Dear Pablo, I understand you. At the beginning I had the same problem in realize the true deep meaning of the percentile, because the concept is not so simple and immediate; so I'll try to explain you how it works. It is important to note that this term is referred to a distribution of values, differently to the average value. In my opinion the rationale of this value is to guarantee a "reasonable" space of certainty that a condition could be satisfied.

In my experience I use the 99th (for instance) percentile of an calculated indicator, to assess the value that may guarantees that in the 99% of cases the imposed condition is satisfied (is true), with a chance of 1% that this is not true.

For further deepening please refer to this link: 

http://onlinestatbook.com/2/introduction/percentiles.html

http://english.stackexchange.com/questions/74329/percent-vs-percentile

http://apmblog.dynatrace.com/2012/11/14/why-averages-suck-and-percentiles-are-great/

I hope this may help you 

Cheers Valerio

I imagine this reflects available toxicological evidence that shows that hourly concentrations at that level can affect the health of vulnerable groups. Such a standard would then ensure that the health of susceptible people is also protected even if the average concentration is acceptable.

Yes Luc, you are on target. In fact I adopted percentile values in Critical Load estimation to asses the ecosystem sensitivity to pollutant depositions, such as nitrogen, acidity and heavy metals.

It is an effort to make the standard protection.  But the choice of percentile often varies across standards and the Environmental Protection Agency does not provide good reasons for that.  Related to this percentile issue is the issue of exceedances.  Some of the standards allow a certain number of exceedances of the standard determined by a particular percentile.  For discussion of these issues, see my article with Michael Livermore, Rethinking Health-Based Environmental Standards.

Dear all,

Thanks a lot for all answers. I will try to clarify my question. I understand that there are short term standards to protect people from accute effects of air pollution. However, my question focus on the use of percentiles. This percentiles would allow to a number of exceendances that would not be considered so that no long term responses have to be taken (let´s say no air pollution control plan). As I understand this is because we are not interested in just one event that is high but that the upper tail of the distribution is high, in other words, it was not just an isolated event or the sum of  many "bad luck" situations (very cold days, very low dispersion days, and such). But, I have had troubles to find "official" literature stating the logic, as Richard comment in his note, EPA does not provide good reason for the use of percentiles. My context is that I am working in a report for our government and I need to explain as clear as possible the basis of using percentiles in regulation. So, any info, paper, book chapter, explaining the logic of using percentiles in regulation would be greatly appreciated.

Pablo  

EPA does not provide good answers and does not use consistent percentiles or a consistent number of exceedances.  In essence, the choice is arbitrary.  My article (referred to in my prior post) analyzes this issue in some detail.

The US/EPA uses percentile forms for its PM10 (99th percentile) and PM2.5 standards (98th percentile).  I agree with Richard's view that EPA does not offer any substantial reasons for choosing these forms.  The 8-hour ozone standard is effectively a 99th percentile form (4th high in each year = 353/365 = 99.2%).

The upper tails of PM distributions can be highly skewed (a few very high values compared to the rest of the distribution).  That is probably why a few values are not considered (about 7 each year for a 98th percentile form).

This leads to some public confusion, because the level set by the standard (such as 35ug/m3) is not explicitly the "safe" level.  The level of the standard and the form of the standard work together as evaluated by the "health experts" to provide a protective standard.  So, in a three year period, there could be 20 days on which the level is exceeded, and that would be safe. 

I think everyone understands how percentiles are better than averages.

However, I have another question. Let me quote from the Philippine's Clean Air Act 1999:

PM-10 24-hr average to be limited to 150 µg/m³ with the rider:

"Maximum limits represented by ninety-eight percentile (98%) values not to exceed more than once a year."

Of how many previous values do I take 98th percentile of? How wide can be my data set? Last 1 month gives me 30 readings (wherein 2nd highest percentile (0.6 readings) makes no sense), last 3 months gives me 90 readings (of which I can exceed in 1.8≈2 readings), or whole year which gives me 365 readings (of which I can exceed in 7 readings). You get the idea.

If there is any reference of this in any national/international regulation, it would be helpful.

Devansh Sharma

Regarding your question about the short term daily limit value of 150 µg/m³ with the averaging time of 24 hours:

Since it is a short term limit value, so it should be considered as a daily average value and it is written that it is an averaging time of 24 hours. So, I would consider that the measurements are done on a higher resolution than 24 hours. It depends on the measurement principle that how high your time resolution can be for the measurements performed. For the gravimetric principle it would be difficult as the values are usually on a daily average basis but for TEOM or light scattering principle it is possible to have a time resolution as low as one second. As far as I understood, from these high resolution values that you obtained throughout the day, you should consider the 98th percentile of these values throughout that day and it should not exceed the limit of 150 µg/m³. However, you are allowed to exceed this value once in a year which might happen on the first day of the year due to new year fireworks. The long term limit value is simply the arithmetic mean of the whole year and it should not exceed the value of 60 µg/m³.

I hope this answer will help you. Please let me know if you find another answer or explanation about that. Thanks.

Abdul Samad , thanks for the response.

I had the same understanding since a long time as well regarding percentile values. However, since our organization is legally responsible for compliance, I am looking for a references and standards in black and white.

If anyone can quote some National/International standards like ISO/EN/EPA/ANSI/GB-T/DENR etc., that would be helpful.