The
Bootstrap
The bootstrap
is a central tool in the validation of predictive models. First,
we will briefly address general characteristics of the bootstrap;
second, we will address the role of the bootstrap in model development.
The bootstrap
is a technique for drawing conclusions about the population
where the sample originated in a nonparametric
way (Efron and
Tibshirani, 1993). A bootstrap sample is a random
sample, drawn with a replacement from the data set under study.
The size of the bootstrap sample is identical to the size of
the original data set. Every patient may be represented in the
sample 0, 1, 2, .. times. On average, a patient has a probability
of 63.2% of being selected at least once in a bootstrap sample.
Conceptually,
drawing a bootstrap sample replicates the situation that the
sample is drawn from the underlying population. Statistics can
be calculated in every bootstrap sample. The distribution of
such statistics is identical to parametric
estimates, when formulas are available.
Two different
cases - Example
#1 and Example
#2 -- further illustrate use of the bootstrap."