Dissenting from J. W. Nienhuys' "Retrospect"
|"Two-time Nobel laureate Linus Pauling never gave up his belief in Vitamin
C, even though clear clinical evidence never materialized," so says Jan Willem
Nienhuys (Nienhuys, 1997, p. 29). Agreed. But do we therefore have to distrust
Pauling's discovery of the helix structure of protein? Some believe that
Mendel's data was biased. Agreed. But do they therefore discard his laws
of heredity? Some of Gauquelin's data is biased. Agreed. But is ti true therefore
that there is "no value...to the hypothesis these data gave rise to?" (Nienhuys,
p. 29). Nienhuys rejects Gauquelin's claim of a Mars effect because he distrusts
all results connected with Gauquelin which were positive (his own, or that
based on data collected by others with his assistance) and he trusts only
skeptics' results which were negative, i.e., results obtained from data collected
by U.S. (CSICOP) and French researchers (CFEPP). Since the skeptic data is
an important foundation for his belief, it is fair to ask: are their data
If the skeptics' data are unbiased, as Nienhuys believes, then their results should not differ appreciable from each other. But the two experiments from which he draws negative conclusions regarding the Mars effect actually differ remarkably.The percentage of Mars in Gauquelin's key sectors (rise/culmination) is 13.5% in CSICOP's sample, and 18.6% in the CFEPP's sample. The difference of 5.1% is significant by Fisher's exact test (p = .017), which demands an explanation. Biased selections are certainly conceivable on the skeptics' side, given these numbers. They might manifest themselves, for example, by researchers admitting mediocre sports figures to their samples more generously than they should. If the Mars effect exists, the key sector percentage for Mars affected by such bias should drop to the chance level accordingly. But CSICOP's Mars percentage is even much lower than the chance level, although not quite significantly so, while the CFEPP's Mars percentage is higher than the chance level, although not quite significantly so. Why is the CFEPP's Mars percentage not larger (an answer will be attempted below), and why is the CSICOP's percentage lower, even significantly lower than the CFEPP's level? The CSICOP's extreme negative deviation from all other skeptics' and from Gauquelin's sample has been dealt with in Ertel, 1992 and Ertel & Irving, 1997 (followed by a rejoinder by Kurtz, Nienhuys & Sandhu, 1997).The CFEPP's less than expected positive deviation has been dealt with in Ertel &Irving, 1996, and Ertel, 1996 and will be taken up below. It has been shown earlier and will be shown below that the CFEPP, just as the CSICOP, did nothing to exclude a most likely objection, namely that their selections were thoroughly biased.
Even when only doubtful samples are used as evidence, Dr. Nienhuys' contention that "what remains of the Mars effect when one starts entirely from scratch "is...nothing" is not well founded. A closer look at the CFEPP's data is due. Let us compare the Mars key sector percentage of top athletes, i.e., the percentage of the crème de la crème born with Mars rising and culminating, with the Mars percentage for lesser figures. If Gauquelin's claim does not hold, the Mars percentage of top athletes within the CFEPP's sample should not differ from that of lesser figures within that sample. Two checks on this will be made individually and in combination.
First, remember that the CFEPP's data was drawn from two sources, Le Roy and L'Athlège (see Benski et al., 1996). The Mars percentage of 295 champions having entries in both sources is 22.4% while the Mars percentage of the rest, N = 771, having an entry in one source only is much less: 17.4% 1) . The difference of 5% is significant by Fisher's exact test (p = .04) 2) . The CFEPP themselves rightly argued (on p. 19 of their report), that athletes having entries in two sources should generally be considered as more eminent than athletes having one entry only, and this is in line with the general principle of using citation counts as eminence indicators. Thus, by a procedure they themselves endorsed, a Mars effect appears within their own data.
Next we look at 464 team athletes (rugby, soccer, basketball, hockey). For each team athlete, the CFEPP published how often a player participated at international games. Using this information, the team sports players may be divided into two almost equal halves, those with less than average (N = 230) and those with higher than average (N = 234) numbers of participations. A significantly (p = .05) higher Mars percentage is found in the "higher than average" subsample (22.2% vs. 15.7%) 3), which confirms Gauquelin's eminence claim once again.
As a final step, high achievers of the above two definitions are pooled, so that those who are either two-source citation athletes or frequent participants in international games form one sample. Their Mars percentage is 21.9% (N = 443) while the Mars percentage of those who did not obtain any of those two qualifications is only 16.5% (N = 623). The difference is significant (p = .017). 4) The CFEPP's own information regarding success in competition (Benski, 1996, p. 48 ff) is thus sufficient to demonstrate the Mars effect and its dependence on eminence. How could they miss this result?
Aside from having not given any consideration to higher and lower achieving subsamples within their total sample of N = 1,066, the CFEPP also missed to adjust it to their standards of selection. Had they done so, the key sector percentage of the total, amounting to 18.6% in their report, would have increased and the error probability (significance level), amounting to p = .148 with uncorrected data, would have decreased correspondingly. First, the CFEPP should have deleted 16 B-Team players, following Gauquelin's objections to B-team players because they are ranked lower than A-team players. 5) The CFEPP actually found that Gauquelin's objection was justified. Their report in fact says "The sports champions not having been selected in A-teams were therefore eliminated" (p. 41). Despite this straightforward statement, the CFEPP did not eliminate B-team players (except Jean-Paul Escale, strangely, leaving 15 still in the sample). Nienhuys checked the sources and found five additional B-team players that should have been eliminated (p.130 f), but they were not eliminated either. Of those 15 + 5 = 20 to-be-eliminated B-team players only one had Mars in a key sector (Nienhuys, Table 5, p. 131). Had these 20 stowaways been thrown off board, the key sector percentage would have raised to 19.0% and the binomial test would have shown a lower chance level of p = .105. 6)
Finally, why did the CFEPP not include French champions born in Maroque, Algeria, Tunisia etc. in their sample? In Gauquelin's database (N = 53), French athletes not born in France showed an astounding Mars percentage (30% , see Benski et al., p. 133). Nienhuys does not contest the rightfulness of Gauquelin's demand, but merely puts it aside, saying "the CFEPP's resources to do much more data gathering had been exhausted at that time" (p. 133). Note that Gauquelin had stressed in a letter of December 6,1990, addressed to Benski, "the data on the births of these persons are particularly easy to access. . . .". Taking the issue up once again in another letter of February 7 , 1991, Gauquelin adds "...all births [of French persons not born in France] are centralized. The attached photocopy is an example of my request; it provides an exact address ...." (Gauquelin in Benski et al., p. 92). Nevertheless, for Nienhuys, "exhaustion" seems to be a legitimate excuse for the CFEPP's having not sent a letter for which the address and text were already done. All they had to do was to add a list of two to three dozen names. When 30 of 53 legitimate French athletes not born in motherland France are included, which is a conservative estimate, and when B-team players are deleted, the Mars key sector percentage reaches 19.4 % which is significant with p = .053. 7)
Thus, the CFEPP's total sample manifests a significant Mars effect after merely making neglected corrections that the researchers themselves found imperative. Why did they not make the necessary corrections? There are ample grounds to submit that the CFEPP were aware of Mars sectors as calculated by Gauquelin in prior investigations. Available data might well have been used to precalculate the Mars percentage. In fact, Nienhuys concedes that the CFEPP's concern for Gauquelin's published Mars sectors was alive right from the beginning: "The CFEPP had started [in 1982] by checking the data in Gauquelin's 1955 publication....," (Nienhuys, in Benski et al. , 1996, p. 121 f). Note that 63% (N = 675) of their final sample of 1,066 was listed in Gauquelin's data publications. Mars sectors calculated by Gauquelin were even explicitly referred to by the CFEPP in Benski's report of 1996 (e.g. on p. 33-37). Whence their knowledge of and interest in Gauquelin's sector information? An awareness of Gauquelin sectors seems to have played a major role throughout the fourteen years of their experiment (from 1982 through 1996. ). Benski et al. do not claim, apparently for good reasons, that they had taken precautions to prevent effects on their sample selection due to their checking with prior Gauquelin results. Nienhuys: "Let's leave it at that and move on to more fruitful research" (1997, p. 29). Should we? Really? 8)
Gauquelin's data does contain bias, too, and Nienhuys is correct in pointing out implications of an earlier discovery of this (see Ertel, 1988). How did Gauquelin's bias come to be? He first collected birth data of athletes ad libitum without too much consideration of their eminence. Only after receiving birth times from town halls did he look at the athletes' records of successes. He wanted to pick very good ones (this data he published) and to discard those of lesser stature (this data remained unpublished). Ertel traveled to Paris and copied Gauquelin´s unpublished data, as he wanted to check Gauquelin´s eminence assessments independently and objectively. He therefore counted the number of reference sources, out of 18 , in which an athlete was referred to. Using these counts, he found indeed that they were much lower for the unpublished as compared to the published samples. This is good news regarding Gauquelin's experimental sample, as he had in fact sorted out low eminence athletes (see Table 1 in Note 9). The bad news is that in Gauquelin's experimental sample the proportion of less eminent athletes, even though small, contained more key sector cases than they should have. When less eminent figures had Mars in a key sector, Gauquelin had been tempted to regard them as still good enough, and to keep them in his "better" sample. This is the essence of Gauquelin´s selection bias. 10)
However, in subsequent analyses Ertel combined Gauquelin's published (N = 2,888) with his unpublished sample (N = 1,496), thus undoing all effects due to his bias-prone procedure. The total resulting sample is less eminent overall, but the worth of it is that it is free from bias. 11) Do birth frequencies of Gauquelin's corrected total no longer deviate from chance expectation? See Figure 1. Birth frequencies across 12 Mars sectors for Gauquelin's corrected sample of N = 4,384 are shown as a dashed line and by hollow circles. A sharp peak at sector 1 (rise) and a much less impressive peak at sector 4 (culmination), taken together, still confirm Gauquelin's claim. The Mars percentage (19.4. %) is still very significant for this sample size (p = .00007). The Mars percentage of Gauquelin's uncorrected published sample of N = 2,888 was 21.8, p = 10 -11 12)
Figure 1 also shows birth frequencies of the CFEPP's sample (solid line). The Mars percentage for the total (rise plus culmination cases) is 18.8 % which is not significant (p = .148). The Mars percentage proved significant only after CFEPP-approved corrections were done (see above). Nevertheless, even with uncorrected birth frequencies of Figure 1, the CFEPP's birth frequency distribution hardly differs from Gauquelin's, and in fact the correlation amounts to r = .85 which is very significant (p = .0003) 13)
Why are the two samples so highly correlated? Due to a Mars effect? This option would be avoidable only if other explanations suggested themselves. How about effects due to data intersection? The two samples do intersect, as we find 21.1% cases of Gauquelin's total sample of N = 4,384 in the CFEPP sample. But this is apparently too little to explain the extraordinary correlation between Gauquelin's and the CFEPP's totals.
Moreover, if Gauquelin's peak at sector 1 (rise) were due to his biased selections (remember, bias effects were undone in the mixed sample anyway), then the CFEPP's curve should be flat at this point, but it isn't.
At Mars sector 4 (Mars culminating), the difference between Gauquelin and the CFEPP is slight, it is smaller, e.g., than the difference at sector 8. Moreover, both Gauquelin and CFEPP frequencies at sector 4 are close to chance expectancy. Why should Gauquelin's selections, if they were biased, not produce another sharp peak here?
The CFEPP's curve is really puzzling. Why should this curve drop abruptly from sector nos. 4 to 6 below the expectancy level, and why should it rise abruptly from 6 to 8/9 above the expectancy level, and why should these ups and downs match exactly the discontinuities of Gauquelin's curve? Dr. Nienhuys, scrutinizing Gauquelin's data, already noticed what he called "secondary" effects, i.e., increases of birth frequencies with Mars setting and reaching its lower culmination. He conjectured that this was merely another indication of Gauquelin's wishful selections. But how could Gauquelin's wishful "secondary" selections (removed anyway, if they existed at all in the mixed sample) affect the CFEPP's sample so as to match Gauquelin´s birth distribution? The CFEPP's sample was assembled "from scratch," as Nienhuys says, without Gauquelin's meddling.
The Mars effect drama has not ended yet and a final "retrospect" should be postponed for those who wish to give the matter dispassionate and objective consideration. It must be based on a much broader base of information, since reasonable questions are still left unanswered. One can agree, though, with Nienhuys' conclusion: "Even the best scientists can be trapped in illusions of their own making" (p. 29) - with the caveat that it applies equally to those who claim an anomaly and to those who claim that it cannot exist. One of us opponents must be trapped in illusions of his own making. The future will tell whether the claim of a Mars effect - or the claim of its non-existence - is the self-made illusion. 14)
An earlier version of this paper was kindly commented on by Alvo von Alvensleben, Ivan Kelly, Jim Lippard, and Rudolf Smit.
1) Fisher's exact test is used throughout for testing Mars percent differences between samples.
2) With Mars in key sectors: 66 of 295 two-source athletes, 134 of 771 one-source athletes.
3) With Mars in key sectors: 52 of 234 international games favorites vs. 36 of 230 non-favorites.
4) With Mars in key sectors: 97 of 443 highly qualified vs. 103 of 623 less qualified athletes. By applying the two criteria together (larger citation count, frequent international participation) the number of athletes needed to test the Mars effect hypothesis is much smaller than the CFEPP's research protocol suggested.
5) Gauquelin had submitted to them his objection in a letter of Feb 7, 1991
6) With Mars in key sectors: 199 of a total of 1046 CFEPP athletes.
7) With Mars in key sectors: 213 of a total of 1089 athletes. Nienhuys counted in Gauquelin's published data base 53 champions not born in European France, 16 (30%) born in a key sector (Benski et al., p. 133). We here assume that the CFEPP should have recruited about 30 of those 53 champions, not much less since Gauquelin´s published athletes, using citation criteria, are very eminent.
8) Nienhuys himself added 54 cases to the CFEPP's sample of N = 1,066, a purported amendment (see his list 3, p. 150/1) "Extra names...", p. 150ff). But he did not amend the sample by deleting those 20 B-team players. Moreover, he does not add athletes from non- European France. Above all, there is nothing in Nienhuys's account to indicate that steps were taken to insure that his adding these "extras" and his omitting justified deletions was done blind, something which would have excluded bias effects due to Mars sector awareness.
Citations 0 1 2 3 4 5 6 >6 Total Publ 1305 668 477 208 97 75 37 21 2,888 Unpubl 937 440 72 43 4 0 0 0 1,496
Example: 1305 athletes of Gauquelin's published athletes were not found in any of those 18 reference sources
10) When Gauquelin suggested adding 146 athletes to the CFEPP's sample and deleting 16, he apparently based his judgment on eminence information, see Table 2, not primarily - as Nienhuys seems to hold - on Mars sector knowledge.
Table 2: The number of Gauquelin's suggested additions and deletions, with a breakdown for number of citations that omits those found in LeRoy, as most CFEPP cases were taken from that source
Citations 0 1 2 3 4 5 >5 Total additions 59 37 10 4 3 3 0 146 deletions 15 1 0 0 0 0 0 16
Example: 15 athletes of Gauquelin's suggested deletions were not found in any of those 18 reference sources.
11) The grand total of 4,384 (published plus unpublished) suffers from low eminence while Gauquelin's published sample of 2,888 suffers from selection bias. A way out of this dilemma is to pick from the grand total athletes with at least one reference in those 18 sources, irrespective of whether Gauquelin had or had not published them. Such a sample might be regarded as bias- and eminence-corrected at the same time. Its N has 2,142 cases, the Mars percentage is 20.1%, p = .0003 . An even better sample, comparable in size to the CFEPP's, is defined by athletes with two or more references in those 18 sources. Its N has 1034 cases, the Mars percentage is 22.1%, p = .00004.
The expectancy level for Gauquelin data is set at 17.2% throughout, the highest value for such calculations prior to the CFEPP, and an unbiased value. To some degree, expectancies are sample-dependent. For the CFEPP data, Ertel obtained an expectancy of 17.5%, which value was exactly replicated by Mark Pottenger using different procedures. Astronomer Alvo von Alvensleben, using controls by shifting birth years (see Note 14) calculated CFEPP data expectancies repeatedly and obtained slightly varying levels, none of them exceeding 17.45% (personal communication). Ertel warned the CFEPP repeatedly that their expectancy value of 18.2% must be wrong, their report was not yet published. Only after its publication, Nienhuys found that it was based on miscalculations (Nienhuys, 1997, p. 25). His correction yielded 17.7%. However, considering the above calculations by three independent researchers, even this value appears still too large and is thus in want of justification.
12) With Mars in key sectors: 851 of 4,384 total cases, 630 of 2,888 published cases.
13) The correlation must be based on deviations from chance expectancies, see dotted line, thus preventing a boosting of the coefficient due to the non-uniform (wave-like) expectancy distribution.
14) Astronomer Alvo von Alvensleben, Kiepenheuer-Institut für Sonnenphysik, Freiburg i. Br., Germany, rejected an editor's invitation to write a review of Ertel and Irving's The Tenacious Mars Effect (1996) because he suspected computational or laymen's errors in celestial mechanics had played a significant role in the Mars effect debate. But since he felt challenged to check the Mars effect claim, he wrote a program of his own, analysed the CFEPP's data together with 100 control samples (with -25 through +53 years of shifted birth years) and faxed his results, adding as summary that the CFEPP's claim that there is no indication of a Mars effect is definitely not true (personal communication, 1997).
Following are additional comments on some specific matters in Nienhuys's reply (Skeptical Inquirer, July/August, 1998, p. 60: "Responding to Ertel: Mars Flukes") to the abbreviated version of the "dissent" shown above, as it was published in SI. That shorter version will be published here at a later date.