06.06.2013 Views

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

184 2 Distribution <strong>Theory</strong> and Statistical Models<br />

The support <strong>of</strong> a distribution may also be changed by censoring. “Censoring”<br />

refers what is done to a random variable, so strictly speaking, we do<br />

not study a “censored distribution”, but rather the distribution <strong>of</strong> a random<br />

variable that has been censored.<br />

There are various types <strong>of</strong> censoring. One type is similar to the truncation<br />

<strong>of</strong> a distribution, except that if a realization <strong>of</strong> the random variable occurs<br />

outside <strong>of</strong> the truncated support, that fact is observed, but the actual value<br />

<strong>of</strong> the realized is not known. This type <strong>of</strong> censoring is called “type I” fixed<br />

censoring, and in the case that the support is an interval, the censoring is<br />

called “right” or “left” if the truncated region <strong>of</strong> the support is on the right<br />

(that is, large values) or on the left (small values). A common situation in<br />

which type I fixed censoring occurs is when the random variable is a survival<br />

time, and several observational units are available to generate data. Various<br />

probability distributions such as exponential, gamma, Weibull, or lognormal<br />

may be used to model the survival time. If an observational unit survives<br />

beyond some fixed time, say tc, only that fact is recorded and observation <strong>of</strong><br />

the unit ceases.<br />

In another kind <strong>of</strong> fixed censoring, also illustrated by observation <strong>of</strong> failure<br />

times <strong>of</strong> a given set <strong>of</strong> say n units, the realized failure times are recorded<br />

until say r units have failed. This type <strong>of</strong> censoring is called “type II” fixed<br />

censoring.<br />

If an observational unit is removed prior to its realized value being observed<br />

for no particular reason relating to that unobserved value, the censoring is<br />

called “random censoring”.<br />

Censoring in general refers to a failure to observe the realized value <strong>of</strong><br />

a random variable but rather to observe only some characteristic <strong>of</strong> that<br />

value. As another example, again one that may occur in studies <strong>of</strong> survival<br />

times, suppose we have independent random variables T1 and T2 with some<br />

assumed distributions. Instead <strong>of</strong> observing T1 and T2, however, we observe<br />

X = min(T1, T2) and G, an indicator <strong>of</strong> whether X = T1 or X = T2. In this<br />

case, T1 and T2 are censored, but the joint distribution <strong>of</strong> X and G may be<br />

relevant, and it may be determined based on the distributions <strong>of</strong> T1 and T2.<br />

2.8 Mixture Families<br />

In applications it is <strong>of</strong>ten the case that a single distribution models the observed<br />

data adequately. Sometimes two or more distributions from a single<br />

family <strong>of</strong> distributions provide a good fit <strong>of</strong> the observations, but in other<br />

cases, more than one distributional family is required to provide an adequate<br />

fit. In some cases most <strong>of</strong> the data seem to come from one population but<br />

a small number seem to be extreme outliers. Some distributions, such as a<br />

Cauchy, are said to be “outlier-generating”, but <strong>of</strong>ten such distributions are<br />

difficult to work with (because they have infinite moments, for example). Mix-<br />

<strong>Theory</strong> <strong>of</strong> <strong>Statistics</strong> c○2000–2013 James E. Gentle

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!