 It's actually very interesting doing this presentation because this presentation wasn't really the key focus of our study, it was more an afterthought and the afterthought also led to one of the problems that this study has got because we included the question that we use quite late in our study but I will come back to that as well. So we were actually interested in finding out if an active labour market policy could actually, how an active labour policy could affect actually happiness. So this doesn't necessarily follow on the first two presentations, it's not on subjective well-being in terms of poverty, it's really a sort of life satisfaction. And very much like my first speaker, why do we care about happiness in the first place? Again we sort of very much from the labour economic side and we also diving into this area only recently, so this is the first time that we are actually presenting these findings and comments are most welcome. But what matters for us in that sense actually, I mean like there is definitely an established correlation between employment and self-reported happiness, okay, and that has been actually shown over and over again, especially sort of we know that people that are employed report to be much more happy than the people that are actually unemployed. So actually could self-reported happiness be somewhat used as a policy indicator, okay, in two ways actually, as a measure of success of a policy but most probably also as an indicator of who to target in the policy. In South Africa for example there is a long lasting debate about the correct definition of unemployment and particularly should the discouraged workers be included in the unemployment rate or not. And it started with a paper with King and the Knight, it's recently sort of updated by paper by Lloyd and the labourer as well and I think the idea is that they show that actually happiness can be used to make a strong argument that people that stopped searching are not necessarily happy, they are actually very unhappy about their state of life. Okay, so again so more on the policy side but what was interesting for us in particular was more the literature that started to develop around the effect that actually the mood will have on economic behavior and I'm talking about a paper by Ifcher and okay I struggled to say the name Zargami, if I said it incorrectly I apologize. And the basic message was that happiness actually can reduce the live for the day behavior so people actually if they're happier will be a little bit more forward looking rather than just sort of live into the day and actually sort of engage in economic activity or behavior that will have long term benefits. So how do we link all this to an active labour market policy? One paper that we looked at was written by Benjamin Crust who used a German active labour market policy, it was a subsidized employment project and he wanted to find out if the subsidized employment project actually affected happiness of the exposed participants, of the participant of the project. He used two methods, he used propensity score matching and actually showed that people that are in that subsidized program were significantly happier than people that were remained unemployed and secondly he showed that when this program came to an end and the employment probability for the people that were in the program dramatically dropped okay because they were not necessarily re-employed straight away, irrespective of the fact that they remained or that they had access to income due to the unemployment insurance their level of happiness decreased significantly. Okay so based on that we actually wanted to see how does that actually relate to South Africa. To say in the theme of the conference, so you are the most excluded group in the South African labour market and we looked at the youth unemployment problem in South Africa, and in particular we are very concerned about the unemployment rates of African youth. These are broad unemployment rates so obviously I buy into the argument that a broad unemployment is a much better measure of unemployment in South Africa and we can see that over the period from 2008 to 2012 African youth the age between 2024 experienced unemployment rates brought unemployment excess of 60% okay up to 66% actually. Colored youth is not necessarily much better off but still significant better off than the African youth so African youth really struggles to get into the labour market and we also know that the 20s is a very crucial time period for people to actually finish education and transition into labour markets. So based on that the Harvard group, the so called Harvard group, one of the experts on at Jim Levinson actually wrote a paper we suggested that South Africa should experiment with the youth wage subsidy or hiring voucher in order to make it cheaper for firms to experiment with young job seekers whose productivity levels are unknown. The research unit, the African economic research unit and sorry I should have, I apologize for that, all the authors on this paper are part of this research unit so it's a very concerted collaborated off effort and our research unit was commissioned by the National Treasury to actually test the logistics and also the mechanisms of hiring voucher for South Africa for that youth group. So the project itself actually looked at 2024 year olds they basically got a voucher for 5000 rent which lasts at least six months which can pay up to half the wage with a maximum of 833 rent and the company basically had to climb back. So a person would go with the voucher to a company and say if you hire me this institution will reimburse you for part of the wage that you will pay for me. Where was the study implemented, sorry so first of all it initially started in 2009 and interviewed around 4000 young people we had a second baseline in 2010 and also the allocation of vouchers was done in 2010 with a follow-up in 2011 and in 2012. The areas that were covered were Houtang, Durban and Polokwane so two more urban areas and one rural area as well as two sampling strategies one was to actually include labor centers so we actually got respondents from labor centers the other one were actually enumerator areas. So what was the effect itself of the wage subsidy? It had a relatively large and robust treatment effect with sort of an up limit of 7% points on the employment probability of the treatment group. So the treatment group definitely experienced quite a large employment effect and that holds for various specifications as well. So the question now clearly is what has happened to the happiness? The question that we use and as I said I mean it was after thought one of our enumerators not enumerators but one of our survey managers decided that would be a nice question to investigate one day so they basically just snuck it in. So in 2011 we don't have a baseline that's the reason why I'm pointing this out. In 2011 we asked our respondents to indicate how they actually feel about the life in general early up in the survey so that it's not influenced by other questions. This follows a typical like it scale with very unhappy to very happy five different outcomes. We transform this into we follow sort of the literature and transform it so we turn very unhappy into one and then all the way up to five for very happy. So what is our prior? Given the treatment effect on employment and given the literature that actually suggests that employment should actually be correlated with higher levels of happiness we should see that on average the treated group should report more higher levels of happiness. So the red ones is the treated group and the blue one is the control group. Well first of all what we find is well this one sticks out. The treated report to be more unhappy in the unhappy classified group the control more indifferent or sort of happy whereas then we've got again a small fraction of treated that report to be very happy. Well I looked at it and looked at it over and over again I don't find a pattern there. Other than there isn't a pattern which is surprising which is surprising. So obviously there is this relatively stronger effect on being unhappy but we also have got this effect that they are more happy on this side. So how much do we actually pick up when we run this as a regression? We're starting off with a straightforward OLS with and without controls and then ordered probe it and now we'll come just now to the propensity score match as well. What we find is we have a negative effect. So again that in itself is actually very surprising because as I said we've got this large employment effect but it's incredibly marginal. It's actually not significant at all and that is sort of quite that's the story that we get across our different specifications with or without controls. So what is the story with propensity score matching? Because we don't have a baseline. It could be that we simply had a larger number of very unhappy people in the treated group who after they now got treated and had experienced more employment might also now come to the same level of happiness as the control group. So we don't know. What we did with our propensity score matching is we matched on characteristics in 2010. So we matched the group in 2011 on characteristics in 2010 that are highly correlated with life satisfaction. So for example including education, well a lot of the ones that were mentioned here we haven't thought about yet but now that I was here now we can include them. But we've got self-reported health, number of children being a parent, being the marital status, individual characteristics, employment status as well. So things that we could think of would affect life satisfaction. So even if we control for that and we used a number of different algorithms to actually test for the sensitivity of the, the result doesn't change much. Okay, so but why could that be? Why could it be that on average we don't pick anything up? Well maybe the story is actually that the people that actually did get employed through the treatment effect have a very positive effect, whereas the people that didn't get employed despite being treated, so despite having a voucher and they still can't find a job, maybe they got incredibly dissatisfied with life. So maybe what we just have here is the average effect of those two groups, of these two effects where treatment actually just had opposing effects to that. So we split the sample and I know that this is not, that's not the correct way to do it. Again we'll talk about this just now. So we split the sample and just wanted to see, okay, are the people that are actually unemployed in 2011, treatment and control group, are they significantly different? So is it really driven by these two very opposing impacts across the, the employment status? Yes, we do pick up actually this strong, not strong, but like again this, this stronger effect of the unemployed in the unhappy group and we sort of get rid of the very happy ones. But again it's not extreme, it's not a very strong effect, right? So we again try to see if it holds when we actually run through various regressions. We have the negative coefficient as we would predict, okay? We again do PSM and that's the reason why I was saying this is clearly not the best way to do it. Because the two groups clearly aren't the same anymore, they cannot be the same, okay? The unemployed in the treatment group are very different now to the unemployed in the control group for the mere fact that anyone who was treated and successfully transitioned into employment would have been potentially in the unemployed group still, okay? So there, the comparison itself is problematic and we are fully aware of that but it was more for us to actually see and unpack if there is something in it. So on the employed side do we then sort of see a stronger effect on the employed side? Again we've got a positive effect but it's not significant, we do the same thing with the pregnancy score. So nothing there, okay? Or very, very little is there. So then we actually thought, okay, maybe it's sort of treatment heterogeneities and we should look at the initial state or the state in 2010 and sort of interact that with the treatment. This is not the treatment itself but this just shows people that were previously employed, the red ones previously employed, so this group is just previously in 2010 and where do they actually move in 2011? People that are unemployed in 2011 compared to people that are employed to 2011 despite the fact that both groups were actually employed in 2010, okay? People that are employed report higher levels of satisfaction, okay? Again, we can't put, could, but it definitely shows that employment is again correlated with higher levels of satisfaction, causally we can't say. The same holds for the people that were initially unemployed, okay? Unemployed in 2010, unemployed in 2010, the blue line are the ones that are employed in 2011 whereas the red ones are the ones that are unemployed in 2010. Again the unemployed report lower levels of satisfaction. We don't know if they were more dissatisfied or unhappy already in 2010, it's impossible for us to unpack it, okay? So, but what we do is just to sort of get somewhere with it as well. So we interact the treatment with their previous employment status. So the employment status in 2010 and we actually want to see, okay, can we pick up any change or differences in their reported happiness on that level? Our comparison group are treated previously unemployed in 2010, okay? We compare them against treatment that are previously employed as well as control previously employed and control previously unemployed. This is the interesting part. These are the controlled previously unemployed and again we don't find any difference in their reported happiness, okay? But the previously employed, irrespective of its treated or control report much larger, much higher levels of life or happiness actually in 2010. Well, one reason for that is we know that employment in 2010 is highly correlated with employment in 2010. Okay? So for us to unpack that, it's actually very, very difficult. Okay, what are some of the problems that we've got and just to come back to that? So the biggest problem clearly is the baseline, the baseline story. Because we don't know if we are balanced in terms of happiness in 2010, it's very difficult for us to actually make a lot of the claims. However, we do find that, I don't know the words, it's switched off. We do find that there are actually balanced on most other characteristics. Actually for 2010 and 2011, the samples, the control and the treatment are really well balanced over the observed characteristics of interest, okay? Non-random attrition. So obviously what could also happen is that with the treatment, because we would expect normally a much larger positive effect. So with the treatment, it could be that people that actually were successful, okay, and now on the drop mark, they don't have a need to engage with us anymore, okay? So potentially actually attrition is on employment status, okay, but from 2010 to 2011. So the ones that were treated but were unsuccessful hope that we can still help them and therefore remain with us, okay? So maybe it is actually a non-random attrition story. We started, so what we find actually does in terms of correlates with attrition. We've got less of a problem with that for 2011 than for 2012. So therefore I'm also not using 2012 at this point for this. But on unobserved characteristics, it's difficult to say, right? So we could work with Lee Barnes to actually just see how sensitive our results are actually to this non-random attrition. And finally, as a limitation, we need to look at the mediator analysis, okay? So following along the lines of my at all, we need to actually think about the causal mechanism through which actually the three variables of interest are interlinked, okay? So treatment, but treatment is supposed to affect employment where employment affects happiness, okay? It could equally be that treatment affected happiness because it did not get employed, which affects how people behave and that actually it becomes a self-fulfilling prophecy. They, for example, search less and therefore are more likely to stay unemployed, okay? So in which way, I mean, which one is now the mediator and which one is now our outcome variable of interest? But we are still playing around with that. So the conclusions. So we clearly show that, again, the employed are more likely to report higher, or are reporting higher levels of happiness compared to the unemployed, but across both groups, treatment and control group. The causality there or the direction is for us not really able to, we can't really pick this up. What does it mean? What do we show for the treatment? We definitely, the fine suggests a positive treatment effect on happiness, okay? But it's only marginal. What we also clearly have is the positive effect for is for employed, but there seems to be also a negative effect for people that actually remain unemployed. So the net effect for us is basically indistinguishable. Why is it so surprising though? It is surprising because of the time period that you're looking at. The intervention was implemented as the financial crisis hit South Africa. The largest group that suffered from the financial crisis, okay, were African youth. So we would have expected that a policy that actually gives access to employment, which happened, should actually have led to a much larger level of happiness. We don't see that. So what could that mean, actually? One possible explanation is that for this particular age group, okay, employment is not as high a determinant of happiness as prudential for other age groups. Maybe it is an experimental phase though, 20 to 25, okay? If I have a job, yes and no. I mean, I would like a job, but maybe it's not as important in order to determine my happiness. I would like to leave it there. Thank you very much.