Establishing and generating the model

 
Figure 2:  Generating the model

Before constructing any influence diagrams, it is necessary to prepare the model. To do this, consider the fragment of code shown in Figure 2, which does the following.

• The ELEMENT:  command is used to introduce the four random quantities together with prior expectations for them.
• The BASE:  command is used to specify the collection .
• The VAR:  command is used to specify the variance-covariance structure over the collection . These are explicit variance specifications.
• The ASSIGN:  command is used to specify the modelled linear relationships. At this point, i is an index. The intention is later to construct the 24 quantities from these assignments.
• The FVAR:  command is used to specify beliefs for the error components. This command is used especially when the random quantities involved are uncorrelated with most other quantities, so that the complete description of beliefs about these quantities, and between these and other quantities, can be specified as simple functions. For random quantities such as these, the default action of [B/D] is to treat as zero any expectations and correlations not explicitly stated.
• The DATA:  command is used to input the data shown in Table 1. The data follows directly after the command. For larger data sets it would be more usual to input the data from a different file.
• The FOR:  command loop transforms each temperature, by subtracting the average temperature and dividing by 100.
• The INDEX:  and COBUILD:  commands construct the 24 quantities , taking into account the previously constructed quantities and the functionally defined error components. The covariance structure between and the 's and 's will also be correctly constructed. The COBUILD:  command, which is an extension of the BUILD:  command introduced in [15], generates collections of quantities from given linear relationships simultaneously.
• Finally, within the FOR:  loop, the DATA:  command is used to attach individual observations to their corresponding random quantities.

See [15] for various tools for examining the specifications. For example, the LOOK:  command with parameters e , b , d , a , v1 , and r1  may be used respectively to view expectations, bases (collections), data, assignments, variance-covariance specifications, and variance-covariance specifications in correlation format.