Wes the CEO of Dotty's Hot or Not app here. Thank you Ethan & IPVM for the article!
One reason readings could be low is there is a sliding window crowd average to compensate for environmental conditions like a cold or hot airplane affecting the entire crowd. We assume the majority of a crowd is healthy and elevated temps are low probability outliers.
If you push too many 99-100F faces into the statistical machine as part of augmented elevated temperature testing it will push the following face temps to read low for a few normal readings until it re-stabilises around normal body temp values. Any temperature above 100F is excluded from the crowd source data considered as true temperature outliers and don't affect the running average.
Not saying this is the reason the values were low, but sharing for deeper understanding of the science and the challenges of artificially inducing temperatures in rapid succession for testing. It's also hard to get a normal crowd sourced average with only a handful of unique faces. More is better in the case of our system.
Thanks again guys for taking the time to dig into our systems, keep up the good work.
Hi Wes, I think your approach is correct. In regard to the second paragraph: 'If you push too many...'. What does this mean to the user? How many is too many? How does a user know that too many range-bound temps were pushed and the system is entering a re-stabilization mode? Given most temps will be normal range bound is this expected to happen at the start of each day? It seems that the system should identify to the user that it's in re-stabilization mode so the user can handle readings taken in this mode or maybe the system stops taking measurements during re-stabilization. Can the system remain stable in high traffic situations? What interarrival rate is supported?
I've borrowed one of IPVMs graphics from an article to help show the latest normal body temp ranges. Anything above 100F we don't count in the statistical machine, we see them as outliers.
Our system relies on the subjects being part of the normal healthy population to actually look like this graph over time (more or less), and then outliers are the potential fevers. The issue when testing for fevers over and over with a hot water bottle or a hair dryer on the face is you may be cooling down from 100.4F through to 99F and that has the normal distribution higher than in reality. Also only 3 or 4 different faces won't create this normal distribution very well. You need a stream of different people.
The more throughput for our system the better. It's just augmented temperatures during testing that needs to be understood, in the real world that's not an issue.
wouldn’t be the first time we’ve added a +1 to our code just because :)
But we actually normalise around 97.5F based on a global medical study. This Stanford medical literature shows a normal distribution at 98F for North America, so we could take it up half a notch for North America. We’ll consider that.