Stephen E. Edgell

Validity of Forced Responses in a Randomized Response Model

An empirical test was conducted to determine the degree to which respondents would follow the directions of the randomizing device in the directed response model of the randomized response technique (RRT). The results showed that a sizable percentage of the respondents would not give the directed response to a question if that response was a socially undesirable one and the question was sufficiently sensitive.

Determining the sensitivity and specificity of common diagnostic tests for carpal tunnel syndrome using latent class analysis.

Background: The accuracy of a diagnostic test used to classify a patient as having disease or being disease-free is a valuable piece of information to be used by the physician when making treatment decisions. If a standard reference test is available, determining the sensitivity and specificity of a new test is straightforward. However, if that reference test is incorrectly assumed to be perfectly sensitive and specific, the errors of the reference test can result in an underestimation of the accuracy of the test being evaluated. Latent class analysis can be applied to determine the sensitivity and specificity of a new test when no standard exists. Methods: Latent class analysis was used to determine the accuracy rates of three commonly used measures of carpal tunnel syndrome: Tinel’s sign, Phalen’s test, and the nerve conduction velocity test. Data included 162 wrists from 81 patients seeking treatment for symptoms associated with carpal tunnel syndrome. Results: Tinel’s sign and Phalen’s test were both highly sensitive (0.97 and 0.92, respectively) and specific (0.91 and 0.88, respectively). The sensitivity and specificity of the nerve conduction velocity test were 0.93 and 0.87, respectively. Conclusion: Estimates of the sensitivity and specificity of these common tests for carpal tunnel syndrome support their widespread clinical use.

A set-theoretic random utility model of choice behavior

Abstract A random utility model of choice was developed by combining the basic ideas of the well-known theories of Thurstone and Restle. The new model has exactly the same number of free parameters as Tverskys Elimination-by-Aspects model. Furthermore, both models were found to fit, with equal accuracy, the data reported by Rumelhart and Greeno, and Tversky. It was concluded that although the two theories are not identical, they may be difficult to discriminate empirically.

Configural information processing in two-cue probability learning

Abstract A linear model transformation technique for analysis was presented and used to reanalyze the Edgell and Castellan (1973) data. The configural effect reported by those authors was shown to be due completely to a correct utilization by the subjects of the configural information. Further, it was found that any relevant dimensional information was also utilized and that the level of dimensional information utilization was independent of the level of configural information utilization. This analysis technique was also applied to a new experiment which showed that this combined dimensional and configural information processing was not due to pattern only processing but rather to mixed processing of both dimensions and patterns.

Delayed exposure to configural information in nonmetric multiple-cue probability learning

Abstract The effects of delayed exposure to relevant configural information were studied. Three groups were run for 480 trials in a two-cue nonmetric multiple-cue probability learning task. All groups had one relevant cue dimension. One group had no relevant configural information, while another always had relevant configural information. The third group began without relevant configural information and was switched to relevant configural information after trial 120. The switched group did learn to utilize the configural information, but not nearly as strongly as the group which always had relevant configural information. A follow-up study explored the effects of the length of the delay before exposure to the relevant configural information. Four groups were run under the above conditions but were switched at trial 40, 80, 120, or 200, respectively. The results replicated the effect for the group switched at trial 120 and showed no differences between the four groups, indicating that the effect of delayed exposure is constant and requires little delay.

Delayed exposure to additional relevant information in nonmetric multiple-cue probability learning

Abstract Five experiments were run to further study the effect of additional information becoming relevant during the course of learning trials in nonmetric multiple-cue probability learning. The first experiment replicated the finding that utilization of configural information was degraded when the configural information became relevant during learning in an environment with a relevant dimension compared with the level of utilization if the configural information was relevant to begin with. The amount of degradation was again found to be constant regardless of the length of delay in introducing the relevant configural information, even to a delay of only 20 trials. The second experiment found little or no degradation in utilization of dimensional information that became relevant during learning in an environment with relevant configural information. The third experiment found some degradation for the utilization of dimensional information that became relevant during learning in an environment with relevant information in the other dimension. The longer the delay in introducing this relevant information, the greater the degradation. The fourth and fifth experiments found lower average utilization of configural information that became relevant during learning in an environment with no other relevant information.

Higher order configural information processing in nonmetric multiple-cue probability learning☆

Abstract Configural information processing in three-dimensional nonmetric multiple-cue probability learning environments was studied. It was found that subjects did utilize configural information of the overall pattern and of subpatterns. This utilization was higher for simpler patterns and for those containing relevant dimensional information. Dimensional information processing was also found and was both stronger than configural information processing and unaffected by the level of configural information utilization. A model by Castellan and Edgell was found not to fit these data.

Separable and unitary stimuli in nonmetric multiple-cue probability learning

Abstract Two experiments were run to study the effects of separable and unitary stimuli dimensions in a probabilistic, decision-making-like environment. Using a nonmetric multiple-cue probability learning task with two relevant dimensions and additional relevant configural information, no differences in utilization were found for these two types of stimuli. However, in an environment that subjects find more difficult, one in which one dimension is relevant along with the pattern of two additional dimensions, an effect was found. Unitary stimuli were found to facilitate utilization of configural information, but no difference was found for utilization of dimensional information.

An empirical test of the unrelated question randomized response technique

An empirical test was conducted to determine the degree to which respondents, interviewed using the unrelated question randomized response technique (RRT), would truthfully answer the nonsensitive question when directed to by the randomization outcome if that answer could be jeopardizing should it pertain to the sensitive question. Depending on the sensitive question, between 2% and 10% of the respondents did not answer the unrelated question truthfully. However, these percentages were much smaller than were previously found for the directed response RRT.

A statistical check of the DECsystem-10 FORTRAN pseudorandom number generator

The pseudorandom number generator in the FORTRAN library (function RAN) of the DECsystem-10 was tested for the statistical properties of an independent uniform random number generator. Five statistical tests were employed: the frequency test, serial test, lagged product test, test of runs up and down, and test of runs above and below the mean. The generator passed all tests. It was concluded that it could be used in applications that require a sequence of uniformly distributed random numbers.