Chapter Contents
Chapter Contents		 Previous
Previous		 Next
Next
INTERVALS Statement

Methods for Computing Statistical Intervals

The formulas for statistical intervals given in this section use the following notation:

Notation Definition
nnumber of nonmissing values for a variable
\bar{X}mean of variable
sstandard deviation of variable
z_{\alpha}100\alpha\!^{{\scriptsize th}} percentile of the standard normal distribution
t_{\alpha}(\nu)100\alpha\!^{{\scriptsize th}} percentile of the central t distribution with \nu degrees of freedom
t^'_{\alpha}(\delta ,\nu)100\alpha\!^{{\scriptsize th}} percentile of the noncentral t distribution with noncentrality
 parameter \delta and \nu degrees of freedom
F_{\alpha}(\nu_1,\nu_2)100\alpha\!^{{\scriptsize th}} percentile of the F distribution with \nu_1 degrees of freedom in
 the numerator and \nu_2 degrees of freedom in the denominator
\chi^2_{\alpha}(\nu)100\alpha\!^{{\scriptsize th}} percentile of the \chi^2 distribution with \nu degrees of freedom.

The values of the variable are assumed to be independent and normally distributed. The intervals are computed using the degrees of freedom as the divisor for the standard deviation s. This divisor corresponds to the default of VARDEF=DF in the PROC CAPABILITY statement. If you specify another value for the VARDEF= option, intervals are not computed.

You select the intervals to be computed with the METHODS= option. The next six sections give computational details for each of the METHODS= options.

METHODS=1

This requests an approximate simultaneous prediction interval for k future observations. Two-sided intervals are computed using the conservative approximations 

		\(
 {Lower Limit} & = & \bar{X} - t_{1- \frac{\alpha}{2k}}
 (n - 1) s \sqrt{1 + ...
 ...it} & = & \bar{X} + t_{1- \frac{\alpha}{2k}}
 (n - 1) s \sqrt{1 + \frac{1}n}
 \)

One-sided limits are computed using the conservative approximation 


		\(
 {Lower Limit} & = & \bar{X} - t_{1- \frac{\alpha}k}
 (n - 1) s \sqrt{1 + \fr...
 ...Limit} & = & \bar{X} + t_{1- \frac{\alpha}k}
 (n - 1) s \sqrt{1 + \frac{1}n}
 \)
Hahn (1970c) states that these approximations are satisfactory except for combinations of small n, large k, and large \alpha. Refer also to Hahn (1969 and 1970a) and Hahn and Meeker (1991).

METHODS=2

This requests a prediction interval for the mean of k future observations. Two-sided intervals are computed as 

		\(
 {Lower Limit} & = & \bar{X} - t_{1-\frac{\alpha}2}
 (n - 1) s \sqrt{\frac{1}...
 ... = & \bar{X} + t_{1- \frac{\alpha}2}
 (n - 1) s \sqrt{\frac{1}k + \frac{1}n}
 \)
One-sided limits are computed as 

		\(
 {Lower Limit} & = & \bar{X} - t_{1-\alpha}
 (n - 1) s \sqrt{\frac{1}k + \fra...
 ... Limit} & = & \bar{X} + t_{1-\alpha}
 (n - 1) s \sqrt{\frac{1}k + \frac{1}n}
 \)

METHODS=3

This requests an approximate statistical tolerance interval that contains at least proportion p of the population. Two-sided intervals are approximated by 

		\(
 {Lower Limit} & = & \bar{X} - g ( p; n; 1-\alpha) s \& & \ {Upper Limit} & = & \bar{X} + g ( p; n; 1-\alpha) s
 \)
where g(p; n; 1- \alpha)= z_{\frac{1+p}2}(1+\frac{1}{2n}) \sqrt{\frac{n - 1}{\chi^2_{\alpha}(n - 1)}}.

Exact one-sided limits are computed as 


		\(
 {Lower Limit} & = &
 \bar{X} - g^'(p; n; 1- \alpha ) s \& & \ {Upper Limit} & = &
 \bar{X} + g^'(p; n; 1- \alpha ) s
 \)
where g^' (p; n; 1- \alpha) = \frac{1}{\sqrt{n}} t^'_{1-\alpha}(z_p \sqrt{n}, n - 1).

In some cases (for example, if z_p \sqrt{n} is large), g^' (p; n; 1- \alpha) is approximated by 


		\(
 \frac{1}a (z_p + \sqrt{z_p^2 - ab})
\)
where a = 1 - \frac{z_{1-\alpha}^2}{2(n - 1)} and b = z_p^2 - \frac{z_{1-\alpha}^2}n.

Hahn (1970b) states that this approximation is "poor for very small n, especially for large p and large 1-\alpha, and is not advised for n < 8." Refer also to Hahn and Meeker (1991).

METHODS=4

This requests a confidence interval for the population mean. Two-sided intervals are computed as 

		\(
 {Lower Limit} & = & \bar{X} - t_{1-\frac{\alpha}2}
 (n - 1) \frac{s}{\sqrt{n...
 ...pper Limit} & = & \bar{X} + t_{1-\frac{\alpha}2}
 (n - 1) \frac{s}{\sqrt{n}}
 \)
One-sided limits are computed as 

		\(
 {Lower Limit} & = & \bar{X} - t_{1-\alpha}
 (n - 1) \frac{s}{\sqrt{n}} \& & \ {Upper Limit} & = & \bar{X} + t_{1-\alpha}
 (n - 1) \frac{s}{\sqrt{n}}
 \)

METHODS=5

This requests a prediction interval for the standard deviation of k future observations. Two-sided intervals are computed as 

		\(
 {Lower Limit} & = & s (
 F_{1-\frac{\alpha}2} (n - 1, k - 1) )^{-\frac{1}2} ...
 ...\ {Upper Limit} & = & s (
 F_{1-\frac{\alpha}2} (k - 1, n - 1) )^{\frac{1}2}
 \)
One-sided limits are computed as 

		\(
 {Lower Limit} & = & s (
 F_{1-\alpha} (n - 1, k - 1) )^{-\frac{1}2} \& & \ {Upper Limit} & = & s (
 F_{1-\alpha} (k - 1, n - 1) )^{\frac{1}2}
 \)

METHODS=6

This requests a confidence interval for the population standard deviation. Two-sided intervals are computed as 

		\(
 {Lower Limit} & = & s \sqrt{\frac{n - 1}
 {\chi^2_{1-\frac{\alpha}2} (n - 1)...
 ...Upper Limit} & = & s \sqrt{\frac{n - 1}
 {\chi^2_{\frac{\alpha}2} (n - 1)}} \ \)

One-sided limits are computed as 


		\(
 {Lower Limit} & = & s \sqrt{\frac{n - 1}
 {\chi^2_{1-\alpha} (n - 1)}} \& & \ {Upper Limit} & = & s \sqrt{\frac{n - 1}
 {\chi^2_{\alpha} (n - 1)}} \ \)

Chapter Contents
Chapter Contents		 Previous
Previous		 Next
Next		 Top
Top

Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.