Quote:
Originally Posted by rtomes
As I understand it they combined together several catalogs to try an make a fully comprehensive sample as possible. I think that these catalogs all predate the use of fibers, so that is not an issue.
|
What do you mean, by "predate the use of fibers, so that is not an issue?" Whether the spectroscopy is performed by fibers, slits, grisms, prisms or gratings does not matter here. What matters is how the objects were chosen for inclusion in the catalog. Just picking as many objects as you can, from as many catalogs as you can, is great for creating a large database, but said catalog's uniformity
isn't (uniform, that is)! As Nereid has pointed out,
Quote:
|
Originally Posted by Nereid
The authors have stated that the "data base" is incomplete and inhomogeneous, wrt that "expected of a sample [of galaxies] with a limiting magnitude of mpg ~15, and a depth of about 75 h-1 Mpc" These are quite explicit and unambiguous.
|
which seems pretty clear to me. And what do they include as galaxies here? Defining a galaxy, spectroscopically, is almost as hairy as defining a quasar. As an example, I've attached a histogram of the redshifts of everything spectroscopically classified as "galaxy" in SDSS DR4 (this includes a good number of duplicate entries). The double-humped shape is due to LRGs and AGNs at higher redshift, since those galaxies are intrinsically brighter and, if I recall correctly, nearby clusters and blue galaxies at low redshift.
Again, this is well understood by those who put together these surveys, and most of the selection function happens before
any assumptions about cosmology are made.
And, as in my previous example with the Bell 2004 paper, you can't just assume that a published paper has correctly identified selection effects, corrected for them and also has a uniform classification criterion. Even in the mainstream papers, sometimes folks get this wrong, because it tain't trivial atall.
Quote:
Originally Posted by rtomes
I do not deny that there will be selection effects. Why I claim is that selection effects will produce a smooth curve with a maximum of 2 or 3 inflections. They will not produce 7 evenly spaced peaks.
|
First of all, the existence of your "7 evenly spaced peaks" is in doubt. Second of all, it is quite possible for a selection function to create several evenly spaced peaks: the SDSS quasar selection function does exactly this, in part because of the nature of the photometric filters that are used to select targets.
Quote:
Originally Posted by rtomes
Quote:
|
Originally Posted by parejkoj
What are the error bars on that graph? To roughly quote my advisor: "I won't do chi-by-eye without errorbars!" It looks to me like many of the "peaks" that you see are just statistical fluctuations.
|
There are no error bars, but in bins with 100, 400 or 1600 items the expected sampling errors amount to 1/10, 1/20, 1/40 of the numbers in the bins or 10, 20, 40. The early peaks are about 400/40 or 10 s.d. (where the data will be the best) and the later ones about 30/10 or 3 s.d. The result is very significant when all the data is taken together. |
No, there
are error bars. It is a physical measurement (number of galaxies with a given redshift), thus, by definition, it has an associated error--I've gotten burned on this before, myself! For something like this, one often assumes Poisson errors, as you have done above (square-root of the number of objects in the bin), but that includes some assumptions about the underlying distribution. Do you know what those assumptions are, and whether they apply to this sample?
Just so you are aware, these are not trivial questions, nor are they tangential to the point at hand. Determining whether something is a peak or not depends on what the function is that you are comparing it to. If the function itself is "peaky" (and selection functions often are)...
Quote:
Originally Posted by rtomes
I simply drew a curve that follows the trend so that about equal amount of space shows between the curve and the line. Its shape is not important and even without it you can just test the difference between the peaks and troughs of the claimed 4330 km/s cycle.
|
Uh-uh. No. Just drawing a line through the points says nothing about whether your claimed peaks are really there or not. You have to have something to quantitatively compare it to, not just a hand-drawn function.
Personally? Except for the first "peak" in the plot you have (and I have a couple of guesses where it comes from), the "peaks" you claim look like ordinary statistical variation, and/or bin size effects.