I have yet to see anybody to a backwards looking analysis on covid related to the data we have now. It's obvious from that chart that the initial blue spike should have been 10-50 times higher, but wasn't due to testing.
Many previous models and predictions used relatively naive models of social contact that didn't take real-world social network effects into account. The node degrees tend to be tail-heavy (a few people who have a lot of contacts), the nodes tend to be heavily clustered (groups of nodes that have lots of contacts within said group but relatively few outside said group), the graph tends to have a surprisingly small diameter ('6 degrees of separation' - it takes surprisingly few hops to get from any one person to another), etc, etc.
So, what do you get when you combine this sort of network with disease models? Interesting behavior. The disease very quickly spreads to the most social people, and you get exponential growth for a while. But the interesting thing is that the disease ends up selectively pruning precisely those nodes that would help it spread between cliques. Which means that R goes down much faster than expected given the overall percentage of people infected, and often ends up hovering around a linear regime after the initial phase.
It would be interesting (although I doubt we'll ever see it) to study and see the responses to 'how many people have you had contact with in the past week' (or somesuch) for people infected with COVID over time. My suspicion would be that you'd see a similar 'rise near the start as it climbs the popularity ladder, then decaying over time' effect.
I have yet to see anybody to a backwards looking analysis on covid related to the data we have now. It's obvious from that chart that the initial blue spike should have been 10-50 times higher, but wasn't due to testing.
The interesting thing is, that's arguable.
Many previous models and predictions used relatively naive models of social contact that didn't take real-world social network effects into account. The node degrees tend to be tail-heavy (a few people who have a lot of contacts), the nodes tend to be heavily clustered (groups of nodes that have lots of contacts within said group but relatively few outside said group), the graph tends to have a surprisingly small diameter ('6 degrees of separation' - it takes surprisingly few hops to get from any one person to another), etc, etc.
So, what do you get when you combine this sort of network with disease models? Interesting behavior. The disease very quickly spreads to the most social people, and you get exponential growth for a while. But the interesting thing is that the disease ends up selectively pruning precisely those nodes that would help it spread between cliques. Which means that R goes down much faster than expected given the overall percentage of people infected, and often ends up hovering around a linear regime after the initial phase.
It would be interesting (although I doubt we'll ever see it) to study and see the responses to 'how many people have you had contact with in the past week' (or somesuch) for people infected with COVID over time. My suspicion would be that you'd see a similar 'rise near the start as it climbs the popularity ladder, then decaying over time' effect.