Am I missing something here? (re: reopening of bars, etc... now)

Where are the numbers for the trajectory you’re talking about? Florida didn’t reopen. The bans that were in place are still in place

Florida’s deaths per million increased to 1,082 while I was typing this.

The only numbers I’m referring to are the Worldometers stats, same as I linked to before.

I’m not sure what bans you are talking about that were in place at one time and are still in place now. My impression is that California is under far more restrictions at the moment than Florida is, and that it’s been that way for a few months.

I’m not seeing the numbers you’re referring to in Worldometers. Could you point them out?

The same bans you’re talking about. Florida may have less restrictions than California, but they still have some restrictions. Those are still in place.

These articles might show some of the trends. Florida’s trend is moving up. California’s trend might be flattening, at least at the moment, hopefully due to the shelter in place restrictions.

Man, I have a hard time keeping up with exactly which argument is convenient. One minute the bodies are stacking up in California, the hospitals are overrun, the schools won’t be open for another semester. The next minute lockdowns are working (though one wonders what took them so long). I guess let’s check back in on it in a couple weeks.

Man, I have a hard time keeping up with which tactic you’ll come up with next to shift away from your insinuations. Both are true. LA is having a rough time right now, but the curve is flattening, hopefully due to the lockdowns.

Well, let’s sure hope so.

You’ve got me curious enough to do some research on Florida. What I’ve read has me believing that businesses are much more open there than in California (or New York, for that matter), that restaurants are open for indoor dining, and so on. If I’m that mistaken, I sure want to figure out why.

Was that in the expanded pdf? I may have to dig in further.

Yeah, it’s in Section 2.2 under Supplementary Methods.

Then, we evaluated whether the relative risks predicted by each formula concorded with the rankings of POI categories proposed by independent epidemiological experts.

As shown in Figure S7, we find that the predicted relative risks match external sources best when we use our original parametric form that accounts for both dwell time and density: restaurants, cafes, religious organizations, and gyms are among the most dangerous, while grocery stores and retail (e.g., clothing stores) are less dangerous. However, when we assume only dwell time matters and remove the density term, we see unrealistic changes in the ranking: e.g., restaurants drop close to grocery stores, despite both sets of experts deeming them far apart in terms of risk. When we assume only density matters and remove dwell time, we also see unrealistic changes:e.g., limited-service restaurants are predicted to be far riskier than full-service restaurants, and gyms and religious organizations are no longer predicted as risky, which contradicts both of our sources. These findings demonstrate that both factors — the dwell time and density — are important toward faithfully modeling transmission at POIs, since the predictions become less realistic when either factor is taken out.

I bolded the part I alluded to before.

See, this whole approach just bemuses me, in that there was no real chance they were ever going to arrive at any results that someone else hadn’t already arrived at, broadly, and in a much, much less scientific or even rigorous way. It seems like such a shallow foundation to build upon.

Probably because there’s some degree of lag in every stage of all the things mentioned- lockdowns take a while to show an effect, and something like hospitalizations and deaths trail the initial infections by quite a bit of time as well.

Well, but which of California’s, or LA County’s, policies have changed recently?

I read that section, and that’s a fair criticism. The “Grocery Stores” category includes the lower density stores in other areas with the high density stores in Philidelphia’s low-income areas. That means the spread of risk for “grocery stores” is wider than one might guess just by labels.

Amusing, but they actually did split restaurants into
“full-service restaurants”, “snack and non-alcoholic beverage bars”, and “limited-service restaurants”. As for the classifications and divisions of the retail category, they used the North American Industry Classification System.

So that’s not something they just made up, that’s the standard for analyzing businesses. I wish they had included the data for “Drinking Places (Alcoholic Beverages)” in their analysis, but they said this:

I don’t see where they say that, but it makes some sense. But you are correct, a mall is a mixed venue interaction in a closed environment, with many patrons visiting multiple establishments of different classifications - like department stores and snack bars. That makes separating the risk tricky, as “the mall” might be a category unto itself.

It’s always standard practice to test your model with counterfactuals or with alternative structures. Separating the parameters is required to understand their different impacts. Comparison of the three structures is critical to validating the model.

The question is what makes the comparisons valid? What justifies the choice of using both parameters? Well, typically one looks at the results and compares against expected results validated by other sources. Okay.

Removing dwell time makes limited-service restaurants and snack bars way more dangerous than dine-in restaurants. Do you dispute that makes zero sense? Nevermind any expert opinions or rankings, just common sense?

So what about looking only at dwell time and eliminating density? See below.

I accept the criticism that the scientific rigor is not visible. It certainly would help to have details on Emanuel et al’s model available for inspection. I’m unable to see the other article, but it appears to be a newspaper article summarizing expert opinions, not any kind of rigorous scientific data source. So that’s a valid concern.

They need better sources for their risk assessment. However, their study is not really intended to assess how restaurants fare against grocery stores against nail salons or whatever. That’s more of a given condition, assumed premise for the study. They wanted to assess conditions that lead to different risk levels to different populations. That’s more about understanding both influencers of transmission.

I did find where they explain SEIR in the article, not the supplementary information.

What I see here is Florida data.

For Total Cases, the slope is increasing.

For Daily New Cases, using 7-day moving average, the plot is flat from April to June, then a spike to 11,943, then a drop to October where it begins rising again now hitting 15,985.

Total Deaths has a fairly steady incline from August from 7,022 to 23,075. Daily Deaths has a similar profile to Daily Cases, but lagging two weeks. It peaked in August at 184 and now is at 141.

California has Total Cases shooting up at a huge rate from 1,340,716 on Dec 5 to 2,758,909 on Jan 11. Daily New Cases 7-day average shot up from Nov 7 to Dec 21, but then leveled off and sits at 42,096. Total Deaths the slope has increases since Dec 13. Daily New Deaths 7-day average is fairly flat, with a peak on August 7 at 141, then a drop, followed by an upswing starting around Dec 1 from 64 to 484 on Jan 11.

Looking at the Worldometers page for the US, we see the stats where
Florida has 70,002 cases/100M and 1,082 deaths/100M, while California has 70,982 cases/100M and 787 deaths/100M.

So, 484 vs 141 seems significant, but by population 1,082 beats 787. Pick your poison.

Yes, I agree with that in full. My problem is with the conclusions that are being, or have been, drawn from the study. You can see this in the way it’s been written about, in the New York Times article for example, but also in their own words. Here, for instance, from their FAQ:

Can you use the model to predict what will happen in the next weeks/months?

In principle we can, but we need to provide the model with updated data on mobility and COVID-19 cases, since the analysis in our paper uses data from the spring. We are working on doing this now.

There’s plenty else in that FAQ to support the notion that they are doing more than just exploring disparate risks across populations, but this in particular is concerning to me. It’s not just that they use assumed premises as the basis for analysis. It’s that they think they can accurately predict the future.

I don’t think it’s difficult to imagine plausibility. If it were to be true that one good breath of virus-filled air is enough for infection, then you surely wouldn’t find any comfort in walking into a virus-filled room for only a few minutes as opposed to an hour. I mean, a lot of people walk in and out of a McDonald’s, and I’m not sure what the ventilation system is like in a typical fast-food joint. They often don’t have high ceilings like grocery stores do. For that matter, workers remain inside for hours on end (and presumably in cramped quarters at that, meaning they are presumably more likely to spread infection to each other), breathing out all the time. What’s hard to imagine about any single breath someone takes in a fast-food place being more risky than the same single breath in a full-service restaurant? In fact, it is the opposite that would seem to me to strain common sense.

As best I can tell, the premise that dwell time is critical has more to do with accumulating opportunities than it does with accumulating viral load. But I suspect that elevators in busy buildings carry a pretty significant risk of their own, despite any one person not spending much time there.

Well let’s consider for a moment what their assumption is - that full-service restaurants are riskier than grocery stores. Sure, rigorous scientific literature isn’t drawn upon to establish that for this study, but it does happen to be the working model of all the epidemiologists - scientists studying how diseases spread. The two cites they pull aren’t science literature, but the credentials of the sources are there. It’s not some random internet yahoos or philosophy professors or mechanical engineers, it’s experts in disease spread.

Second, the model established that it reflects the patterns of spread fairly well. Not perfectly, but fairly well. Naturally, any model that conforms to the existing data can be used to predict what will occur. How accurate that prediction turns out to be may vary, may not match the accuracy of the previous results for various reasons. That then becomes data to evaluate the model.

Anything that has a reasonably proven performance against prior data may be a useful tool for what to expect. Or not. But it’s better than no tool.

And they admit they need to update the data. That would definitely shake out model performance to a better extent prior to any projections. If the assumptions are not valid, there will be errors that surface as they update the data.

Again, this model is not designed to determine the relative risk levels of different types of businesses individually. It’s designed to try to figure out the distribution of spread.

So you have to assume that “one good breath of virus-filled air is enough for infection” rather than any look at accumulating viral load, or considering accumulating opportunities for exposure. Yet severity of infection does seem to correlate with viral load exposure.

Furthermore, their study does not look at workplaces. They explicitly state their data can’t separate workers from customers, but workplaces constitute their own level of hazard from exposure duration and other factors - like shared break rooms and masking within the staff zones. So looking at McDonald’s workers is a bit unfair without considering the staff at regular restaurants.

And regular restaurants are not any less crowded than McDonald’s kitchens.

Maybe you have a point with respect to a crowd standing around to order without any kind of crowd limits. No attempts at metering or social distancing can leave some crowded lines. But then dine-in restaurants can get awfully crowded, too, when they are open. Haven’t you seen places with regular waiting times of 20 to 30 minutes and peak waits of over an hour?

As for ventilation, I rather suspect your fast-food place actually has pretty good ventilation to keep smoke and odors from the kitchen from proliferating into the customer area. But that’s speculation on my part.

I just don’t buy that the exposure for your average one breath is higher in a fast-food joint than a dine-in place. That you find it plausible says we are definitely on different pages.

But even if you are correct, consider that all restaurant controls seem to selectively impact that very feature. Full-service restaurants allow more people for longer. Fast-food places are being metered and often have the dining room completely closed, just using drive through.

I was asking SayTwo about the trajectory which is what he was claiming was steeper in the case of CA. I’m not sure which trajectory he meant.

You picked a random day to compare the death rate on a widely varying statistic. If I picked Jan 2 instead of Jan 11, FL would be 217 and CA would be 239 with CA having double the population of FL. That would be less significant.

I’m not sure that’s the fairest analysis. They built their model on the existing data. In other words, they chose the parameters that the data told them to choose. You did catch, right, that each city has its own infectiousness factors? This means that new car lots in Philadelphia are more infectious than new car lots in Chicago, and so on. Now, maybe there really is something to all the points of interest in one place, in total, being measurably different than all the same points of interest in another place. Or, maybe that’s a convenient way to get the model to work in the first place. And by ‘work in the first place’, I of course mean ‘match the existing data’.

There could be other explanations, though, besides that grocery stores in one city are built differently or get different audiences and are just inherently more prone to infection. For example, there could be climatic influences, as are seen with viruses that are seasonal. They could have built a model that attributed variance to regional climates, but they did not. If they had, they might have used the same existing data and arrived at differing predictions for the future.

I’m not claiming it means we should automatically dismiss this study out of hand, but I think it’s pretty well established that when you choose parameters and build a model based on how well it fits existing data, you aren’t the only one who can build such a model and get the same results.

Maybe. I don’t think the science on that is settled. There are people who believe that reasons exist for why a small initial load can outpace the immune system’s attempts to slow its growth and end up at the very same result.

My point was not that the study looked at restaurants as workplaces, but that employees (of any business) spend more time there than customers do. You asked how I thought it passed the common sense test that a fast-food place could be as infectious as a full-service restaurant. I said that if an employee at Burger King was infectious and breathing virus particles out all day, those particles might find a way to infect someone who walks in and places an order and leaves. In other words, it’s not like two joggers passing on the street.

But at any rate, we can ignore the workers if we like and just focus on the customers. People do sit and eat in a lot of fast-food restaurants. Presumably they are breathing and spitting droplets while they do. I mean, if you ask me how it makes any sense that you might catch the virus at a Burger King, I don’t think you have to look hard to find a reasonable answer.

On the whole, I just don’t get this notion that seemingly a whole category of places is basically risk-free. It’s like people think that because they only popped in the grocery store (or convenience store, let’s say) to buy milk, and they were only in and out, and they wore a mask, and so did the other customers, and so did the workers, then they didn’t get the virus there. Like, how does that work out in contact tracing? Can anyone, and I mean anyone, point to a quick trip to the grocery store as where they got the virus? How could they possibly know?

Same thing with the bank, the post office, the oil change place. I mean, over the course of a few weeks or months you will have had countless such little interactions, if you aren’t sheltering. Each one may seem relatively risk-free, certainly as compared to something like going to a club for a couple hours, but is that really the case? I mean, when a virus this contagious rampages through a community, isn’t it likely that it uses interactions like those as vehicles?

But that’s not a convenient analysis, because it’s not as neat as ‘shut down indoor dining’. But from the point of view of the virus, I’m not sure things are all that different.

Well, I’m not sure you’re considering all the elements. Look, for one thing, it’s a bit of a fool’s errand to paint all members of these categories with the same brush. You might have a crowded hour-long wait at Chili’s and your party might spend a couple hours there. Or you might have an intimate dinner with your wife at a white-table Michelin-starred place. (Maybe the French Laundry!) And you might have a crowded McDonald’s dining room, overrun with screaming kids, versus a drive-through-only joint. I’m not sure the blunt tools are the best way to go about this.

But even if they are, I’d question your point that sit-down places allow ‘more people for longer’. If it’s longer, meaning the average party sits for a longer time than at a fast-food place, then it might be fewer people that any one diner is exposed to, if there’s any merit to the idea that aerosolized virus particles can remain in the air for a fair amount of time. In other words, you might share a room with 50 other people if you sit for an hour in a small-ish restaurant but you might be exposed to what was in the breaths of a hundred people if you walk into a busy fast-food place. I mean, I’m not sure the proper analysis is so obvious, if you start to consider additional factors.

And to get back to the point about painting with a broad brush, I’m also not sure that every restaurant deserves to be seen as a crowded place where lots and lots of people gather. Many of them are far from that by sheer design. And many more of them are far from that by lack of success, let’s say. It’s not an easy business to thrive in, as I think we all know.

You’re ignoring (purposely it seems) how many people are in each establishment are there AT A GIVEN TIME. The point of a nightclub is to pack the place and stay there for a few hours. The point of a grocery store is to get your groceries and leave. Curbside pickup Is popular now. I don’t know why you’re not understanding this.

I’m not ignoring that in the least. I’m talking about total spread in the community, collectively, not risk of infection for any one person. The relative individual risks of the different venues according to things like how packed they are and how much time people spend there are already baked in. I’m talking about adding up all the infection in the community caused by people going to one category and comparing it to all the infection in the community caused by another category.

So, let me try it again and see if I can make it more clear. A venue could be a thousand times riskier than another, like let’s say a crowded nightclub versus grocery store curbside pickup, yet still contribute less to community spread if a hundred people go to nightclubs and ten thousand people go to grocery stores.

You do understand saying one venue is a lot less risky than another is not saying that people don’t get infected at the less risky place, right? That if curbside pickup is a thousand times less risky than something else, it means that one person gets infected during curbside pickup for every thousand people that get infected at the other place, and not that no one gets infected. Right?

Sorry, I used the wrong figures for example. Here’s a simple example that shows how places with different risk could result in the same number of infected people in the community:

Risk of infection for those who go there: 10% (1 in 10)
Number who go there: 1000
Number infected: 100

Grocery Stores
Risk of infection for those who go there: .1% (1 in 1000)
Number who go there: 100,000
Number infected: 100

…haven’t we already established that these numbers are completely made up and don’t actually demonstrate anything at all?