James C. Naylor

Characteristics of the human inference process in complex choice behavior situations

Abstract Ten subjects were assigned to each of six experimental two-cue inference conditions created by varying the validity of the first cue across levels of .40 and .80, and by varying the validity of the second cue across levels of .20, .40, and .60. In each case the two cues were orthogonal. All performance indices closely approximated the dictates of a probability matching strategy. Subject consistency did not deviate greatly from the predictability available in the stimulus system, and subjects exhibited a high degree of ability in matching their equations to those defining the environmental complex. The value of a second ecological cue was a function of both the validity of the cue itself and the validity of the cue it was paired with. Pairing an additional cue to one of low validity was always facilitating, while adding an additional cue to one of high validity was always detrimental.

The Use of Simulated Stimuli and the "Jan" Technique to Capture and Cluster the Policies of Raters

he is being required to evaluate. For example, if we ask k judges to rank a group of n individuals with regard to general job competence, the way in which they are ordered by each judge would represent that judge’s policy. To the extent that two judges did not order people in the same fashion their policies would be considered to be different. Also, the extent to which one could predict the rank order for a given judge using information about the individuals being ranked would represent the degree to which that judge’s policy has been captured. The process of capturing policies and of determining their general similarities and differences are two distinctly separate problems and have received considerably different emphasis on the literature. Generally speaking, almost nothing has been done concerning policy capturing, while a fair amount of research is available pertaining to clustering raters in terms of the similarities of their judgment. The intercorrelation and factor analysis of this type of data is the most

Generating multiple samples of multivariate data with arbitrary population parameters

A method of generating any number of score and correlation matrices with arbitrary population parameters is described. EitherZ scores or stanines are sampled from a normal population to represent factor scores by an IBM 1620 program. These are converted to variates from a population with an a priori factor structure. The effectiveness of the method is illustrated from research data. Some further modifications and uses of the method are discussed.

The influence of cue redundancy upon the human inference process for tasks of varying degrees of predictability

Abstract Nine groups of ten subjects each performed in a 200 trial typical multiple-cue learning situation. The groups were defined in terms of three levels of cue redundancy between the two cues ( r ij = .00, .40, .80) and three levels of total system predictability ( R e = .50, .70, .90). While both absolute and relative achievement was found to be significantly and directly related to r ij and R e , the effects of r ij were much greater at high R e levels. This interaction was attributed primarily to subjects increasing inability to match the total ecology at low R e levels as cue redundancy increased. Subject consistency also varied directly with r ij and R e , while subjects ability to develop proportional strategies varied only as a function of r ij . The obtained importance of cue redundancy as a variable in multiple cue inference tasks thus strongly supports Brunswiks argument for “representativeness” in probabilistic inference research.

Comparison of Two Approaches—Jan and Prof—for Capturing Rater Strategies

IN an earlier paper the present authors (Naylor and Wherry, 1965) have described the use of the Bottenberg-Christal JAN technique (Bottenberg and Christal, 1961) for the isolation of rater policies. In that paper we reported in detail the JAN outcome for one of four air force specialties analyzed. Actually four specialties were studied and the present paper is based upon the complete study. The four specialties consisted of two supervisory levels of a mechanical specialty and two versions of an administrative (housekeeping) type of specialty, both versions being at the same supervisory level. The JAN technique is based upon defining the capturing of rater policy as the extent to which one can predict the actions of a rater

A Cautionary Note Concerning the Use of Regression Analysis for Capturing the Strategies of People

A number of recent articles on multiple cue probability learning and profile judgment analysis have involved situations in which the stimulus cue R matrix was not an identity matrix (e.g., see Naylor and Schenck, 1968; Slovic, 1966; and Dudycha and Naylor, 1966). A common finding among the studies in which the cues possess greater than zero intercorrelation has been that subjects’ responses become more systematically a linear function of the cues (or profile dimensions) as the amount of intercorrelation between the cues increases.

ρ m As an "Error-Free" Index of Rater Agreement

are all instances of the application of regression analysis to the evaluation of performance strategies in probability learning tasks. Similarly, Naylor and Wherry (1965), Wherry and Naylor (1966) and Madden (1963) have used the multiple regression model to capture the strategies or policies of judges in attitude research. There is an important distinction, of course, between these two kinds of research investigations. The major distinguishing feature lies in the presence or absence of a criterion of &dquo;correct&dquo; behavior.

Some comments on the accuracy and the validity of a cue variable

Abstract The concepts of cue accuracy and cue validity are defined in the context of multiple-alternative tasks using the correlational model. The interdependence of cue validity ( ϱ XY ) and cue inaccuracy ( σ D 2 ) are illustrated and the importance of cue variance ( σ X 2 ) is demonstrated. Conditions under which σ D 2 is at a minimum are indicated and possible tradeoff decisions between accuracy and validity are commented upon.

An Empirical Comparison of ρα and ρm As Indices of Rater Policy Agreement

was made that the policy of a judge in a typical rating situation be defined by his least-squares regression equation. This definition is possible whenever the judge bases his ratings of a series of n stimuli on k known and quantifiable characteristics or traits (Xi). One of their main reasons for approaching judgment policies in this manner was to introduce a new method of expressing agreement or similarity between different judges’ respective policies. Perhaps the most traditional method for grouping raters has been