‡ RCI is the relative c-index for the image calculated versus only the other images in the same top-level gallery. For example, images in "Photography->Castles" and "New Images->Photography->Flowers" are in the same top-level gallery "Photography".
Sounds like the RCI calculates image rankings on the basis of image types/categories so that, if the image is a Photograph, it's ratings are statistically compared only to other images in the Photography category, rather than to images in all categories. So, if you're a fractal artist, your fractal scores are compared only to the scores of other fractals and your RCI would appear to be a RANKING of your image relative to the ranking of other images of the same type. I'd be interested in knowing how it's actually calculated. Wen
That's pretty close Wen. I think that a strict ranking system would go 1,2,3... This one is a bell curve (Gaussian) with the same shape as the full c-index distribution. That way our normal understanding of what constitutes a "good" score is the same for both CI and RCI.
Makes sense, but I have a sincere question about the decision to use a normal bell curve as a distribution for images on this site, Caedes. I know it's been touched on before, but perhaps you can make it clearer to me. It seems to me that a normal distribution would represent the gamut of artistic talent and skill - - from very poor to extremely good. My question concerns the generally high quality of the bulk of images here as compared to a random population of artists and non-artists. It would seem to me that the curve for the population of those who post here would be significantly skewed to the right...meaning that there would be more scores above 5 than below. Do you really think that forcing scores here into "normal distribution" accurately represents the quality of images on this site? Wen
It is an arbitrary decision: just as arbitrary as if I were to choose it to be at a higher value or with a different distribution. The point is that you have to choose something, it is just a matter of definition. My choice of a Gaussian distribution is based on my understanding of the statistics of large random samples.
Can someone explain why, without a single exception across a couple hundred images, the difference between the indexes for my images is always 3. The RCI‡ is always lower.
If your images are all photographs, then the transform used to turn your c-index into the RCI is going to be the same for all of them. They will all move the same amount if your images all score about the same. For instance, two photos which score 75 and 80 will move down about the same amount, however two fractals which score the same c-index will tend to move up the same amount (but a different amount than the photos moved down).