Decreasing Errors For Binary Star System Masses

On Monday, October 27, 2014 3:21:53 AM UTC-4, Robert L. Oldershaw wrote:

[Mod. note: reformatted. I would characterize this whole thread as
'harping on trivial issues' given that either the whole dataset or the
subset you arbitrarily chose rule out your preferred model at
extremely high confidence levels. I am tempted to close it here unless
there is anything more interesting to add. -- mjh]


1. I dispute your argument that the existing 49 data-point subset is
sufficient to "rule out" my prediction.

2. While not statistically convincing yet the existing subset's mass
distribution looks more like what I would predict than what you would
predict.

3. Are you absolutely positive that when the subset reaches 200
data-points it will have shifted away from my predictions and verify
your expectation?

I find the subject of radial velocity measurements fascinating so I
hope the moderator allows a couple of more comments.

[Mod. note: new insights are always welcome: it's the rehashing of
tired old argument that I am trying to avoid -- mjh]

1. The Southworth catalog includes stars of range 0.21 to 27 solar
masses. The rotational broadening present in higher mass stars can
limit the accuracy of the radial velocity determination and the
sharpness of the eclipse.

2. The catalog includes stars with periods ranging from 0.8 to 771
days. The accuracy of the data improves as more eclipses are observed
but this is more difficult with longer period orbits.

These factors are taken into consideration when using the quoted error
bars, and are *not* taken into account when selecting by publication
date.

On Monday, October 27, 2014 3:21:53 AM UTC-4, Robert L. Oldershaw wrote:

I also said that at some point in the new millennium I noticed a
marked and general improvement in stellar mass estimates. This is
reflected in the decreased error values and the decrease in
conflicting mass estimates by different research groups.


No you did not notice this. 1. The newer data does not have smaller
error values (see my previous plot). 2. The catalog gives only only
mass estimate for each star, so how could you notice conflicting mass
estimates? It seems that you are being dishonest.

The choice of a start date of 2012 can be characterized as arbitrary,
but most scientific tests based on observational data involves a
choice of the data to be used in the test.

Scientific tests do not use an arbitrary choice. I would hope a paper
that only quoted results from left-handed astronomers would be
rejected. Your criteria are barely better than this.

since my explicitly chosen sample
is an open-ended and growing sample, the choice of a 2012 start date
is a trivial issue that becomes evermore trivial with time and sample
growth.


Yes, if we only select data from left-handed astronomers then we would
eventually get the same results we would have gotten by not making
this arbitrary selection. It just takes longer to collect sufficient
data.

--Wayne

In article ,
Robert L. Oldershaw wrote:
1. I dispute your argument that the existing 49 data-point subset is
sufficient to "rule out" my prediction.

It's not an argument: it's a fact. You can apply simple
undergraduate-level statistical tests to the data to verify this for
yourself. Or you can see the earlier discussion where I did it for
you. The data are conclusively inconsistent with your model and they
are consistent with being uniformly distributed within the mass range.

2. While not statistically convincing yet the existing subset's mass
distribution looks more like what I would predict than what you would
predict.

No it doesn't: that's the whole point of the test. (I have never
predicted anything, of course: I've simply tested the prediction that
you provided and found it inconsistent with the data.)

3. Are you absolutely positive that when the subset reaches 200
data-points it will have shifted away from my predictions and verify
your expectation?

I don't have an expectation to be verified. What I can say is that,
based on the data we now have, it would be absolutely astonishing if
new data suddenly started to favour your model to the extent that the
confidence levels of rejection on statistical tests started to make it
viable again,

Martin
--
Martin Hardcastle
School of Physics, Astronomy and Mathematics, University of Hertfordshire, UK
Please replace the xxx.xxx.xxx in the header with herts.ac.uk to mail me

On Tuesday, October 28, 2014 5:12:15 AM UTC-4, Martin Hardcastle wrote:

It's not an argument: it's a fact. You can apply simple
undergraduate-level statistical tests to the data to verify this for
yourself. Or you can see the earlier discussion where I did it for
you. The data are conclusively inconsistent with your model and they
are consistent with being uniformly distributed within the mass range.

But they certainly do not prove that the distribution is uniform
either.

The first subsample of 34 binary star masses showed a clear preference
for my predicted multiples and a very clear avoidance of the gap
between the peaks.

You attributed this to a statistical fluke and said that more data
would lead to the uniform distribution you expect.

The newer 14 member subsample has a similar distribution to the first
34 member subset. I suppose you will claim that this is another fluke.

But here is the important question. If we get to 200 masses and the
distribution still favors multiples of 0.145 solar mass and disfavors
inter-peak masses, would this still be considered a fluke?

In other words, is there *any* point at which you would think that
maybe something scientifically interesting and unexpected has been
identified?

Or is it always a confederacy of flukes?

[Mod. note: reformatted -- mjh]

On Tuesday, October 28, 2014 4:54:17 AM UTC-4, Robert L. Oldershaw wrote:
It is interesting and informative to apply your reasoning to the
entire history of the research effort to determine the value of the
Hubble constant.

Precision is all well and good, but it is profoundly trumped by
accuracy. Lately we have seen bold claims of 7-sigma detections go up
in flames, or get buried by dust. Clearly statistics are a
double-edged sword.

Measuring the Hubble constant is extremely difficult. Measuring masses
of eclipsing binaries is extremely easy.

You slandered the eclipsing binary community when you said without
providing evidence that there were eclipsing binary mass measurements
that disagreed with each other. Could you please withdraw this
statement or provide evidence for it?

The eclipsing binary measurements are more analogous to measuring the
brightness of stars. Ever since the development of linear detectors
(CCDs), there is no reason to doubt the accuracy of stellar brightness
measurements. The precision can improve, and one can now observed
fainter stars, but it is ridiculous to bring up the subject of
systematic errors for these simple measurements.

Or perhaps you think that the pre-2012 researchers missed a factor of
two when applying Kepler's law? --Wayne

[Mod. note: reformatted -- mjh]

In article ,
Robert L. Oldershaw wrote:
But they certainly do not prove that the distribution is uniform
either.

Why, no. I assume you're familiar enough with statistics to realise
that 'consistent with X' and 'proves X' are different things. It's
impossible in principle to prove uniformity to some arbitrarily
low level with any given finite dataset.

The first subsample of 34 binary star masses showed a clear preference
for my predicted multiples

No, you don't understand. It was consistent with being uniform. That
means you cannot claim there were any 'clear preferences' for anything
else. The null hypothesis has to be the uniform distribution.

You attributed this to a statistical fluke and said that more data
would lead to the uniform distribution you expect.

I didn't make any claim about what more data would do, if I recall
correctly, but certainly the most plausible hypothesis based on the
existing data is that they will continue to be consistent with being
uniform.

The newer 14 member subsample has a similar distribution to the first
34 member subset. I suppose you will claim that this is another fluke.

I won't claim anything until I or someone else has done an appropriate
statistical test. Have you?

But here is the important question. If we get to 200 masses and the
distribution still favors multiples of 0.145 solar mass

'still' is of course not true: the data do no such thing.

and disfavors
inter-peak masses, would this still be considered a fluke?

In other words, is there *any* point at which you would think that
maybe something scientifically interesting and unexpected has been
identified?

Sure, that kind of goes with being a scientist: you have to believe
what the data say. If a statistical test ever shows that those
residuals are non-uniform to a high confidence level, I will be as
interested and surprised as anyone else. But I will accept no
substitute for a high-confidence result on a standard statistical test
based on an appropriately selected sample: certainly not statistically
naive handwaving about small cherry-picked samples. Since tests on the
existing Southworth catalogue or any sensibly selected subset of it
show no significant deviations from uniformity, I see no reason to
expect that that's going to happen.

Meanwhile, let's ask the opposite question. Supposing the data are
completely inconsistent with quantization -- and, so far, they are --
what will you do? Is there any observation that would modify *your*
beliefs? Are you willing to commit to some action once we have some
specified number of observations, and abide by it? (That would have
the advantage that we could hope to lay this topic to rest until the
data were taken.)

Martin
--
Martin Hardcastle
School of Physics, Astronomy and Mathematics, University of Hertfordshire, UK
Please replace the xxx.xxx.xxx in the header with herts.ac.uk to mail me

On Wednesday, October 29, 2014 6:55:34 AM UTC-4, Martin Hardcastle wrote:

Meanwhile, let's ask the opposite question. Supposing the data are
completely inconsistent with quantization -- and, so far, they are --
what will you do? Is there any observation that would modify *your*
beliefs? Are you willing to commit to some action once we have some
specified number of observations, and abide by it? (That would have
the advantage that we could hope to lay this topic to rest until the
data were taken.)

Bottom line first: I am willing to admit I am wrong about the M1+M2
stellar mass distributions for detached binary star systems if no
preference is shown for the predicted preferred masses.

Here is the current situation as I see it. No one else is required to
see it this way.

I make a simple histogram of deviations from the predicted peaks with
3 equally sized bins. For the 49 system set there are 24 systems
closest to the peak, 5 systems in the inter-peak bin and 20 systems in
the intermediate bin.

This distribution was found for the 34 member set and the newer 14
member set. These results encourage me to think that when 3 more 50
member subsets are available, each subset might repeat the same
pattern and the full 200 member set of systems would show a clear
preference for the predicted peaks at n(0.145 solar mass).

I repeat that others may reject this strategy, but it is the only one
I can fully trust at the present time.

[Mod. note: reformatted -- mjh]

In article , Martin Hardcastle
writes:

Meanwhile, let's ask the opposite question. Supposing the data are
completely inconsistent with quantization -- and, so far, they are --
what will you do? Is there any observation that would modify *your*
beliefs? Are you willing to commit to some action once we have some
specified number of observations, and abide by it? (That would have
the advantage that we could hope to lay this topic to rest until the
data were taken.)

Probably not, as a "definitive prediction" (direct quote from the
abstract) of a DSR paper has, now that experimental accuracy and
precision is much better than when the paper was written, been
falsified. There is no evidence of substructure of the electron at the
predicted level. Measurements at DESY and elsewhere testing QED at
small scales and assuming point-like fermions are completely consistent
with theory. Any deviation would have been seen.

In article , "Robert L.
Oldershaw" writes:

I make a simple histogram of deviations from the predicted peaks with
3 equally sized bins.

Why 3?

For the 49 system set there are 24 systems
closest to the peak, 5 systems in the inter-peak bin and 20 systems in
the intermediate bin.

You once posted an example with more details. IIRC the widths of the
bins were such that it was most likely to get a spurious signal. 0.145
is quite small; any reasonable stellar mass will be close to SOME
multiple of it.

This distribution was found for the 34 member set and the newer 14
member set. These results encourage me to think that when 3 more 50
member subsets are available, each subset might repeat the same
pattern and the full 200 member set of systems would show a clear
preference for the predicted peaks at n(0.145 solar mass).

If I flip a coin 3 times and it comes up heads three times, I am not
going to bet much on it coming up heads 10 times in a row.

In article ,
Robert L. Oldershaw wrote:
Bottom line first: I am willing to admit I am wrong about the M1+M2
stellar mass distributions for detached binary star systems if no
preference is shown for the predicted preferred masses.

Excellent, because no statistically significant preference is
currently shown in the full Southworth dataset or any subset of it,
including the one you've chosen (quite arbitrarily, as Wayne has made
clear elsewhere in the thread).

This distribution was found for the 34 member set and the newer 14
member set. These results encourage me to think that when 3 more 50
member subsets are available, each subset might repeat the same
pattern and the full 200 member set of systems would show a clear
preference for the predicted peaks at n(0.145 solar mass).

They may encourage you to think that, but they should not. In fact, as
you were shown in several different ways last time this came up, the
data for your 34-member subsample are consistent with a uniform
distribution and so you have no grounds for believing anything else.

As the late John McCarthy used to say, he who refuses to do arithmetic
is doomed to talk nonsense. Or, to put it another way, if you want to
convince anyone of anything in science, you need to learn to
understand what is statistically significant and what isn't.

I see we are pretty much repeating arguments from 9 months ago here.
If there are no further new points to make then perhaps this thread
should be closed at this point.

Martin
--
Martin Hardcastle
School of Physics, Astronomy and Mathematics, University of Hertfordshire, UK
Please replace the xxx.xxx.xxx in the header with herts.ac.uk to mail me