Data and Test Details
The app asks what size the room is. How Xiaomi uses that data is a little opaque, but it would be logical that settings for larger rooms will run the purifier harder, so we used the highest allowable setting, 34-37m2.
It should be noted that the Xiaomi also has a turbo mode, which I’ve heard people describe as “sounding like a jet taking off.” This really isn’t meant to be used while people are in the room. Instead, it’s meant to be used temporarily before returning home.
Tests in a Different Room
We’ve used this same bedroom in Chaoyangmen, Beijing for many of our tests—of the Original DIY, Cannon, Philips, Blue Air, and IQ Air. So I doubt there’s something weird about this room that somehow hurts the Xiaomi but not other machines.
But you never know! So to be sure, we conducted an overnight test in the Smart Air office near Sanlitun. The results are shown in the noise test above. They show the same pattern as the other room tests. Thus, the results don’t seem to be something weird happening with that particular test room.
Tests with a Different Particle Counter
We’ve used the Dylos Pro for most of our tests, so this is constant across tests. But particle counters can break or lose accuracy over time. Thus, we carried out the noise/particle test above with a different Dylos Pro. The fact that the pattern of the results is the same suggests the results are not because of any problems with the particle counter. (That conclusion is also hinted at by the fact that the Xiaomi performed well for the first three hours while on high.)
How Bad Was Outdoor Air?
It’s important to analyze how bad outdoor air was during the tests because my analyses have shown it’s harder to achieve a high percentage reduction when outdoor air pollution is bad. (Or put another way, it’s easier to achieve a high percentage reduction on relatively clean days.) For example, here is the relationship between effectiveness and outdoor air pollution for the Blue Air 203:
However, theoretically, the Xiaomi auto mode shouldn’t be affected by outdoor air pollution. If it can accurately detect air pollution levels and turn on the fan in response, the results shouldn’t be affected by outdoor air pollution levels (until we get to levels that are too high for even constant high mode to clean). But to be conservative, I analyzed hourly outdoor air pollution data from the US Embassy, about 7 kilometers from Anna’s home.
During the auto mode tests, outdoor air averaged 128 micrograms. During the high-mode tests, outdoor air averaged 246 micrograms. Both of these are higher than the average in Beijing (90-100 micrograms). So it’s worth seeing how the Xiaomi did on days with lower pollution.
I analyzed the four days with lowest outdoor air pollution (average 89 micrograms; 11/29, 12/2, 12/5, 12/7). On these days, the Xiaomi averaged a 56% reduction in 0.5 micron particles and 87% in 2.5 micron particles. Thus, the poor results in the main tests do not seem to be because of outdoor air pollution levels.
Outdoor Air Fluctuations
Besides the baseline level of outdoor air pollution, it’s also important in these real-world tests to look for large fluctuations in outdoor air pollution. If outdoor air gets a lot worse during the test, it can look like the purifier is not cleaning the air very well. And on the flip side, if outdoor air pollution goes down a lot during the test, it can look like the purifier did a great job.
First, I analyzed the data after removing any test day where outdoor air fluctuated more than 100 micrograms from beginning to end (4 days total). That left 8 test days. The result was very similar: on these stable days, the Xiaomi average a 67% reduction in 0.5 micron particles and 86% in 2.5 micron particles.
Next, I analyzed all days where outdoor PM2.5 micrograms changed no more than 50% from baseline at any point during the test. This “at any point” criterion is more stringent, and it left three test days (12/11, 12/12, 12/18). Again, the results were nearly identical: a 69% reduction in 0.5 micron particles and 83% in 2.5 microns. In sum, the poor results did not seem to be caused by fluctuations in outdoor air pollution.
% Hours of Unsafe Air
To calculate the percentage of unsafe air for the Blue Air 203/270E (3,600 RMB), and Philips AC4072 (3,000 RMB), I used the data in my previously published tests. For the Cannon, I used three series of tests: my original tests, tests where tested whether adding a pre-filter affects performance (below), and tests comparing performance after adding a carbon layer.
That gives a lot of data! Here’s exactly how many hours of data I had:
|# Hours of Unsafe Air (>25ug)||
Average PM 2.5 During Tests (ug/m3)
% Unsafe Hours: How bad was outdoor air?
It’s important to compare just how bad outdoor air was during the tests. If one machine has lots of hours where air was just above 25 micrograms, it would be a lot easier for the machine to clean the air. So, I calculated the average outdoor micrograms for all of these unsafe hours. For reference, Beijing’s air has averaged about 90-100 micrograms for the last 7 years according to the US Embassy.
In the chart above, I analyze what the average outdoor PM2.5 concentration was for the different tests. The Xiaomi really stands out, and that could unfairly lower its results. Thus, I re-analyzed data only looking at days with lower outdoor air pollution, with an average of 138 micrograms—lower than the IQ Air. The result was similar to the original analysis: 83% of hours were unsafe.
To be even more conservative, I analyzed the two days with the lowest outdoor concentration, averaging 108 micrograms. That is lower than all of the other machines except the Philips. On these days, 68% of hours were unsafe. Thus, the Xiaomi was leaving far too much unsafe air, even on days with lower outdoor AQI.
Nerd note: In calculating the percentage of time of unclean air in the main text, I accidentally excluded the shortest test day (12/11). So I re-did the analysis with this data included, and the new result was that 85% of hours were unsafe. This oversight makes little difference, but I think it’s important to note for completeness.
The Xiaomi Test Compared to Other Recent Tests in the Same Room
I’ve published dozens and dozens of days of test data using a Dylos particle counter in this exact same room (1, 2, 3, 4, 5, 6). That makes me pretty confident these poor results are not some basic flaw of the test design. However, there is always the possibility that the particle counter will break or lose its accuracy slowly over time. Or maybe some neighbor is setting up a secret chuan’r stand nearby. Fortunately, I have data from just two weeks earlier in the same room with the same particle counter.
If something is wrong with the particle counter, we should get weird results for these tests. But over several tests with a new-and-improved Original DIY 1.1, the results are pretty much what I’d expect—a modest percentage higher than results for the DIY 1.0. Here’s a sample test day for the DIY 1.1 two weeks before the Xiaomi test:
That test data looks normal to me. And that suggests that there’s nothing strange going on with the room or the particle counter recently.
Converting Particle Counts to PM2.5 Micrograms
The tests used a Dylos Pro particle counter. To convert to PM2.5 micrograms, I took the 0.5 micron particles divided by 100. This formula comes from Dylos, and our prior tests show it’s pretty accurate compared to the US Embassy—a correlation of r = .90.
Original Data: Xiaomi
Below is the original data for the Xiaomi tests, including the outdoor PM2.5 levels as recorded by the US Embassy. My comparison data for the Blue Air, Philips, IQ Air, and DIY Cannon are available through in my earlier post. To request a copy of the original data in Excel format, send an email to [email protected].
Original Data: Noise + Particle Count Test
Here is the original data for the test done in a different room with a different Dylos while decibel readings were taken:
Original Data: The Xiaomi Can’t Stay on High?
Below are the transcripts from Xiaomi customer service where we asked whether or not the Xiaomi 2 can stay on high mode continuously.
Original Data: Cannon with HEPA + Pre-Filter
As a part of the calculation of the percentage of hours with unsafe air, I included tests of the Cannon with a HEPA + pre-filter. When I ran these tests, the question I wanted to answer was whether adding a pre-filter decreases effectiveness (because pre-filters add to air resistance, with the upside of lengthening HEPA lifespan). The data showed only a minor decrease in performance: 2% lower average room effectiveness for 0.5 micron particles and 1% lower performance for 2.5 micron particles.
Thomas is a new Assistant Professor of Behavior Science at the University of Chicago Booth School of Business and the founder of, a social enterprise to help people in China breathe clean air without shelling out thousands of dollars for expensive purifiers.