Brad,

Just taking a CLINPLOT timeout (getting close to a first CVS commit
Tim).

I'm not at all surprised that the derived shapes are different.

> I'm trying to determine an average ellipticity of objects in a field as
> one of the quality control parameters for ORAC-DR. If the average
> ellipticity is high, then this is an indication that there's probably
> something wrong with the telescope optics.

This is all objects, right?  You've not selected by size or stellarity.

> To do this, I tried using KAPPA/PSF. I passed in a catalogue of objects

KAPPA/PSF is geared to compact stellar objects.  The word "star" is in
the purpose.  Yes it does fit a Sersic (1968) and such a function
describes certain galaxies and components of galaxies well (e.g. Caon et
al, 1993, Young & Currie 1994, Graham & Prieto 1999).  It calculates the
marginals in small regions.  Did you adjust the ISIZE parameter with the
object dimension?  What's the size of the box in arcseconds and how does
that compare with the size distribution of objects?

Now I don't now the details of how EXTRACTOR and the unnamed pipeline
derives shape characteristics, but it's normally found using the moments
(proper ones, not those ACSIS pseudo moments (-:).  If there's intensity
weighting, that makes a difference too.  Other software could do ellipse
fitting.  Either way, you're likely to be fitting to lower isophotes
over a larger region than with PSF.  PSF may not be reaching sky in the
marginals it uses.  Such factors are going to introduce errors and hence
a broadening of the histogram.

Then there's overlap considerations.  How crowded is the field?

> ellipticities there because I didn't calculate the average ellipticity for
> a field as measured by EXTRACTOR properly so they're skewed too high. The
> bottom histogram is pretty much what we'd expect from a good night at
> UKIRT.

The shape looks like many I produced for my thesis at the external's
request.

> I then took a closer look at two objects. The first is located roughly at
> 512,143. EXTRACTOR says it has an ellipticity of 0.029, and PSF says it
> has an ellipticity of 0.0826. If I use the Pick Object functionality of
> GAIA I get an ellipticity of 0.045. If I take two slices in X and Y
> through the object and put both of these through FIGARO/FITGAUSS I get an
> ellipticity of 0.032.

Did you try PSF with the exponent constrained to 2, i.e. Gaussian?

The size area surrounding the object is relevant if you'recomparing PSF
with GAIA/Pick Object.

> The second object is located roughly at 1168,256. EXTRACTOR's ellipticity
> is 0.008, PSF's is 0.254, GAIA's is 0.039, and FITGAUSS' is 0.049.

I'll take a closer look at the examples you cite once the first CLINPLOT
code is nestling in CVS.

> I suppose my question is: can anybody recommend which method to use to
> give a robust estimation of the ellipticity?

For a selection of objects of various shapes, I'd go for the moments
approach.  The way the image segmentation is performed, you shouldn't
have a size bias.

You might want to try playing with the EXTRACTOR parameters and
thresholds to see what gives least noise.  As you're interested in
shapes deleivered by the optics, then you surely want to concentrate
on the brighter but unsaturated stars.  TOPCAT is good for
matching and comparing such catalogues (Pasadena ADASS Currie 200?).
[BTW what's happened to the ADASS 2004 Proceedings?]

> Right now I can't trust PSF to give me a decent answer. Is PSF broken
> in some way, or is this just down to the different methods of
> determining ellipticity?

I think it's the latter, but the unseen 0.254 example is a worry.

Malcolm