Do Public Work Schemes Deter or Encourage Outmigration? Empirical Evidence from China

How does the introduction of rural public work schemes impact individual incentives to migrate? This paper examines this question in the context of rural public work program (Yigong-daizhen) in China, and unveils empirical evidence that suggest that the introduction of Yigong-daizhen projects in fact stimulates outmigration at the village level, after controlling for village characteristics and project types. By furthermore accounting for the endogeneity of Yigong-daizhen placement, the impact of such projects is found to be even larger. These results are consistent with household migration behavior in the presence of significant cost of migration, and credit market imperfection.


I. Introduction
A distinctive feature of the Chinese economic growth experience has been the massive increase in rural migrant work force living in China's cities since the onset of economic reforms in the 1980's. By 2007, the estimated number of rural migrants exceeded over 135 million (Meng et al., 2010). The mobilization of this sizeable workforce has wide-ranging consequences, including contributions to the growth of export industries (Chan, 2008), shifts in income distribution between urban and rural China (Ha et al., 2009), and changes in educational and health outcomes over time between migrants and non-migrants (Lee, 2011), for example. The depth and breadth of these research studies concerning the consequences of rural-urban migration in China contrast sharply with the relatively limited number of studies on the effectiveness of migration policies in China. Indeed, much of the policy discourse on China's internal migration policy has focused on the hukou system of household registration (Chan, 2008). As a first objective of this paper, we examine the role of rural public work schemes as an alternative migration regulatory mechanism in the Chinese context.
To the best of our knowledge, this is a first attempt at an empirical assessment of the role of Yigong-daizhen programs on the pattern of migration in China. As a contribution to policy analysis, in view of the massive influx of rural migrants into China's urban cities, whether these rural public work schemes serve as deterrents that mitigate the size of the migrant flow, or in fact further intensify out migration is a question of critical policy importance. As a contribution to the economics of migration, this paper brings together three strands of the literature not often discussed together: labor market consequence of public work schemes, determinants of regional migration, and behavior in the presence of credit market imperfection.
By a public work scheme, we refer to the public provision of employment opportunities resulting in the creation of public goods, such as roads and schools. These employment opportunities serve as a form of social safety net, at a prescribed wage for those unable to find alternative employment. Worldwide, the implementation of public work schemes spans transition countries, developing countries and developed countries (Betcherman et al., 2004).
A number such public work schemes, such as the Maharashtra Employment Guarantee Scheme (EGS), and more recently the Mahatma Gandhi Employment Guarantee Schemes in India, have attracted much academic and policy attention, (Acharya, 1990;Ravallion et al., 1993;Gaiha, 1996aGaiha, , b, 2000Basu et al., 2009). Research on the labor market consequences of these more well-known public work schemes has so far focused on employment, earnings, and targeting. 1,2,3 By contrast, the labor market consequence of public work schemes in China is a far less well-understood topic. Specifically, Yigong-daizhen --to offer job opportunities instead of sheer relief --is a public work scheme initiated in the mid-1980s a part of the Chinese government's poverty reduction programs (Rozelle et al., 1998). Research on the effectiveness of these programs has been very limited. Two exceptions are Park et al. (2002) and Zhu and Jiang (1995), emphasizing respectively the targeting effectiveness, and the earnings impact of Yigong-daizhen programs. 4,5 While offering valuable insights, neither of these studies identify program-specific effects due mainly to data limitation. In addition, identification also requires proper accounting of endogeneity of program selection among participating villages or county -these are issues that we will pay particular attention to in our analysis in the sequel.
This paper is also related to a rapidly growing literature on the determinants of inter-regional migration in China. Some of these determinants include the potential role of farmland shortages and availability of household labor (Zhao, 1999a), earning differences between destinations and origins (Zhao, 1999b;Zhu, 2002;Zhang and Song, 2003), as well as the cost of migration (Zhang and Song, 2003) whether monetary or psychic (Sjaastad, 1962).
At the household level, participating in migration has been shown to significantly raise per capita income as high as 16 to 43 percent (Taylor et al., 2003 Finally, this paper is also related to the literature on behavior in the presence of credit market imperfection. In a wide variety of settings, credit market imperfections have been shown to fundamentally alter the properties of equilibrium, giving rise for example to non-monotonic, and / or distributionally sensitive comparative statics responses (Galor and Zeira, 1993;Matsuyama, 2008). In the context of migration, credit market imperfection is expected to play a critical role particularly when the upfront cost of migration is significant.
Indeed, several studies (Du et al., 2005;McKenzie and Rapoport, 2007) document an inverted-U shape relationship between household endowments and migration likelihood. That is, households with middle wealth are more willing to and able to migrate.
Following the lessons that can be drawn from these three strands of literature, we argue that Yigong-daizhen programs introduce two opposing forces on the village level propensity for outmigration. By increasing available employment opportunities and raising earnings and production efficiency through public goods provision in general, Yigong-daizhen programs narrows the expected earnings gap between destination and sending locations, thus potentially discouraging outmigration. Going in the opposite direction, by raising individual / household's ability to pay for the cost of migration, due either to the improvements in earnings thanks to Yigong-daizhen programs in the face of credit market imperfection that has previously deterred attempts to migrate, or to public construction that directly decreases the cost of migration by improving transportation and building roads, public work schemes can in fact stimulate outmigration. On balance, the net outcome of Yigong-daizhen programs on village level outmigration will in the end depend on the relative strength of these two effects. 7 In this paper, we take this issue to the data using the Center for Chinese Agricultural Policy  Du et al. (2005) and McKenzie and Rapoport (2007) and the implications of models with significant cost of migration and imperfect credit markets, we also find an inverted U-shape relationship between average per capita income and outmigration at the village level. Finally, to address the potential heterogeneity of Yigong-daizhen programs across villages, we work with four restricted samples: (A) villages with productivity improving public projects, such as irrigation system improvement, drainage system improvement, soil improvement, small-scale water conservation and terrace construction projects, (B) villages without productivity improving public projects, (C) villages with migration cost decreasing public investments through roads and bridge construction projects, and (D) villages without roads and bridge construction projects. The impacts of Yigong-daizhen programs on outmigration in these specifications continue to be positive and significant, and indeed, the impact of Yigong-daizhen in villages with newly constructed roads is positive and larger than the other villages. These findings suggest the potentially critical roles that cost of migration and credit market imperfections play in the determination of the inter-regional flow of migrants in China.
The plan of the rest of this paper is as follows. Section II describes the institutional background of Yigong-daizhen program in China. Section III provides the summary of statistics of the data used in this paper. Section IV discusses about the identification strategy.
Section V shows the empirical findings and Section VI concludes.

II. Country Background
The Yigong-daizhen program in China was initiated in the mid-1980s by the Chinese government as part of the rural poverty-reduction programs. Table 1 shows the types of projects, the amount of investment and the achievements of Yigong-daizhen projects from 7 rural land construction and irrigation system construction, road construction, drinking water facilities improvements and small-scale water conservation. Forestry and meadow maintenance, river and lake conservation are also included in the Yigong-daizhen project schemes from 1985 to 1995.
The amount of investment on each category of Yigong-daizhen projects is also presented in Table 1. Rural land construction, irrigation system construction, as well as road construction take up around 60 per cent of the total investment in 1985-1995, and around 80 per cent of the total investment in 1995-2000. Total investment in Yigong-daizhen projects is substantial, amounting to 2.62 billion yuan, or around US$ 0.38 billion.
The Chinese central government provided both monetary and in-kind investments. Table 2 shows the types of in-kind investment made by the government, such as cereals, cloth, edible oil, grains and medium-and low-grade consumer goods. In the year 2005, the Regulation of The main difference between the Yigong-daizhen programs in China and other public work schemes, such as the EGS program in India, is the wage rate paid to the employed workers and the potential impact that this has on inter-regional migration. It is documented that the wage paid to the unskilled day laborers in the local villages was around 10 yuan (around 1.5 USD) per day during the year 1998 to 2002. This was much lower than the wage of migrant workers (Luo et al., 2007). For example, in the rural area of Sichuan province in 1995, an out-migrant helped to increase an average of 2,388 yuan more of household earnings than a non-migrants per year, relative to an average net income per capita of 1354.66 yuan and an average household size around 4, or around 10 yuan per day assuming that a worker works five days a week (Zhao, 1999a). Thus, a rural-urban wage gap continued to exist, and remained large despite the introduction of Yigong-daizhen programs. By contrast, the EGS in Neelamangalam, India, for example, pays the minimum wage at 80 rupees a days, a figure very close to the day wages of unskilled migrants, which could be less than 100 rupees a day. (Naomi, 2008).

III. Data Description
The village level data used in this paper is obtained from Center for Chinese Agricultural In China, the key items of public investment in the rural villages are often related to basic infrastructure improvement, such as road construction, irrigation system improvement, school construction and so on. Table 3  As a subset of these public investment projects dedicated to alleviating rural poverty in China, Yigong-daizhen is a public work scheme initiated in1984. 9 The funding of the Yigong-daizhen projects is allocated to the local governments for local infrastructure construction using local laborers, where the payment to the laborers accounts for around 20 per cent of the total funding, at a wage around 10 yuan (around 1.5 USD) per person per day (Luo et al., 2007). Table 4

Heterogeneity by Income Quartiles
Our empirical estimation tests the impact of Yigong-daizhen on village level outmigration. As To capture the potential heterogeneity in program impact by average per capita income in a village, we will provide quartile specific estimates of the impact of Yigong-daizhen projects on out-migration.

Heterogeneity by Program Characteristics
As discussed earlier in Section I, different public investment projects should be expected to have different implications on the urban-rural wage gap, and the cost of migration in the Yigong-daizhen villages. In particular, productivity improving projects such as irrigation system improvement, drainage system improvement, soil improvement, small-scale water conservation and terrace construction projects may be viewed productivity improving, potentially narrowing the urban-rural wage gap. Other programs such as road and bridge construction may be seen more as migration cost reducing. To account for program-specific heterogeneity across Yigong-daizhen projects, we will examine four restricted panels. The first includes only villages with five types of 'productivity improvement' projects. The second is the complementary set with villages that lack 'productivity improvement' projects. The third panel includes only villages with road and bridge construction projects, and as such, are more likely to experience migration cost reduction. The fourth panel includes the complementary set of the third restricted sample, i.e., villages without road and bridge construction projects during the survey periods.

Endogenous Program Selection
Naturally, whether Yigong-daizhen projects are implemented in a village is likely to depend on village characteristics. Based on information available from our sample, Yigong-daizhen projects are more likely to be implemented in villages with more surplus laborers and less income per capita if it is correctly targeted. Table 5 shows the differences in village characteristics by Yigong-daizhen status. Results from t-tests between the two groups show that villages with Yigong-daizhen projects on average have significantly lower net income per capita, larger village size (higher total population), higher proportion of land steeper than 25 degrees, longer distance from the village committee (usually locating near the center of the village) to the nearest tarred road and more fellow villagers working at township. These suggest strongly that the endogeneity of program placement needs to be accounted for in our econometric model.

Unobserved Heterogeneity
Due to data limitation, we do not have information on the implementation of Yigong-daizhen projects in the surveyed villages prior to year 1997. Thus, it is likely that labor market equilibrium in the villages was affected not only by the Yigong-daizhen projects implemented during year 1998-2002, but also by the ones implemented before 1997. To mitigate the impact of these unobservable differences across villages, we opt to take the difference between the outcome variables in 2002 and 1997 using the balanced panel data. Assuming that the long term impacts of such projects do not vary significantly across years, the resulting estimates should provide an unbiased assessment of the impact of Yigong-daizhen projects.
In view of these econometric issues and identification challenges, we will provide first a series of baseline OLS regressions of a difference in difference model with and without provincial fixed effects. The Difference-in-Difference specification is as follows: Province, k controls for provincial fixed effects. ε i is an error term. 13 In order to deal with the endogeneity of the non-random placement of Yigong-daizhen projects, we will additionally provide estimates based on Two-Stage Least Square to instrument for the placement of Yigong-daizhen projects. Specifically, the first stage is: such as the number of households with access to tap water, the number of households with access to telephone, the distance from the village to road. As remote villages are more likely to be poor and lack infrastructure, they are more likely to become the target of Yigong-daizhen programs.
The second stage is: where YGDZ ı � is the predicted value of Yigong-daizhen status from the first stage. Other variable definitions are the same as the Difference-in-Difference specification. Table 6 presents the empirical results for the impact of Yigong-daizhen projects on the migration patterns in the villages. Columns 1-4 display the results of OLS estimation of the difference in difference setup. Column 1 is the parsimonious specification without any control variables, Column 2 controls for village characteristics, Column 3 further controls for project type fixed effect and Column 4 controls for both project type and province fixed effects.

V. Empirical Findings
Column 5 displays the result of the 2SLS estimation. It can be seen from the first row of Table   6 that the introduction of Yigong-daizhen projects had a positive and significant impact on the number of out-migrants in the village. After controlling for village characteristics in the year 1997 and the types of public project in the village, the introduction of Yigong-daizhen projects on average led to an increase of around 28 local laborers migrating out in the subsequent years as shown in Column 3. The significance and magnitude of coefficients decrease after controlling for provincial fixed effect. But the p-value of the coefficient on Yigong-daizhen projects is only slightly above 0.10 (p=0.102.) While for the 2SLS estimation, the coefficient on Yigong-daizhen is much larger than the OLS estimates, which may be attributed to a relatively low F statistic (7.17) in the first stage. But the coefficient is significant at the 0.1 level after controlling for types of project and province fixed effect.
In addition to the main coefficient related to the impact of Yigong-daizhen, the coefficients on village characteristics are of interests as well. For example, the linear term of per capita income in year 1997 is significantly positive, while the squared term of per capita income is significantly negative, which suggests that there exists an inverted U-shape between per capita income and migration probability. The estimation of turning point is consistently around 615 yuan according to column 3 and 4. Furthermore, and as should be expected, higher population leads to more out-migrants in the village. Better road access also encourages out-migration. here. The coefficients for Yigong-daizhen are positive and significant for the second and third quartile by income in 1997 without controlling for provincial fixed effects. In addition, the magnitude of the coefficient for the third quartile is larger than its counterpart in the whole-sample estimation. These estimates provide some evidence on credit constraints as the households neither too poor nor too rich are the most likely ones to migrate out given the cash income from the public work projects. A potential drawback of the by-quartile regression is that sharp decrease in the number of observations per regression. Indeed, once separated by quartile, the Yigong-daizhen coefficient is no longer significant after controlling for provincial fixed effects, though the sign of the estimated coefficient remain positive for all except the second quartile. Table 8 presents the results concerning 'increasing productivity' and 'reducing cost'

hypotheses. Panel A and B test the hypothesis of 'increasing productivity' as a mechanism of
Yigong-daizhen projects. In Panel A, we use a restricted sample including villages with five types of 'productivity improvement' projects: irrigation system improvement, drainage system improvement, soil improvement, small-scale water conservation and terrace construction projects. We exclude projects which may not have effects in the short run, such as school and clinic construction. In addition, we exclude the Grain for Green projects as they mainly benefit the downstream villages. In Panel B, we use the complementary set of the restricted sample in Panel A as a comparable group.
The findings in Panel A suggest that the impact of Yigong-daizhen on the number of out-migrants in the villages is uniformly positive and significant in the four specifications.
Comparing panels A and B, the coefficients do not seem to differ significantly in the two panels without controlling for provincial fixed effects. However, the impact of Yigong-daizhen is not significant in Panel B after adding the provincial fixed effects, in stark contrast with the estimation in Panel A. This lends some credibility to the hypothesis that productivity improvement is a key mechanism by which public work projects impact the behavior of potential migrant workers. Specifically, an increase in local productivity will lead to an increase of domestic wage, which may help release the credit constraint of the poor, thus enabling more migration.
Panel C of Table 8 only includes the villages with road and bridge construction projects during 1998-2002. These villages are more likely to experience migration cost reduction through better connectivity to neighbor cities and villages. The coefficients on Yigong-daizhen in the villages with newly constructed roads are positive and larger than the other villages, though the coefficients are not significant for both groups after controlling for province fixed effect. While in panel D, which includes only the villages without such migration cost reduction projects, the impact of Yigong-daizhen on outmigration is not significant at all. Thus, the above estimates provide us with some confidence that cost reduction may be a channel for the positive impact of Yigong-daizhen projects on the flow of out-migrants.

VI. Conclusion
In this paper, we present the impact of public work schemes in China Specifically, we find evidence suggesting that the impact of Yigong-daizhen on migration is most important for the middle class, which is in consistent with the inverted-U shaped relationship between migration and income level in the presence of capital market imperfect.
The robustness of the positive impact of Yigong-daizhen remains upon accounting for heterogeneity in program characteristics, most important in villages with productivity improvement projects (such as construction of irrigation system) and cost reduction projects (such as road construction.) Our findings also reveal two observations that suggest that the present set of results should be interpreted with caution, and that additional research with better identification techniques and broader data coverage should be encouraged. In particular, in regressions that do not control for the endogeneity of program selection, the significance (but not the sign with one singular exception in Table 7) of the impact of Yigong-daizhen on out-migration is sensitive to the introduction of provincial fixed effects. However, upon accounting for endogenous program selection, the impact of Yigong-daizhen on outmigration continues to be positive and significant. Future research with ideally a broader data coverage should devote particular attention to possible province-specific effects of the impact of Yigong-daizhen on outmigration. In addition, for the 2SLS regression in Table 6, the first stage F statistic is 7.17, which is slightly lower than the commonly accepted criterion of 10 (Angrist and Pischke, 2008). As well, with a substantially reduced number of observations in each income quartile and each project panel, the 2SLS estimations cannot be applied in the restricted (income quartile specific / program characteristic specific) samples for the F-statistics are too small.
Alternative identification techniques with better instruments and broader data coverage, for example, should be applied in future research.

Acknowledgements:
We are very grateful to Xiaobo Zhang and Yongmiao Hong for their extremely helpful comments. We also thank Jijie Zhao for very helpful comments and discussions. All remaining errors are ours. 2. Suggested by Betcherman et al. (2004), of the two studies with impact evaluation on earnings (Benus and Rodriguez-Planas, 2002;Jalan and Ravallion, 2002), the evaluation evidence regarding to the impact of public works on earnings is mixed for the transition and developing country programs. For example Benus and Rodriguez-Planas (2002) find that in Romania, the Public Works Community Job Creation Program has no impact on wages in the period of 1999-2001. While according to Jalan and Ravallion (2002), the Trabajar program in Argentina significantly lifts up the net income of the poor participants, where the percentage net gain for the poor 5 per cent is 74%. Gaiha (1996b) analyzes the impact of EGS on the wages of the poor in Maharashtra, India. Program participation is shown to bring significant positive effect on agricultural wages of the poor possibly since EGS enables them to bargain for higher agricultural wages by improving their fall-back position. Finally, the study also finds that EGS program has an income stabilization effect in agriculturally slack periods.
3. Gaiha (1996b) examines the targeting precision of the EGS program. It is expected that EGS program is designed to help the poor by providing them job opportunities. However, he finds that the targeting of EGS is no better than the general labor market. In other words, the share of the poor among EGS participants is close to the share of the poor in the labor force. This evidence is further confirmed in Gaiha (2000). In addition, there are more male participants in EGS with a significantly higher wage than the female participants. Another interesting finding with respect to the participation of the EGS is that the poor people turn to depend less on EGS, or to be more likely to withdraw from EGS when the overall economic condition gets better.
4. Park et al. (2002) studies the targeting effectiveness of the three main poverty reduction programs, namely the Yigong-daizhen program, the subsidised loan program and the budgetary grant program. They find that for both Yigong-daizhen program and the subsidised loan program, the amount of fund allocation to poor counties is not significantly correlated with income levels. Only the budgetary grant program is progressive. In addition, they find that being designated as a poor county increases the growth in rural income per capita by 2.28 per cent per year during the period of 1985-1992 and 0.91 per cent during the period of 1992-1995.
5. Zhu and Jiang (1995) is the only paper that estimates the impact of Yigong-daizhen program in China with data from three counties. This study finds that Yigong-daizhen projects have improved the income of participating households. However, they simply compare the difference of average income per capita between participating and non-participating households without controlling for village characteristics and accounting for program placement endogeneity.
6. This question has been addressed by Ravallion (1991) for example in the Indian context.
Specifically, the initial purpose of the Maharashtra Employment Guarantee Scheme was to discourage worker migration in the slack seasons and drought affected years since some workers would not return in the harvest seasons. Also see Naomi (2008) for evidence on the potential of the National Rural Employment Guarantee Scheme in India on outmigration propensities.
7. Kanbur (1981) suggests that a rural development program which increases the rural income may indeed increase migration in the villages with imperfect credit market, as the poor now have more money to spend on migration if the gain from migration is greater than the earnings from local employment. Qin (2011) narrows down the concept of rural development programs to public work schemes and provides a theoretical framework to analyze the impact of such programs on interregional movements of labor.
8. The objective of Grain for Green projects is to increase forest cover and prevent soil erosion on sloped cropland (Uchida et al., 2005). The distance from the village committee seat to township government seat measures how isolated a natural village is, which is likely to affect the placement of public projects.
13. The 19 types of public projects are: roads and bridges construction, school construction, clinic construction, drinking water facility provision, irrigation system improvement, drainage system improvement, electricity infrastructure construction, telephone installation, cable TV or loudspeaker installation, soil improvement project, small scale water conservation, terrace construction, environment improvement project, forest closure project, public forest planting, Grain for Green project, meadow construction, recreational center construction and others.
14. Angrist and Krueger (2001) suggest that using probit or logit as first stage in two-stage least squares is not necessary and may even do some harm. Specifically, if the probit or logit model does not reflect the correct first-stage functional form, the second stage estimation will not be consistent. Instead, using a linear regression for the first-stage estimates generates consistent second-stage estimates even with a dummy endogenous variable.  1985-1995; 1995-2000  Source : Zhu and Jiang (1995)    Notes. *** denotes significant at the 0.01 level; ** denotes significant at the 0.05 level; * denotes significant at the 0.1 level. Robust standard errors are reported in the parentheses.
OLS (1) OLS (2) OLS (3) OLS (4) Yigong  Notes. *** denotes significant at the 0.01 level; ** denotes significant at the 0.05 level; * denotes significant at the 0.1 level. Robust standard errors are reported in the parentheses. The specifications are the same as reported in Table 6. Only the coefficient on Yigong-daizhen is reported due to space constraints.
Highest quartile by net income in 1997 (N=600) Yes Notes. *** denotes significant at the 0.01 level; ** denotes significant at the 0.05 level; * denotes significant at the 0.1 level. Robust standard errors are reported in the parentheses. The specification is exactly the same as reported in Table 6. Only the coefficient on Yigong-daizhen is reported due to space constraints.