34
acquisitions rather than both mergers and acquisitions.
When studying acquisitions among insurance companies, we define an event as an
insurance company got acquired, a dependent censoring as an insurance company
filed bankruptcy, and an independent censoring as otherwise.
Due to the existence of both dependent and independent censoring, conventional
survival analysis approaches, which only address the independent censoring, are no
longer adequate for this research. The root reasons for such inadequacy are twofold:
First, the degree and direction of the correlation between events and censoring (de-
pendent censoring) can lead to biased estimations of survival rates. Specifically, the
survival rates will be overestimated under the conventional approach when there is a
positive correlation between events and censoring, and vice versa. Second, the per-
centage of observations being dependently censored will also affect the magnitude of
the bias. Unless there is only a negligible amount of dependently censored data, the
dependent censoring tends to have sizeable influence on the estimation bias.
3.1.2 Sensitivity analysis
The coexistence of dependent and independent censoring can be solved if the addi-
tional data regarding the dependent censoring can be obtained. However, in practice,
more often than not, such data is unavailable. For example, as far as mergers and ac-
quisitions are concerned, many of the genuine motivations behind the deals may only
reside in companies’ internal documents that are strictly confidential to outsiders.