Southworth Det Eclips Binary Catalog

On Tuesday, January 21, 2014 3:42:50 PM UTC-5, wrote:
[Mod. note: stuff. Misattributed quote deleted, reformatted -- mjh

-------------------------------------------------------------

This is my second attempt to respond to your post. The first was
perhaps too sarcastic, especially regarding your "help".

I am surprised by the persistent efforts to deny the validity of the
sample I have chosen as a first test of the preferred total masses of
stellar systems hypothesis. I cannot see any scientific reason to
discount this sample, but look forward to the time that it will be one
of several test samples - making such arguments moot.

Having followed developments in physics/astrophysics for decades, my
distinct impression is that the nature of the gauntlet that a theory
or idea must run is dependent on who the authors are and whether or
not the new theory or idea fits in well with the prevailing paradigms.

I am troubled by the non-answers to my questions (3) and (4). You say
the computer deleted your tests, but would you forget the approximate
sigma values? Saying you have done tests and know the answers but the
results have disappeared would not hold up in a court of law or
science.

The non-answer to (4) is most troubling. Firstly you say again that
the data is "not at hand" and then you introduce the red herring of
the general stellar mass function (which is not really germane to the
issue). I claim that there is no more evidence for your assumption of
no preferred stellar masses than there is for my hypothesis of
preferred stellar masses. If you think I am wrong let's have an
unbiased person use Southworth's 2012-present data to test the two
hypotheses head-to-head, and report the full results.

Your pejorative comments about "quashed your research" are also
troubling. My comment on the "extraordinary claims require
extraordinary proof" was completely general comment and did not
exclusively referred to my research.

I welcome and appreciate any and all help in my research so long as it
is not directed at arriving at preconceived conclusions.

Robert L. Oldershaw
Discrete Scale Relativity

In article , "Robert L.
Oldershaw" writes:

I am surprised by the persistent efforts to deny the validity of the
sample I have chosen as a first test of the preferred total masses of
stellar systems hypothesis.

It is not a "first test". You have been playing with this for 30 years
and said yourself that you chose this data set because it looks
promising. Rule #1: first choose the data set then test for
significance, not vice versa.

I cannot see any scientific reason to
discount this sample,

There is no reason to discount the sample per se, but there is reason to
criticize the fact that you picked it because it looked good to you.

but look forward to the time that it will be one
of several test samples - making such arguments moot.

There already are many such samples.

On Saturday, January 25, 2014 3:24:28 AM UTC-5, wlandsman wrote:
Robert,

Let us know when you have a 2-sigma detection. Craig's conclusion from

his statistical analysis was



" my "results" are consistent with the measurements being a 50-50

random occurrence. This is not a viable scientific result."



Perhaps somewhere else in this long thread, there is evidence

presented for a 2 sigma detection, but I haven't been able to find it.

------------------------------------------------------

In CM's post of 1/9/14 you will find the following unsupported verbal comment.

"The end result was that in *no* case was there an individual excess
power present with better than 2 sigma significance."

I have asked CM to provide the exact chi-squared and sigma values, but
he said he deleted them right after doing the tests. From memory he
should know if, and by how much, the highest sigma value compared with
2.0-sigma.

The tests I have recommended involving the 18/3 disparity surely
involve significance at the 2-sigma level, and I have explicitly
stated that I think CM's assumed alternative hypothesis does far worse
on this particular test.

I am not claiming convincing evidence for my hypothesis, but only that
it deserves recognition as a valid hypothesis that can and will be
tested more thoroughly in the future.

RLO

On Saturday, January 25, 2014 3:26:53 AM UTC-5, Robert L. Oldershaw wrote:
On Tuesday, January 21, 2014 3:42:50 PM UTC-5, wrote:

I am surprised by the persistent efforts to deny the validity of the
sample I have chosen as a first test of the preferred total masses of
stellar systems hypothesis. I cannot see any scientific reason to
discount this sample, ...

Astronomers are routinely expected to justify their sample selection
criteria. What "objective" reason guides your selection? A selection
that looks good to you is not an objective reason.

Having followed developments in physics/astrophysics for decades, my
distinct impression is that the nature of the gauntlet that a theory
or idea must run is dependent on who the authors are and whether or
not the new theory or idea fits in well with the prevailing paradigms.

Regardless of whether that is true, observational tests of a theory
should routinely be able to justify its assumptions and statistical
basis and account for statistical biases like number of trials.

I am troubled by the non-answers to my questions (3) and (4). You say
the computer deleted your tests, but would you forget the approximate
sigma values?

What I said what that I was working at an interactive session (=
command line) with results stored in memory. When I rebooted the
computer that memory was not preserved. (I wouldn't say it was
deleted.) I recall that one of the ten statistical tests had a single
trial significance of 0.047, and after accounting for ten trials this
would be a significance of 0.47 or 47% chance of rejecting the null
hypothesis. That is a not a credible statistical detection.

... [Y]ou introduce the red herring of
the general stellar mass function (which is not really germane to the
issue).

Your originally proposed hypothesis, "there are no preferred masses,"
which is not a quantifiable hypothesis because "preferred" is a
subjective term by itself. Did you mean that "masses are not quantized
at multiples of 0.145 Msun?" Did you mean "masses are not quantized
multiples of any mass?" Did you mean "any mass has an equal
probability of occurring?" Because of your ambiguity, I picked an
example to demonstrate a point: nature does prefer certain lower
masses more than higher masses. Statistical tests are routinely
expected to put forth a well defined, quantifiable null hypothesis.


... If you think I am wrong let's have an
unbiased person use Southworth's 2012-present data to test the two
hypotheses head-to-head, and report the full results.

Several other people besides myself in this thread have attempted
other statistical tests, and none of them has produced a significant
result. These are impartial methods and tests that were applied, in
the sense that we didn't come with a preconceived notion. How would a
new "unbiased" person help? An unbiased person working with a sample
that you yourself admitted was biased to your personal preferences (a
sample you "liked") will not produce an unbiased result.

I say again, we're expecting a basic level of statistical robustness
because that's what we expect from any observational test of a new
theory. There's no reason to give you a special exemption from this
expectation.

CM

[Mod. note: apologies for delay in posting this and other articles.
Your moderator's day job has been demanding his attention more than
usual -- mjh]

On Tuesday, January 28, 2014 3:49:31 AM UTC-5, wrote:
On Saturday, January 25, 2014 3:26:53 AM UTC-5, Robert L. Oldershaw wrote:

On Tuesday, January 21, 2014 3:42:50 PM UTC-5, wrote:



I am surprised by the persistent efforts to deny the validity of the

sample I have chosen as a first test of the preferred total masses of

stellar systems hypothesis. I cannot see any scientific reason to

discount this sample, ...



Astronomers are routinely expected to justify their sample selection

criteria. What "objective" reason guides your selection? A selection

that looks good to you is not an objective reason.

------------------------------------------------------------

I only have time this morning to answer this first issue. A more
detailed response should be forthcoming.

Perhaps my choice of words was too cavilier and vulnerable to attack.

I learned about Southworth's binaries catalog from another paper
posted to arxiv.org that said good things about the catalog.
Southworth has probably never heard about DSR so he is not biased. The
catalog is not fossilized but grows. He is careful to only allow mass
estimates with relatively narrow error bars.

Before I even looked at the catalog, I thought that the 2012-present
data would be a good test of the preferred masses hypothesis.

I choose to only use 2012-present data because I sincerely believe that our accuracy in the very difficult estimates of stellar masses is improving with time. This is a basic principle of science. I cannot understand why you have such a problem with this sine qua non assumption of science. Do you think accuracy gets worse? If so then science is a fool's game.

You may not like the sample I chose for your own private reasons, but
it is a good sample that was chosen objectively and will become
increasingly diagnostic with time. Even if there was some bias in my
choice, and I formally declare that there was NOT ANY BIAS, the fact
that the catalog "grows" in size with new data, would eventually erase
any problem with the objectivity and scientific value of the evolving
sample.

Can we drop this issue?

On Tuesday, January 28, 2014 3:49:31 AM UTC-5, wrote:

Here are the rest of my additional responses to your 1/28 posting.

(1) I have already posted (and reposted!) a response to your comments
about my choice of sample and possible bias. I will keep reposting
this important response until your misunderstanding of the sample
situation has been corrected.

(2) Now you say sigma values are 0.047 to 0.47. Then what were you
talking about when you said: "The end result was that in *no* case was
there an individual excess power present with better than 2 sigma
significance." Why mention 2-sigma if the tests never came close to
that? I really do not know what to think about the disparity of
numbers being bandied about. It seems a bit inconsistent. That is
why I think we need testing by people who are not emotionally involved
to a high degree.

[Mod. note: as you generally claim someone is emotionally involved
whenever they find a result that disagrees with your preconceptions,
it's hard to see how that can be done to your satisfaction. Any
suggestions? -- mjh]

(3) Here is the hypothesis, as it was and evermore will be, please
write it down for future reference. DSR predicts that the estimated
total masses for binary stars will cluster around discrete multiples
of 0.145 solar mass. There will be peaks at the predicted multiples
and deep valleys at about M +/- 0.07 solar mass. The narrowness of the
peaks will depend on the accuracy of the estimates.

[Mod. note: this hypothesis is ruled out at extremely high confidence
level by observation, as discussed at great length over the past few
months. I am tempted to say that this discussion should end here
unless anyone has anything new to add -- mjh]

(4) I am looking for help from people with the following qualities.
(A) Good statistical skills. (B) No positive or negative bias in this
specific matter or towards DSR in general. (C) Someone who will
candidly report the most positive results as well as the most
negative, so we get a more comprehensive feeling for the overall
situation.

RLO
Discrete Scale Relativity

On Friday, January 31, 2014 4:22:57 PM UTC-5, Robert L. Oldershaw wrote:

I choose to only use 2012-present data because I sincerely believe that our accuracy in the very difficult estimates of stellar masses is improving with time.

It is *not* difficult to estimate stellar masses from eclipsing binaries -- in fact, there can hardly be a more straightforward astrophysical measurement. One applies Kepler's law to the radial velocity curves, and uses the light curve through eclipse to compute any small correction from a 90 degree inclination.
Compared to exoplanet studies (which have been around for almost 20
years now) the radial velocity amplitudes in eclipsing binaries are
enormous (by the ratio of a stellar mass to a planetary mass).

The difficult aspect of eclipsing binary studies is *finding* these
rare stellar systems. As we enter the age of large-scale synoptic
surveys, many newer and fainter systems are being discovered. For
example, the 2013 data in the Southworth catalog includes 14 stars
from the OGLE survey of the Magellanic Clouds. But it is perverse
to include this data but reject the ~10 magnitude brighter classical
eclpsing binaries just because there hasn't been any publicaton about
them in the past 3 years. Suppose we were considering masses within
our Solar System: most of the recent publications would be about newly
discovered outer Kuiper Belt objects, but that would not negate the
accuracy of mass determinations for the inner planets and asteroids.

Robert had it right in the very first sentence of this thread: "[The
Southworth catalog is] "a very good catalog of detached eclipsing
binary stars with mass determinations "accurate to 2%"." There
are 161 stars in this catalog -- why not use them?

On Friday, January 31, 2014 4:22:57 PM UTC-5, Robert L. Oldershaw wrote:
I choose to only use 2012-present data because I sincerely believe that our
accuracy in the very difficult estimates of stellar masses is improving with
time. This is a basic principle of science. I cannot understand why you have
such a problem with this sine qua non assumption of science. ...

That's not necessarily a valid assumption for newly discovered systems
(e.g. see Wayne Landsman's excellent post). If you are interested in
increased accuracy, then why not select those with smallest error
bars? The more recently reported sample measurements are *not*
necessarily the most accurate.


On Friday, January 31, 2014 4:27:58 PM UTC-5, Robert L. Oldershaw wrote:
On Tuesday, January 28, 2014 3:49:31 AM UTC-5, wrote:
(2) Now you say sigma values are 0.047 to 0.47. Then what were you
talking about when you said: "The end result was that in *no* case was
there an individual excess power present with better than 2 sigma
significance." Why mention 2-sigma if the tests never came close to
that? I really do not know what to think about the disparity of
numbers being bandied about. It seems a bit inconsistent. ...

A 2 sigma event occurs by chance approximately 1 in 20 times or 5%
chance of occurrence. But if I sample times, I have a 10 x 5% = 50%
chance of seeing a single such event. When I look at more than one
sample, I increase the chances of seeing a rare object just by chance.
I must account for this. This is accounting for number of trials.

The best *significance* probability value I reported for a single
chi-square trial of 0.047 = 4.7%, is approximately equivalent to 2
sigma for a single trial[*]. However, I did ten different sample
trials, so the final significance is 0.047 x 10 = 0.47 = 47%
probability of occurring by random chance. This is not significant.

I mentioned this in my original post: I saw a 2-sigma value in a
single trial (~5% significance) but must reduce the significance
because of number of trials.

Again, I ask, why don't you take responsibility for understanding the
statistics?

CM
[*] - in the physical sciences we often readily interchange between
probability and "sigma". 0 sigma is 50%/50% significance; 1 sigma is
~31%; 2 sigma is ~5%; 3 sigma is ~0.3%; 4 sigma is ~0.006%; 5 sigma is
~0.00006%; etc. The "confidence" is 100% - (significance).

[Mod. note: I'm posting this as it was plausibly sent before I closed
the thread, but followups should be sent directly to the poster.
Further postings in this thread will not be accepted. -- mjh]

wrote:
chi-square trial of 0.047 = 4.7%, is approximately equivalent to 2
sigma for a single trial[*]. However, I did ten different sample
trials, so the final significance is 0.047 x 10 = 0.47 = 47%
probability of occurring by random chance. This is not significant.

A minor nit (which doesn't affect your overall conclusion):

It seems to me that if each trial has a 0.047 chance of a positive
result, then the chance of at least one positive result in 10 trials
should be
1 - (chance of no positive results) = 1 - (1-0.047)**10 = 0.38

--
-- "Jonathan Thornburg [remove -animal to reply]"
Dept of Astronomy & IUCSS, Indiana University, Bloomington, Indiana, USA
"There was of course no way of knowing whether you were being watched
at any given moment. How often, or on what system, the Thought Police
plugged in on any individual wire was guesswork. It was even conceivable
that they watched everybody all the time." -- George Orwell, "1984"