Bias and Its Remedy in Interval-Censored Time-to-Event Applications - Interval-Censored-Time-to-Event Data: Methods and Applications - page 320

FIGURE 11.3: Histograms for the estimated treatment parameter 1
respectively. Both cover zero, indicating that treatment effect is not statistically sig-
nificant.
Table 11.2 summarizes all combinations for the treatment effect parameter 1 ,
where the row in bold print corresponds to Figure 11.3. In this table, \1" denotes
statistical nonsignicance and \0" denotes statistical signicance. From this table,
it may be seen that there are several cases where \Cox.Right" and \Cox.Mid" yield
statistically nonsignificant treatment effects, but statistically significant treatment
eects for \IntCox."
In summary, the common practice of using the midpoint or the \rst observed"
approximation for interval-censored time-to-event data yields biased parameter es-
timates. This could lead to statistically nonsignificant treatment effects (when they
are significant) and inflate Type II errors while lowering the power of statistical tests.