17 PSF size vs magnitude

I choose the exposures from g,r,i,z,Y bands from 12/27/2012 and measure the M20 using weighted moments (kernel size of 2 pix) and then corrected for the weights, as well as the non-weighted moments from the sextractor output. The magnitude is from my run of sextractor, which is not corrected for the zero points. The red dots and error bars are the mean and standard deviation (not to the mean) of each bin. The plots are made from the selected stars from all CCDs. Clearly, the there is a magnitude dependence if one use the non-weighted moments.

In terms of the peak value of the pixels of the stars: