 percent of the probability of data being observed. A Wald test confirmed that those variables are jointly statistically different from zero (the P-value is 0.000).
However, this still leaves a large percentage of attrition (around 92 percent) unexplained. 13 Therefore, we note that the results in the following section should be interpreted with caution.
We present results in the next section for four specifications. Specification 1 presents OLS ii) Does assignment to NVSP-supported projects increase selected youth's soft skills? More specifically, compared to the youth who were not selected to benefit from NVSP support, 12 In specific, the set of baseline covariates are: spoke more than one language, lived for at least one year outside the district they lived in at the time of the baseline survey, education level of intermediate certification, education level of completed vocational or complementary education studies, previous volunteering experience, previous technical/professional training, previous soft skills training, previous volunteering training, coming from Beirut, coming from South Lebanon, mother's highest education as elementary, mother's highest education as secondary, father's highest education as elementary, father's highest education as secondary, father's highest education as university, unemployed, inactive, self-employed, wage employed, offered a paid job in the month before the baseline survey, leadership skills score, communication score, confidence score, and tolerance of other opinions' score. 13 Many research studies have identified the quality of the interview or enumerator quality as strongly associated with attrition (Thomas et al. 2012). Unfortunately, our study did not collect such information and we acknowledge that it is an area of improvement for future work.
14 do selected volunteers increase their teamwork/leadership and their communication skills? Do they increase their self-esteem/self-satisfaction?
iii) As a result of their assignment to NVSP volunteering experience, are selected youth more likely to find a job than non-selected ones? This is interesting given that at baseline, non-selected youth had a stronger sense of belonging than selected youth (by 0.2 S.D., see table 1). The worsening of results for both groups could be explained by the broader country environment, including the continued deterioration of the Lebanese economy, poor service delivery, as well as ae strong perceptions of limited governance and accountability, all of which can contribute to alienating Lebanese youth and their sense of identity with the state. Assignment to the NVSP program, through inter-community civic engagement opportunities, appears to have mitigated those negative perceptions in a statistically significant manner, albeit by a small margin as selected youth's perceptions also got worse over time.  The leadership skills measure appears to have worsened for both selected and non-selected youth over time, which is somewhat puzzling given that both groups are active volunteers and members in their communities (see the estimate and the estimate for column 1). Any changes for the communication and confidence scores one year following NVSP were not statistically significant for both selected and non-selected youth (see the estimate and the estimate for Lack of results also holds for specification 3, where imputing missing data with upper and lower bound estimates to account for the potential bias introduced by attrition did not alter the lack of impact of the program, as well as for specification 4 (see the estimate). The majority of NVSP volunteers in focus group discussions and interviews mentioned that they would have welcomed more advanced trainings on soft skills, as well as on technical topics and job-relevant skills that can support their employability. These may include skills on how to write a CV, prepare for job interviews, business and entrepreneurial skills to start a business, etc.
In this regard, the mechanism for improving social cohesion values appears to have come from inter-community volunteering activities, rather than improvements in soft skills.

(c) IMPACTS ON LABOR MARKET OUTCOMES
While the NVSP was designed primarily to improve social cohesion values among participating Lebanese youth, it was also hoped that engaging them in volunteering activities, coupled with soft skills training, would enhance their employability and thus increase their chances of employment.
At baseline, half of the selected and non-selected volunteers were active and searching for a job.
Among them, 49 percent reported being unemployed, 31 percent wage employed, 13 percent employed in unpaid jobs, and 7 percent self-employed (see table 1). Those active volunteers were older in age than the rest of volunteers who reported being inactive in the study's sample (with an average age of 21 and closer to labor market insertion). One year later, it appears that many of them found jobs. youth are less likely one year later to be unpaid employees, presumably because they were more successful in obtaining wage employment (see the estimate and the estimate for column 4). Finally, selected and non-selected youth were less likely to be offered a job in the last month (see the estimate and the estimate for column 5), potentially due to many of them having found employment during the same period.
However, it appears that NVSP is not responsible for the increase in employment levels among participating youth, as reported in table 5 across the two specifications (see the estimate in columns 2 & 6). In other words, there is no evidence that because of their assignment to the NVSP, selected youth were more likely to improve their employment status in the labor market, when compared to non-selected youth. Additionally, it appears that NVSP had no impact on youth employability, as measured by the probability of being offered a job in the last month and labor market participation rates (see the estimate for columns 1 & 5).
Lack of results on employability also holds for specification 3, where imputing missing data with upper and lower bound estimates to account for the potential bias introduced by attrition did not alter the observed lack of program impact, as well as for specification 4 (see the estimate).
The NVSP lack of impact on employability and labor market outcomes is not surprising given that the project's primary objective was to improve social cohesion values. The employability component was very small and consisted of offering soft skills training, which as previously discussed, was not effective in further improving their skills. Increased volunteering opportunities, which was also perceived as an intermediary channel to enhanced employability, appears to be ineffective and unlikely to work in isolation without a comprehensive approach to addressing both demand and supply-side driven youth employability challenges in Lebanon.
While we find no impact that selected youth who have had improvements in their social cohesion values would also be likely to improve their employment outcomes because of their assignment to NVSP, another question worth examining is whether selected youth who were wage employed at  4). Yet, it is still pertinent to assess whether assignment to the NVSP played a role in further strengthening those attitudes and perceptions towards volunteering among selected youth. Table 6 presents results on the 10 measures of attitudes and perceptions towards volunteering, which were ranked on a seven-point scale (seven being 'totally agree' and one being 'totally disagree'). Each measure is internally standardized with a mean of 0 and S.D. of 1 for non-selected youth at baseline. Results indicate that for both selected and non-selected youth, perceptions on volunteering got paradoxically worse over time on some of the measures that were ranked very favorably at baseline, namely "volunteering allows those who volunteer to participate in helpful activities", "volunteering enables those who volunteer to learn how to use their time more effectively", "volunteering allows those who volunteer to develop skills such as teamwork and leadership", and "volunteering contributes to improving the self-satisfaction of those who volunteer" (see the estimate and the estimate for columns 3, 4, 5 & 10). Those perceptions 20 This research question was posed by the World Development Report 2013: On Jobs. 21 To examine this question, we estimated the following equation for selected youth only: + + + * + . The coefficient of interest, , can be interpreted as: where is the outcome of interest at follow-up, is the outcome of interest at baseline, and is a dummy variable that takes on the value 1 if employed at baseline and 0 if not employed at baseline. In other words, it can be understood as the difference between selected youth employed at baseline and selected youth non-employed at baseline in their differences in outcomes post-pre-assignment for NVSP. The estimated coefficient in specification 4 was -0.11 (0.14) with a P-value of 0.568.

20
are related somewhat to measures of soft skills and may be consistent with earlier findings on the lack of NVSP impact on improving those skills.
Results for both specifications 1 and 2 show that non-selected youth's perceptions got worse on two more measures, namely "volunteering contributes to the happiness of others" and "volunteering may contribute to increasing understanding across communities" (see the estimate for columns 1 & 6). Selected youth's perceptions witnessed a decline on one additional measure, namely "volunteering may contribute to improving social cohesion in Lebanon" (see the estimate for column 9). Those perceptions may be consistent with earlier findings on the worsening of social cohesion values among both groups over time, but at a much higher magnitude for non-selected youth compared to the selected ones.
In terms of improvements, for both specifications 1 and 2, both groups strengthened their perceptions on the role of volunteering in improving the chances of volunteers to find a job, which scored somewhat less favorably at baseline (see the estimate and the estimate for column 8). Selected youth also witnessed an increase in their perceptions on two more measures, "volunteering contributes to expanding the professional networks of those who volunteer" and "volunteering contributes to increasing the wealth of communities" (see the estimate for columns 2 & 7). Improvements in those perceptions may be picking up the effect that both groups are more likely to be employed one year following NVSP implementation, although as previously discussed, NVSP assignment had no impact on employment and employability measures.
In terms of the impact of NVSP assignment on those measures, it appears that selected youth were more likely to strengthen their perceptions towards volunteering compared to non-selected youth on two main measures only: "volunteering contributes to the happiness of others" and "volunteering allows those who volunteer to develop skills such as teamwork and leadership" (see the estimate for columns 1 & 5). However, the impact no longer holds for specification 3. In short, it appears that NVSP did not have much of an impact on strengthening a broad range of attitudes and perceptions among selected youth at follow-up. Specifically, this second NVSP pilot is enabling Lebanese youth to volunteer together with Syrian refugees for improved social stability and social service delivery in targeted host communities. It is hoped that this second pilot yields similar results as the first one in terms of improved social stability throughout Lebanon.

22
The lack of results on improving employability and employment outcomes among youth highlights the need for a comprehensive approach to addressing both demand and supply-side driven youth employability challenges in Lebanon. A review of the evidence from fragile and conflict-affected contexts seems to suggest that skills training and microfinance have shown little impact, whereas capital-centric programs --cash, capital goods, or livestock --seem to work in stimulating selfemployment and raising long-term earning potential (Blattman and Ralston). Integrating some of those interventions proven to work in fragile settings into the NVSP framework may be one possible next step. However, strengthening the NVSP's link to better labor market outcomes requires first understanding the multiple barriers facing youth in the labor market and for policy makers to pilot small-scale pilots before launching large programs.
Finally, it is hoped that the positive results of the NVSP can influence the decision of other MENA countries facing similar social challenges to adopt volunteerism for improving social cohesion.
Many MENA countries are facing increasing social tensions as a result of different economic and political shocks. Volunteerism is certainly not a panacea, but it is a mechanism that, in the shortterm, could alleviate social tensions and, in the medium to long-term, serve as the basis for youth citizenship development. Developing responsible citizens, especially among youth, is vital for the social, political, and economic development of MENA countries and beyond. Note: *significant at 10%. ** significant at 5%. *** significant at 1% Table 2. Estimates of a probit model regressing attrition on the following interaction terms (dummy covariates at baseline with a dummy for being assigned to treatment) The interaction terms can be interpreted as the differences in estimated probabilities of attrition for the various covariate dummies between the treatment and comparison groups. For example, the coefficient on the interaction term, male * treatment, is understood as:

TABLES
where is the estimated probability of attrition for males, is the estimated probability of attrition for females, is a dummy variable for males that takes on the value 1 if assigned to treatment and 0 if assigned to comparison group, is a dummy variable for females that takes on the value 1 if assigned to treatment and 0 if assigned to comparison.  Specification 1 is an OLS estimation that includes assignment to treatment dummy, post-intervention year dummy, interaction term between those two variables, and NGO fixed effects (equation 1). Specification 2 is an individual fixed effects estimation Specification 3 presents OLS estimates from equation 1 but includes lower and upper bounded-imputed data for individuals who could be tracked at follow-up. Specification 4 is an OLS estimation that includes the same independent variables as specification 1,but weights the observations by the inverse probability of being observed at follow-up. Note that in specification 4, the sample is only restricted to the complete cases that were tracked at follow-up (i.e. N=1,336)

. Attitudes and Perceptions of Selected and Non-Selected Youth Towards Volunteering at baseline
Note: The scale of each indicator is as follows: 7 = totally agree, 6 = strongly agree, 5 = somewhat agree, 4 = neutral, 3 = somewhat disagree, 2 = strongly disagree, 1=totally disagre