Analysis Settings

 

In addition to data settings, the Data and Analysis Control contains inputs for the following analysis settings:

 

Output Prefix: A name for the analysis output. This will be the prefix appended to the SQL tables generated by/for this analysis.

 

Credibility: Provide the credibility method to be used as well as the parameter for that method. The Exposure Method assigns credibility based on exposure amount and the t-Statistic Method assigns credibility based on the consistency of observed relationships. For the Exposure Method, enter a 50% Credibility Level corresponding to an amount of exposure that will be considered 50% credible. For the t-Statistic Method, enter a Confidence Level (between 0 and 1) at which you would not reject the null hypothesis (no relationship). For more information on the two credibility methods, as well as the application of credibility where Smoothing across bins is involved, see the section on Credibility.

 

Convergence Threshold: The amount at which a change in a factor for any characteristic would be deemed insignificant. The analysis will complete once the change in all factors (from the previous iteration) is less than the convergence threshold (regardless of whether the maximum number of iterations has been attained).

Number of Iterations: The maximum number of iterations to run. The number of iterations necessary for convergence will depend on the complexity of the model being run. The greater the complexity of the model (more characteristics and values of characteristics), the more iterations will be necessary. If the convergence threshold is met before the maximum iteration has been reached, the analysis will complete before running the remaining iterations. Vice Versa, if the maximum number of iterations has been reached, the analysis will complete regardless of whether or not the convergence threshold has been attained.

Convergence Accelerator: Select whether or not to use the convergence accelerator. This option may help the model reach convergence faster and thus can be helpful in cases when analyses take a long time to run. However, there is a potential for greater iteration instability. For more information on this option, see the Convergence Accelerator section.