A Study On The Effect Of Rural Labor Training In Western China

In recent years, with the sustainable development and inclusive developmentconcept, the anti-poverty way in rural China was given a new meaning. Namely, theoriginal transfusion poverty alleviation policies using fiscal transfers payment turnsinto education and training which enable farmers to get fair policy of equaldevelopment opportunities. Although the Government has spared no effort to carryout the labor force training in rural area, the proportion of trained farmers is not high.Among all sample households in Yunnan Province, the proportion of trained ruralworkforce is around21.95%; wherein the proportion in ethnic areas is22.33%, but inpoor area the proportion is only16.67%.There are many problems in the past study of the effect of training, including:（1）byusing the heterogeneity of instrumental variables model to estimate the yield oftraining, only individual heterogeneity and selection bias are considered, theheterogeneity of the village is not;（2） emphasis on the contribution of training toindividuals, ignoring its impact on the entire rural households;（3）empirical researchdata developed mainly for the domestic area, without the concern that China’santi-poverty focus of the western minority areas;（4）emphasis on the estimationanalysis of yield of training, with less attention on farmers’ participation in trainingprograms.Therefore, the paper fully considered all these issues above, emphasizing the ruralhouseholds must be treated as the basic research units. Then it can conduct acomprehensive analysis of training program in Yunnan, including3elements:（1）analysis of labor force participation in the training program of rural households;（2）households’ choice of training;（3） estimation methods and empirical analysis of thetreatment effects from training. Taking into account the uneven development ofdifferent areas, each element above is studied basing on province, ethnic areas andpoor areas. Study data is from Yunnan’s rural households survey,2008.First, analysis of rural labor force participation in the training program mainly describes the status of rural household and the acceptance of training in different area.Based on the number of household, the proportion of trained households was29.40%,while ethnic areas was28.52%, but the proportion of trained households in poor areas（which is22.12%） was significantly lower than others. Training programs weremainly organized by governments. The reason of untrained mainly lies on the absenceof local skills training, as well as other. A high proportion of "other" indicates thatthere are some unobserved factors which make rural households did not receivetraining.Secondly, the choice of training in rural households is studied using modeling. Theaim is to recognize important variables to the choice of training. Also, a summarize ofthe households who are willing to be trained is preferred. As the option of training is acategorical variable, except for the use of the classical modeling Logistic regression,two kinds of machine learning methods are also used, namely, DTL（decision treelearning） and random forest method. The study found the presence of the following4important variables affect the training option: L5average years of education ruralhousehold labor force, K2the proportion primary industry takes of productive fixedassets, F1whether households members have village cadres, V1₁and V1₂villageterrain.Finally, the treatment effects is estimated based on Neyman-Rubin counterfactualframework. Under different assumptions, the paper uses OLS estimation, Heckittreatment effect model and matching estimator methods.3kinds of treatment effectsis estimated, including: average treatment effect （ATE）, average treatment effect of thetreated （ATT） and the average treatment effect of the untreated （ATU）. OLS estimationbrings a low estimate of the treatment effect because the method does not considerselection bias and heterogeneity; while matching estimator method only considersselection bias without heterogeneity, thus the treatment effect may be too high; asboth the selection bias and heterogeneity is considered, Heckit treatment effect modelbrings a more reasonable estimate. Based on estimates of the Heckit treatment effectmodel, ATE of the province was15.37%, while ethnic areas was11.36%and poorareas was9.02%. This indicate that the rural labor force training in Yunnan Province indeed increase the farmers’ income. In addition, due to a high ATT in ethnic areas andpoor area, it may indicate that the trained households in these two regions are thebiggest beneficiaries of the training program. Thus, it prefer to develop more trainingsin these areas.