Propensity Score Matching Research And Exploration And Application Of Law

In the actual experimental studies, estimates and evaluate the role of some kind of treatment effect, using randomized controlled trials (randomized controlled trial, RCT) is undoubtedly the best. RCT is not only the "gold standard", and is the basis for the actual test statistical analysis，but the application of the RCT is also subject to some restrictions, such as the high cost of research and practical difficulties, and ethics factors and is not apply to the research that have a long cycle. Non‐randomized study have some limitations, such as is the unbalanced nature of the distribution of prognostic factors between groups and a variety of bias exists, will produce biased treatment effect estimates. Statistical research is often used in the multivariate analysis model matching method, hierarchical method and other methods to control bias, but multivariate analysis model does not apply to the low incidence of outcome, and confounders many cases; matching method and hierarchical method does not apply to mixed many factors.Propensity scores is a new method of non‐randomized study to control bias. In recent years, the propensity score method for its research step high degree of standardization, easy to understand and advantages have attracted much attention of researchers, and widely used in various of non‐randomized study fields,. Rubin and Rosenbaum proposed the propensity scores in1983, its concept is given a set of covariates premise, any one of the subjects assigned to a treatment group or control group conditional probability. Propensity score can represent multiple covariates combination of the results; it can control the bias by adjusting the covariate balance between the treatment group and the control group. Propensity score method is mainly used in the two groups has not been extended to the field of multi‐packet data, packet data applications, there are a number of key issues yet to be resolved, such as the choice of matching calipers value, sensitivity test and balance evaluation methods.Probit model, estimate propensity scores through the gradual introduction of rural households characteristic variables, check the cultivation of tobacco users and non‐tobacco planting households tend to score balanced model Pseudo‐R^2value options to meet balance requirements Pseudo‐R^2value combinations of variables for the final propensity score estimates. Probit model estimates Pseudo‐R^2in2007‐2010are0.047,0.062,0.052and0.054, the choice of variables to meet the balance requirements. Probit model to estimate the variables selected for farmers to grow tobacco. From the specific significant impact in the four survey years, the smaller the proportion of farm household labor, the greater the likelihood of tobacco cultivation; proportion of farmers Mountain area bigger and more likely to choose to grow tobacco; the farmers family size is more, choose to grow the greater the likelihood of tobacco.Propensity score matching method to assess planting tobacco farmers income effect analysis found that tobacco cultivation to the increase of farmers income has a strong positive effect. Assessment using propensity score matching method to tobacco farmers to plant different from the use of descriptive statistical analysis of the results. First, the two methods examined planting of diminishing difference in the trend. Kinds of tobacco users and non‐tobacco planting household income descriptive statistical analysis: smoke household income and non‐species differences in smoke household income (times) in2007‐2010, four study years showing a larger trends (from1.16to1.44and then1.18, and finally to1.29); analysis using propensity score matching method is able to see that kind of overall showed a increasing trend (from1.041to0.997, to1.248, and finally to1.387). Second, the traditional descriptive statistical analysis to overestimate the effect of planting tobacco farmers’ income. In tobacco cultivation effect assessment methods, this paper uses a different matching algorithms assess the effects of the kinds of tobacco farmer’s income. The size of the effect of different matching algorithm proceeds there is a difference, the reason lies in the type of matching method to produce the type of co‐supporting region, thereby causing a different quantity of the loss of sample households and the type of match quality.