Covid-19 in OKC (coronavirus)

[LocoAko]

This is an interesting graph. I noticed it last night but wasn’t sure that I could trust it. It actually shows our total cases have peaked and are actually dropping. Is that correct? Interesting if true.

I find these charts highly misleading. Results frequently take 5-10 days to come in, and this plots things as a function of symptom onset. So, given a) the delay in getting and reporting test results and b) the fact that it takes symptoms a variable number of days to come on, it will always be the case that the peak in symptom onset was a number of days ago with much fewer cases of reported symptom onset within the recent past. That chart has always bent sharply downward no matter when you look at it, with the data for a recent date "filling in" as time goes on. Even the doctors at the OKC City Council meeting called it potentially misleading, although it has value for epidemiological reasons, of course.

[Pete]

^

For that graph to be valuable, you need to chop off at least the last two weeks given the lag in testing and other factors.

[OKC Talker]

I find these charts highly misleading. Results frequently take 5-10 days to come in, and this plots things as a function of symptom onset. So, given a) the delay in getting and reporting test results and b) the fact that it takes symptoms a variable number of days to come on, it will always be the case that the peak in symptom onset was a number of days ago with much fewer cases of reported symptom onset within the recent past. That chart has always bent sharply downward no matter when you look at it, with the data for a recent date "filling in" as time goes on. Even the doctors at the OKC City Council meeting called it potentially misleading, although it has value for epidemiological reasons, of course.

The value of the chart is that it's accurate data that is updated and backdated as cases are reported. Politicians don't like it because they can't spin day by day variability to their benefit, but that's not the point of the data. We need to see accurate information to make decisions.

[LocoAko]

The value of the chart is that it's accurate data that is updated and backdated as cases are reported. Politicians don't like it because they can't spin day by day variability to their benefit, but that's not the point of the data. We need to see accurate information to make decisions.

I can assure you I've looked at this exact chart for weeks and months now and it has always been the case that it bends sharply downward in the last week or so because of the aforementioned reasons. This is not political bias, it's an accurate understanding of the data being presented.

But go off, I guess.

[pw405]

These large endpoints, like today’s numbers, can really negatively affect trend analysis. Any derivative you calculate including today’s numbers will have a very large affect. I know the common way to analyze this data has been to apply a 7 day running average which smoothes the curve; however, our tests are delayed up to 10 days in some cases. I wonder if a 10 day running average would be better.

Even better than running averages, I wonder what the data would look like if you binned it instead. My suggestion is to pick a day. Any day. Doesn’t matter. I propose Friday’s as the numbers decrease over the weekend and are captured in Tuesday’s. Take Friday, and sum all the numbers for the previous week. So you’ll bin the numbers by week. Week 1, week 2, etc. Plot that curve. I bet it will look nicer than the 7 day running average.

Didn't try binning, need to read instructions first. But... did a combined plot of 7 day, 14 day, and 21 day case averages. It does tend to iron out some of the lumpiness with recent volatility. I thought about doing a 10-day trend, but sticking with multiples of 7 allows equal # of variances due to the "weekly" nature of the data. (Low Sun/Mon, high Tues.)




If you look at the Day over Day (DoD)% Change of the 7-day case average, and fit to a logarithmic trend (Excel), it HAS trended downward. However... I don't really like this measure. I was just curious to see what it would look like over time.

It simply calculates New-Old / Old where New = Today and Old = Yesterday. While this could be a valid measure on a per-day basis, trending it over time simply tells us: "Today's cases changed X% from yesterday's cases". It is also sensitive to the irregularities in daily reporting.

Still, a consistent long term value of this measure below zero would be a very good sign.

[OKC Talker]

I can assure you I've looked at this exact chart for weeks and months now and it has always been the case that it bends sharply downward in the last week or so because of the aforementioned reasons. This is not political bias, it's an accurate understanding of the data being presented.

But go off, I guess.

I don't think you're understanding the graph. Of course there will be fewer people reporting that they have symptoms of Covid-19 starting yesterday than a week ago. As those cases are reported in the following week though, yesterday's number is changed to reflect the new cases. That's how it can show us accurate data and trends.

Edit:
I apologize if my comment about politicians offended you but people have been cherry picking low (and high) daily case reports to further their narrative and it's frustrating.

[Pete]

I don't think you're understanding the graph. Of course there will be fewer people reporting that they have symptoms of Covid-19 starting yesterday than a week ago. As those cases are reported in the follow week though, yesterday's number is changed to reflect the new cases. That's how it can show us accurate data and trends.

But not until at least two weeks have passed for every data point on that graph.

Which also means the last two weeks are always going to be way under-reported.

[pw405]

But not until at least two weeks have passed for every data point on that graph.

Which also means the last two weeks are always going to be way under-reported.

Excellent visual way to demonstrate this problem. The same problem occurs with deaths and death reporting. They very often do not occur the day they are reported by the state, making it always appear they are "going down":

Screenshot from:
https://www.docdroid.net/snNC3wY/cov...sa-pdf#page=10

[C_M_25]

Dang! You guys are stepping up your game. I’m now watching the winter soldier with my son, so I’m not going to comment much until I can take a deeper look. Great work and great conversation!!

[OKC Talker]

But not until at least two weeks have passed for every data point on that graph.

Which also means the last two weeks are always going to be way under-reported.

The purpose of the graph is to report what day symptoms for known cases were reported, which can more accurately identify what day the person was infected (a week or so earlier) and understand the spread. If it helps though, yes you can just chop off the last week of data and pretend that number is the number from today's executive report. It will likely still give you a more accurate idea of what's going on than trying to manually account for reporting delays and other variability.

[Pete]

The purpose of the graph is to report what day symptoms for known cases were reported, which can more accurately identify what day the person was infected (a week or so earlier) and understand the spread. If it helps though, yes you can just chop off the last week of data and pretend that number is the number from today's executive report. It will likely still give you a more accurate idea of what's going on than trying to manually account for testing delays and other variability.

Fair enough!

But really, the last two weeks are always going to be incomplete because there are plenty of reports of testing taking 10 days or more.

[C_M_25]

Here are the weekly totals (defined as Monday-Sunday) since the start of this thing, up through and including today. Data is from CovidTracking, which seemed to be the most reliable data I could find (not sure how they're handling the weird multi-day report compared to other sites, though).

Attachment 16281

Derivatives on data like this will be quite sensitive to how you define the width of the bin and I didn't want to bother with the assumptions involved with fitting polynomials to the data, but summing the data to be weekly should help negate the day-to-day biases. In this case, here is the change in weekly totals:

Attachment 16282

So, there's some movement, but nothing I would call an obvious or substantial slowing down, yet.

Finally, here is our weekly change in cases normalized by the total number of cases (which is less meaningful than it appears given "recoveries", but I digress). Other than the large numbers early on due to the very low number of existing cases, our rate the last month or so has been remarkably steady at about 0.19 (that is, we're adding about 20% of our total cases week-over-week).

Attachment 16283

Sorry for the poor graph quality. Blame Google Sheets, lol.

Great analysis. On that last plot, if you max out your y-axis to something about half of what it is, I bet we could see more variability in recent weeks. There’s so much noise in this data that it’s really hard to make any detailed analysis without fine tuning the bin size for the derivative calculations.

[soonerguru]

Dale Bratzler of OU weighs in:

https://m.facebook.com/story.php?sto...00000551045034

[C_M_25]

Didn't try binning, need to read instructions first. But... did a combined plot of 7 day, 14 day, and 21 day case averages. It does tend to iron out some of the lumpiness with recent volatility. I thought about doing a 10-day trend, but sticking with multiples of 7 allows equal # of variances due to the "weekly" nature of the data. (Low Sun/Mon, high Tues.)




If you look at the Day over Day (DoD)% Change of the 7-day case average, and fit to a logarithmic trend (Excel), it HAS trended downward. However... I don't really like this measure. I was just curious to see what it would look like over time.

It simply calculates New-Old / Old where New = Today and Old = Yesterday. While this could be a valid measure on a per-day basis, trending it over time simply tells us: "Today's cases changed X% from yesterday's cases". It is also sensitive to the irregularities in daily reporting.

Still, a consistent long term value of this measure below zero would be a very good sign.

Really great plots. I would like to see this going back a couple of months to remove any bias from recent misreported data, but I understand if you don’t have the time. Btw, what software are you using?

Edit: I would be curious to see how the second graph looks in April/May when our cases were declining before ramping up. That may be a good place to ground truth the interpretation.

[soonerguru]

Lol at everyone ignoring Bratzler’s comments about the data. Says a lot.

[OKC Talker]

Lol at everyone ignoring Bratzler’s comments about the data. Says a lot.

Sorry sooner, there was no exerpt and I usually don't click on links in this thread because there's no telling where you might end up! =P

Which part of Dr Bratzler's comments were you referring to? The following jumped out at me with all the efforts people are taking to "correct" the daily reports and my crusade to point out that it's already being done for us:

"I am often asked to comment on whether some event (Memorial Day, Rally, Protests, July 4) could be linked to the new spike in cases. Honestly, without knowing when the tests are being done, it is exceedingly difficult to correlate events with spikes. We know clearly now that there are often lengthy delays between when a test is done on a person, and when those results may be reported as negative or a new case by OSDH in their daily reports."

[brian72]

https://www.news9.com/story/5f1df58e...due-to-covid19. Sad situation all the way around.

[RustytheBailiff]

How to stop Covid-19:


https://threadreaderapp.com/thread/1...499965441.html

It's really not that hard, with the right leadership.....



STAY SAFE WEAR MASKS

[RustytheBailiff]

https://www.news9.com/story/5f1df58e...due-to-covid19. Sad situation all the way around.

The solution to any real or perceived mental health crisis is to control the spread of the virus. The solution to the very real economic and employment collapses is to control the virus. Controlling and largely eradicating the virus is the only way our country will return to a semblance of normality.



STAY SAFE WEAR MASKS

[bsmall]

Lol at everyone ignoring Bratzler’s comments about the data. Says a lot.

You posted a facebook link at 9pm, and you're upset that people hadn't replied to your post by 2am. How many people do you think are reading this thread between those hours?

[TheTravellers]

Lol at everyone ignoring Bratzler’s comments about the data. Says a lot.

Work blocks Facebook and Twitter (and Instagram).

[soonerguru]

Gotcha. I wish it had been posted to a venue other than FB but the discussion seemed timely.

[Pete]

Google just announced they are giving most employees the option of working from home until July 2021.


I really wonder how much the pandemic will change the way companies work. The move to telecommuting has been very slow but this might end up being a huge tipping point.

As an employer, I can tell you we spent a small fortune on rent and utilities plus other office-related costs. I personally like going into an office every day and working with people face-to-face, but it may be paying for all that might not be the best investment.

[FighttheGoodFight]

Google just announced they are giving most employees the option of working from home until July 2021.


I really wonder how much the pandemic will change the way companies work. The move to telecommuting has been very slow but this might end up being a huge tipping point.

As an employer, I can tell you we spent a small fortune on rent and utilities plus other office-related costs. I personally like going into an office every day and working with people face-to-face, but it may be paying for all that might not be the best investment.

There are a lot of employers who are still "old-school" they want butts in chairs even if it is just people sitting at a computer all day. Pretty ridiculous.

[Pete]

There are a lot of employers who are still "old-school" they want butts in chairs even if it is just people sitting at a computer all day. Pretty ridiculous.

I call it "arse in seat" or A.I.S.

But only when it applies to meetings when others are waiting on someone.

Monday's meeting is 10AM A.I.S!!