Here are the recent numbers for Massachusetts.
The columns are: date, total confirmed cases to that date, increase factor of total confirmed cases for the day, and new confirmed cases for the day.
"S" indicates Saturdays and Sundays, which might affect the timing of reporting and thus shift the numbers of cases between days.
"*" and "+" indicate a new all-time high for new confirmed cases on that day.
Code:
3/21 525 1.27 112 S
3/22 646 1.23 121 S
3/23 777 1.20 131
3/24 1159 1.49 382 + <--- business closures / lockdown
3/25 1838 1.59 679 +
[COLOR="SeaGreen"]3/26 2417 1.32 579
3/27 3240 1.34 823 +
3/28 4257 1.31 1017 * S[/COLOR]
[COLOR="RoyalBlue"]3/29 4955 1.16 698 S
3/30 5752 1.16 797
3/31 6620 1.15 868
4/01 7738 1.17 1118 +
4/02 8966 1.16 1228 +
4/03 10402 1.16 1436 *[/COLOR]
4/04 11736 1.13 1334 S
4/05 12500 1.07 764 S
[COLOR="DarkRed"]4/06 13837 1.11 1337
4/07 15202 1.10 1365
4/08 16790 1.10 1588 +
4/09 18941 1.13 2151 *
4/10 20974 1.11 2033
4/11 22860 1.09 1886 S
4/12 25475 1.11 2615 * S[/COLOR]
[COLOR="SeaGreen"]4/13 26867 1.05 1392 <--- 1.5 hr. change in data closeout time
4/14 28163 1.04 1296
4/15 29918 1.06 1755[/COLOR]
The interesting and unexpected behavior in this data is pointed out by the coloration, showing stable periods where the rate of increase of cases is roughly constant (which causes an exponential rise of new confirmed cases per day at that rate, so as to eventually exceed any previous maxima) separated by rapid "step-downs" after which the rate of increase drops to a new lower range and the new cases per day, at least at first, decrease sharply.
I'll consider the overall "leveling off" to have occurred when the overall trend in the cases per day is no longer increasing, taking the rough day to day variations into account. Or equivalently, when "step-downs" occur before rather than after new highs are set within a current regime.
The step-down pattern occurs consistently in each of the largest and medium sized individual MA counties, as well and nearly as clearly as in the state aggregate. The main differences between counties are that less populous counties and also wealthier counties have consistently shown slightly smaller growth rates during a given stable period. The differences are a matter of plus or minus one percentage point from the aggregate state figure, which made little difference early on when the growth rates everywhere were above 10% (1.10), but in recent days that same spread in percentages results in the least populous counties having likely leveled off a few days ago (their data is of course relatively noisier), and medium-density counties might be leveling off now, while the most populous counties are still rising noticeably day to day (but not yet exceeding previous highs) during the current "green" period.
I'm not really sure how to understand these phenomena. The measures being taken haven't really changed since March 24th, with the exception of increased usage of (mostly homemade) face masks during the most recent week. It's understandable that cases contracted before the lockdown would take a week or more (average 5-6 days) to become confirmed, so at first the increase rate would continue to be high. Which the data shows, for the early black and green periods. But why the later step-downs weeks afterward? That is to say, why do they happen at all, but not happen sooner?
One way to model it would be to assume a large number of asymptomatic carriers with a long statistical tail of durations before they either become symptomatic or recover completely, and that pool has to dwindle away before the new cases can level off. But in that case, why wouldn't the greater number of new cases occurring during those weeks make the asymptomatic pool even larger? (It's been true all along, and still true, that only people with symptoms have been getting tested. We're not really doing contact tracing yet either.) And, research so far doesn't suggest that the number of asymptomatic cases is a very high ratio.
Another possible model is that herd immunity is taking over, within a small sub-population of "behaviorally vulnerable" people -- which would have two main components: careless people, and people still working in public-facing "essential" jobs.
The epidemiologists who make forecasts must have some other hidden variable in mind, because even though their predictions haven't been very accurate (e.g. about specific dates when peaks will occur), they've been consistent with the general pattern we're seeing everywhere -- that is, that cases will level off at a peak relatively soon, rather than increasing at some percentage per day until herd immunity kicks in.
