Interesting. And encouraging to see that at least one zooite is still actively working on Quench*.
Some quick comments (sorry, not too much effort put into checking things out):
- Bamford+ (2008) (I think that's the correct ref), working on GZ1 data, developed a (supposedly) robust method of estimating 'true' morphology, at least out to z=0.085, based on things like size, surface brightness, and redshift (basically, the smaller/fainter/etc a spiral is, the more likely it is to be classified as an elliptical). Given the rather large redshift range (0.02-0.10) and, more importantly, the difference in distribution of size (QS vs QC) independent of z, how robust are estimates of 'true' incidence of bars?
- a not-insignificant proportion of objects classified as 'not features-or-disk'->'cigar-shaped' are, in fact, disk galaxies ('spirals'); perhaps a look at those (sharp-eyed) zooites who went down the other branch of the tree, in terms of 'bar fraction' might also be interesting
- there shouldn't be a systematic inclination bias (QS and QC objects are, over the whole of subset 2, as likely to have an inclination of θ1 as θ2, for randomly selected θ), but one thing I've learned to be very wary of is untested systematic biases; fortunately, the relative inclination bias - between QS and QC - should (ha!) be fairly easy to determine/estimate.
*I am too, but in a different part of the electromagnetic spectrum (hello Ivy); stay tuned! 😛
Are you referring to the bias correction with redshift that was applied to GZ2 and (I think) GZ1 classifications?
No, those applied to GZ1. But, as you say, with such a well-matched control sample, if there's a bias, it'll likely be pretty subtle
Also, and I just checked this, there is no significant difference in angular sizes as measured by PetroRad_r or PetroR50_r between subset 2 of QS and the matched controls.
Me too (I checked); for the ~1150 QS (and their matched QC) galaxies, the size distros are very close. That it's not (or may not be) for a different subset isn't relevant.
It seems to me any definitive analysis of the "robustness" of classifications would require having a gold standard to compare against. Lacking that I'm just copying and pasting objects into the SDSS image list tool and seeing what my own lying eyes tell me. That's why I posted those tables, so anyone who stops by can have a look for themselves.
One opportunity - beyond our (you 'n me) control - is to set up a semi-closed classification exercise, and (blindly) invite a couple of dozen super-classifiers to take part. Kinda like what waveney did (own engine, own classification trees), with the added help of having someone in GZ blindly (as in, not get to ID individuals, at any level) pick super-classifiers. The Supernova Zoo paper (can't find link now) clearly showed that zooite classifiers can be cleanly split by their acuity (at least for supernovae, but it should also work for galaxies), so apply a similar algorithm to GZ classifiers and pick 'the best' tribe.
Something like this has certainly been talked about - e.g. at the Taipei workshop - and will surely one day be implemented; whether it's made as a semi-open tool, for enthusiastic zooites themselves to pick up and use, is ... well, it's not even discussed (as far as I know).