ملخص البحث :

Abstracts. It is necessary to find or search for a way by which the important variables are selected to be included inthe model to be studied. especially when the study data suffers from a cut-off point that occurs as a result of anabnormal interruption of the phenomenon studied, which leads to the division of the experimental units into twogroups, where this division leads to a gap Or a jump in the values of observations of the response variable, so wepropose in this paper a new method for the process of estimating and selecting important variables by combining theRegression Discontinuity Designs (RDD) with the (Smoothly Clipped Absolute Deviation (SCAD)) Penalty method.Local linear regression (LLR) method was used to estimate the effect of processing on the cut-off region of theobservations within the optimum bandwidth selection for the RDD design to obtain the best model, since (LLR ) is thebasis of the ( RDD ) model . Three methods were used to determine the IK (Iembens and kalyanman) bandwidth,cross-validation (CV) method, and The CCT (Calonico, Cattaneo & Titiunik) bandwidth. The problem of the paper isthat the design (RDD ) is used to estimate the causal effect of the phenomenon studied, as the effects of treatment areestimated using the covariates included to improve efficiency. Where the treatment is estimated with a small numberof observations. Therefore, this paper aims to employ the method (SCAD ) which is one of the methods of selectingthe variable in estimating RDD to improve accuracy with the covariates. A simulation study are conducted toinvestigate the performance of the proposed method. The mean squared errors (MSE) is used to choose the bestmodel. To illustrate the use of SCAD with RDD, a simulation study with the R program is used. .Keywords. Regression Discontinuity Designs (RDD), (SCAD) Penalty, variable selection, Local linear regression,bandwidth selection, IK, CV, CCT.

ملخص البحث :

Variable selection is an important topic in linear regression analysis. In practice, a large number of predictors are usually introduced at the initial stage of model construction to mitigate potential model biases. On the other hand, to enhance the ability to predict and select important variables, regularization techniques are one of the great and proven methods for dealing with a large number of variables. In previous years, statisticians made great efforts in developing regularization procedures to solve the problems of V.S. These actions automatically facilitate Variable selection (V.S) by setting specific coefficients to zero and reducing coefficient estimates, providing useful estimates even if the model contains a large number of variables. In this paper, two methods of regularization were proposed to estimate and select the appropriate variables at the same time in the multiple regression model, which are the Standard Error Adjusted Adaptive LASSO (SEA-LASSO) and Minim ax Concave Penalty (MCP) method and selection of the best. The paper problem focuses on using the best regularization method that works on estimation and appropriate selection of important variables at the same time and addressing the problem of multiple linearity using the SEA-LASSO and MCP method. To get the real model. This paper aims to evaluate the performance of the method (SEA-LASSO) and method (MCP) in terms of the process of estimation and appropriate selection of important variables and treatment the problem multicollinearity through the simulation study. A simulation study was conducted to compare between these two methods, which included different cases of the factors and testing the effect of the levels of these factors on the performance of these two ways, as well as determining the value of the control parameter () and the criterion for selecting the best value for it and the basis on which to evaluate the performance of the two methods. The simulation results showed that (SEA-LASSO) method is superior to (MCP) method in terms of percentage of operation to reach the real model measured by (PCT), and it is also better in terms of mean squares error (MSE) because it achieves less (MSE) in most cases. A simulation study was used with the program R.