Statistical conundrum

[colin todd]

Big Al, while all that waffle (no disrespect intended Old Timer!) might be true, it is not what I was getting at.

 It is also true that, in medicine and biology, the 5th -95th percentile is deemed to be normal, so it is the case that, by definition, 10% of the population are abnormal.  (Most of them working in the civil service and my bestest not favourite FRS in the world).  All you would have to do is change the definition of normal from 1st-99th percentile and you would reduce at a stroke the number of abnormal people in the world.

The 75dB(A) used in fire alarm systems in rooms in which people sleep is based on the 95th (or it might have been the 99th) percentile of American students woken by this SPL in the 1970s.

But none of this was the point I was getting at, which I cannot reveal to you yet WeeB, as the bells have not  yet tolled to end the year.

I will have to give you a clue-think fire engineering (though the more I do think about it these days the more I become disillusioned by the smoke and mirrors).

[kurnal]

I guess I should have known better than to try. I should have remembered from experience  that Dottys arms are in the 5th shortest percentile and his pockets are within the deepest 95th percentile.
 

[wee brian]

I suppose, being abnormally tall just means you have to duck through doors.

But if your slower than the walking pace included in a fire engineers escape sums then your might die.

Saying that, you walk through doors all the time but might never be in a burning building....

This is all very deep, you'll start confusing yourself.

[kurnal]

And if a number of the occupants are overweight  they will take up more space extending the time taken  for the occupants to pass through exits and staircases leading to an extended evacuation time for everyone.
If they cant see very well they will move more slowly.
If they cant hear well their response time will be affected.
If they have learning disabilities this may affect response.
If they have consumed too much of Dotty's wine....no that's never going to happen ......

One thing leads to another....... and the beat goes on.

[colin todd]

Brian, you are getting warm, but are not quite there yet.

[wee brian]

I get bored easily

[colin todd]

Ok, let me put this point to you,  if the 1st percentile for pre-movement time was say 30 seconds, and the 99th percentile were 90 seconds, why on God's earth would any sane fire engineer regard total pre-movement time as 120 seconds and design accordingly.

 This would assume that the 99th percentile was the group of people between the 1st percentile and the 99th percentile (assuming you were trying to design for 99% of people), when in fact the 99th percentile already includes the people in the 1st percentile. 

Alternatively, it would assume that the 99th-100th percentile was the same as the 0-1st percentile, which clear is unlikely to be the case (as it is unlikely to be a normal distribution) and that you were trying to cater for 100% of people, which is not possible.

[Phoenix]

Because most "sane" fire engineers know nothing about statistics.

Most people, including fire engineers, think they understand statistics and can employ them usefully to some degree.  They're virtually all wrong.  Particularly journalists (and fire engineers).

You clearly have a particular case in mind here.  120 seconds might sound like it possesses a feasible safety margin above the 99th centile value of 90 seconds but rigorous sensitivity analysis of what might reasonably be expected to cause delays would be the only correct way to establish the best safety margin.

[wee brian]

I often get complaints from fire engineers (young ones) that building codes are full of magic numbers made up by committees with no basis in science etc.

I like to explain to them that the numbers they use aren't much better.

[colin todd]

Ashes, it is even worse than a particular case. PD 7974-6 Table C1, says that total pre-movement time equals the time for the 1st percentile to move plus the time for the 99th percentile to move.  A number of the worked examples then use the two added together (although some of them use different figures so maybe that is another error).

I agree with your view on people's handling of stats, but I was serious when I said that a Scottish O grade statistics would give you enough understanding of percentiles to appreciate what seems to me to be a schoolboy howler in the PD

I also agree with wee b (who is a cool dude for a civil servant).  I rarely admit it, but I was one of a cohort of 4 students who undertook the first ever university education in fire engineering in the UK. The stuff I see young (and sometimes not so young) fire engineers turn out today makes me ashamed.  A lot of it is smoke and mirrors designed to do little else than prove that black is white, often to suit a client or a court case.  Bring back old fashioned fire prevention officers and building control officers, but it will never happen.

[Tom Sutton]

I accept this is way above my head, but is there good available data to make all these assumptions it looks like C1 doesn't think so, as they say garbage in garbage out?

[colin todd]

Exactly, Tam.

[kurnal]

The stuff I see young (and sometimes not so young) fire engineers turn out today makes me ashamed.  A lot of it is smoke and mirrors designed to do little else than prove that black is white, often to suit a client or a court case.  Bring back old fashioned fire prevention officers and building control officers, but it will never happen.

Precisely Colin, for once I find myself in complete agreement with you. Have said it several times myself and have completely dissociated myself from several projects when I have smelt a rat. I am not a fire engineer, but always found basic  firemans  questions enough to confirm my concerns.

However getting the developer and client to agree at the design team meeting to spend more money when the greenhorn fire engineer ( sometimes working for the same group as the AI) actually believes everything his modelling shows him and the AI is happy to sit on the fence and accept the fire engineers report without reading it or understanding it  ( it's backed up by their PI insurance isn't it so if the engineering is wrong they will be sued first) .

That's the time to walk away from the client, there are plenty of others out there who expect and appreciate a more diligent approach.

We are where we are as a direct consequence of the privatisation of building control in the early 1990s. Vested financial interests will often take priority over good fire safety design.
I fully agree  with your final comment "Bring back old fashioned fire prevention officers and building control officers, but it will never happen" , so here's a New Years resolution for you Colin.

 It's clearly time to ditch your avatar message "civilianise enforcement- you know it makes sense"

Happy new year

[Phoenix]

Yes, I see what you mean, Colin.  It's not right but I can see that there would be some sort of pragmatic applicability of this approximation if the distribution of occupants' individual pre-movement times was symmetrical.  But, as you have already pointed out, these times are not symmetrical about a mean and, indeed, Figures C1 and C2 immediately above Table C1 show a number of distributions, all of which are skewed. 

Of course, the distributions approach the x-axis asymptotically which means, in practice, that we can never be sure how long it will take to get that last centile moving.

But there are solutions to the problems that can be used to achieve useful predictions and they depend upon the type of premises under investigation. In densely populated spaces where queues can be expected we're only interested in the first few centiles and in sparsely populated spaces we cannot predict the last centiles so we have to impose management regimes to control pre-movement times.

All parts of 7974 tend to over-theorise their respective subject matter, often with errors.

As for the general quality of fire engineering in the commercial world...well...I don't know where to begin...

[colin todd]

Ashes, it a basic fact that where the lowest centile of a distribution can be zero but cannot be less than zero and there is no theoretical limit to the upper centiles, the distribution is bound to be skew wiff.  The maximum possible pre-movement time is infinity, as people might never bother to evacuate. I did this myself once, while 'er indoors went and stood dutifully in the rain and ruined her hair do.

As for fire engineers, the reports I love are those than involve thousands of pounds of computer modelling to show that B1 is met in a supermarket with code compliant travel distances, so allegedly proving that sprinklers are not required, without even a notion that sprinklers would be necessary for B3 not B1.  I saw this from a large, well-known practice of fire engineers.

And what about using a soot load related to PU foam in a warehouse full of paper. Duh.

Enforcing authorities and BCBs should never trust fire engineers, but alas all you have to say to them is "its a fire engineering solution , guv" and any old tat gets approval.