Then that just means that we are likely underestimating R0--possibly by a huge amount. Okay, I agree with you.
Infections aren't LESS than the diagnosed cases and positive test results. So, all the calculations from data about the estimates of R0 are all a lower bound, not an upper one.
As such, our estimates of R0 are likely painting a much rosier picture than reality. Probably not the point you were trying to make.
If we somehow magically started accounting for those cases the R0 would immediately go up (more cases per unit time means increased R0) unless the number of those cases is zero.
Thus, the R0 estimates we have are currently all underestimates. Covid is likely more contagious than we are calculating, not less.
Now, this doesn't matter to the people who recover, of course. And it effectively is correct in that people who simply recover aren't a big problem and so the R0 is roughly "correct" in the aggregate.
However, that incorrect R0 DOES matter when you start talking about vulnerable populations and how far you need to go to protect them.
I'm going to share some quotes from the code.
> Our goal is to translate positive test counts to the dates where they likely occured. Since we have the distribution, we can distribute case counts back in time according to that distribution. To accomplish this, we reverse the case time series, and convolve it using the distribution of delay from onset to confirmation. Then we reverse the series again to obtain the onset curve.
> We assume a poisson likelihood function and feed it what we believe is the onset curve based on reported data. We model this onset curve based on the same math in the previous notebook
> We calculate this by figuring out how many cases we'd expect there to be yesterday total when adjusted for bias and plugging it into the first equation above. We then have to re-bias this number back down to get the expected amount of onset cases observed that day.
I don't want to invalidate efforts to build models because models are important, but it's also important to frame models as being statistical models and in this attempting to resolve R0 values from self-reported data, across districts with dramatically altered reporting methodology is extraordinarily fraught with peril. Even the model authors indicate this and have made several improvements, and models will naturally evolve over time.
Then let's touch on some of the stats we're looking at. Remember we are dealing with a sampled phenomenon - as our sampling rate changes (the amount of testing we're doing), our case rate will naturally rise. We want more testing because it helps us determine more accurate numbers, but it also means we need to understand we may see spikes in our daily reported cases. We therefore want to be careful with using small samples (3 or 4 days).
You'll also notice we have natural declines that align with the weekends - it's not because the disease takes the weekends off. One of the dangers of reading the worldometers graphs is that they're inclusive of all the data, including data from March/April which is probably underestimate.
Texas started lessening quarantine restrictions on May 16th, and indeed until about the 29th we see a pretty consistent decline (well past any incubation time). From the 30th to today's data (and always take the current day with a grain of salt because reporting isn't always instant) we do see a growth, primarily due to several abnormal spikes in the last week. When we compare that against deaths we don't see those same spikes represented - and in the last two days of the series (excluding today) we see a declining pattern again (but two days a pattern does not make).
So it's wrong to say "Texas has about 1500 new cases per day now. They've had a steady increase in new cases per day since mid April, when they were about 750 new pos cases/day. They are really growing.". The 7 day average keeps us at 1200 or less, with the exception of the last few days.
I pulled the hospitalization data from [^2] (xlsx format unfortunately) and while hospitalizations are technically increasing as a whole, that is almost entirely in the TSA of Houston (though Dallas/Ft Worth and Longview/Tyler show some increase as well) with 19 other TSAs declining or remaining steady. So, I wouldn't be overly concerned for your family, the data indicates things are generally going quite well.
I would be careful in how we read data.
They do have per state R values. This does show that NY is at about 0.8. Texas is 1.02 currently. The one negative with their calcs is their R values seem too sensitive to daily conditions. It's a great visualization of infection rates too, please look at that.
To look at the daily infection numbers, find Texas, click on the "show new cases" tab. Texas has about 1500 new cases per day now. They've had a steady increase in new cases per day since mid April, when they were about 750 new pos cases/day. They are really growing. I'm not sure where your stats about Texas doing well are. It looks to me like the positive cases are unfortunately increasing there steadily, and it's a precursor to increasing hospitilization. I have family there, I'm increasingly concerned.
They get their data feed from https://covidtracking.com/api/v1/states/daily.csv which is downloadable. Look at TX (some days have multiple updates, you have to calculate the delta yourself). US date style, positives: 6/7 74978, 6/6 73553, 6/5 71613, the increases are very clear in daily positive cases; those counts are total pos counts to that date of course.
Another place to see per state (such as Texas) is https://www.worldometers.info/coronavirus/usa/texas/. Go down to 'daily new cases in texas' click on 7 day moving average. They've really been growing per day in new positive cases. One note, different websites show the current day's data differently (some show values that up dated in the middle of the day, others show one day late, so looking at the very latest rhs data point is sometimes confusing).
Texas absolutely shows the potential of opening up increasing infection rates. Alabama is another state that is hurting but reached a top, see Arkansas too. Florida was down but is now showing highest per day infection rates historically (7 day average, looks like new records beating 2 months ago). Some states are better, plenty are getting worse.