Page 306 - Contributed Paper Session (CPS) - Volume 6

CPS1937 Xu Sun et al.
In this paper, a novel automatic log-linear modeling method is introduced
for understanding and fitting mobility processes. Eigenspectrum
decomposition community detection approach (Newman, 2010) is used in
association analysis. By thinking about class categories in a mobility table as
nodes, and people as the weighted relations between them, novel insights are
drawn that illuminate the association between rows and columns of social
mobility tables. A community detection analysis identifies which categories
share members at rates above chance, and which should belong together. First
of all, community membership may be included in the “independence” model
without interaction. Subsequently, if the log-linear model include community
membership is ill-fitting, then the second community discovery was initiated
to remain after the log-linear model has been estimated. If data still is ill-
fitting, then implement the third and fourth times until the good-fitting is
achieved. The results from the community detection analysis parsimoniously
inform the analyst of associations that remain after the log-linear model has
been estimated.
The approach described above offers a couple of advantages relative to
other similar methods. Falguerolles & Leeuw (1989) and Melamed (2015)
analysis of the residuals from an ill-fitting log-linear model, suffers from
uncertainty of the results. Because the different initial model of mobility
processes (e.g., quasi-independence, quasi-symmetry, unidimensional social
distance etc.) which does not fit the data will affect the final modeling results.
The modeling results of Moses & Holland (2010) or Tibshirani (2011) or Yuan
et al. (2011) are particularly complicated to understanding the underlying
mobility processes. The strength of the novel modeling method that is
detailed below is that an objective function is maximized to identify the “best”
way to combine categories, and subsequently a single within-community term
may improve any log-linear model fit. The results are therefore certainty, clear,
automatic, and relatively straightforward.
Below, this new approach to interpreting social mobility that draws from
recent advances in community detection algorithm in social networks. Then
the approach is applied to social mobility tables that were derived from the
General Social Survey (GSS; Smith, Marsden, Hout, &Kim, 2005). The
community structures of multiple models of social mobility respectively is
identified for female, male, and all respondents, the results of which reveal
interesting substantive findings. Last, the proper model for female, male, and
all respondents respectively are provided.

2. Communities of intergenerational mobility tables
Within network science, a mode refers to a set of objects for which relations
may be measured. A person-to-person network is a one-mode network, while
a person-to-groups network is a two-mode network. An intergenerational

295 | I S I W S C 2 0 1 9