Poverty and Employability Effects of Workfare Programs in Argentina

In 1993 Argentina began implementing workfare programs, and workfare has become a central public policy starting 2002 when the government increased the number of beneficiaries from 100,000 to 2 million people in a country of 38 million. We explore targeting, poverty and employability effects of workfare before 2002 based on the permanent household survey (EPH). We find that the program was pro-poor although more than one third of participants did not satisfy the eligibility criteria. Our estimates suggest that the income of participants increased during treatment - particularly for women - indicating beneficial short run poverty effects. However, the long run effects of the program are not obvious due to selection on treatment completion. We present evidence suggesting that - for a large fraction of participants - the program generated dependency and did not increase their human capital.


Introduction
Argentina suffered a deep economic, social and political crisis in the last few years.
The economy shrunk by about 11 percent in 2002, and due to the currency's depreciation, GDP per capita dropped off to approximately US$ 3,000 (down from US$ 8,000 at its peak in 1998).
The crisis sharply aggravated the country's already difficult social situation. During 2002 poverty and unemployment reached their maximum historical level: more than 50 percent of Argentine households were below the poverty line, and almost 20 percent of the labor force was unemployed. Unemployment was particularly severe among the least-skilled workers, the rate being higher than 30 percent. This extremely negative context also had an impact on the education and health sectors where there is growing evidence of deterioration in service delivery. The combined effect of all these factors was an increasingly volatile social situation with high levels of violence and protests (see Fiszbein et al. 2002). In 2002, one of the main policies implemented by the government to deal with the crisis was to significantly increase the budget allocated to active labor policies. The number of beneficiaries of workfare programs increased from 90,000 in December 2001, to 1,200,000 in October 2002, and to 2,000,000 in 2003. The recent decline in unemployment and poverty has been presented by the government as evidence of the positive income and employability effects of workfare programs (Ministerio de Trabajo, 2003).
Allocating more funds to social sectors, and particularly to labor programs, seems to be an adequate policy considering the current difficult situation. However, several questions have been raised in Argentina regarding the fairness and effectiveness of workfare programs. These programs have been pointed out as a source of political clientelism and corruption1, and many analysts argued that their employment effects are questionable. In spite of the topic's importance, most of the arguments are based on anecdotal evidence: there are very few empirical evaluations of these programs. Our research objective is to contribute to the debate by providing an econometric evaluation of the poverty and employability effects of workfare programs in Argentina, using the Encuesta Permanente de Hogares (Permanent Household Survey, hereafter EPH).
While our focus is on the Argentine case, we consider that the study is relevant to other countries, particularly those in Latin America, where active labor policies have been advocated as a way to soften the shocks generated by market-oriented reforms (Heckman et al., 1998;Goldbert L. and C. Giacometti, 1998;Marquez, 1999).
The paper is organized in five sections. The next section briefly describes the characteristics of workfare programs in Argentina. The third section presents our research objectives, a review of the empirical evidence and the knowledge gaps. The fourth section presents the methodology and the data. The fifth section presents the results, and the last section cites our conclusions.

Brief background of workfare programs in Argentina
Currently, Jefes de Hogar is the main workfare program. It was implemented a few weeks after president Duhalde took office in February 2002. However, workfare programs in Argentina have been implemented since 1993, and while the program names have changed 2 , they all have all the same basic characteristics and objectives 3 . In this paper we 1 Ronconi (2001) surveys the main Argentine newspapers and finds that most of the press reports related to workfare programs mention the existence of political clientelism and corruption in the funds allocation process. 2 In 1993 it was called Programa Intensivo de Trabajo, from 1995 to 2001 Programa Trabajar, and since 2002 Programa Jefes de Hogar. Provincial governments also implemented their own workfare program with similar characteristics to the federal ones. In terms of magnitude the most important program was Barrios Bonaerenses implemented by the provincial government of Buenos Aires. evaluate workfare programs during 2000 and 2001 (i.e. before the implementation of Jefes the Hogar). The main program during these years was called Trabajar. The common features of these workfare programs were as follows: • These programs targeted the least-skilled unemployed workers, preferably the heads of household. People who receive unemployment insurance benefits or a pension, or hold a job (even if it is in the informal sector) were not allowed to participate 4 .
• Participants received a monthly benefit below the minimum wage 5 , during a certain period (between three and six months) paid by the government 6 .
• During that period participants received training and had to work between twenty and forty hours per week 7 on communitarian projects at public or non-profit organizations 8 .
• The objectives of the program were: To act as a short-term safety net, and to increase employability among the least-skilled unemployed workers.

Research objectives and knowledge gaps
The following three components help define our research objectives:

Program targeting
A review of workfare programs in OECD and some developing countries found that public-service jobs are well targeted at low-income unemployed workers when the wage rates have been set very low (See Dar and Tzannatos, 1999).
In Argentina, the benefit is below the minimum wage so we might expect selftargeting as argued by Jalan and Ravallion (2003). However, the state has low enforcement capacity, hence some benefits might be assigned to individuals who are not unemployed, but already hold a job. However, a second and more important concern is that, due to lack of and not economic need (e.g. benefits assigned to friends, relatives or clienteles of influential politicians). Furthermore, anecdotal evidence suggests that these beneficiaries do not comply with the workfare work requirement. Therefore, in a poor institutional environment such as the one that characterizes Argentina, it is not necessarily true that imposing work requirements and setting the benefit below the minimum wage implies self-targeting. Kremenchutzky (1997) and Ministerio de Trabajo (1999) have surveyed a small number of workfare program participants (60 and 159 respectively), and they find few cases (less than 10%) where participants do not meet the eligibility requirements to receive the benefit (i.e. the participant is well-educated or already holds a job) 9 . Ronconi (2001)

Poverty effects
A second concern is related to the poverty effect of workfare programs. Even in the case where the program is well targeted, it is necessary to measure the income gain conditional on income in the absence of the program, to assess its impact. Common practice has been to estimate the gains by the gross wages paid, assuming that the labor supply to the program came only from the unemployed and from people who were out of the labor force. But, even if a participating worker was unemployed at the time she joined the program, it does not mean that she would have remained unemployed had the program not existed.

6
Ministerio de Trabajo (2003) argues that the Programa Jefes de Hogar helped 29.3 percent of households that were below the indigence line to move out of indigence, and 6.5 percent of households that were below poverty to become non-poor. This 'estimation' is done assuming that benefits are targeted towards the poorest, and that the income gain of participating in the program is equal to the benefit. For the aforementioned reasons this analysis is not very informative. Jalan and Ravallion (2003)  In this paper we compute the average net income gain of workfare programs 13 , using a different database (i.e. the Permanent Household Survey) and a matching pairs approach 14 . The data and our empirical approach allow us to estimate the short and medium run poverty effect of workfare programs 15,16 .

Employability effects
Finally, we assess the employability effects of workfare. According to Bartik (2001), public service programs significantly increased the long-run earnings of participants in the US, since they provide some work experience and the needed soft skills. Do we observe this 12 Jalan and Ravallion's (2003) results are based on a sample of 2,802 participants. Is this a representative sample of workfare participants? The authors randomly selected 350 projects, and mention that some participants were dropped from the random sample because their addresses could not be found, or because they did not want to respond. Assuming that the average number of participants per project is 20 -which is a conservative estimate-implies that 350 projects includes 7,000 participants. Ronconi (2001) questions how representative are the analyzed 2,802 participants. 13 The EPH allows distinguishing participants from non participants, but it does not inform in which specific workfare program the participants are participating. During the period we analyzed (i.e. 2000-2001) the main program was Trabajar, although other federal and provincial workfare programs were in place such as Programa de Emergencia Laboral and Barrios Bonaerenses in the province of Buenos Aires. 14 Since we do not follow a general equilibrium approach, we ignore indirect effects such as an increase in income due to the increase in aggregate demand generated by the program. These indirect effects were probably small before December 2001 because the number of participants was 1 percent of the labor force. However, they presumably have become important after 2002 when the government increased the number of participants to almost 15 percent of the labor force. 15 We estimate if the direct income gains generated by the program helped the participants to move out of poverty and/or indigence. We use the official poverty and indigence lines which are described in the next section.
effect in the Argentine case? How did participants perform in the labor market after treatment? Are participants more or less likely to be employed than individuals in the control group after program completion? Does participation affect the odds of getting a formal job 17 ?
Do ex-participants receive higher wages due to treatment? Or is the workfare program a disguised income transfer? Furthermore, did workfare have any negative impact, such as 'signaling' or stigma effects on participants? Did the program generate dependency among participants?
None of these questions have been appropriately answered in Argentina. As far as we are aware, there are no statistically reliable evaluations of the employment effects of workfare programs. Our objective is to contribute towards filling this gap, exploiting the panel characteristic of the EPH and implementing a matching pairs approach to construct the control group 18 .
We also analyze the predisposition of employers to hire workfare program participants, based on a poll conducted during 2002 by the Ministerio de Trabajo (Encuesta de Indicadores Laborales).
To summarize, our research objective is to analyze how well targeted is the program, and how effective it is in reducing poverty and increasing employability.

-Methodology and data sources
The empirical strategy adopted in this study is the result of the research objectives advanced in the previous discussion and the characteristics of the available data.
The main analysis is based on the Permanent Household Survey (EPH There are two additional advantages of using the EPH as the source of information to estimate the effects of the program: First, the same questionnaire was administered to both participants and non-participants. Second, the EPH contains information regarding the urban agglomerate where the individual works, allowing us to construct the comparison sample with individuals who reside in a similar local labor market as program participants. As Heckman et al. (1998) point out, these two characteristics of the data prevent important bias from arising. Furthermore, they show that bias due to the use of different surveys and differences in the distribution of participants and comparison groups across the local labor market is often large relative to selection bias.

Targeting
To assess how well targeted workfare is, we analyze several socioeconomic

Poverty Effects
A simple way to analyze the income effect of workfare programs during treatment is to compare the income of participants during treatment relative to their income before In order to compute the program's short-run poverty effects, we need to measure the income gain conditional on income in the absence of the program (Heckman et al., 1998).
The "with" data is provided by the EPH (i.e. we observe the income of participants Let X i be a vector of variables that helps predict participation in the program; and P(X) = Prob(D=1/X) is the probability of participating conditional on X, the propensity score.
We calculate the propensity score for each individual in the Permanent Household Survey using standard probit model. In order to ensure that participants and their matched non-participants are affected by the same local labor market conditions we run six separate regressions, one for each region 21 .
Then, for each region, we select from the group of non-participants the five individuals who have the most similar propensity score to each participant. These selected non-participants constitute the comparison group. 20 The reason we use a non-experimental method to estimate the effects of the workfare program is because experimental data is not available. Also, we use a control group drawn from "external" sources (i.e. non-participants) because there is no available information to distinguish between those non-participants who applied but were not selected from those who did not apply. Bell et al. (1995) provides a clear exposition about the pros and cons of using a control group drawn from 'external' versus 'internal' sources.
One way to measure the mean income effect of the program during the period when participants are receiving the benefit is by estimating µ: (1) This estimator is based on a cross-section of individuals so it does not confront the problem of attaching changes in the macroeconomic situation to the program. Also, by construction, the estimator controls for observed characteristics of individuals. However, a potential problem with this estimator is that program participants and their respective nonparticipating nearest neighbors may differ according to unobservable characteristics, and if those unobserved characteristics also affect labor performance the estimates would be biased.
However, by considering the income of both participants and the control group in a period before treatment and computing a difference-in-difference estimator, it is possible to remove the bias generated by time invariant unobserved factors. The difference-in-difference estimator of the income effect of the program during treatment is given by coefficient δ: is the change in income of individual i between periods t and t-1, and treatment takes place during period t. The sample is also restricted to participants and the comparison group.
Finally, we compute the percentage of participant households who moved out of poverty using the average income gain estimated via equation 2. While this strategy does not allow us to claim any long-term effects over poverty reduction, at the least it assesses the importance of workfare programs as short run safety nets.

Employability Effects
Our third objective is to estimate the employability effects of workfare. We evaluate if participants are more or less likely to find a job in the formal sector, earn a higher wage and a higher hourly wage after program completion. As in the previous case, the fundamentally unobserved data is the employment performance and salaries of participants 'without' treatment. Again, our empirical strategy is to use propensity score matching methods to draw a comparison group to workfare participants from the large number of non-participants available in the EPH.
But unlike the previous case where we were interested in the income effect during treatment, now we need after-treatment outcomes. Taking advantage of the panel structure of the Permanent Household Survey we compute 'before and after' treatment difference-indifference estimators. For each of the dependent variables of interest, we run a model similar to equation (2) except that in this case we focus on changes in Y i between t+1 and t-1 (Recall that t refers to the period when participants receive treatment).
Considering that the rolling panel structure of the EPH allows to follow the same individual for four waves (i.e. two years), we can estimate the employability effects of the program six and twelve months after program completion.
At this stage an important point should be re-emphasized: During the period under consideration, the overall state of the Argentine economy suffered major changes. GDP per capita decreased 25 percent, the unemployment rate went up 3 percentage points, and the share of informal employment increased from 37 percent to 50 percent. Under such a crisis, it would be incorrect to attach all the negative changes in participants' outcomes to the workfare program. In other words, a 'before and after' estimator based exclusively on a sample of participants should be discarded, or at least taken with extreme caution. However, since we also work with a comparison group of non-participants, and under the assumption that the crisis had a similar effect over the outcomes of participants and their respective nonparticipants nearest neighbors, we can isolate the workfare program's effects from the economic crisis by computing a difference-in-difference estimator.
Summing up, we follow the conventional evaluation literature. The value added of our paper is that we use these standard methods to explore a database and answer several questions that have not been analyzed yet 22 .
of observable factors. Obviously, unobserved heterogeneity is a potential problem, and we discuss it below. 22 Let us mention that there are several potential extensions to the methodology we follow in this paper. For example, instead of taking May 2001 as the unique treatment period, a more flexible methodology would be to construct a larger dataset including those individuals who received treatment during different periods of time; in which case, time dummies should be included to account for changes in the macroeconomic performance. A second plausible extension would be to construct the comparison group not only based on the characteristics of participants and non-participants during treatment, but also on their characteristics before treatment allowing controlling for Ashenfelter's dip. While this alternative clearly improves the quality of the matching, we decided not to implement it because it significantly reduces the number of observations in our sample. (Recall the rolling panel structure of the EPH). In any case, these may be constructive extensions to this paper, and we leave them for further work.

Targeting
In this section we analyze how well targeted towards the low skilled and unemployed workers were workfare programs in October 2000 and May 2001, before the large expansion occurred in early 2002 (i.e. before the Jefes de Hogar program was implemented).
During the periods we note that the number of beneficiaries was a very small share of those who needed support. While beneficiaries were approximately 100,000, the number of unemployed people was 1.5 million, and the number of people living in households below the poverty line was 9 million. Therefore, we expect to find that a large share of poor and unemployed people did not receive the benefit. But were the scarce benefits allocated properly? Tables 2, 3 and 4 present basic socioeconomic characteristics for program participants and all non-participants.  We also observe that 18.2 percent of workfare participants not enrolled in school report at least one of the following sources of income in addition to the workfare benefit: a formal job, an informal job, self-employment income, pension or unemployment insurance.
This is preliminary evidence that a large share of the benefits have not been assigned to the poorest unemployed and least skilled workers as established in the normative. The inadequate allocation of benefits becomes more evident when we analyze household income per capita: Only 22.1 percent of participants were below the indigence line, 35.4 percent were below the poverty line but above the indigence line, and the remaining 42.6 percent were above the poverty line 23 . Table 4 shows that the average participant was poorer that the average non-participant, but the differences are not that large. On the other hand, only 4.8 percent of the total number of indigent households in Argentina had a member participating in the program. Note: The figures are for individuals. However, since indigence and poverty are defined according to household income per capita, we categorized a household as 'participant' if at least one member is participating in the workfare program. In October 2000, the number of participating households was lower than the number of participating individuals because 13 percent of participating individuals were members of a household that had two or more members participating in the workfare program. 23 These figures are computed including the benefit as a component of income. The alternative extreme assumption is to compute poverty based on the income of participants without including the benefit. As we mentioned before any of these alternatives is adequate: In order to properly analyze how well targeted was the workfare program according to family income, it is necessary to compute the income of participants in the absence of the program. This analysis is done in the next section. See table 12.
We also observe that, while most of participants are members of households located in the poorest quintiles, one fourth of beneficiaries are members of a household that ranks in the top 50 percent of the income per capita distribution in both October 2000 and May 2001.   Note: The sample is restricted to individuals 18 to 65 years old with complete information in all the variables used to define eligibility.
Summing up, the evidence suggests that the limited number of benefits was not appropriately distributed 25 . This evidence is consistent with the argument that setting work requirements and the benefit below the minimum wage are not sufficient conditions to ensure self-targeting in countries with a lack of sound political institutions such as Argentina 26 .

Poverty effects
The first and critical step in estimating program poverty effects is to find a comparison group of non-participants who has sufficiently similar characteristics to participants except for not participating in the program.
We run standard probit models -one for each local labor market 27 -to estimate propensity scores. The vector X i of individual and household characteristics includes the 25 As a caveat it should be mentioned that, while the INDEC assures the confidentiality of the collected information, it is always possible that some individuals do not report accurate information.
For example, an individual who participates in the program and does not meet the eligibility criteria might have the incentive to report that he does not participate, in which case the true targeting of the program would be worse that the figures presented above suggest. We suspect this is not a large problem because the eligibility requirements are seldom enforced by the government. 26 As Besley and Kanbur (1990) argue, perfect targeting (i.e. benefits allocated to the poorest) is not optimal in practice since it implies high administrative costs (i.e. the government has to spend too much money collecting information and monitoring if it tries to reach the poorest). However, we consider that the extent of misallocation of workfare funds cannot be justified by administrative costs. 27 We use the region as an indicator of local labor market. There are six regions in the EPH: Northwest, Northeast, Cuyo, Pampeana, Patagonica and Greater Buenos Aires. A more disaggregated definition of the local labor market is to use the urban agglomerate. There are 28 urban agglomerates in the EPH. We choose the region because there are several urban agglomerates with very few workfare participants impeding an adequate identification of a comparison group. However, we include indicators of urban agglomerates as controls in the estimation of propensity scores. 28 Individuals are categorized into three groups: Foreign born, born in a different province, and born in the same province where they reside.
In May 2001 the EPH reports 655 individuals who were participating in the six month workfare program. The group of non-participants includes almost 50,000 observations. From this large group we extracted the five nearest neighbors for each participant, creating a comparison group totaling 3,275 individuals. The following table presents the probit regressions, one for each region.   The average propensity score across regions for those who were participating in workfare is 0.038 while for non-participants it is 0.012, which indicates that these two groups have different observable characteristics. The propensity score for the comparison group (i.e. the group formed by non-participants that we selected as nearest neighbors) is also 0.038 making us confident to carry on our strategy 29 . It should be mentioned that this is not always the case. In those cases where the program is well targeted at a particular socioeconomic group and most of eligible individuals receive treatment it may not be possible for the researcher to draw a reliable comparison group from the sample of non-29 Appendix 3 presents basic statistics for participants and the comparison group. Both groups present very similar observable characteristics as expected. participants 30 . However, during the period being analyzed, the number of beneficiaries was a small proportion of the objective population of the program, and the scarce benefits were allocated to different socioeconomic groups as shown in the previous section. This situation allowed us to construct a comparison group with similar observable characteristics to program participants.
The first question we consider is by how much the monthly income of participants changed during participation in the program due to treatment. In other words, which would have been the income of participants during May 2001, if they were not beneficiaries? The first estimates we compute are based on equation (1) 31 . Table 9 presents the results. We find that, on average, participants had an income in their main occupation that is US$26 higher than the comparison group but a total income that is US$8 lower -but statistically not different from zero. We also find that participants have an hourly income in their main occupation that is US$0.87 higher. These estimates differ significantly across gender. On the one hand, female participants had an income in their main occupation US$77 higher, a total income US$39 higher and an hourly income US$1.2 higher than females in the comparison group. On the other hand, male participants had an income in their main occupation US$50 lower, a total income US$80 lower and an hourly income US$0.4 higher than males in the comparison group.
There are several possible interpretations for these results. One possibility is that participants and the comparison group differ in unobservable characteristics and the results reflect not only program effects but also selection bias. However, assume for a moment these are unbiased estimates and discuss selection bias later. 30 We suspect this would be the case if we had chosen October 2002 -or a more recent survey-as the base period since by that time workfare programs were almost universal. 31 The vector W i of control variables includes the same variables used to estimate the propensity score plus regional dummies.
It appears that females benefit more from program participation than males. This result is not surprising given the fact that program benefits are the same for both genders and women presumably have fewer opportunities than men in the labor market. The higher incentive that females have relative to males to participate may also explain the fact that more women than men actually participate in the program.
Another finding is that treatment appears to increase the hourly earnings and income in the main occupation, but has a negative/zero 32 effect on total earnings. This last result is driven by male participants, who present a lower total income than the comparison group. Why would men thus choose to participate if they could earn a higher total income in the labor market? An  Table 10). Therefore, a potential explanation is that male participants choose to participate because they may prefer to earn a lower total income but work less hours instead of working more hours in the informal sector. An alternative explanation is that they choose to participate because otherwise they would have remained unemployed. Table 10 also suggests that a large share of female participants were presumably 'inactive' (i.e. not working or looking for a job in the labor market) before joining the program.  (2) we find that (see column 1 in Table 11): Program participation is related with an increase on total monthly income equal to US$44.9, increase on monthly income in the main occupation by US$53.2 and on hourly earnings US$0.67 34 . The program's positive income effect during treatment could be explained by considering that many participants would presumably have remained unemployed or inactive, and hence without income, in the absence of the program. However, the estimated income effect is smaller than the benefit probably because many participants would presumably have gotten a job and worked more hours in the absence of the program.
We also observe that 154 individuals appear to be participating in the program during both May 2001 and October 2000, representing 43 percent of participants in May 2001. We discuss this issue later, but so far we want to emphasize that such a high rate of dependency is unexpected, since the normative establishes that the length of participation in the program is between three and six months.
An additional exercise is to compute poverty effects excluding all those individuals who actually participated in the program, but according to their socioeconomic characteristics were not eligible to participate. As we discussed in the targeting section more than one third of participants did not meet the eligibility criteria.   Considering that during October 2000 the number of beneficiaries was 100,000 we estimate that as a consequence of the program 6,000 households moved out of indigence, and 3,500 moved out of poverty. We expand these figures by average household size, leading to an estimated reduction in the number of people below the indigence line of 38,000 and a reduction in people below poverty of 19,700, implying approximately a 1.2 percentage point reduction in Argentina's indigence rate and a 0.2 percentage point reduction in the poverty rate. 35 Their estimation is for the year 1997 and they used a different database. These two factors might explain the discrepancy in the results. However, we suspect that the difference could also be due to a potential bias on the sample used in Jalan and Ravallion (2003) as discussed in footnote 12. 36 According to the news reports surveyed by Ronconi (2001), another plausible reason that explains why the program effect on income is smaller than the benefit is that a fraction of the benefit (between 5 and 50 pesos) was actually not received by the participant, but kept by local political bosses in exchange for giving people access to the program.

Employability effects
In this section we analyze the labor performance of participants after treatment. We estimate program effects using equation (2) as described in the methodological section. In other words, we study the labor market performance of participants after program completion, both with respect to their labor performance before entering into the program, and with respect to the comparison group.
We begin studying labor outcomes five months after treatment (i.e. October 2001).
Out   (1)  2001 in order to properly measure before and after treatment outcomes 38 . The estimates for this restricted sample are in column 2. We find that treatment is correlated with an increase in total monthly income by US$29.4, monthly income in the main occupation by US$24.1 and hourly earnings by US$0.09. Treatment is also correlated with a higher probability of being employed 3.7 percent, higher probability of having a formal job 4.3 percent, higher probability of participating in the labor force 9 percent and higher probability of being unemployed 5.3 percent 39 .
Before proceeding to discuss these results we should notice that, except for the estimate of the program's impact on the probability of participating in the labor force, all the other estimates are not statistically significant. Three different interpretations for the lack of statistical significance of the estimates are plausible. First, by taking into account that the labor market is not frictionless, it could be argued that ex-participants actually improved their skills during their participation in the program but were not able to take advantage of those new skills few months after program completion. According to this hypothesis it is necessary to analyze the performance of participants several months after program completion to uncover the employability effects of the program. A second interpretation is that the program actually had an impact on the performance of ex-participants even a few months after participation, but the results we find are statistically insignificant due to the low number of observations. A third interpretation is that participants did not improve their human capital during participation and the results are simply reflecting that.
Besides this discussion some results are worth considering. A few months after program completion ex-participants appear to have a statistically significant higher propensity to participate in the labor market. The labor force participation rate of the comparison group did not change much (it was 63.9 percent in October 2000 and 63.1 percent in October 2001), but the labor force participation rate of participants increased by 8.1 percentage points during that period. This result is mainly driven by females.
We now study program effects by analyzing the performance of participants twelve months after treatment. The May 2002 survey includes 116 individuals out of the 655 individuals who were program participants during May 2001 and were also surveyed in 37 When we compute the estimates by gender we find that females benefited more than males as before. The estimates (and t-values) for females are: US$35.2 (2.05) effect in total income, US$30.8 (2.16) effect in income in main occupation, and US$0.21 (1.08) in hourly earnings. For males the figures are: US$13.6 (0.45), US$15.1 (0.56), and US$0.02 (0.10) respectively. 38 Dropping participants biases the estimates as we discuss below. However, we consider worth presenting the estimates obtained using this restricted sample since they literally compare the income of participants before and after treatment with respect to the comparison group. 39 The seemingly contradicting result that treatment increases both the probability of being employed and unemployed is simply explained by the finding that treatment increases the probability of participating in the labor force by a larger amount than finding a job.  Table   14): An increase in total monthly income equal to US$59.7, an increase in income in the main occupation equal to US$48.5, and an increase in hourly earnings US$0.31. This result is unexpected since the normative establishes that the length of the program is between three and six months. While renewal of benefits was not explicitly prohibited, there was an implicit solidarity objective in the program. The idea was to distribute the scarce benefits among as many poor people as possible. Hence, those applicants who did not participate before have priority over those who did participate andas we already showed -during the period under consideration there was a large number of low skilled and unemployed people who never received the benefit. While this result does not prove the existence of political clientelism in the allocation of benefits, it is consistent with that presumption. Furthermore, it is consistent with the claim that a poorly implemented social policy would lead to dependency.
Since the estimates in column 1 are not literally 'before and after treatment' estimates, we proceed to compute truly 'before and after treatment' estimates by keeping only all those individuals who participated during May 2001 but did not participate during While these are truly 'before and after estimates' they are presumably not true treatment effects. The reason is that dropping from the sample those participants who stayed longer than the established program length biases the estimates if those who left the program are different from those who stayed. To the extent that workfare has operated more as permanent unemployment insurance than as a fixed term program, then restricting the sample to those who left the program overestimates treatment effect: Once individuals are able to enter into the program they remain participating until being offered a sufficiently good job in the labor market. The fact that the estimated total monthly income effect of the program excluding those who stayed participating is 40 percent larger with respect to the estimate obtained including them (i.e. column 2 versus column 1 in Table 14), supports this interpretation. We leave the need to control for selection out of the program for future research.
Another obvious limitation of the available data is that we can only follow the same individual for a relatively short period of time: only two years. Presumably, a more accurate assessment of the program would be obtained if data for several periods before and after treatment were available.

Are Employers willing to hire workfare participants?
Before concluding, it is interesting to analyze the opinion of employers regarding their predisposition to hire participants of workfare programs.  The survey does not provide the ideal information to asses the employability effect of workfare programs. Employers are asked about their predisposition to hire participants -in which case they would receive a subsidy. However, we would like to know if employers have any preference for ex-participants relative to comparable workers who never participated in the program. Second, the questionnaire only allows employers to express their motives choosing between the options: "Yes, if" and "No, because". More adequate options would have been: "Yes, because" and "No, because". In any case, and taking into consideration these caveats, we interpret the opinions of the employers as evidence that workfare programs may not improve the skills of participants. The fact that 22 percent of the employers mentioned that they would not hire program participants despite the US$150 subsidy supports that interpretation. Moreover, the results suggest that participating in workfare programs may have, to some extent, a negative effect on the employability of workers because some employers believe that workfare participants are prone to conflict.
However, this stigmatization effect does not appear to be large since only 10 percent of all employers reported that concern.

Concluding Remarks
A mixed picture arises after analyzing targeting, poverty and employability effects of workfare programs in Argentina during 2000 and 2001.
On the one hand, an active labor policy targeted at the least skilled seems particularly appropriate considering the increase in poverty and unemployment that Argentine society suffered during this period. We observe that the policy was actually propoor, and the average participant had less human capital than the average non-participant.
Targeting towards the poor also improved over time. More women than men received treatment, which is presumably a positive characteristic of the program considering that women have fewer opportunities in the labor market. The program was effective in increasing participants' income and reducing poverty particularly during treatment: We estimated that during treatment income increased by approximately US$50, helping 3.5 percent of participating households move out of poverty and 6 percent out of indigence, implying that 38,000 people moved out of indigence and 20,000 out of poverty thanks to workfare programs. Moreover, we estimate that in the absence of the program one third of participants would have had remained inactive or unemployed. After program completion, we observe that ex-participants perform worse than during treatment but better than before treatment, which is remarkable considering the deterioration of the economy. The program also increased the propensity to participate in the labor force. These results are consistent with the hypothesis that at least some ex-participants improved their human and/or social capital during treatment.
On the other hand, we observe that a large share of the scarce benefits was suggesting that some participants may have chosen to participate in the program instead of working longer hours, probably in the informal sector. We also observe that more than half of those who entered the program stayed for more than six months, which was the program's length established in the normative. Furthermore, 34 percent of participants received the benefit for at least 19 consecutive months. These facts, together with the inadequate allocation of some benefits, are consistent with the hypotheses that some benefits were allocated on the basis of political patronage and that the program generated dependency among recipients.
Finally, the fact that some participants were able to stay for more than the time established in the normative obscures the positive after treatment effect mentioned in the previous paragraph. It seems plausible that the estimated income gain not only reflects an improvement in the human and social capital of participants, but also reflects that those participants who were offered a good job in the labor market are the ones that chose to leave the program. This is consistent with the claim that workfare in Argentina operated more as unemployment insurance than as a training program. Finally, according to the opinions of employers, they do not express much interest in hiring program participants despite the subsidy they would receive. Moreover, 10 percent of employers expressed their reluctance to hire workfare participants because they consider them prone to conflict.
Regarding the statistical model, there are several plausible alternatives and extensions to the model we used that were already mentioned. Presumably, the most important extension is to model the fact that some participants select when to exit the program.
Finally, from a policy perspective, we want to emphasize that nowadays approximately 2 million individuals -equivalent to almost 15 percent of the labor force-are participating in the Jefes de Hogar workfare program. The magnitude of the figure is impressive. While our estimates refer to the workfare programs that preceded Jefes (e.g.

Trabajar)
, and extrapolating is always risky, these programs are very similar in several dimension making us confident that the estimates obtained in this paper are informative enough to discuss the adequacy of the Jefes de Hogar program.
In our opinion, a drastic reduction on the number of beneficiaries is not recommended considering the impact it would have on the well being of participants and their dependents, which in most cases are below poverty. However, it is hard to imagine a healthy future for the Argentine society if such a large percentage of its members continue depending on workfare subsidies. First, the productivity of participants during treatment seems to be very low -actually, many of them are not working at all-affecting growth prospects. Second, we are worried about the effects of workfare on political liberty. So far, the program has been mainly controlled by the executive branch of government and not by an independent body. Incumbent politicians seem to be more interested in maintaining power instead of improving societal welfare, leading to a clientele usage of workfare funds.
How can workfare participants vote freely if their main source of income is a workfare benefit received from the executive power and used in exchange for supporting an incumbent politician?

Appendix 1. Encuesta Permanente Hogares -Permanent Household Survey-(EPH)
The EPH is a sampling survey implemented by the Instituto Nacional de Estadistica y

Indigence
The concept of "indigence level" (or indigence line), IL, aims to assess whether the households earn enough income to purchase a food basket that satisfies a minimum threshold of energetic and protein needs. Thus, a household that does not meet that threshold is considered indigent. The procedure is based on the use of a Canasta básica de alimentos -basic food basket-(CBA), determined as a function of the consumption patterns of a reference population defined according to the results of the 1985-86 Household Expenditure and Income Survey. The procedure also takes into account the prescribed kilocalories and protein requirements for that population (as specified in the "Basic Food Basket for the Equivalent Adult", included below). Once the CBA components have been established, their prices are assigned according to the Indice de Precios al Consumidor (CPI) for each measurement period.
Since human nutritional requirements vary according to age and gender, INDEC adjusts for each person's characteristics, taking as reference the requirements of a male aged between 30 and 59 years old. This reference unit is called the "equivalent adult" and is assigned the value 1. The table of equivalences of energetic requirements for each consumer unit in terms of equivalent adult is presented in the table below.
Each household's composition in equivalent adults determines a specific CBA value for that household. (For example, in October 2000, the CBA value for an equivalent adult was $62,4). As a final step, the specific value of each household's CBA is compared to the household's total income. If the total income is less than the household's CBA, the household and its members are considered to be under the indigence level.

Poverty
The measurement of poverty by the poverty level or "poverty line" (PL) method is based on determining, from the household's reported income, whether the household in question is able to satisfy -through the purchase of goods and services-a set of nutritional and non-nutritional goods considered essential. In order to calculate the poverty line the INDEC determines the CBA value and compounds it with the inclusion of non-nutritional goods and services (clothing, transportation, education, health care, etc.) so as to obtain the value of the Canasta basica total -total basic basket (CBT).
For the purpose of compounding the CBA value, the so-called Engel coefficient (EC) is used. The EC is defined as the ratio of food expenditures to total expenditure observed in the reference population in the base year . Thus: Engel coefficient = Food expenditures / Total expenditure.
In each period, both the numerator and the denominator of the Engel coefficient are updated with the price variations obtained from the CPI. According to the relative price variation, the EC is determined each month for the purpose of measuring poverty. In order to compound the CBA value, in practice its value is multiplied by the reciprocal of the Engel coefficient: CBT = CBA x 1/Engel coefficient.