i don’t think normal folks have access to get that level of information effectively. via /v1/observations in the API, we have the ability to use an identified=true or =false parameter, which can help distinguish between observations with 0 identifications and with >0 identifications. theoretically, you could use some combination of identifications=some_agree, =most_agree, and =most_disagree to further separate observations with >1 identifications.
if you just want to get a rough idea of percents in each of your categories, /v1/observations does provide an identification count per observation, and you can do a random sample of observations.
here’s the distribution for a random sample of n=2000 that i just pulled:
| # of ids | # of obs | % of total |
|---|---|---|
| 0 | 49 | 2.5% |
| 1 | 554 | 27.7% |
| 2 | 872 | 43.6% |
| 3 | 350 | 17.5% |
| 4 | 113 | 5.7% |
| 5 | 37 | 1.9% |
| 6 | 20 | 1.0% |
| 7 | 4 | 0.2% |
| 8 | 1 | 0.1% |
| total | 2000 | 100.0% |
just to see if the above numbers are reasonable, here are the actual counts for identified=true and =false:
identifications=false at 2.6% roughly matches 0 identifications at 2.5% percent. so that seems to indicate that the figures in the first table should be reasonable.