On 2016 Dec 04, Lydia Maniatis commented:

Let's say we ask the proverbial “man in the street” the following question: Do you think chickens will be better at discriminating between the colors of tiled food containers if the tiles are many or large, or if they are few or small?

I think that most, without too much thought, would answer the former. “More and bigger” of anything is generally more salient then “Fewer and smaller.” Would those who made correct guesses be licensed to claim that their pet theory about chicken vision had been corroborated? It should be obvious that predictable predictions do not constitute rigorous tests of any hypothesis. This is the type of hypothesis-testing Olsson et al (2017) engage in in this study.

Furthermore, the hypothesis that the authors are supposed to be testing doesn’t consist of a straightforward, coherent, intelligible set of assumptions, but of a hodgepodge of uncorroborated assumptions and models spanning over fifty years. The “success” of the authors’ simple experiment implies corroboration of all of these subsidiary models and assumptions. Obviously, the experiments are being tasked with far too much heavy-lifting, and the conclusions that hinge on them are not credible.

Here is a sampling of the assumptions and models that chickens greater sensitivity to “more and bigger” are presumed to corroborate:

The main hypothesis: “Chickens use spatial summation to maintain color discrimination in low light intensities.”

Supporting models/assumptions: “Color differences delta S in the unit of just-noticeable differences (JND) were calculated using the receptor noise limited (RNL) model (Vorobyev and Osorio, 1998) as….” I.e. the RNL model is assumed to be valid.

“Spectral sensitivities, R, were derived by fitting a template (Govardovskii, Fyhrquist, Reuter, Kuzmin, & Donne, 2000)…” I.e. the model template is assumed to be valid.

“We assumed the same standard deviation of noise for all cone types such that the Weber fraction for the L channel was 0.06, based on the color discrimination thresholds measured in a previous study (Olsson et al 2015)” Note that, according to a Pubmed Commons comment by the lead author, Olsson et al (2015) “figure out the equivalent Weber fraction which describe these limits. Whether that is actually caused by noise or not we can not say from our experiment..." Yet the main hypothesis of Olsson et al (2017) uncritically assumes a “noisy” process.

“The same simple model (SM) of calculating the absolute quantum catch as in a previous study (Olsson et al 2015)” Again, the authors cannot say that the results of that previous study were “actually caused by noise or not”, i.e that the “simple model” is actually modeling what they are claiming.

“We modeled increasing levels of spatial summation, assuming that absolute quantum catches….are summed linearly…” Should we even ask why?

“From ….cone densities in the dorso-temporal retina of chickens (Kram et al, 2010) we estimated the number of cones that viewed a single color tile of a stimulus.” This last assumption obviously doesn’t consider the fact of chicken eye movements, which would make the number of cones involved much larger. The idea of simple pooling is also problematic from the point of view that chickens do exhibit constancy under varying illumination, so in the context of sunshine and shadow, pooling across an illumination boundary would arguably produce unreliable estimates that would undermine constancy.

“We derived intensity thresholds by fitting a logistic psychometric function to the choice data of each experimental group of chickens and individual chickes using the Matlab toolbox palamedes (Prins & Kingdom, 2009).” We assume that Prins and Kingdom’s hypothesized quantitative link between choice and thresholds, as well as all of those authors’ underlying assumptions, e.g. that signal detection theory is an appropriate model for vision, are valid.

I would note, finally, that the authors’ current description of the findings of Olsson, Lind and Kelber (2015) differs significantly from those implied by the title of the latter publication (“Bird color vision: behavioral thresholds reveal receptor noise”). As mentioned above, the lead author of that study has acknowledged that the title goes further than was licensed by experiment. Here, the Olsson et al (2015) study is described as having shown that “the intensity threshold for color discrimination in chickens depends on the chromatic contrast between the stimuli and on stimulus brightness.” This result, i.e. that “higher contrast, brighter = more salient”, is, if anything, even more predictable than the prediction of Olson et al (2017).

On 2016 Dec 04, Lydia Maniatis commented:

Let's say we ask the proverbial “man in the street” the following question: Do you think chickens will be better at discriminating between the colors of tiled food containers if the tiles are many or large, or if they are few or small?

I think that most, without too much thought, would answer the former. “More and bigger” of anything is generally more salient then “Fewer and smaller.” Would those who made correct guesses be licensed to claim that their pet theory about chicken vision had been corroborated? It should be obvious that predictable predictions do not constitute rigorous tests of any hypothesis. This is the type of hypothesis-testing Olsson et al (2017) engage in in this study.

Furthermore, the hypothesis that the authors are supposed to be testing doesn’t consist of a straightforward, coherent, intelligible set of assumptions, but of a hodgepodge of uncorroborated assumptions and models spanning over fifty years. The “success” of the authors’ simple experiment implies corroboration of all of these subsidiary models and assumptions. Obviously, the experiments are being tasked with far too much heavy-lifting, and the conclusions that hinge on them are not credible.

Here is a sampling of the assumptions and models that chickens greater sensitivity to “more and bigger” are presumed to corroborate:

The main hypothesis: “Chickens use spatial summation to maintain color discrimination in low light intensities.”

Supporting models/assumptions: “Color differences delta S in the unit of just-noticeable differences (JND) were calculated using the receptor noise limited (RNL) model (Vorobyev and Osorio, 1998) as….” I.e. the RNL model is assumed to be valid.

“Spectral sensitivities, R, were derived by fitting a template (Govardovskii, Fyhrquist, Reuter, Kuzmin, & Donne, 2000)…” I.e. the model template is assumed to be valid.

“We assumed the same standard deviation of noise for all cone types such that the Weber fraction for the L channel was 0.06, based on the color discrimination thresholds measured in a previous study (Olsson et al 2015)” Note that, according to a Pubmed Commons comment by the lead author, Olsson et al (2015) “figure out the equivalent Weber fraction which describe these limits. Whether that is actually caused by noise or not we can not say from our experiment..." Yet the main hypothesis of Olsson et al (2017) uncritically assumes a “noisy” process.

“The same simple model (SM) of calculating the absolute quantum catch as in a previous study (Olsson et al 2015)” Again, the authors cannot say that the results of that previous study were “actually caused by noise or not”, i.e that the “simple model” is actually modeling what they are claiming.

“We modeled increasing levels of spatial summation, assuming that absolute quantum catches….are summed linearly…” Should we even ask why?

“From ….cone densities in the dorso-temporal retina of chickens (Kram et al, 2010) we estimated the number of cones that viewed a single color tile of a stimulus.” This last assumption obviously doesn’t consider the fact of chicken eye movements, which would make the number of cones involved much larger. The idea of simple pooling is also problematic from the point of view that chickens do exhibit constancy under varying illumination, so in the context of sunshine and shadow, pooling across an illumination boundary would arguably produce unreliable estimates that would undermine constancy.

“We derived intensity thresholds by fitting a logistic psychometric function to the choice data of each experimental group of chickens and individual chickes using the Matlab toolbox palamedes (Prins & Kingdom, 2009).” We assume that Prins and Kingdom’s hypothesized quantitative link between choice and thresholds, as well as all of those authors’ underlying assumptions, e.g. that signal detection theory is an appropriate model for vision, are valid.

I would note, finally, that the authors’ current description of the findings of Olsson, Lind and Kelber (2015) differs significantly from those implied by the title of the latter publication (“Bird color vision: behavioral thresholds reveal receptor noise”). As mentioned above, the lead author of that study has acknowledged that the title goes further than was licensed by experiment. Here, the Olsson et al (2015) study is described as having shown that “the intensity threshold for color discrimination in chickens depends on the chromatic contrast between the stimuli and on stimulus brightness.” This result, i.e. that “higher contrast, brighter = more salient”, is, if anything, even more predictable than the prediction of Olson et al (2017).