pminimax

I wouldn't worry about one datapoint out of 35,000 messing up the stats.

Author here. I added fields so you can specify your first language (relevant link: https://en.wikipedia.org/wiki/Blue%E2%80%93green_distinction...) and colorblindness.

FAQ:

* I can't know your monitor's calibration, your ambient light, or your phone's brightness. Obviously, this will affect the results. However, I am tracking local time of day and device type, from which we should be able to infer whether night mode and default calibration has any aggregate effects. Anecdotally, thus far, I haven't found any effects of Android vs. iPhone (N=34,000).

* The order is randomized. Where you start from can influence the outcome, but methodologically it's better to randomize so the aggregate results average over starting point. You can run the test several times to see how reliable this is for you.

* It's common practice in psychophysics to use two alternatives rather than three (e.g. blue, green, something in the middle). It would be a fun extension, which you can handle with an ordered logistic regression. The code is open if you want to take a shot at it: https://github.com/patrickmineault/ismyblue

* I will release aggregate results on my blog, https://neuroai.science

* I am aware of most of the limitations of this test. I have run psychophysics experiments in a lab on calibrated CRTs during my PhD in visual neuroscience. *This is just entertainment*. I did this project to see if I could make a fun webapp in Vue.js using Claude Sonnet, and later cursor, given that I am not highly proficient in modern webdev. A secondary point was to engage people in vision science and get them to talk and think about perception and language. I think it worked!

hsl

The mask is 200 ms long, which is a bit on the long side compared to most psychophysics experiments. I can try to crank it up to 300 ms, but beyond that I think it'll start feeling slow.

The site records local time of day when you hit submit so I can track whether this has any effect. I have 7,000 answers thus far, I should have enough by tomorrow to determine whether there are any systematic effects.

That's fun! I bet people would tend to nudge the threshold toward the middle of the scale. Or you could do a sorting interface, etc.

You can fork this here: https://github.com/patrickmineault/ismyblue

People have done this. See, e.g. Brouwer and Heeger (2009), Decoding and Reconstructing Color from Responses in Human Visual Cortex.

Yes, you can absolutely get better at discriminating different colors, orientations, etc. though unfortunately improvements tend to be highly specific to the stimulus. There's a great book by Barbara Dosher called Perceptual Learning that extensively overviews the literature.

The order is randomized. Hit reset and you'll get a different sequence. The sequence is also adaptive (not a binary search---it's hitting specific points of the tail of a sigmoid in a logistic regression it's building as you go along). Try it a few times and you'll see how reproducible it is for you.

It of course depends on the calibration of your monitor. One of the reasons I did this project is I wanted to see if there were systematic differences in color names and balance in the wild, for example, by device type (desktop vs. Android vs. iPhone), time of day (night mode), country (Sapir-Whorf), etc.

It is common practice in psychometrics to use two levels in a forced choice and model responses as a logistic regression, which is what's done here. Adding an N/A option turns the thing into an ordered logistic regression with unknown levels, which is tricky to fit, but it's possible. Having done a lot of psychophysics, having more options generally doesn't make the task easier.

Author here, yes, it tests a mix of your monitor calibration and colour naming. The two types of inferences you can make with this are:

1. If two people take the test with the same device, in the same lighting (e.g. in the same room), their relative thresholds should be fairly stable. 2. If you average over large populations, you can estimate population thresholds, marginalizing over monitor calibrations.

The most interesting thing for me is that while cyan (#00ffff) is nominally halfway between blue and green, most people's thresholds, averaged over monitor calibrations, imply that cyan is classified as blue. I was not expecting that the median threshold (hue 174) would be so deep into the greens.

Author here, it started out with:

``` Can you help me make a website called is my blue your blue? I want to make a website that is in vue.js that allows one to determine the boundary between their perception of blue vs. green. It should use a golden ratio search to find the midpoint between blue and green. It should have the color be the color of the background, and it should have two buttons, blue and green. If they pick green, you should show something bluer, and vice versa. ```

It offered a starter with vue and tailwind, then I asked to add a supabase backend. I took maybe 5 hours to get the original version, which I tweaked until I got about 800 initial responses so I could show a population curve. Later I modded it with cursor to add an about section, fit a proper GLM rather than a simple golden ratio estimation method, and the d3 animation at the end.