The replication crisis in science begins with faked data. I discuss here a well-known recent case, Hans J Eysenck. An enquiry at King’s College London and scientific journals concluded that multiple publications by Hans J Eysenck’s are ‘unsafe’ and must be retracted. These recent events suggest that the entire edifice of Eysenck’s work warrants re-examination. In this post I examine some early experimental research by Eysenck and his students at the Institute of Psychiatry during the 1950s and 60s.
Hans J Eysenck was a chameleon-figure in the science of psychology. Eysenck doctored data from the very beginning of his theorising. Time and again HJE proved that he was a master of camouflage. I examine here some historically significant data that HJE used to promote his biological theory of personality, data that were used by HJE in a misleading way to promote his theories.
The evidence suggests that HJE massaged data to give them more ‘scientific’ appeal. HJE’s biological theory had predicted that introverts would condition more quickly than extraverts. The original data were collected by Cyril M Franks who had worked for his PhD under Eysenck’s supervision at the Institute of Psychiatry, London. Even Franks would later turn upon the master for his misleading methodology and data analysis. However, HJE dismissed and vehemently attacked all of his critics, claiming they were wrong, foolhardy and unreasonable.
HJE’s version of Cyril Franks’ data was originally published in the British Journal of Psychology (Eysenck, H. J. (1962). Conditioning and personality. British Journal of Psychology, 53(3), 299-305) and again, the next year, in Nature (Eysenck, H. J. (1963). Biological basis of personality. Nature, 199(4898), 1031-1034 and in multiple other publications. HJE doctored the data to make the introverts show a more rapid increase than the extraverts. These data were a crucial step in his theory published in his 1957 book, The dynamics of anxiety and hysteria.
HJE used a series of questionable practices (QPRs) that raised many eyebrows including insiders at the Institute of Psychiatry. Eysenck’s theory of personality became the subject of scathing criticism. Chapter 5 of Playing With Fire by Rod Buchanan provides the full details.
My personal skepticism about HJE began as an undergraduate student when a lecturer, Vernon Hamilton, told me that HJE had ‘cheated’ with his data – see Hamilton’s critique here. Other telling criticism was published by Storms and Sigal here and in another article with Franks: see Sigal, Star and Franks here.
In spite of all of the controversy, which he seemed to rather enjoy, HJE became one of the most influential psychologists of all time. His Nature paper has been cited 6331 times.
In light of the recent exposure of Eysenck as person who carried out serial publication fraud, it is informative to take a close look at Cyril Franks’ PhD research that in HJE’s creative accounting became a foundation stone of HJE’s first theory of personality.
EYSENCK’S DOCTORED CURVES
An almost perfect set of findings, one might assume – too good to be true even. My detailed scrutiny suggests that this was indeed the case. When one examines the data HJE used to generate these two curves, we see anything but smoothly increasing scores.
JAGGED-LOOKING ORIGINAL DATA PUBLISHED BY FRANKS (1956):
Franks tested a hypothesis attributed to Pavlov: “Neurotics of the dysthymic type form conditioned reflexes rapidly, and these reflexes are difficult to extinguish; neurotics of the hysteric type form conditioned reflexes slowly, and these reflexes are easy to extinguish”. Franks chose data from 20 dysthymic patients (having rejected data from 8 others), 20 hysteric patients (having rejected data from 7 others), and 20 non-patients …in a specially constructed soundproof conditioning laboratory.” The results for the dysthymic and hysteric groups were as follows:
Not unreasonably, Franks concluded that dysthymics give significantly more CR’s than hysterics. Buoyed by his initial success, Franks carried out another study to examine the factor of extraversion/introversion in the same eye-blink conditioning task. In this instance, Franks hypothesised, following HJE’s theory, that the introverts conditioned more quickly than the extraverts.
MORE JAGGED-LOOKING ORIGINAL DATA PUBLISHED BY FRANKS (1957):
Franks’ 1957 data again show the rates of classical conditioning in eye-blink responses in this case for 15 introverts and 15 extraverts. According to Eysenck’s theory, the former group should show more rapid conditioning than the latter. The maximum score was 15.
EYSENCK COMBINED DATA FROM FRANKS’ TWO STUDIES
HJE combined the data from Franks’ two studies in a rather creative and unconventional manner. HJE combined data from groups of introvert non-patients with patients diagnosed with dysthymia and he combine data from a groups of extravert non-patients and patients categorised as hysterics. The data from the two Franks studies were a hotchpotch that needs untangling.
1) Eysenck combined the data from the extraverts with the data from the patients classified as hysterics and the data from the introverts with that collected from the dysthymics This rather odd amalgam smoothed out many of the jagged edges in the two data sets.
2) There was no justification for assuming that the CR rates began at zero because all four groups had minimum scores above zero. This fact was pointed out by Vernon Hamilton. Yet HJE made the data look this way by imposing curves that started at a zero origin.
The next figure shows the data after they had been combined, groups D and I together, and groups H and E together. I show the combined data with HJE’s smooth curves and the data points as HJE reported them.
EYSENCK ACHIEVED THE LOOK HE WANTED USING CHILDISHLY SIMPLE METHODS
HJE’s 4-step approach to a successful scientific outcome proceeded as follows:
- First, HJE combined data from 4 different groups to create two new groups even though there was no scientific basis for doing so.
- Second, although HJE’s and my computations of the combined data points show a fair degree of consistency, HJE appears to have ‘adjusted’ a few data points that didn’t fit the curve.
- Third, HJE’s gave his smoothed curves zero starting points, contrary to the actual data, which indicate above-zero baseline scores. HJE attempted to disguise the fact that the groups had radically different, non-zero starting points.
- Fourth, HJE ignored the fact that two lines with identical slope fitted the data equally well.
Using these devices, HJE promoted the data as respectable science fit for publication in the most reputable journals. This analysis suggests something rather different – that HJE was an out-and-out charlatan.
The left panel of the diagram below shows HJE’s published curves with his cunningly averaged data-points converted to percentages (small dots), and the same averaged data-points obtained by this author (DFM; o’s and x’s). In most cases, HJE’s and DFM’s data points coincide. In at least 4 instances, however, HJE’s points lie closer to the theoretical curve than the correct figures would suggest.
The right-hand panel shows the fit to the same two data sets using linear plots with identical slope. Neither of the fitted functions look anywhere near perfect, but there is no prima facie case for preferring the curvilinear to the linear fit.
HJE constructed a curvilinear association between eye-blink classical conditioning rates and questionnaire measures of extraversion-introversion. These curves were artificially doctored to suggest that introverts conditioned more quickly than extraverts, as HJE’s theory had predicted. By combining data that did not belong together, HJE was able to smooth the data sets, which when considered separately did not fit the predictions quite so well. HJE avoided a feasible alternative (null) hypothesis that the two groups produced identical rates of conditioning. In so doing, HJE helped to establish his first biological theory of personality. This was not only bad science, it was faked science, the work of a chameleon.
I thank Rod Buchanan for his input and advice.
Follow me at: