In progressive joint source-channel coding of images, experiments show that optimizing rate allocation according to PSNR criteria and according to rate criteria produce essentially similar results ? a very important observation because optimization by rate is often much easier. This paper presents analysis to uncover the reason for ? and the generality of ? these empirical results. We first examine the typical shape of the distribution of PSNR in light of the distribution of the first uncorrectable error in the bitstream. Then we analyze the sensitivity of rate allocation to the operational R-D curve, via parametric modeling of the source and the channel. We demonstrate that rate allocation is locally insensitive to variations in the R-D characteristics. This analysis provides a justification for lowcomplexity rate allocation algorithms, such as the one by Chande and Farvardin, that work independently of the operational R-D curve.