...follows
The specifications are the result of a genuine attempt to express plausible beliefs for the problem, but are not the judgements of an expert. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... tex2html_wrap_inline8550
This is because we consider tex2html_wrap_inline8582 as representing the sum of a base level, tex2html_wrap_inline8580 , plus a difference, tex2html_wrap_inline8600 ; and tex2html_wrap_inline8546 is informative for tex2html_wrap_inline8582 via tex2html_wrap_inline8580 . 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...sense.
For example, for a traditional regression of tex2html_wrap_inline8818 on tex2html_wrap_inline8820 , various linear combinations of the tex2html_wrap_inline8822 's give estimators for tex2html_wrap_inline8818 . Each such estimator possesses properties which we assess in order to determine whether the estimator is likely to yield good estimates, and so forth. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...it
We identify any quantity having zero variance with zero. Thus, the number of mutually uncorrelated quantities in the structure represents the number of underlying different axes of variation. For example if a belief structure, tex2html_wrap_inline8920 , consisted of specifications over the three quantities tex2html_wrap_inline8922 , and if we knew that tex2html_wrap_inline8924 , and had specified variances coherently, then there would be at most two different axes of variation over tex2html_wrap_inline8920 . 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... tex2html_wrap_inline8582 .
A property of the adjusted expectation is that it is a linear operator, so that for any scalars tex2html_wrap_inline9124 and tex2html_wrap_inline9126 we have

displaymath9114

. 

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...adjustment:
When it is the case that the current adjustment variance is already a small fraction of the initial variance, a further variance reduction might represent only a small partial resolution. However, this might also represent a substantial reduction relative to the current adjustment variance. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...reasons,
For example, we might remove uninformative quantities, or quantities that are relatively unimportant and expensive to observe. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...restrictions
There are various restrictions like this which vary from machine to machine. The principal restrictions can be found by issuing the LOOK: (program) command. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...letter
Many of the punctuation symbols have a special meaning in [B/D]. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...program.
A summary of the possible arguments to the LOOK:  command, together with their meanings, can be seen by issuing the command with a question mark as argument: LOOK: (?). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...bases
We can use wildcard facilities to generate complicated bases quickly. In particular, the symbol `$' after a name means all those elements whose names match as far as the `$' symbol; for example we could have used ``B$'' as shorthand for ``B1, B2'' above. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...1
[B/D] allows several different belief stores, indexed by store numbers, intended for alternative belief specifications. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...defined
Recall that ``B'' and ``D'' are the names of our two bases. Had they been called ``bacon'' and ``eggs'', we would have issued the command ADJUST: [bacon/eggs]. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...configuration
[B/D] has a default configuration which allows us to analyse straightforward problems without worrying too much about esoterica. In general, though, we have to consider the settings of the various [B/D] controls which control the action of the program. We will see some of these controls in use later. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...HREF="node45.html#figgrid">14.
We can issue the command without the ``+'' symbol, yielding only the canonical resolutions. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...retain
Much of [B/D]'s output is, or can be, retained for further use, perhaps as fresh input to the program. Some quantities are retained automatically: these include summaries of the latest adjustment, for example. Other quantities can be retained if you set the relevant control. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...construction
The assignments Y1 and Y2 still exist, but we can forget them for the time being. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...command
The standard LOOK:  command thus consists of a list of options in parentheses, optionally followed by a list of items, meaning that detail is to be restricted to these items. Such restrictions are relevant to only a subset of the possible options. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...it.
There are other ways of asking [B/D] to perform sequences of adjustments, suited to different contexts. There are also facilities which help to automate the partial adjustment process for sequences in a manner similar to some traditional stepwise regression approaches. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

David Wooff
Thu Oct 15 12:20:04 BST 1998