33
the performance of the proposed method.
3.1 Dependent censoring and sensitivity analysis
3.1.1 Dependent censoring
Censoring is a particular issue requiring additional attention in survival analysis. A
common assumption made in survival analysis is that the censoring is independent of
the event of interest. This assumption may hold for the non-informative (independent)
censoring, but it hardly makes sense for the informative (dependent) censoring.
The time horizon we considered for this research is from 1994 to 2005, which
covers 6 years before the passage of the GLB Act and 6 years after. The data used
in this research was originally collected in 2006 and was updated in 2007.
When analyzing the GLB Act’s impact on the survival of insurance companies, we
define an event as an insurance company filed bankruptcy, a dependent censoring
as an insurance company got acquired (identified as being involved in a Mergers and
Acquisitions (M&A) transaction and abandoning its original company name subse-
quently), and an independent censoring as otherwise. The reason for such defini-
tion of dependent censoring is, when the event is defined as filing bankruptcy, being
acquired (giving up one’s name in an M&A transaction) can be reasonably regarded
as being dependent on the event of interest. In our real data set, the majority of
the M&A cases is companies being acquired. Thus, in the application, we focus on