Page 305 - Contributed Paper Session (CPS) - Volume 6

CPS1937 Xu Sun et al.

Data mining of mobility table
Based on community discovery methods
2
Xu Sun , Xiao-hui Li
1
1 School of statistics, Dongbei University of Finance and Economics, Dalian, China
2 College of public administration and humanities, Dalian Maritime University, Dalian, China.

Abstract
Based on community discovery methods, a new approach to the modeling
social mobility data is presented. Community detection algorithm to identify
communities of social classes within which social classes share members at
above expected rates. This approach, when applied to mobility data, may be
used to substantially improve the fit of models of social mobility. To illustrate,
the community effect model of social mobility is analyzed using data from the
General Social Survey.

Keywords
Intergenerational mobility tables; Log-linear model; Community detection;
Eigenspectrum decomposition

1. Introduction
Intergenerational mobility is an important perspective of social mobility
analysis, and have a variety of log-linear at their disposal with which to analyze
the structures and patterns embedded within mobility tables (e.g.,Hout, 1983).
In empirical analysis, in many cases, the structure in mobility tables is so
sufficiently complicated that parsimonious models do not capture the
observed patterns. In such circumstances, there are ever more complicated
models that may be fit to the data. For example, Moses & Holland (2010)
compared 12 statistical strategies which included significance tests based on
four chi-squared statistics proposed for selecting log-linear models. Tibshirani
(2011) proposed Lasso (Least aboslute shrinkage and selection operator)
method for estimation in generalized regression model. Yuan et al. (2011)
purposed an automatic data mining method of contingence table based on
multinomial processing tree model. Likewise, if a preferred model (e.g., quasi-
symmetry) does not fit the data, one can estimate a correspondence analysis
on the residuals to “see” the associations left over in the data (Falguerolles &
Leeuw, 1989). Melamed (2015) drawn on the idea that mining the residuals
and uses community detection methods to “see” the associations left over in
the data. These methods provided a good-fitting log-linear model, however,
the results are often particularly complicated and an understanding of the
underlying mobility processes may be obscured by the complexity of the
model.
294 | I S I W S C 2 0 1 9