The Marginal Benefit of an Active Labor Market Program Relative to a Public Works Program: Evidence from Papua New Guinea

Abstract Policymakers typically try to address youth unemployment in developing countries through either active labor market programs (ALMPs) or labor-intensive public works programs (LIPWs). We examine whether there is any additional benefit for unemployed youth from participating in a comprehensive ALMP compared to a LIPW. We exploit an unanticipated intervention in the largest employment program in Papua New Guinea, which resulted in one intake of the program completing a LIPW and missing out on a comprehensive ALMP. We conduct a difference-in-difference analysis between participants in the intake that missed out on the ALMP component of the program and participants in the intakes immediately before and after. In contrast to most impact evaluations of ALMPs, we show youth that completed the comprehensive ALMP were around twice as likely to be employed in the formal sector 9–12 months after the program compared to similar youth in the intake that only completed a LIWP. This effect was entirely driven by 20% of youth who participated in the ALMP staying with the employer they were placed with following the end of the program. Surveys of these employers illustrate that they use the ALMP as a low-cost, low-risk, and relatively low-effort way of hiring new employees.


Introduction
The vast majority of young people in developing countries do not have a formal sector job and earn an income through a range of activities classified as "vulnerable employment" by the International Labour Organization. To address this issue, billions of dollars have been spent by governments and aid donors on programs that typically take the form of either active labor market programs (ALMPs) or Labor Intensive Public Works Programs (LIPWs). The former aims to improve the long-term employability of youth by providing training, work placements and job searching assistance, whereas the latter is a form of social protection through the creation of temporary employment opportunities. The key difference is that ALMPs tend to be more resource intensive, as they aim to create employment for participants beyond the life of a program, as opposed to LIPWs where the benefits are expected to be short term. However, there is a growing evidence base indicating that ALMPs fail to achieve their aim and tend to have little to no impact on employment (McKenzie 2017;Blattman and Ralston 2015). This raises the question as to whether policymakers should only invest in LIPWs if there are no additional employment benefits from ALMPs beyond those of LIPWs. This is the first study to explore this question and we do so by rigorously testing Specifically, we conduct a difference-in-difference analysis between eligible youth in the intake that missed out on the ALMP (they only completed the LIPW component and a one-week basic training course prior to the LIPW) with the youth in the intakes immediately before and after that completed both the LIPW and ALMP components.
We show that the youth that completed both program components were around twice as likely to be employed in the formal sector 9 to 12 months after the program compared to the eligible youth in the intake that only completed the LIWP component. This effect was driven by 20 per cent of ALMP participants staying with the employer they were placed with following the end of the program. Surveys of employers illustrate that partnering with UYEP provides them with a low-cost, low-risk and relatively low-effort way to screen potential new employees.
These findings are contrary to the bulk of existing evidence that shows ALMPs tend to have little or no effect on employment (McKenzie 2017;Blattman and Ralston 2015).
We provide evidence suggesting this may be due to the comprehensive nature of the ALMP component of UYEP. Most ALMPs only provide youth with one of the following: subsidized job placements, vocational training, or job search and matching assistance. In contrast, the ALMP component of UYEP includes all three of these aspects (in addition to the initial LIWP component) and lasts for six months. Employer surveys illustrate that each of these aspects of UYEP contribute to them offering youth the opportunity to continue working for them following the conclusion of the program. This paper is structured as follows. Firstly, we present the related literature on ALMPs and our contribution followed by a discussion of the context the program is operating in and provide details about the program. This is followed by a description of the difference-in-difference methodology we employ and evidence is provided in support of the "equal trends" assumption. We then present the main findings, some suggestive evidence from an employer survey, and discuss the implications of the results.

Related literature and our contribution
There has been a significant number of impact evaluations of ALMPs in a variety of contexts and the vast majority have shown they tend to have little to no effect (McKenzie 2017). In particular, programs that only include one component of an ALMP (either subsidized job placements, vocational training, or job search and matching assistance) tend to have no impact on employment. The direct employment benefits from subsidized job placements, whereby participants complete internships and their wages are subsided, tend to only be short lived as the effect quickly diminishes once the program is completed (McKenzie 2017). In addition, in some cases subsidizing job placements has been shown to negatively affect employment opportunities of individuals who do not participate in the program, which suggests that the aggregate number of job opportunities in the labor market is not being increased (Groh et al. 2016). Blattman and Ralston (2015) show that providing participants with vocational training rarely generates jobs (unless a cash grant is provided) and almost never passes a basic cost-benefit test. Training conducted by private institutions that tailor the education to meet the demands of the private sector have been shown to be slightly more effective, however even in these cases the benefits drop off quickly over time (Hirshleifer et al. 2016). Another type of ALMP consists of providing job search and matching assistance, which is substantially cheaper than wage subsidies and training (Caria and Lessing 2019). However there is very little evidence that this type of ALMP has an effect on employment in developing countries (McKenzie 2017).
There is some evidence to suggest that ALMPs that combine two or more of these components are more effective. In Latin America, a number of programs (such as entra21, Jovenes, and Juventud y Empleo) combine vocational training with subsidized work placements and they have been shown to have a positive effect on formal employment but not overall employment (i.e. both formal and informal) (e.g. see Alzua et al. 2016;Attanasio et al. 2015;Diaz and Rosas 2016). A recent study in Uganda shows that combining vocational training or subsidized work placements (in the form of internships) with job matching assistance resulted in large employment benefits that lasted four years after the program (Alfonsi et al. 2017). The authors of this study argue that vocational training was more beneficial than subsidized work placements as participants learnt more sector-specific as opposed to firm-specific skills.
A major shortcoming of most existing impact evaluations on this topic is their inability to disentangle whether the impact they are detecting is due to youth simply being part of an employment program or the actual content of the program (McKenzie 2017).
Specifically, Beam et al. (2016), Galasso et al. (2004), and Levinsohn and Pugatch (2014) have shown that simply being part of a program, as opposed to the nature of the program itself, can have an effect on formal sector employment. Many impact evaluations on this topic are unable to overcome this issue because they use comparison groups that are not part of the program, such as a control group in a randomized controlled trial whose members do not participate in the program. In these cases, studies may overestimate the impact of the particular type of program as they are estimating the effect of a combination of being part of any employment program and the effect of the particular type of program they are evaluating.
Against this background, we make four contributions to the existing literature on this topic. This is the first study to rigorously investigate the marginal effect of an ALMP beyond a LIWP. Existing research has tended to focus on the impact of participating in just one type of program relative to not participating in any program. Secondly, we investigate the effect of one of the most comprehensive ALMPs in the developing world as it provides participants with vocational training, job matching and fully subsidized work placements over a period of eight months. Previous impact evaluations have typically analysed the effect of more limited ALMPs, such as those that only provide vocational training or wage subsidies. Thirdly, we overcome the challenge faced by most other impact evaluations by having a "placebo" comparison group of program participants that do not fully participate in the program compared to the treatment group (they only do the LIWP component). This ensures that the effect we detect is driven by the content of the program as opposed to simply being part of any program (however we cannot rule out the possibility that differences between the treatment and control groups are partly due to the duration of exposure to the program). Finally, this study is one of only a small number of rigorous impact evaluations that have been completed in the very challenging context of Papua New Guinea. By doing so our study helps to enable evidence-based policymaking in this part of the world.
3 Details about the context and the program design

Identification strategy
To determine the impact of the ALMP component of this program (beyond just the LIWP component) we conduct a difference-in-difference analysis and exploit a one-off, unexpected change in the implementation of the program. Specifically, during intake 13 of UYEP there was an unanticipated intervention in the program and the standard ALMP component was removed part way through the LIWP component. Instead, participants were offered the opportunity to be paid to be a volunteer at the 2016 Under 23 Women's FIFA World Cup. Youth who undertook this placement were only engaged for around one-third of the period of time as youth who undertook the standard ALMP (2 months compared to 6 months). It seems unlikely that if the duration of the volunteer program had been longer, employability of the youth in this intake would have been improved. The volunteer placement only involved basic manual labor, limited training opportunities and there was no potential for ongoing employment with the employer as this was a one-off event. As such, program participants in intake 13 who were eligible to complete the ALMP component because of their academic performance in the screening test and satisfactory completion of BLST did not complete an ALMP. Following intake 13, the usual program resumed whereby people who were eligible completed the ALMP component as was the case in earlier intakes. Importantly, the change in the program for intake 13 was not made public until after youth for intake 14 had expressed interest in UYEP and all youth in the treatment who were eligible to participate in the ALMP component of the program took up this opportunity.
The youth who were eligible for the ALMP component in intake 13, but missed out due to the temporary change to the program ("placebo" control group), are comparable to the youth in the intakes just before and just after (intakes 12 and 14) who completed the ALMP component (treatment group) for five reasons ( Figure 1). Firstly, all youth in each of the three intakes had not worked in the formal sector in the last six months and on average between 75 and 80 per cent of the youth had never worked in the formal sector throughout their life. Secondly, intakes 12, 13 and 14 occurred at a relatively similar time (the baseline surveys occurred in April, July and November 2015, respectively).
Thirdly, the youth are sourced from the same areas in Port Moresby and expressed interest in the program through the same channels. Fourthly, the removal of the ALMP component of the program during intake 13 was not known by participants at the time the intake began and they were only informed of this change part way through the LIWP component. Finally, statistical balance calculations show that there were almost no statistically significant differences between the demographic characteristics of treatment and control groups (see section 4.2). The main outcomes of interest are whether the treatment group outperformed the control group in terms of employment in the formal sector, earning a cash income and seeking employment. The exact wording of each of the questions is shown in Table 1 below and they are based on what has been used in other impact evaluations on this topic (McKenzie 2017). Sought job in last three months Have you sought any kind of paid job in the last 3 months?
Note: These questions were typically asked in the local dialect, Tok Pisin.
The baseline survey was conducted from April to November 2015 when youth were registering to participate in the program. The follow-up survey was conducted in July to August 2017, which was around nine months after all youth from these three intakes had completed the program.

Sample size and statistical balance
There were 355 program participants who met the criteria outlined above and completed the baseline survey. Among these respondents, 215 were re-interviewed in the follow up survey (113 in the control group and 102 in the treatment group). Respondents in the treatment group were only three percentage points more likely to complete the follow up survey than respondents in the control group. We show in Appendix A1 that on average there was no difference in the demographic characteristics between respondents who were re-interviewed and those that were not, except in regards to education. Respondents in the treatment group who were re-interviewed were more likely to have lower levels of education at baseline (there were no statistically significant differences between respondents in the control group who were re-interviewed and those that were not). The relatively small sample size (compared to other impact evaluations of ALMPs) means we are only adequately powered to detect statistically significant differences between treatment and control groups of at least 8 percentage points.
Statistical balance calculations show that there were no statistically significant differences (at a p-value below 0.05) between the demographic characteristics of respondents who completed both surveys in the treatment and control groups. Table 2 below shows that the only statistically significant difference between these groups is that those in the treatment group were slightly older. Standard errors (SEs) in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01. N refers to the number of observations # One people in the control group did not provide an answer to this question

Addressing the Equal Trends Assumption
Sections 4.1 and 4.2 outline how the treatment and placebo control groups were clearly comparable at the time of the baseline survey, however the key assumption of a differencein-difference analysis is that in the absence of the program the comparability of the groups would have lasted over time (i.e. there would be "equal trends"). We take two steps to illustrate that this equal trends assumption is reasonable. Firstly, we show that youth in the treatment and control groups experienced very similar employment and job-seeking histories prior to the program. While we only have two rounds of survey data (before and after the program), the baseline survey includes a number of questions that provide insight into respondents' lives at various points in time (see Table 3). There are no statistically significant differences between the treatment and control groups across a range of characteristics that capture their employment and job-seeking histories. This provides good reason to believe they would have had very similar trajectories for employment and job-seeking outcomes in the absence of the program. Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01.
In addition, there were few opportunities for these unemployed youth to enter the formal sector. As discussed above, the levels of formal sector employment in Port Moresby declined by 20 per cent from 2013 to 2017 and most employers were not hiring new staff (BPNG 2017). As such it is highly unlikely that opportunities in the formal sector for the youth in these intakes that are three months apart would have varied dramatically in the absence of the program.
The second step we take to illustrate the credibility of the equal trends assumption is to follow Duflo (2001) and conduct a placebo test. This involves selecting groups of individuals in the same intakes as the treatment (intakes 12 and 14) and control groups (intake 13) who were not affected by the program, and conducting difference-in-difference analysis. If there are no statistically significant differences in outcomes this provides further evidence that in the absence of the program the trajectory of outcomes of the actual treatment and control groups would have been comparable. The design of this study naturally lends itself to conducting a placebo test using UYEP participants who were not eligible to complete the ALMP component due to their performance in the academic test in intakes 12, 13 and 14. We show in Table 4 below that there are no statistically significant differences between these groups across demographic characteristics. Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01. # Some respondents did not provide an answer to this question

Empirical strategy
To estimate the effect of the ALMP component on the main outcomes in Table 1, we conduct a difference-in-difference, Ordinary Least Squares (OLS) regression in the form of a linear probability model that can be written as follows: where Y it is a dummy variable for each of the outcomes of interest discussed in Table   1, T reatment i is a dummy variable that takes the value of one if the respondent belongs to the treatment group and the value of zero if the respondent belongs to the control group, Af ter t is a dummy variable that takes the value of one for the follow-up survey and zero for the baseline survey and ε it is the model error term. β 3 captures the average difference in differences in regards to the outcomes of interest (Y it ) between respondents in treatment group and the control group (i.e. the treatment effect).
The estimates of the OLS regression above are compared to those obtained from a binary logit model and they are qualitatively similar (see Appendix A2). In addition, the OLS regression above is conducted controlling for baseline characteristics and the results are very similar (see Appendix A2).

Effect on employment and job-seeking behavior
The ALMP component has a large effect on formal employment, with a 41.2 percentage point increase in the share of participants in the treatment group who are currently working or worked in the last six months compared to a 23.0 percentage point increase in the control group (see Table 5). This is from a starting point of no formal employment in the last six months at the time of the baseline survey, which means the 18. The large effect on employment from the program is due to youth continuing to work with the employer they were placed with after the end of the program (see Figure 2).
In the treatment group, 20 per cent of participants were employed with their ALMP employer at the time of the follow-up survey, compared to no one in the control group as this option was not available for them. In comparison, there was no statistically significant effect from the program on formal sector employment with a different employer, as 21 per cent of youth in both treatment and control groups were employed somewhere other than their ALMP employer. This suggests the effect of the ALMP on formal sector employment is a level shift in outcomes (i.e. increase in the likelihood of having employment immediately following the program) as opposed to an increase in the rate at which an individual might be employed in the future. Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01. Note: # in formal sector employment Figure 2: Impact of the program on formal employment by type of employer Note: Two respondents in the control group claimed to be employed with their ALMP employer even though they did not complete the ALMP. This is why in the figure above the control group does not have 0 per cent employment by ALMP employers.
Similar to the results of other impact evaluations of ALMPs (McKenzie 2017), the program did not lead youth in the treatment group to be more likely to have earned cash or searched for a job compared to youth in the control group (see Table 5). However the program does have a positive effect on job seeking behaviour if we restrict our analysis to youth who did not have a job in the formal sector in the last six months. Among these youth there was a 24.2 percentage point difference between changes in job-seeking behavior between treatment and control groups.

Placebo test for employment and job-seeking behavior
We illustrate the robustness of the results in Section 5.1 above by conducting a placebo test of the equal trends assumption. As discussed Section 4.3 we compare the change in outcomes for individuals in the same intakes as the treatment (intakes 12 and 14) and control groups (intake 13) who were not eligible to complete the ALMP component due to their performance in the academic test. These individuals only completed the LIWP component of the program. The results of the placebo test are shown in Table   6 below. There are no statistically significant differences in outcomes, which provides further evidence that the equal trends assumption is reasonable. Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01. # in formal sector employment The outcomes shown in Table 6 are not directly comparable to those shown in Table   5 because the time between the follow up survey and when youth exited the program is substantially different. Youth in the treatment and control groups did not exit the program until a much longer period of time had passed (in some cases over a year) than youth in the placebo analysis above (see further discussion in Appendix 3). As such the follow up survey was around 18-24 months after the youth in the placebo analysis exited the program, while the follow up survey was around 9-12 months after the youth in the treatment and control groups completed the program. This means that the youth included in the placebo analysis (those that were ineligible for the program) had spent on average almost twice as long in the labor market following the conclusion of the program as eligible youth in the same intakes. As a result they have had substantially more time following the program to find alternative employment opportunities than youth in the treatment and control groups.

Supplementary evidence from an employer survey
The findings above show the employment benefits for youth from UYEP are predominantly driven by employers continuing to engage ALMP participants beyond the end of the program. To verify this result and to provide a qualitative understanding of the degree that each component of UYEP contributed to this, we surveyed a sample of 65 out of 100 employers involved in the program. The sample consisted of all the largest employers (those that had more than 10 ALMP placements over the last two years) and a random selection of the remaining employers who had at least one ALMP placement over the last two years. This survey showed a very similar rate of post-program employment for youth who completed the ALMP (20 per cent) as was reported by program participants themselves (discussed above). Employers repeatedly highlighted the importance of each of the three components of UYEP (the provision of a subsidized wage, the training provided to participants, and the job matching process) in playing a key role in why they participate and continue to employ youth beyond the ALMP placement. The findings of the survey that relate to each of these components are discussed below.

Subsidized wage
Employers involved in UYEP did not need to pay the ALMP participants throughout the five-month placement (their wage was provided by the program) and this appears to be essential for at least three reasons. Firstly, most employers would not have participated in the program to the same extent in the absence of a subsidy. We captured employers' willingness to pay by asking how many ALMP placements they intended to offer over . Finally, the subsidy allowed some employers to reduce the total labor costs of their business as 28 per cent reported that they would have more staff in the absence of UYEP (compared to 10 per cent that said they would have fewer staff). This suggests a displacement effect from the program as fewer jobs may have been available in the labor market for non-participants. However, the extent of the displacement effect is unknown and it may not fully offset the employment benefits created for the program participants.

Training
The importance of the training provided to youth to overcome the skill shortages in the labor market was consistently highlighted by employers. One of the main barriers to employers hiring more staff is that "training new staff is too costly", with this option being listed as one of the top three reasons provided by 52 per cent of employers. The popularity of this response is even higher than "economic conditions" (listed by 49 per cent of employers), which is noteworthy given the downturn in the non-mining sector in PNG. In light of this, it is not surprising that 51 per cent of employers listed the training provided to youth prior to ALMP placements as the most important aspect of the program. Further, 49 per cent of employers wanted the amount of training provided to youth prior to the placement increased, which was the highest response provided in regard to a question about how to improve the program.

Matching
Employers appear to have used the ALMP placements as a low-cost way to screen youth prior to deciding whether they wanted to offer them a job. More employers stated that the main reason they offered jobs to ALMP youth was due to the performance of the youth undertaking the placement as opposed to any need for additional staff (52 per cent compared to 37 per cent). This is reinforced by the finding that one of the main barriers to hiring more staff employers face is that "staff with the right skills are difficult to find" (52 per cent of employers included this as one of the top three answers). Further, employers indicated that UYEP was the third most common way they recruited new staff, after newspaper advertisements and job agencies.

Discussion of the results
This study has shown that youth who participated in the ALMP component of UYEP were around twice as likely to be employed in the formal sector 9 to 12 months after the program compared to those youths who just completed the LIWP component. This is The ALMP component of the program is substantially costlier than the LIWP, which means that even though the employment benefits are larger this might not justify the additional investment. The marginal cost of funding an ALMP placement beyond a LIWP placement in UYEP was around USD1500 per person. A back of the envelope calculation suggests that the large employment effects above would need to persist for well over three years for the additional income gained by program participants to outweigh the cost of the program. This means that despite the relatively high post-program employment rates due to the ALMP component, the benefits might still not be as large as the cost. This is consistent with the existing evidence that shows ALMPs rarely have cost benefit ratios greater than one (McKenzie 2017, Blattman and Ralston 2015).
The findings from this impact evaluation also suggest two limitations of the ALMP component of the program, which relate to who the youth are employed by and who benefits from the program. Firstly, the large effect on employment from the ALMP component is due to youth continuing with the employers they were placed with through the program. This would suggest that the skills these youths gained were mainly firm specific, and may not necessarily transfer to other firms and industries. This limitation of subsidized work placements has been shown in other studies (Alfonsi et al. 2017).
As such there is little scope for scaling up the program because there is limited demand from the small number of firms in Port Moresby where an ALMP placement is feasible.
A second limitation is that even though we show large employment impacts for youth in the treatment group, the design of our study means we cannot rule out the possibility that the ALMP component did not increase the total number of jobs in the labor market. A displacement effect could have occurred whereby ALMP employers would have hired more people from elsewhere in the absence of the program and as a result total employment could have remained unchanged. Few studies have been able to rigorously test whether these general equilibrium effects exist (Groh et al. 2016).

Conclusion
This study illustrates that youth who participated in both ALMP and LIPW components of UYEP are substantially better off in the labor market than those who only completed

Acknowledgements
The authors are grateful to seminar participants at the Australian National University and the University of Papua New Guinea. We are also very grateful for comments provided on an earlier version of this paper by Sonya Woo, Stephen Howes, Mathias Sinning and an anonymous reviewer.

Availability of data and materials
The data that support the findings of this study are available from the National Capital District Commission (NCDC) of Port Moresby but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of the NCDC.

Competing interests
The IZA Journal of Development and Migration is committed to the IZA Guiding Principles of Research Integrity. The authors declare that they have observed these principles.

Funding
There is no funding source that needs to be acknowledged in this manuscript.

Disclaimer
This paper presents independent analysis and is solely the responsibility of the authors.
The views presented in this paper should not be attributed to the World Bank.

Authors contributions
CH analyzed the data and drafted the paper. DN designed the surveys and managed the data collection. All authors read and approved the final manuscript.  Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01.

Further examination of the equal trends assumption
In the body of the paper, we illustrate that the employment outcomes of youth in the control group are substantially worse than the treatment group at the time of the follow up survey (Table 5), however there are relatively little differences between the employment outcomes of youth in the control group and either group in the placebo analysis (Table 6). At first glance this may draw into question the validity of the equal trends assumption between eligible youth in intakes 12 and 14 (the treatment group) and eligible youth in intake 13 (the control group). Assuming eligibility is positively associated with performance in the labor market, we would have expected that eligible youth in intake 13 would have achieved employment outcomes substantially better than ineligible youth in intakes 12, 13 and 14 (the groups used in placebo analysis). We show in Tables A4 and A5 below the point estimate for differences in employment outcomes are not very large (and rarely statistically significant).  Standard errors in parentheses. * p< 0.10, * * p < 0.05, * * * p < 0.01. Note: # in formal sector employment We attribute the limited differences in employment outcomes (that are largely statistically insignificant) to differences in the timing of the follow up survey and when youth exited the program as opposed to raising a significant concern about the equal trends assumption. Eligible youth in intakes 12 and 14 (the treatment group) and eligible youth in intake 13 (the control group) would on average wait around 6 months from completing the first stage of the program (the PWP) until beginning the second stage (the ALMP for youth in the treatment group and the alternative opportunity offered to the control group). This is because the timing of the recruitment of intakes do not perfectly match the availability of positions in the second stage of the program. Therefore eligible youth in a given intake do not exit the program until a much longer period of time has passed (in some cases over a year) than ineligible youth in the same intake. As such the follow up survey was around 18-24 months after ineligible youth exited the program, whereas the follow up survey was around 9-12 months after eligible youth completed the program. This means that the youth included in the placebo analysis had spent on average almost twice as long in the labor market following the conclusion of the program as youth in the treatment and control groups in the same intakes. As a result, they had substantially more time following the program to find alternative employment opportunities than youth in the treatment and control groups who exited the program without an ongoing job.
This issue of timing is particularly acute for eligible youth in intake 13 as they did not have any potential for future employment from the alternative opportunity offered to them as it was a one off event. Therefore all eligible youth in this intake were unemployed immediately following the conclusion of the program. In contrast, at least 20 per cent of youth in the treatment group (this is what is captured in the follow up survey 9-12 months after the program) had a job with their ALMP employer at the conclusion of the program. We show in Figure 2 there are no statistically significant differences between employment outcomes for youth in the control and treatment group who did not have a job following the program (i.e. among youth that did not continue to have employment with the employer they were matched with as part of the program), which is consistent with the equal trends assumption. This suggests the effect of the ALMP on formal sector employment is a level shift in outcomes (i.e. increase in the likelihood of having employment immediately following the program) as opposed to an increase in the rate at which an individual might be employed in the future.
To provide further evidence in favour of the equal trends assumption we conduct two additional types of analysis whereby we use ineligible youth from intakes 12, 13 and 14 as control groups as opposed to the eligible youth in intake 13 that are used as the control group in the main regression analysis. Specifically, we conduct the identical analysis as presented in Table 5 however in one case we use ineligible youth in intake 13 as the control group and in the other case we use ineligible youth in intakes 12 and 14 as the control group. We present the findings of these additional types of analysis in Tables A6 and A7 below. The main results hold using these alternative specifications. Table A6: Differences in outcomes between youth in the treatment group and youth who were ineligible in intakes 12 and 14 (1) (2) t-test Ineligible youth in Treatment (2)