Cost-Effectiveness of Air Purifiers against Pollution

post by Lukas Trötzmüller (Lukas T) · 2020-07-30T08:37:48.908Z · EA · GW · 6 comments


  Quantifying human health impacts of air pollution
    role of ultrafine particles
    vs Outdoor Pollution Levels
    Exposure Levels
    in estimating the effects of PM2.5 reduction
      Obstacle 1: Measurement of mortality vs years of life lost
      Obstacle 2: Is average PM2.5 level the right kind of analysis?
      Linearity Assumption
  Air purifiers with HEPA filters for PM2.5 mitigation
    the achievable reduction in personal exposure
      Step 1: Simple Model
      Step 2: Comparison with real-world personal exposure data
    Benefits delivered by this reduction
    and Filter Cost
      Device Cost
      Filter Replacement Cost
      Electricity Cost
    to improve the cost-effectiveness
  Negative side effects of this intervention
  Open Questions
    about air pollution in general
    on health effects
    on the practicalities of air purifiers
  Conclusions and Further Work

The goal for this post is to give an introduction into the human health effects of air pollution, encourage further discussion, and evaluate an intervention: The use of air purifiers in homes. These air purifiers are inexpensive, standalone devices not requiring any special installation procedure.

This particular intervention was selected out of personal interest, not because I believe it’s particularly effective. It’s plausible that other interventions against air pollution would be much better - for example, providing more people with clean energy for cooking.

We will investigate what it costs to significantly reduce personal exposure to the most damaging form of particulate matter (PM2.5) using these devices. A first analysis suggests that the cost-effectiveness of this intervention is two orders of magnitude worse than the best EA interventions. However, it is still good enough to qualify as an “effective” or even “highly effective” health intervention according to WHO criteria.

The model corresponding to this post can be found here:

Epistemic status: I am confident that this post identifies most of the big questions and uncertainties. Given that I have no background in public health, it is possible I’ve missed major pieces of the puzzle, and it’s likely that the specific numbers are off.

Quantifying human health impacts of air pollution

Air pollution is a significant risk factor for cardiovascular disease, cancer, respiratory infections and COPD (WHO). For further discussion, we will look at particulate matter pollution (PM) only - which, by itself, caused a loss of 106 million DALYs (disability adjusted life years) in 2016 - 51% of that in China and India (State of Global Air Report 2018). A Finnish study looked at NO2 and O3 in addition to PM, and found that PM contributes 85% of the total disease burden among those pollutants, with the finest particles (PM2.5, particles with a diameter of less than 2.5µm) producing the vast majority of the harm. Therefore, for the remainder of this post we will only consider PM2.5, which is measured in µg/m³.

Air pollution is a significant problem even in regions with relatively good air: The WHO states that “even in the European Union, where PM concentrations in many cities do comply with guideline levels, it is estimated that average life expectancy is 8.6 months lower than it would otherwise be, due to PM exposures from human sources.”

The role of ultrafine particles

According to some sources, ultrafine particles - which are significantly smaller than 2.5µm - might produce a significant portion of the harm. However, for our investigation here it is enough to consider PM2.5 measurements only, without gathering separate data on ultrafine particles. This is for two reasons: First, we can assume that all kinds of PM are roughly correlated with each other (most literature uses PM2.5 measurements only). Second, the HEPA filter intervention we will suggest is good at filtering ultrafine particles - so if PM2.5 is successfully filtered, ultrafine particles are removed too. [1]

Existing work

A large amount of studies investigate the relationship between PM2.5 exposure levels and health effects. The consensus seems to be that this relationship is sublinear when looking at a large range: at high concentrations, each additional µg/m³ of PM2.5 contributes less harm. Here are some studies which examine this nonlinear relationship: 1, 2, 3, 4, 5, 6, 7.

All of those studies calculate extra mortality or rate of illness. This data is not sufficient to estimate years of life lost or DALYs [2]. Other studies do make estimates of these, but they do not take into account the nonlinearity:

Indoor vs Outdoor Pollution Levels

Indoor air pollution correlates with outdoor air pollution relatively closely, ranging from “50% lower” to “as high if not higher”, depending on human activity (source). People using open stoves for cooking will experience the highest levels of air pollution - far above the amount of PM2.5 even in bad city air (source). Researchers in Germany measured a ratio between indoor and outdoor PM2.5 concentration of 0.33-0.78 - the lower value for closed windows, the latter value for tilted windows. However those measurements were only performed in an uninhabited room in one building in one city.

The specific ratio between indoor and outdoor PM2.5 concentration might depend on a variety of factors:

A comprehensive meta study looked at 61 articles investigating different kinds of buildings. Over 40% of articles found higher indoors than outdoors PM2.5 levels. The authors of the meta study do not attempt to arrive at a consensus for the ratio between outdoor and indoor levels, but we can guess that it's plausible to assume indoor levels equalling outdoor levels (from Figure 3a in the meta study). For smoking households, this assumption is likely to be too low, for non-smoking households with low air infiltration from the outside this might be too high.

For our calculation we don't even need to consider indoor levels explicitely, for reasons described in the next section.

Personal Exposure Levels

The relevant quantity for health effects is the personal exposure a person experiences during a certain period of time. We will not consider this quantity directly. Instead, we will estimate the health effects for certain outdoor PM2.5 levels using real-world data. This data already includes the fact that people spend their days in a variety of indoor and outdoor settings. Therefore, the relationship between outdoor and personal exposure is already implicitly taken into account.

The same can be said for indoor levels, so we don't need to estimate those either.

Difficulties in estimating the effects of PM2.5 reduction

We will propose to place air purifiers in subjects homes, which gives them significantly reduced PM2.5 levels for parts of the day. How can we estimate the health benefits of this intervention? There are two obstacles here.

Obstacle 1: Measurement of mortality vs years of life lost

We know that health impact scales sublinearly with PM2.5 levels. In our summary of existing work, we have seen that most studies that evaluate this nonlinear relationship estimate the mortality risk only and do not attempt to quantify the years of life lost. This is an obstacle to our analysis. In order to arrive at cost-effectiveness estimates, we would very much like to know the additional years of healthy life that can be generated per $ invested in our intervention. The studies evaluated do not provide enough information to calculate that. None of the studies considered calculates the gain in years of life at each PM2.5 level while correctly taking into account the sublinear nature of the relationship.

This could be resolved in several ways:

  1. Do a proper analysis, incorporating additional data. [4]
  2. Take reliable DALY values from one source and multiply it with the shape of the mortality risk curve taken from other sources.
  3. Use a simple rule of thumb: For example, this UK study suggests multiplying the number of premature deaths by 12 to arrive at the years of life lost (of course, this rule-of-thumb is valid only in the particular case of air pollution).

For our investigation here, we choose neither of these approaches and simply ignore the nonlinearity for reasons that will be outlined below.

Obstacle 2: Is average PM2.5 level the right kind of analysis?

Even if obstacle 1 did not apply and we knew the relationship between average PM2.5 and years of life lost perfectly, this data would not be directly applicable to our analysis. This is because we are not reducing average PM2.5: We are suggesting to place air purifiers in subjects homes, thereby giving them an environment that is much lower in PM2.5 for some portion of the day, and normal levels for the rest of the day. The effects of this on human health might be quite different than simply exposing them to a constant exposure, even if the averages are the same. The effects might be more positive than the calculation based on averages would suggest. [5]

Finally, outside PM2.5 levels can vary significantly depending on weather and season. Perhaps any analysis based on average PM2.5 levels will always paint an incomplete picture.

Linearity Assumption

Because of the great uncertainties mentioned above, we will assume linear scaling of health effects based on average PM2.5 reductions. We will only use data from one source. That source indicates 0.22 days of life lost from extra exposure of 1µg/m³ for one year, it does not take into account disability. We can be fairly confident in these numbers because they are roughly consistent with the sources from the UK and India.

Air purifiers with HEPA filters for PM2.5 mitigation

A standalone air filtering device, using a HEPA (high-efficiency particulate air) filter, can reduce indoor PM2.5 levels significantly:

Calculating the achievable reduction in personal exposure

To estimate the benefits of air purifier use, we need to know the achievable reduction in personal PM2.5 dose - which includes people going about their daily lives and not spending all day at home next to the device.

Terms used here:

In order to calculate the reduction in personal exposure, we will make a simple calculation based on some assumptions. Then, we will compare it with real-world data from studies in which air purifiers were placed in participants homes and personal exposure was measured using portable devices.

Step 1: Simple Model

We assume an air purifier is used in the main bedroom only, and that the bedroom is occupied for 10 hours each day. We also assume that the windows are closed and that the room is fairly small - therefore we will use the more optimistic estimate of a 72% reduction in PM2.5 versus baseline conditions (referenced previously). This yields a total reduction in personal exposure of 30%, or a ratio between personal mitigated and personal baseline of 0.7.

Step 2: Comparison with real-world personal exposure data

We’ll look at studies with people carrying portable measurement devices, and compare them with our calculation of a 0.7 ratio:

Health Benefits delivered by this reduction

With the reduction ratio calculated above, we can calculate the health benefits. As described above, we will assume a linear relationship between average PM2.5 exposure and health effects.

From here onwards, we will consider two locations as examples: Schwechat, a suburb of Vienna, with an average of 15µg/m³ PM2.5, and Muzaffarnaga, India with an average of 89µg/m³ PM2.5 (vales from For the health effects, we use the values from the EU and India studies listed under “existing work” above.


Device and Filter Cost

Device Cost

Air purifiers can cost as little as $80. The studies cited above demonstrate that even cheap devices are sufficient for good results. For device lifetime, we assume 10 years.

Filter Replacement Cost

Performance of the air purifier depends on the condition of the HEPA filter - it needs to be replaced regularly, depending on the amount of particles the filter has already removed. I could not find specific information on how long a HEPA filter lasts, most sources say “about one year”. A test in Beijing revealed a -50% effectiveness drop after 200 days of use, which would mean one year might be too optimistic in high-pollution situations. For our two scenarios of India and Austria, we will assume filter lifetimes of 0,5 years and 1,5 years, respectively.

A HEPA filter costs around $10. It is possible to buy cheaper ones for $4.25 but they perform significantly worse.

Electricity Cost

Air purifiers might use 30-50 Watts (source). In our model, we use local residential electricity prices of our two sample locations, India and Austria.


Putting these numbers together, we arrive at $5230 per DALY for India and $15200 per DALY for Austria (model here).

Placing these numbers in context:

For multiple people living in the same space, cost would go down accordingly. For example, for a bedroom shared by two people, cost per DALY would halve. For a five-person family sharing a small flat, with the purifier placed in the main living area, cost effectiveness might be even better.

Ways to improve the cost-effectiveness

  1. Homemade devices: An air purifier is basically just a fan and a passive, replaceable filter. It is possible to build a perfectly adequate air purifier with just that. Someone has tested this setup in China and gotten very favourable measurements. However, since good air purifiers are available for $80, homemade devices may not change the calculus much.
  2. Cheaper HEPA filters. However: according to this blogpost, none exist that are much cheaper yet perform well.
  3. Targeting the right people: By installing devices in the homes of people who suffer from chronic respiratory disease and who stay home more than the average, we could somewhat increase the health benefits delivered.
  4. Timing: Pollution varies significantly depending on weather and season. Cost savings might be achieved by using the air purifier only on days of very high pollution. This is questionable though, because of the sublinear scaling of health effects and because HEPA filters degrade based on the amount of particles trapped (so using them only on high-pollution days would not reduce filter cost by much).
  5. Location: By placing standalone air purifiers in offices or schools, significantly better cost-effectiveness might be achieved.
  6. Integration with ventilation systems: Ventilation systems of public buildings could suck all incoming air through HEPA filters. This is already done in some places, but not everywhere. I’m unsure how much better the cost-effectiveness would be compared to standalone devices.
  7. Placement: A HEPA filter can remove more than 99.9% of particles. Therefore, outgoing air from air purifiers is almost completely free from any particles ( If someone were to sit or sleep directly in front of the device, it’s plausible PM2.5 levels could be reduced down to almost zero. Possibly a low-powered air purifier, directed at a persons face while they sleep or work, might deliver very large reductions in PM2.5 while requiring very little power. In the summertime, many people use fans for personal cooling - a very cheap intervention would be to strap HEPA filters to those ventilators. This will naturally reduce the cooling effects, but it might be a very cost-effective way to roll out air purification. Furthermore, HEPA filters could be made mandatory for air inlets in cars.
  8. Power consumption could be improved by building a more energy-efficient device.

Virus removal

HEPA filters are good at removing very small particles. This includes viruses like SARS-CoV-2. It is unclear whether this would have any significant protective effect in practice, if a household is shared with an infected person.

Removal of viruses, bacteria and spores might yield additional positive health effects not already considered in our calculation. Possibly a long-term large-scale air purifier study would be required to measure these effects.

Negative side effects of this intervention

Open Questions

... about air pollution in general

... on health effects

… on the practicalities of air purifiers

Conclusions and Further Work

Air pollution is one of the biggest public health problems of our time. Simple air purifiers are surprisingly effective at reducing the harm. In our sample calculation, the intervention easily meets WHO criteria for a “highly effective” intervention in Austria, and the criteria for an “effective” intervention in India. With just a few small improvements to cost-effectiveness, it would qualify as “highly effective” in India too.

There are many ways in which effectiveness could be improved: If the bedroom is shared by two people, effectiveness doubles. Our calculations were made for 10 hours per day of use. Many people stay home for longer than that, and would correspondingly benefit more from an air purifier in their home. It is plausible that we could find more energy-efficient devices and optimize location, placement and timing. Furthermore, devices could be preferentially given to individuals which are at special risk of pollution-induced illness.

Buying air purifiers for people to place in their homes is probably not a promising EA intervention: Cost-effectiveness is two orders of magnitude worse than GiveWell-recommended charities. That being said, there might be much more cost-effective ways of helping people get access to air purifiers. We might lobby governments to subsidize those devices, or to make HEPA filters mandatory for public buildings and vehicles.

Beyond air purifiers, we could probably find other interventions for mitigating air pollution that are significantly more cost-effective.

I’ve been quite surprised by the results. It seems that using an air purifier has solid health benefits, both in very polluted and in averagely polluted locations. It is surprising that in affluent countries, where people can easily afford these devices, air purifier use is not commonplace. The health benefits are clear and well-studied. I have installed a homemade device in my bedroom, together with a PM2.5 sensor, and plan to place a second device in the office.

If you're interested in air pollution, air purifiers, or would like to collaborate on future research please get in touch.


For their comments and feedback, I'd like to thank Andrés Gómez Emilsson (who previously mentioned HEPA filters in his post on Cause X), Boyang Xia, Cameron Earl, Gernot Ohner, Hannah Metzler, Helene Kortschak, Lorenz Krüger, Matthew Dahlhausen and Matthias Samwald.

  1. Details on HEPA filter efficiency on various particle sizes can be found here: 1, 2 and a simple graph can be found on Wikipedia. ↩︎

  2. Disability-adjusted life years, consisting of years of life lost plus a weighted sum of years spent in disability. ↩︎

  3. With a life expectancy in India of 69 years, this works out to 1.6*365/69/89 = 0.095 days of life lost per year and per extra µg/m³. When compared with the EU values above, which are for a PM2.5 range that is much lower than in India, this demonstrates sublinear scaling of health effects at higher levels. ↩︎

  4. In order to estimate years of life lost, the extra mortality risk is not enough. We would need two additional pieces of information: The age distribution in the population, and the extra mortality risk per age group ↩︎

  5. We have seen that reducing average exposure by x% reduces health effects by less than x%, because of the sublinear scaling. However, in our proposed intervention, we’re not only reducing average exposure, we’re also shifting the distribution of exposure over time: The hours spent at home will be in an environment of greatly reduced PM2.5, the hours spent elsewhere will be at unchanged exposure levels. If we assume that health effects accumulate linearly hour after hour and that the sublinearity of effects applies to each hour individually, this might mean that an average reduction of x%, delivered in this way, would reduce health effects by more than x%. ↩︎


Comments sorted by top scores.

comment by algekalipso · 2020-09-11T07:55:40.109Z · EA(p) · GW(p)

Now that the California fires are raging, it may be time to bring up a few additional reasons why HEPA filters make a lot of sense. I don't know how much this changes the cost-benefit analysis, but I think it is important to take into account:

1) Right now the PM2.5 outside my apartment is 230. Inside it's 40. A week ago the PM2.5 was 100, and inside it was 8. By having a HEPA filter inside, I've been seeing reductions of PM2.5 between 80% and 90%. I also saw this two years ago, and it's been a rather consistent pattern.

2) The idea that non-linearity makes the benefits strictly less than linearity, and therefore that assuming linearity will lead to an optimistic assessment is questionable. In particular, I grant this is true with "diminishing returns" curves. But it's not true with S-shaped curves. So, if it is true that the negative health effects of PM2.5 are concave below 20 and convex above 20, then the assumption of linearity will lead to an underestimation of the positive health benefits of HEPA filters for places with relatively clean air.

3) As a special case of (2), I would expect that giving your lungs "time to breath" (so to speak) might be really good to let them heal, and also allow your cardiovascular system to recover from inflammation. So there may be some extra benefits to being in places that have close to 0PM2.5 for at least some periods of time. And lastly,

4) I do think that the case for massively reducing the economic cost of HEPA filters should be considered more thoroughly. If subsidized at the governmental level, how cheap could these filters become? My suspicion is that they can become extremely cheap, turning them into a utility.

Thank you for the analysis and for bringing this topic to attention of EAs (whose saved micromorts may, well, ultimately have compounding benefits for all). Cheers!

comment by Denkenberger · 2020-08-09T18:38:43.026Z · EA(p) · GW(p)
Putting these numbers together, we arrive at $5230 per DALY for India and $15200 per DALY for Austria (model here).

This is actually a very good deal in developed countries, where they typically pay ~$100,000 per DALY. This would imply that nearly all non-rural buildings in those countries should have HEPA filters, especially because there are economies of scale. The fact that they don't I think indicates that the economics are not nearly as good when you take into account ventilation (related to Ben's comment). Also, I think there is very large uncertainty in the health benefits.

comment by Benjamin_Todd · 2020-07-30T17:02:54.296Z · EA(p) · GW(p)

Thanks! I'd be keen to hear what reduction in levels you can measure in your home, with an without the filter over a couple of weeks. I worry that the studies will be overly optimistic.

I also worry that sleeping with the windows shut is a bad idea due to this: [LW · GW]

comment by amandastone ( · 2020-08-27T05:41:06.673Z · EA(p) · GW(p)

That's a very nice article. Thanks a lot!

You've mentioned Honeywell 18155 ( )

"Amount of reduction: -55% in winter, -65% in summer"

Its coverage area is about 200 sqr.ft

I'm looking at the other model - HPA300 ( )

Its coverage area is twice bigger.

Question is: will the amount of reduction be lesser or bigger?


comment by martin_glusker · 2020-07-31T21:28:20.656Z · EA(p) · GW(p)

Thanks for the write up, super informative and interesting. I’m an undergrad tentatively planning on writing my undergraduate economics thesis on air pollution, perhaps focusing on indoor air pollution. In my previous informal research I gathered that air pollution wasn’t cost effective relative to other EA interventions, as you conclude above. The problem is compounded by the fact that that those use indoor open fires for cooking, and are therefore typically the most impacted by indoor air pollution, are often lacking access to electricity and therefore supplying air filtration devices becomes significantly more expensive and complicated (Citation needed).

One line of research that I’m particularly interested in is focusing on the cognitive impacts of air pollution. Given previous EA arguments on the value of increasing worldwide IQ and the high estimated value of increased IQ to the economy, incorporating the cognitive impacts of air pollution into any cost-effectiveness estimate seems important. This is obviously difficult to measure, and there is only preliminary research on the cognitive impacts of air pollution, mostly in the form of test scores, which don’t necessarily translate easily into IQ points. One particularly interesting working paper on indoor air pollution is Gilraine 2020 which exploits a natural experiment that put air filters in a subset of schools of the LA unified school district and finds some gains in test scores because of it. Another potentially cost effective education related air pollution intervention is retrofitting highly polluting diesel school buses in the US, as seen in Beatty and Shimshack 2011. If I recall correctly, both papers offer rough cost-effectiveness estimates as it relates to student achievement that may be helpful.

I didn’t have time to write up a full post, so apologies for not being totally thorough. Would be interested in collaborating!

Replies from: InquilineKea