Hit-Contingent Response Bias in Helmut Schmidt's Automated Precognition Experiments
John A. Palmer · Journal of Parapsychology · 1997
In a previous paper (Palmer, 1996), I presented the results of testing a proposed conventional interpretation of ostensible precognition effects in experiments with a random event generator (REG) that were reported by Helmut Schmidt (1969). The apparatus for the experiments, described fully in the Schmidt (1969) report, was a four-button quantum mechanical random number generator (RNG) encased in a box. For each trial, the subject pressed a button on the box to predict which of the four target alternatives the RNG would select immediately thereafter. One of four colored lamps then lit up to provide immediate feedback of the correct target. Schmidt presented data from four subjects in two separate experiments. One subject, K. R., treated the task as one of PK by responding with a 4 on each trial. Obviously, any test of response bias with this subject's data would be pointless. As in my previous paper (Palmer, 1996), I segregated the high-aim and low-aim trials, and the basic ESP results for the other three subjects, none of whom manifested long strings of the same response, are presented in Table 1. A critic, James Alcock (1990), suggested that subjects in Schmidt's study might have recognized locally nonrandom target sequences during practice trials and exploited them during immediately succeeding official trials. Consistent with this interpretation, it was found that in the records of two of the three subjects for whom the analyses were applicable, there was a strong tendency for the subjects to match their response frequencies to local excesses of particular targets in the target sequence. I determined that these matching biases could account for all the statistical significance in the data where the aim was to score high (the high-aim files) but not the low-aim files. However, because powerful statistical analyses uncovered no evidence of either global or local nonrandomness in the target sequences, Alcock's hypothesis was ultimately rejected. TABLE 1 SCHMIDT'S PRECOGNITION DATA Exp. S Trials Dev. Prop.Hits CR High aim: 1 OC 22577 +289 .263 +4.44(***) 1 JB 16246 +91 .256 +1.64 2 OC 5000 +66 .263 +2.19(*) 2 JB 5668 +123 .272 +3.77(***) Total 49491 +569 .262 +5.91(***) Low aim: 2 JB 4332 -127 .221 -4.46(***) 2 SC 5000 -84 .233 -2.74(**) Total 9332 -211 .227 -5.04(***) * p [less than] .05, one-tailed. ** p [less than] .01, one-tailed. *** p [less than] .001, one-tailed. In the course of scanning the raw data files of these experiments, I happened to notice another kind of response bias. It appeared that subjects had a strong tendency to repeat their calls after hits. If a subject, for instance, responded correctly with a 1 to a 1-target on a given trial, he or she would be very likely to repeat the response of 1 on the next trial. Because this effect was noticeable by visual inspection, I suspected that it was a strong one. I decided to undertake statistical analyses to verify and determine the magnitude of this response bias and then to see whether or not any of the ostensible precognition effect could be attributed to it. As in my previous paper (Palmer, 1996), I segregated high- and low-aim trials. The basic ESP results for the three subjects are presented in Table 1. METHOD AND ANALYSIS OF DATA I began by examining globally how subjects distributed their responses with respect to the target on the previous trial. These data are presented in Table 2, where the numbers of primary interest are those in the first column of proportions. A randomly responding subject should follow a hit with the same response about 25 percent of the time. That clearly is not what happened with these data. …