SED navigation bar go to SED home page go to SED publications page go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages


contents     previous     next


3.4.5 Inference on a Common Mean in an Interlaboratory Study

Mark G. Vangel and Andrew L. Rukhin
Statistical Engineering Division, ITL

Data on a quantity measured by several laboratories often exhibits non-negligible between-laboratory variability, as well as different within-laboratory variances. Also, the number of measurements made at each laboratory can differ. A question of fundamental importance in the analysis of such data is how to best estimate a consensus mean, and what uncertainty to attach to this estimate.

We have been engaged in a detailed investigation of this problem, and its generalizations and applications. Some recent work has focused on hierarchical models for one-way ANOVA and for two-way tables, and their use in interlaboratory studies.

For the following two-stage hierarchical model for one-way ANOVA

$\displaystyle p(x_{ij} \vert \theta_i, \sigma_i^2)$ = $\displaystyle {\rm N\,}(\theta_i, \sigma_i^2)$  
$\displaystyle p(\sigma_i)$ $\textstyle \propto$ $\displaystyle 1/\sigma_i$  
$\displaystyle p(\theta_i \vert \mu, \sigma^2)$ = $\displaystyle {\rm N\,}(\mu,\sigma^2)$  
$\displaystyle p(\mu)$ $\textstyle \propto$ 1  
$\displaystyle p(\sigma)$ $\textstyle \propto$ 1.  

the joint posterior of $\mu$ and $\sigma$ is proportional to the prior $p(\sigma)$ times the product of the densities of $U_i, i=1,\dots,k$, where

\begin{displaymath}U_i = \sqrt{\frac{s_i}{n_i}} T_{\nu_i}
+\bar{x}_i +\sigma Z,
\end{displaymath}

$\{\bar{x}_i\}$ and $\{s_i\}$ are the sufficient statistics, and $T_{\nu_i}$ and Z are independent Student-t and standard normal random variables, respectively. This is implies that

\begin{displaymath}p(\mu,\sigma\vert \{x_{ij}\}) \propto
p(\sigma) \prod_{i=1}^{...
...x}_i-\mu}{s_i/\sqrt{n_i}};\frac{2 n_i\sigma^2}{s_i^2} \right),
\end{displaymath}

where

\begin{displaymath}f_{\nu}\left(x; \gamma\right) \equiv
\frac{1}{\Gamma(\nu/2)\s...
...1+\frac{x^2}{\gamma y
+\nu}\right]}} {\sqrt{\gamma y +\nu}} dy
\end{displaymath}

is the density of $T_{\nu}+Z\sqrt{\gamma/2}$. This result is intuitively appealing and computationally useful. We've investigated generalizations to two-way mixed models, but the numerical integrations required become much more difficult. The figure illustrates posterior calculations for interlaboratory study data on arsenic in oyster tissue.




\begin{figure}
\epsfig{file=/proj/sedshare/panelbk/99/data/projects/inf/interlab.fig.ps,width=6.0in}\end{figure}

Figure 28: Marginal posterior densities for the consensus mean and between-laboratory standard deviation for an interlaboratory study on arsenic in oyster tissue. Calculations were done by numerical quadrature, using the hierarchical model given above, with a uniform prior on the between-laboratory standard deviation



contents     previous     next

Date created: 7/20/2001
Last updated: 7/20/2001
Please email comments on this WWW page to sedwww@nist.gov.