# Ineffective Response to COVID-19 and Risk Compensation

post by Davidmanheim · 2020-03-08T09:21:55.888Z · LW · GW · 23 comments

## Contents

  Argument
None


UPDATE: This post was written and reflects an earlier set of my beliefs. I have updated significantly in a number of ways since it was posted, based on both external events and research, and no longer endorse it.

Epistemic status: I have a mental model that I think separates my view of response activities from that of the majority of what I see on Lesswrong and associated places. If it is incorrect, I'd be happy to update, but I think this is an area I have considered more than most other posters. I want to write a short post explaining this to allow others to update, and seeing if someone has an argument that changes my mind.

Put simply, my claim is that bringing attention to likely ineffective personal methods for reducing risk is not net neutral with a large upside if they work, it is instead likely to be on net fairly harmful, albeit with a large upside if they work.

## Argument

First, we have incredibly effective and vastly underutilized ways to prevent spread of COVID-19, namely handwashing and not touching your face. Given that, if I propose an intervention like making homemade masks from fabric which reduced handwashing compliance by 1% (perhaps due to distracting people or making them think handwashing is less critical,) it would need to be astonishingly effective to be net positive. And most such approaches being discussed are, as far as I can tell, nowhere near that level of effectiveness.

Second, most readers of Lesswrong and effective altruism blogs and facebook groups aren't hardcore rationalists, and even hardcore rationalists aren't immune to Akrasia. On top of that, people like Scott Alexander have huge readerships and sometimes link people to Lesswrong. Many people reading posts here aren't washing their hands enough as it is, and aren't going to rationally evaluate the relative effectiveness of handwashing versus other interventions.

Third, evidence exists that risk-compensation is a meaningful issue. Actions that make people feel safer usually lead to less attention paid to more annoying / more intrusive measures. (There is evidence, such as Vrolix's paper*, that risk compensation reduces the size of the positive impact, but does not make interventions net negative. This is conditioned on the impact being significant and positive, however, and seems not to apply to speculative interventions like those being proposed.

This is not an argument that we should not look into better options for response. It's an argument that we should be more careful in vetting them before encouraging people to do them just in case they work.

*) Vrolix, Klara (2006). "Behavioural Adaptation, Risk Compensation, Risk Homeostasis and Moral Hazard in Traffic Safety" )

comment by Wei_Dai · 2020-03-08T19:45:39.904Z · LW(p) · GW(p)

First, we have incredibly effective and vastly underutilized ways to prevent spread of COVID-19, namely handwashing and not touching your face.

Handwashing isn't "incredibly" effective. See this meta-analysis which concluded that for flu, (telling people to use) medical procedure masks combined with (tell people to do) hand hygiene achieved RR of .73 while hand hygiene alone had a (not statistically significant) RR of .86.

I'm not sure if there are studies about telling people to not touch their faces, but it's also probably not "incredibly" effective, judging from this.

Given that, if I propose an intervention like making homemade masks from fabric which reduced handwashing compliance by 1% (perhaps due to distracting people or making them think handwashing is less critical,) it would need to be astonishingly effective to be net positive.

This doesn't seem true judging from the above numbers. If you don't trust those numbers, can you give your own, even if they're just guesses? Right now I don't see what background assumption you might have that could make this statement true.

comment by Davidmanheim · 2020-03-08T20:19:43.906Z · LW(p) · GW(p)

Yes, telling people to do things isn't an effective intervention. I usually like citing Pearls' paper on this - just knowing that washing hands prevents infections doesn't mean that telling people to do so will reduce infections. And we've known that for a long time - you need something to convince them. Plausibly, say, a pandemic. Which is why I think that this might actually work. (And I've seen a marked uptick in people actually washing hands thoroughly in bathrooms, which is small-n observation but I found very surprising.)

Studies on doctors in hospitals shows that actually getting them to wash their hands. which many hospitals still struggle with, does drastically cut down on infections - and only recently and in some places have hospitals actually gotten people to do it sufficiently. That's why I feel strongly that we need effort and resources devoted to that, instead of distracting people.

comment by Wei_Dai · 2020-03-08T21:09:12.474Z · LW(p) · GW(p)

I still don't understand your model. Do you think that almost all COVID-19 infections occur by face-touching, and almost none by droplets directly landing on someone's face? (If so what's your evidence?) Because masks protect against both, whereas handwashing only protects against the former. It seems if we have 100% compliance with masks that would be more effective than 100% compliance of even very high frequency of handwashing (which would also be a lot more costly for people to follow). Do you disagree with this? (If so, this might be our major crux.)

ETA: Masks don't protect against rubbing eyes, but we can fix that by telling people to also wear goggles. Also it seems a lot easier to achieve 100% compliance with masks/goggles (at least when supplies are adequate) because it's visible and we can attach social stigma to people who don't comply.

comment by Davidmanheim · 2020-03-09T07:19:55.426Z · LW(p) · GW(p)

"Do you think that almost all COVID-19 infections occur by face-touching, and almost none by droplets directly landing on someone's face?"

Not almost none, but very few, yes. Aerosol transmissions are a thing, but for non airborne transmissible diseases, these seem to be rare compared to close-personal interaction transmissions, such as direct hand contact and touching other people or shared surfaces.

EDIT TO NOTE: I have changed my understanding of this. The problem was that I misunderstood terminology - papers saying there is not "aerosol" transmission don't include droplet transmission via droplets that float in the air for up to an hour, from potentially several meters away, which I misunderstood.

Mask usage is great for reducing the rate that already sick people infect others, but not recommended for healthy people, because it's just not that effective. Sick people should be quarantined, which is far more effective. Perhaps 100% compliance with mask wearing would be a viable secondary measure, if there were sufficient supply for at least sick people (there aren't even that many,) and if it were plausible to be fully compliant - but mask usage is necessarily impossible when eating and when drinking, which people need to do routinely, and hand-washing if going to be still useful and particularly critical at those times. (And no, homemade masks aren't as effective, and masks shouldn't really be reused.)

EDIT TO NOTE: Because presymptomatic and asymptomatic transmission is a thing, mask wearing is likely useful to prevent transmitting the disease to others. It's still not clear that there is any benefit for healthy people, but since we don't know who is healthy, that's basically irrelevant.

comment by steve2152 · 2020-03-23T02:23:12.509Z · LW(p) · GW(p)

Can you say why you believe "very few" infections occur by droplets directly landing on someone's face? The CDC says:

"The virus is thought to spread mainly from person-to-person, Between people who are in close contact with one another (within about 6 feet), Through respiratory droplets produced when an infected person coughs or sneezes. These droplets can land in the mouths or noses of people who are nearby or possibly be inhaled into the lungs.

...It may be possible that a person can get COVID-19 by touching a surface or object that has the virus on it and then touching their own mouth, nose, or possibly their eyes, but this is not thought to be the main way the virus spreads."

i.e. (as I read this) handwashing would help with a hypothetical, secondary mode of transmission, whereas properly-fitted masks / goggles would presumably help reduce the primary mode of transmission. This is consistent with a couple other sources I've seen, but I haven't delved into the literature.

comment by Davidmanheim · 2020-03-25T06:26:04.396Z · LW(p) · GW(p)

I have updated strongly towards agreeing with you given research in the past 2 weeks, but transmissions are clearly happening both ways - it's not hypothetical.

comment by steve2152 · 2020-03-25T11:20:51.712Z · LW(p) · GW(p)

comment by Wei_Dai · 2020-03-10T02:46:08.595Z · LW(p) · GW(p)

Not almost none, but very few, yes. Aerosol transmissions are a thing, but for non airborne transmissible diseases, these seem to be rare compared to close-personal interaction transmissions, such as direct hand contact and touching other people or shared surfaces.

They said the study proves the importance of washing hands and wearing face masks in public places because the virus can linger in the air attached to fine droplet particles.

“Our advice is to wear a face mask all the way [through the bus ride],” they added.

[...] The researchers also found that none of those passengers in the two buses who wore face masks were infected.

They said it vindicated the decision to ask people to wear a face mask in public.

I think they can if done properly.

comment by steve2152 · 2020-03-28T17:22:30.986Z · LW(p) · GW(p)

FYI, that first article you linked was later retracted without explanation.

comment by Pablo_Stafforini · 2020-03-09T00:16:38.848Z · LW(p) · GW(p)
I've seen a marked uptick in people actually washing hands thoroughly in bathrooms, which is small-n observation but I found very surprising.

To add another anecdote, Tyler Cowen also noticed this.

comment by Viliam · 2020-03-08T17:37:21.039Z · LW(p) · GW(p)
Given that, if I propose an intervention like making homemade masks from fabric which reduced handwashing compliance by 1% (perhaps due to distracting people or making them think handwashing is less critical,) it would need to be astonishingly effective to be net positive. And most such approaches being discussed are, as far as I can tell, nowhere near that level of effectiveness.

This argument depends a lot on the correctness of your model. How do you know which proposals reduce handwashing compliance by 1%? Without numbers, it becomes a fully general argument against doing or even debating anything (other than washing your hands).

comment by Davidmanheim · 2020-03-08T20:25:07.497Z · LW(p) · GW(p)

Yes, we need to discuss our models. In this case I have some Bayesian priors, informed by a review of relevant literature, and reinforced by observations of how people are actually reacting. I've seen many people actually washing their hands recently - not all, but most. And this isn't usually true, I've been paying attention for a couple years now, since I started working in bio-risk. You don't need to trust my model, but we have fairly strong, if indirect, evidence that many of the proposed interventions - like mask wearing by people who aren't trained to do so, which I keeps seeing, and it would be funny if it weren't infuriating - are marginal at best.

comment by jimrandomh · 2020-03-09T00:33:53.674Z · LW(p) · GW(p)

That paper (which I notice you're first author on) is pretty good and you might want to post it as a top-level post. Note that you estimate handwashing optimistically reduces transmission by 50%. This paper estimated an R0 in Wuhan of 3.86 prior to social distancing measures, so if the situation in other cities is similar, then high compliance with hand-washing would only cover about half of the required reduction, with the rest left to other measures such as reduction in gatherings and travel.

comment by Davidmanheim · 2020-03-09T07:02:53.684Z · LW(p) · GW(p)

I posted in to the EA forum, and have now crossposted that to Lesswrong. (I clearly overestimated the proportion of people who read both.)

Regarding transmission reduction from handwashing, two points. First, reducing flattens the curve in ways that make the epidemic far more manageable for emergency services, saving lives. Second, prior to any social distancing, the R_0 was far higher than it is when we have the growing norms around reduced hand-contact, etc., which could very plausibly combine to push far closer to 1.

comment by habryka (habryka4) · 2020-03-08T21:49:09.092Z · LW(p) · GW(p)

Just curious, does anyone have a link to a paper estimating the effectiveness of masks when not worn properly? My intuition says that I would expect them to be somewhat less effective (maybe 50%), but that a lot of the effect would still be present.

I've spend 15 minutes looking for a good paper, but haven't found one. This is a claim I've seen cited many times, so it seems good to ground it out.

comment by habryka (habryka4) · 2020-03-08T22:41:53.765Z · LW(p) · GW(p)

After researching this for another 40 minutes, I still don't see where this claim comes from. All studies that I found are talking about adherence in the form of "was wearing the mask most of the time", and I can't find any analysis that wearing a mask is super complicated, and that people are wearing the masks in an ineffective way (many sources say that it's hard to get people to wear the masks at all, but that's a different issue).

When you gave a group of people a set of masks, the people who comply to the level of "are wearing the mask most of the time" appear to be experiencing a significant reduction in infection rates. This is true even if the masks weren't fitted. Here is one of the variety of studies I found about this.

I currently don't believe the "masks worn by people who aren't trained to do so aren't effective" claim. I also further believe that the effect of masks on untrained people is on the same order of magnitude as hand-washing (though definitely smaller, my gut says something like a 4th to a 5th as effective), and as such is not marginal. Though the latter claim is definitely not super well-sourced and I haven't done the appropriate fermi estimates, and would greatly appreciate more evidence on the relative importance of these interventions.

comment by Davidmanheim · 2020-03-09T07:39:32.478Z · LW(p) · GW(p)

I think it's also worth pointing out that there just aren't enough masks for what you are suggesting. That's why my example was telling people to make homemade masks, not using N95 respirators. If they are following instructions for mask usage, people should be replacing masks multiple times every day and not reusing them - as soon as you remove the mask, you touch both the inside and the outside, moving droplets around. Suggesting widespread use of the types of masks you're talking about, then, seems unhelpful in any case.

comment by habryka (habryka4) · 2020-03-09T08:28:12.613Z · LW(p) · GW(p)

Oh, I am not at all advocating for widespread usage of masks in this particular pandemic. I am just trying to figure out whether they are effective.

However, on a more object-level, see this [LW(p) · GW(p)] comment for an estimate of the potential increase of global mask production in the next few months.

comment by Davidmanheim · 2020-03-09T08:42:58.434Z · LW(p) · GW(p)

That's a very useful data point, and I'm happy to see that I was pessimistic about how quickly factories could ramp up production. Hopefully we'll see the supply crunch reduced in the near future, (without a collapse in quality,) and at that point I'd be very happy for people to advocate more widespread mask wearing using actual disposable masks.

comment by Davidmanheim · 2020-03-09T07:32:48.702Z · LW(p) · GW(p)

There is research on loosely fit mask like surgical mask effectiveness compared to fit respirators, and they are much less effective. My model here is that loose fitting masks that prevent 50% - or even 95% of the particles from reaching your mouth is near-useless if someone sneezes on you, since you don't need more than a very small number of particles to infect you. And otherwise, masks for non-sick people aren't accomplishing much without handwashing after every time they touch the mask - especially when people still touch their mouths behind their masks, as people often do even when they know they are aerosolizing dangerous pathogens in labs. To reinforce this, anecdotally, biosafety lab workers go through training, and still need routine retraining to make sure they wear masks properly and don't do things like eat with masks on by moving the mask out of the way.

comment by habryka (habryka4) · 2020-03-09T08:06:52.480Z · LW(p) · GW(p)

Do you have any source for this? I remember multiple claiming that this research exists, but as I mentioned, I have so far failed to find it. The "near-useless" claim also appears to be contradicted by the study I linked in the other comment:

https://www.ncbi.nlm.nih.gov/pubmed/19193267

Which said:

We estimated that, irrespective of the assumed value for the incubation period (1 or 2 days), the relative reduction in the daily risk of acquiring a respiratory infection associated with adherent mask use (P2 or surgical) was in the range of 60%–80%

Which seems like a really large amount, and was for non-fitted masks.

comment by Davidmanheim · 2020-03-09T08:33:17.483Z · LW(p) · GW(p)

Thanks for that, and I have updated towards masks being more useful for non-infected people to prevent getting sick than I previously thought - though not much for the usefulness of advocating mask wearing for the public, given shortages.

I do suspect that the poorly fitting masks were useful for other reasons, like reducing hand-to-mouth contacts and increasing handwashing - which was reported as 10% higher for adherent mask wearers than non-adherent ones, potentially contributing to the effect. (It was unfortunately not reported for the control group.)

comment by romeostevensit · 2020-03-09T08:24:55.533Z · LW(p) · GW(p)

It seems like risk compensation and moral licensing are the same, or very similar mental moves. Pointing this out because it can help 'catch it in the act' so to speak.