An Exploratory Analysis of Immigrant Assimilation: Incorporating Decision Trees and Random Forests in Social Science

• In Event: 576 - Section on Methodology Paper Session. Integrating Multiple Methodologies into Data Collection and Analysis

Abstract

Regression models, frequently employed in traditional empirical social science analyses, broadly share common features such as a parametric nature and a linear relationship between dependent and independent variables. Decision trees and their extension random forests, statistical learning algorithms developed in the fields of statistics and computer science, are nonparametric methods that allow for flexible and efficient knowledge discovery. Combining both methods during the analytical stage allows researchers to interpret linear relationships of variables in the context of their average predictive power. Using data from the 2007 Spanish National Immigrant Survey, this investigation explores the relationship between immigrant assimilation and organization membership first with traditional generalized linear models and then with decision trees and random forests to compare results. While organization membership is positively associated with assimilation in linear models, the application of statistical learning techniques shows that their contribution to predicting immigrant assimilation, measured by the decision to bring additional family members to Spain, is found primarily at the margins. This analysis shows the added benefit of employing statistical learning to traditional analysis both as a test of robust findings and as a method of knowledge discovery.

Author

• Christopher Scott Inkpen, Pennsylvania State University