To: All Msg #39, Oct2692 07:49AM Subject: Re: +quot;Mars Effect+quot;: JWN replies Ertel
From: Jan Willem Nienhuys
To: All Msg #39, Oct2692 07:49AM
Subject: Re: "Mars Effect": JWN replies Ertel's 23/10 post (pt 2)
Organization: Eindhoven University of Technology, The Netherlands
From: wsadjw@rw7.urc.tue.nl (Jan Willem Nienhuys)
MessageID: 6049@tuegate.tue.nl
ReplyTo: wsadjw@urc.tue.nl
Newsgroups: sci.skeptic
In article <24OCT199209243794@skyblu.ccit.arizona.edu>
lippard@skyblu.ccit.arizona.edu (James J. Lippard) writes:
#The following is from the BITNET SKEPTIC discussion list.
#
#Date: Fri, 23 Oct 1992 16:13:48 MEZ
#From: Suitbert Ertel
#Subject: JWN and MARS EFFECT
#Sender: SKEPTIC Discussion Group
#Replyto: SKEPTIC Discussion Group
#Messageid: <01GQAYA8T8SI8WVZHG@CCIT.ARIZONA.EDU>
#
#
# 2.2 Control samples by yearwise shifts
#
#Dr. Nienhuys apparently rejects testing for planetary effects by
#examining the effects of shifts by units of years. He points out
#that fixed stars take the same positions in the sky every year at
#the same time and by the same token, Mars is purported to recur
#every year in similar positions. As the planet can only move
#within the restricted limits of the belt of the ecliptic,
#variations of position are deemed to be small.
#
As announced I will try to be more explicit now.
When we shift back by exactly one year the apparent position of
the ecliptica with regards to the observer will hard have changed.
However, on this ecliptica Mars will be in another place. Where?
Well, Mars takes 2.16 years to complete an orbit (between two
conjunctions), if I'm not mistaken. If this apparent orbit would be
traversed with uniform speed, then 1 year would mean that it is about
165 degrees advanced (or retarded). That is, in terms of houses or sectors,
about 16 or 17 sectors (36 sector division). In other words, shifting
all 1076 athletes one year back, would mean shifting all their Marses
by the same amount over the ecliptic. The whole distribution over the
36 sectors is cyclically shifted by approximately 16 sectors.
For other shifts the same thing holds, mutatis mutandis.
Now I know of course that Mars does not move with apparent uniform speed:
that speed is near opposition about 5 times quicker than near conjunction;
moreover, Mars sectors have varying width depending on "Mars seasons".
When Mars is in a wintry position on the ecliptic (wintry= where the sun
is in winter), the relevant Mars sectors are short, and in a summerlike
position they are long. On top of all this, Mars doesn't even move
uniformly in its own real orbit, in connection with its excentricity.
All the same, there is ample reason to believe that shifting birth years
of athletes all by the same amount will for many athletes give about the
same phase shift. Certain multiples of 1 year will give a better
approximation to "exactly same phase shift for all athletes" than other
multiples. For instance, a shift of 79 years will result in a phase
shift of almost an integral number of 360 degrees for almost all
athletes (I guess).
Now the question: how many and which multiples will introduce
such phase shifts that they may be considered as independent samples
of size 1076 from a comparable universe of athletes?
I don't know. I suggest: none, but I would be willing to believe 9:
phase shifts that are for each athlete 40 degrees or more differing
from each other phase shift, I would accept.
The burden of proving that such shifts may be considered as independent
samples is on the one who proposes this method.
#I don't know whether Dr. Nienhuys begging a convincing gist
#in his argument would accept my clarifying paraphrase of it:
#"Time series consisting of G proportions obtained once every
#year on the same day are autocorrelated, i.e. successive
#measurements are not independent, therefore they cannot be
#utilized, as Ertel did, as controls."
That is not what I meant, at least not when "autocorrelated"
is not very carefully described.
Let's call two Mars sector distributions cyclically correlated
when one can be obtained from the other by cyclically shifting
all individuals by the same number of sectors plus or minus a few
sectors. In that case that "same number of sectors" I propose to
call the shift parameter. Obviously two cyclically correlated
sector distributions can't be considered independent if the shift
parameter is between 3 and +3 sectors, nor when it is bewteen
6 and 12 sectors or between 12 and 6 sectors.
My argument is that I would be surprised if 50 yearly shifts would
result in 50 mutually independent distributions (each pair either
not cyclically correlated, or cyclically correlated with permitted
shifts).
# If that was his point and
#if his premise were true his conclusion would be valid. His
#objection might even hold empirically, for some reason or other,
#irrespective of its erroneous deduction.
#
#Therefore I calculated an autocorrelation function across N = 51
#shifts of G% arranged in yearbyyear order. I did not find any
#significant signal, r's for lag = 1 and lag = 2 are 0.1 and .10,
#respectively, both insignificant  autocorrelations should peak
#with lag = 1 and 2 if events observed in t(i) depend on events
#observed in t(i1) and t(i2).
I don't see what this type of autocorrelation, where (1) cyclic
shifts have not been considered and (2) only 1 and 2 years are
looked at. One year would result according to my estimate in at
worst 16 or 20 sector shifts, and two year in +4 or 4 cyclic shifts,
that is, if for these periods the nonuniformity of Mars motion
would not mix the results too bad.
# At that time (1991)
#in Holland optimism culminated. When the EUROSKEPTICS met in
#Amsterdam one of the Dutch researchers, Mr. Koppeschaar, announced
#an "unmasking" of Gauquelin's Mars effect; and the Dutch newspaper
#VOLKSKRANT having obtained pertinent information through de Jager,
#Koppeschaar, and Jongbloet, referred to the Mars effect as an
#artefact due to biological rhythms. The "Gauquelin bastion" was
#"crashing", one newspaper headline proclaimed. Such jumping to
#conclusions and their public dissemination, was it necessary?
One can't do much about newspapers getting hold of conference
abstracts, and then interpreting them in such a way that maximizes
reader interest. Fortunately, the practice of quoting daily
newspapers in scientific disputes has not spread much (yet).
On the other hand, taking isolated scientific findings (that
have not been discussed fully) out of their context, and then
overinterpreting them is what's going on all the time in circles
that look for evidence in favor of homeopathy, astrology, Erays
and so on.
But I don't think that any believer in astrology has wavered for
a second: those results had been "found" by skeptics, which makes
them suspect per se. Among astrologists it is inconceivable that
one's a priori beliefs are not the prime determinant of the
outcome of any investigation. So no damage is done to the tender souls
of occultists. What other damage can have been done?
[here the real JWN shines through the veneer, of course]
#
# 2.4 "Warning bell": recurrent values
#
#Dr. Nienhuys might have asked me whether I could explain
#recurrence among the 51 percentages  but he preferred to
#forget where the decimal numbers came from and to calculate an
#10^9 impossibility ending up in ringing the bell.
#
#I would have been pleased to point out to Dr. Nienhuys the
#following: The total number of athletes is 1,076. The proportion
#of athletes having Mars in key sectors varies between 25.19% and
#21.10% across 51 samples (one genuine, 50 controls). That is, the
#range of *absolute* frequencies span between 271 and 227. With 51
#observations ranging between 271 and 227, there are 45 possible
#results: 271(1), 270(2), 269(3)... 227(45). Numbers must therefore
#recur: First, of necessity, there are 6 more observations than
#there are distinct possible results.
[etc.]
Touch'e. Stupid of me. I should have thought of that.
Completely right.
#
# 2.5 Inferential statistics.
#
#Dr. Nienhuys came up with z = 1.23 as deviation of observed Mars
#born athletes (N = 271/1,076) from chance expectation which he
#estimated as N = 247/1,076 (G% = 22.93%). "Not impressive", he
#says. Error probability would be p = .11, so his statement could
#be rephrased by "not significant" ,i.e., not reaching
#p = .05, the conventional significance level.
I protest. *I* will only use the word "significant" when it
refers to the outcome of an experiment with a null hypothesis and
an alternative hypothesis well formulated before the experiment,
where the experiment should be designed in such a way that all
necessary precautions have been taken to prevent experimental
artifacts favoring one or the other hypothesis.
And even then, a judgement is necessary on the probability P that
the experimenter has overlooked a source of errors large enough
to account for an important part of the observed effect. Only when
that probability P is below the claimed significance level, the
significance level means something.
After all, any judgement like "well, let's believe there is
something there" rests basically on a more general type of
judgement: reject the implausible in favor of the more plausible.
As Professor Ertel will recall, I estimated the standard deviation
at about 14 absolute, no matter what the exact value was for G%.
However, the z = 1.23 was computed not from the 22.93 estimate,
but from another one, namely the middle value 23.6 of Ertel's shift
simulations. (Which I told Ertel, on his request). I clearly stated
(I think) that I don't know the "true" expected value.
#
#I thought he had taken them from Zelen's comprehensive canvass
#because Zelen's large N = 16,756 consisted of "ordinary" controls
#for French athletes. But I found that setting out from that study
#(details published by Gauquelin, 1977) he must have come up with
#an expected G% = 21.84% Mars (N = 235).
How are the percentages compared to the 24,961 births of the
general population of Gauquelin, 1972? Are there any theoretical
computations (like the 16.67% > 17.2% of earlier Gauquelin
computations) of the expected values?
#
#I checked G% with another special file of ordinary people in my
#archive, (N = 1,713), ordinary controls in that file have one
#Gauquelin athlete each as "birth twin" (born on the same day or up
#to 4 days earlier or later). For them G% is 21.72%, close to
With a margin of 1% plus or minus, so 2 more decimals are meaningless.
#Zelen's expectancy of 21.84%. Now, if we use as control 21.84%
#obtained by unsuspected skeptics and essentially confirmed by my
#"replication", the indicator z for CFEPP's Mars G% with athletes
#(N = 271) goes up: z = 2.658 , p = 0.0039. That is, even if we
#follow Dr. Nienhuys' statistical approach and do it correctly the
#result strongly supports the Gauquelin hypothesis.
#
#Nevertheless, the Nienhuys "parametric" test, even though feasible
#in principle, is second to what I have been proposing with using
#controls from yearwise shifts. Here we do not need new data as
#estimates for chance expectancy ("ordinary people") nor do we have
#to rely on "parametric" assumptions.
But the test relies heavily on unproven and implausible independence
assumptions. So: first a theoretical independence proof, and after
that (and Benski's final report, convincing us sufficiently that
his chosen standards are independent from data colleted by him) we'll
see.
#A second addendum refers to Dr. Nienhuys' alleging a statistical
#error on my part in another study:
Let's drop that (even though I am convinced that the track record
of a researcher has bearing on significance claims).
JWN
EMail Fredric L. Rice / The Skeptic Tank
