Daily cases are falling, daily deaths may be close to peaking, but there are problems with the data. Those are the takeaways for today.
Cases are falling faster than I would have expected nationally. Just as the rise was driven by California, Texas, and Florida, the decline is also. The fact that the national decline is faster than expected prompted me to do a bit of digging to figure out why. I think it was just a case where my expectations were set incorrectly. Here’s a bit of background on how that happened.
Here is the long term national graph as of today:
As you can see, the current decline is much steeper than the decline through April. I was expecting the previous pattern to more or less repeat. But the national numbers are really driven by the state numbers, which in turn are driven by county and city level infections. The national graph aggregates all of the local level data into a single whole. The effect of aggregating data in this way depends on how the low level data are related. If each local outbreak is relatively independent of the other, aggregating them will tend to smooth out the curve. If the local outbreaks are more connected, aggregating them can amplify changes.
Think about rolling up a D&D character. For those unfamiliar with it, this involves rolling three dice and adding them, yielding a number between 3 and 18. When you do this, it’s pretty rare to get a 3 or 18, but pretty common to get a 10 or 11. Even though each die has an equal chance of rolling any number from 1 to 6, when you add them up, you end up with different chances for different results. This type of pattern happens all the time – aggregating a bunch of independent numbers will tend to yield similar results each time – the larger the group, the greater the tendency. With some caveats and a lot more precision, this is one of the central ideas in statistics.
On the other hand, think about apples ripening in an orchard. Each apple ripens based on the conditions it encounters. But an important condition is the tree sending a hormonal signal to the fruit telling it to ripen. Because the tree sends a similar signal to all the apples on it, they tend to ripen at close to the same time. Unlike dice, apples ripening on a tree are not independent events. When the tree sends the signal depends on environmental factors such as sunlight, water, and temperature. Trees in the same orchard are exposed to very similar environmental factors, so they all tend to ripen at about the same time. However an orchard in Maine may ripen at a different time than one in California.
What does this tell us about the graph? Well, to the extent that local outbreaks are independent, we expect them to balance each other out, to act like dice. One state’s numbers may rise, another’s may decline, the net result is not a lot of change. This seems to be what happened in April. Each state was largely doing it’s own thing and New York’s outbreak didn’t have much impact on Florida or Texas. So, as New York started to decline, other states were rising or steady and the net result was to blunt the effect of New York’s falling numbers. In fact, if we just look at New York’s graph in April we see a decline that looks a lot like what we’re seeing now at the national level. Here it is:
New York’s case numbers declined just somewhat more slowly than they had risen while the national case numbers declined only a little. The fact that we’re now seeing a similar decline at the national level indicates that states are behaving less independently of each other than they were in April and May. This time around, Texas, Florida, California, and other states are behaving more similarly than they did a few months ago. What accounts for that, given that there has been no appreciable change in strategy at the federal level? I don’t really know the answer to that, but here are some plausible guesses.
First, the media narrative has become more consistent. In April we knew less about the virus and many states were relatively unaffected. As a result, we had conflicting media narrative about the degree of danger and about the best methods for responding to it. In particular, there was a very large red/blue divide in the state level messaging. With red states such as Texas, Florida, and Utah experiencing more severe outbreaks than they had, the messaging about the risk levels has become more unified; the gap between what Newsom and Abbot (governors of CA and TX) say is a lot less than it was.
Second, it helps that we also know a lot more. In April there was still some debate about the degree to which mask wearing would help. Now, there is clear consensus among experts and the “debate” has been relegated to fringe groups.
Third, I think that people in general have become more accustomed to the idea that the virus isn’t just going to go away, so we may be seeing higher levels of compliance. In April, not only was there speculation that the virus would expire in the heat of summer, we also felt the restrictions more acutely. The change in our average daily experience from mid-March to mid-April was huge. It’s hard to make big changes. As anyone who has gone on a diet or started a similar lifestyle change knows, the hardest part is early on. We start with commitment, but soon miss the old habits. Until those habits get replaced by new ones, it’s cussedly difficult to stay on track. That was the US in April – Hey! I’ve been on this diet for three days now and I’m hungry and not losing weight! By now, we’ve adjusted to some of the new behaviors and are overall better at following guidelines.
I want to be clear, these reasons are just my speculation. They fit with my experience, but I don’t have hard evidence to support them. So, back to data.
What we’re seeing is declining case numbers across a wide range of states. CA, FL, and TX are all declining, as are most of the states that had been averaging over 1000 cases per day. There are exceptions, Illinois chief among them, but a very large majority of states are now showing decreasing case numbers. You may recall that my concern was that as CA, FL, and TX started to decline, other states would continue to increase, leading to a long, high plateau. That seems to not be happening, and for that we should all be grateful.
At the same time, a majority of these states are continuing to report increasing daily death numbers. California, Texas, and Florida are among these. This is to be expected as deaths from COVID follow infections with a few week lag. Based on that, we may see flattening or declining case number this coming week or perhaps the week after.
The biggest unknown at this point is the validity of the data we are seeing. California and some other states have reported glitches in their data reporting. California believes this has led to an underreporting of cases. None of this appears to be related to switching the national data from the CDC to DHS, these are state level glitches. I haven’t seen reporting about major issues from that angle. Also, while these glitches may be substantial, they are unlikely to reverse the overall trends we’re seeing. Even if California had been underreporting cases by 30% over the last two weeks, the national numbers would still be declining, they would just be a few thousand cases per day higher. While the difference for California would be huge, it would be much less so against a backdrop of 60,000 cases per day at the national level.
Looking forward
Depending on the tracker you follow, we have either just passed, or are about to pass, the 5 million case mark. We also just passed the 160,000 deaths mark. When we pass 6 million and 170,000 depends largely on how far we continue to decrease the daily case numbers. It makes a big difference whether those stabilize at 50,000 or 20,000. I don’t feel confident enough to guess where within that range we will land. I do think that we will likely see rising numbers again in mid or late September as people spend increasing time indoors and the flu season ramps up. I would suggest preparing for that.
Here’s a guess at what the next 10 days may bring.
| Day | Cases | Deaths |
| 8/7 | 4,941,603 | 161,344 |
| 8/8 | 4,988,006 | 162,360 |
| 8/9 | 5,034,409 | 163,405 |
| 8/10 | 5,080,812 | 164,449 |
| 8/11 | 5,127,215 | 165,494 |
| 8/12 | 5,173,618 | 166,538 |
| 8/13 | 5,220,021 | 167,583 |
| 8/14 | 5,266,424 | 168,627 |
| 8/15 | 5,312,827 | 169,672 |
| 8/16 | 5,359,230 | 170,716 |
| 8/17 | 5,405,633 | 171,761 |
Be safe, stay well, and thanks for reading.