Mike Hoerger, PhD MSCR MBA

@michael_hoerger · Twitter · 2026-08-07 20:39 UTC

You will often see scary maps like this with people screaming "COVID SURGE." I take COVlD very seriously, but I want to explain why this particular map is low quality. It's the CDC COVlD "forecast." Why is it low quality? 1) It is based on Emergency Department (ED) visits. ED visits are not transmission. Wastewater data provide the best data on transmission. Through vaccinations and mortality, SARS-CoV-2 has less obvious and severe acute effects. 2) It's not even raw ED counts. Instead, it's based on a bastardized stat, the proportion of ED visits that are due to COVlD. Nobody with any meaningful training in psychometrics would ever use such a variable. If there are a bunch of car accidents, high flu/RSV levels, firework explosions, etc. clogging up the ED, the proportion of COVlD ED visits goes down, even if COVlD is up a little. Reverse, if there's not much else going down, the proportion of COVlD could rise, not because it's rising, but because everything else is going away. There's a long psychometric history of people using these types of hodgepodge variables (e.g., Rorschach scoring), and the moral of the story is "don't." 3) It's a biased and noisy indicator. It assumes testing is happening. Often, it's not. There are also biases in who gets tested and why. As *perceived* COVlD levels rise, more people may get tested than typical, and more people will test positive; this could happen if people perceive or expect a wave, even if levels are flat or going down a little. If a wave starts earlier than anticipated, it may be slow to rise. If a wave starts later than anticipated, it may tick up too early. 4) It's hardly a "forecast." It's simply the change in the proportion of ED visits due to COVlD from one week to the next. It's a historical metric. When I was 12, we'd trade baseball cards, and we quickly learned that if Hakeem's Olajuwon's card value went up 25% in the almanac one month, that didn't necessarily mean it would the next. That's an imperfect analogy because of the differences between finance vs transmission, but I think you see the point that past does not necessarily indicate a future trend. In this context, the changes from the prior week are VERY important, but they are also only one metric that is very important. Our forecasting model uses a lot more data than that. The CDC's (version here, they have others) is quite unsophisticated. 5) Limited validity in predicting transmission. The map has a lot of "false alarms." State-level transmission often has somewhat of a roller coaster pattern, rather than steadily increasing or decreasing. You'll look at this map and think "oh sh*t" my state's levels are rising. Look next week at the wastewater data, and they went down. 6) People get confused between levels vs change. Deep purple looks bad, but that just means something is increasing. If levels are very low, that might not mean much. In contrast, if it says levels are decreasing but they are already very high, what's more important is that levels are very high, not that they might be going 5% lower at the moment. 7) No magnitude of effect. You can get some magnitude information in the pop-up boxes, but the color scheme does not tell you much. Growing 5% is very different from growing 80%. In most places, the growth is pretty slow, but positive. That matters. I view this map every week. It's one indicator, maybe the 10th best source of info. I find it helpful in teasing apart if something seems "odd" in other indicators. For example, in mid January, wastewater levels often tick up marginally after having gone down from the peak. These data often show a similar trend. I'd also find it useful if wastewater levels were rising in say May or June, and this showed the same. BUT it's just one source. ' You'll see a lot of epis and MDs over-relying on this. Mask up, just not because of this map.