WALKER-DUNCAN 

 ALGORITHM 



RISK, the computer version of this algorithm, uses weighted least squares param- 

 eter estimation. Weights are estimated as the inverse of the standard deviation. A 

 first-degree Taylor series expansion around a guessed value of B provides the linear 

 function of the parameters required to make least squares a manageable estimation 

 procedure . 



An alternative form of the logistic function is given by 



In [P/(l-P)] = x'B 



Although this form of the logistic function can be fit by any linear regression program, 

 it has disadvantages. The independent variables must either be discrete or must be 

 partitioned into discrete classes. The set of all possible combinations of these 

 classes defines a set of categories to which each observation is assigned. P is the 

 proportion of the observations in each category for which the dependent variable is 

 1. Another disadvantage is that each category must contain enough observations to 

 insure a meaningful value of P for that category. Thus, the form of independent vari- 

 ables is not only restricted by the alternative model, but also the data requirements 

 are frequently greater than if equation (2) were used. 



6 



