|
Originally Posted by DaveReidUK
(Post 10477890)
It's perfectly possible that the ability distribution is actually skewed, in other words the peak isn't halfway between the two extremes, so that for example there could be a concentration of pilots towards the upper end of the ability scale. Or, perish the thought, the lower end. :O With a sufficient number of thousands of individuals, the distribution always seems to more or less follow the above curve, whatever the criteria, be it body height, fish size, driving or piloting "abilities". The notion that some professional pilots could not be up to "average" may seem a bit disturbing, and yet we all know some colleagues that are "outstanding". And unfortunately, some others are not as good, so this de facto creates a distribution where some are "above average" and some are below... |
Originally Posted by tdracer
(Post 10477514)
Now, I'm not suggesting we're to the point where we can design the aircraft so we don't need pilots - that's still decades away. But when pilots become completely overwhelmed and demonstrate the inability to even remember to pull the throttles back so they don't overspeed when something goes wrong, it moves us one more step in that direction.
However, Do I have the confidence in the implementation quality and regulatory enforcement which makes that technology actually safe ? . . . No chance. There is a 'race to the bottom' in the software business which largely makes it fundamentally incompatible with the creation of high end safety critical solutions. The more complex and critical those systems are the more incompatible that approach becomes. I see it regularly in many industries and it is both concerning and somewhat eye opening when you see what is really going on under the skin, as may be becoming apparent with this aircraft. |
Originally Posted by DaveReidUK
(Post 10477890)
An equally valid question would be why ability (at anything, not necessarily confined to piloting) should be assumed to follow a normal distribution (normal in the mathematical sense, i.e. symmetrical about the mean/median/mode).
It's perfectly possible that the ability distribution is actually skewed, in other words the peak isn't halfway between the two extremes, so that for example there could be a concentration of pilots towards the upper end of the ability scale. Or, perish the thought, the lower end. :O But in reality, it's very likely very close to normal. Take an example: how is the height of NBA players distributed? We pick the best, tallest athletes from the whole population, for sure they will all be best of the best, right? There are just 500 NBA players, out of the population of 300,000,000, that's top 0.000167%. For sure their height can't be normally distributed?? Well, spoiler alert, their height is basically normally distributed. https://cimg9.ibsrv.net/gimg/pprune....9b15db5924.png Ok, actually this is showing % of minutes played by different heights, so one could expect it to be even more skewed, but it's not. And that's just 500 players. There are 600x more airline pilots, and with the sample size like that, the distribution is going to be normal. |
Originally Posted by derjodel
(Post 10477919)
And that's just 500 players. There are 600x more airline pilots, and with the sample size like that, the distribution is going to be normal.
|
Originally Posted by DaveReidUK
(Post 10477998)
OK, I must have missed the lesson in statistics class demonstrating that large distributions are always normal.
1. How exactly does the type of distribution make my argument invalid? Plug in whatever you like, it's still the same. 2. Yes, indeed you might have missed the Central Limit Theorem. And because skill is sum of many different factors, the pilot skill is even more likely to be normally distributed. |
I am not a pilot, but I am a Chartered Engineer, Chartered Statistician, a Chartered IT Professional, a European Engineer, and a Fellow of the Royal Statistical Society, with a Doctorate in Operational Research and Mathematical Statistics.
Not all natural distributions are Normal. However, very many are, because of the Central Limit Theorem. In essence, if a physical measurement is the sum of smaller figures, it is very likely to be Normal. That's why heights of people are approximately Normal, IQs are roughly Normal, etc. It would be very surprising if the performance of pilots differed much from a Normal distribution. A Normal (or Gaussian) Distribution is entirely described by its mean and variance. So, unless the distribution of performance of pilots is very unusual (such as a Laplace Distribution), the only relevant features of the distribution of pilot performance will be the mean (or average) and the variance (which determines how many are in the "tails" and how long the "tails" are). |
The word average (skilled) pilot should be avoided for the purpose of this discussion. If we are true to the mathematical meaning, adding even a small number of extremely well proficient people to a group, the previously "average" pilots of that group become less-than-average.
Less than average skilled sounds like "not competent enough to fly an aircraft designed for an average pilot". Which itself is an element of the debate, thus the word "average" makes a big mess of everything here. My suggestion is that "the average pilot" in the certification realm means "any pilot who passed the licence proficiency criteria with at least bit of a margin, is able to achieve that or better performance continuously (as his experience builds but memory fades), and is not having a bad day" |
Recent discussions have revolved around the man and machine. The inconclusive (different) views reflect the difficulty in defining the problem, together with the natural human dislike of uncertainty. These are further restricted in forum communication, how best to convey judgements with text, especially where these depend on the task being assessed and operating environment, each with a range of interpretations. There are views of the man or machine, and then the ‘vague’ interface between them. Each attempts to quantify parameters of reliability and/or performance; similarly the concept of average seeks quantity, but most cases the real world the divisions are subjective - uncertain - qualitative judgement. (Skill is not a number, it’s a subjective rating). Numerical or statistical assessments are meaningless without first constraining the contributing parameters. This is like an experiment judging human performance for a given task (detect trim runaway and act), but where the environment defining the situation is primarily determined by the test subject - the experiment is unbounded. We cannot measure what the crew perceived or ‘felt’ about the situation, or know what they knew - or recalled, or why they acted as they did. All that might be concluded is that the performance of the single ‘entity’, the human and machine (humach?) in the circumstance was insufficient for the task. The balance of contributions in improving ‘humach’ capability is a judgement, best made within the guidelines of certification and the wide range of experience in that process. The recent accidents suggest that the overall process for this judgement (certification, design, and test) was flawed; the resultant uncertainty in system failure could not be managed by ‘humach’. The certification process is being reviewed, thus should identify technical and human aspects to be reassessed, as should be the resultant uncertainty from system failure. We may not be able to judge any of the above without greater understanding of the original certification, the accident system failure, proposed modification, and most of all the associated justification. With that, the subjective discussion in this forum and elsewhere could be founded on fact - reducing some of uncomfortable ‘uncertainty’ and need to resort to quantities to describe uncertainty. |
Statistics
Sorry, I don’t buy that graph. Too convenient. If we have over 7 billion people in the world and just over 50% are in working age, and there are 300,000 odd pilots in the world, then there is one pilot per 14,000 people roughly - a small number, even if the figures are da-neben. Let’s say there is a similar graph of 7 odd billion people. Where do you put these few pilots on it? And I am talking about selected airline employed professionals. What do you measure on such a graph? Intelligence? Co-ordination? Leadership? Team spirit? Drive? Fitness? Imagination? All these factors bear on a pilot’s suitability as well as other, conflicting things such as: Ability to concentrate on an item vs ability to keep the big picture Ability to stick to procedure vs ability to throw away the rules and improvise Ability to command vs ability to take advice Ability to insist vs ability to be patient. Ability to consider carefully vs ability to act fast and many other conundrums which don’t mean you are especially clever - but suitable. If you could quantify and measure all those qualities I very much doubt you would get a nice neat rounded peak for the general population as you would for an intelligence alone graph and even if you did, if you take a bunch of people out of it and scan them, the shape of the new graph would depend on where they were on the old one. Still a nice hill in the middle? I don’t think so. Now about getting a licence - that is one small part of getting an airline job - one which you have to have but not one which will cause an airline to take you - they are looking for the items above. So to design a machine for anyone who has managed to get a licence is not what is required. The operative word is professional. Design a machine for the pros and train them on it, then you will have safety. Anyway the Jodel is a good ship - and as we know old men think there are lies, damn lies and statistics! ^_^ B |
Originally Posted by derjodel
(Post 10478011)
You are trying to make the whole arument invalid on a premise of "pilot skills don't necesseraly follow normal distribution.
1. How exactly does the type of distribution make my argument invalid? Plug in whatever you like, it's still the same. 2. Yes, indeed you might have missed the Central Limit Theorem. And because skill is sum of many different factors, the pilot skill is even more likely to be normally distributed. To become a *commercial* pilot you need ability to take orders, ability to learn, to do CRM, to be methodical etc etc. And a certain ability to fly the plane. And all these factors are tested for. But on the day HAL decides to take a vacation or turn hostile, you suddenly need grace under pressure, a capacity to retain situational awareness, an ability to hand fly the plane in a configuration which has *not* been trained for, a will to find some way to survive. See Sully.. The problem with the aircraft design may be that the designers assume that the pilots have been selected adequately because of the shape of the large distribution, and find out that the distribution of skills necessary in an emergency is actually a bell shaped but much sharper curve with few individuals matching the criteria necessary for survival. Edmund |
Originally Posted by FlightDetent
(Post 10478028)
The word average (skilled) pilot should be avoided for the purpose of this discussion. If we are true to the mathematical meaning, adding even a small number of extremely well proficient people to a group, the previously "average" pilots of that group become less-than-average.
Less than average skilled sounds like "not competent enough to fly an aircraft designed for an average pilot". Which itself is an element of the debate, thus the word "average" makes a big mess of everything here. My suggestion is that "the average pilot" in the certification realm means "any pilot who passed the licence proficiency criteria with at least bit of a margin, is able to achieve that or better performance continuously (as his experience builds but memory fades), and is not having a bad day" That's why I'm saying, designing airplanes for average pilot is very, very bad idea. Planes need to be designed for any certified pilot. And to make discussion clearer, perhaps it would be good idea to stick to those terms. Because once you accept that planes must be designed for all certified pilots, then you can not, by definition, blame pilots to not react properly, specially in this situation. It's actually the other way around - this crash is giving us a data point how a certain crew of certified pilots reacted in under specific circumstances. Actually we have a few more data points: penultimate Lion Air crash, ultimate Lion Air crash and ET crash. All are certified pilots. They reacted differently. Was the system designed to accommodate any certified crew? And I'll go further and ask: isn't it the fact that the crew can not be responsible. Either they should not have been cleared to fly (e.g. stripped of their license, fail the MAX rating). Either they meet the minimums or they don't. If they don't and they still got licensed, it's not their fault. It's the regulator's. If they crashed because the plane was not flyable by any certified crew, it's boeing's fault. The crew could only be at fault if they were under influence, if they cheated to pass the tests or hide any medical conditions etc... |
d.j., your last paragraphs are missing the case of a properly qualified and competent pilot severely underperforming on a certain occasion. At this point, I am not connecting any dots to the MAX incidents and crashes as far as the crew performance goes, though a general remark that a poorly designed and implemented aircraft system (such as the MCAS on the MAX) will tip the scales.
|
Originally Posted by compiforce
(Post 10478024)
I am not a pilot, but I am a Chartered Engineer, Chartered Statistician, a Chartered IT Professional, a European Engineer, and a Fellow of the Royal Statistical Society, with a Doctorate in Operational Research and Mathematical Statistics.
Not all natural distributions are Normal. However, very many are, because of the Central Limit Theorem. In essence, if a physical measurement is the sum of smaller figures, it is very likely to be Normal. That's why heights of people are approximately Normal, IQs are roughly Normal, etc. It would be very surprising if the performance of pilots differed much from a Normal distribution. A Normal (or Gaussian) Distribution is entirely described by its mean and variance. So, unless the distribution of performance of pilots is very unusual (such as a Laplace Distribution), the only relevant features of the distribution of pilot performance will be the mean (or average) and the variance (which determines how many are in the "tails" and how long the "tails" are). Had statistics class in college about 3 decades ago..... |
Statistics are used by most people - especially politicians - just as drunks use lamp posts:-
... more for support than for the light they shed! |
Originally Posted by derjodel
(Post 10478099)
That's why I'm saying, designing airplanes for average pilot is very, very bad idea. Planes need to be designed for any certified pilot. And to make discussion clearer, perhaps it would be good idea to stick to those terms.
Because once you accept that planes must be designed for all certified pilots, then you can not, by definition, blame pilots to not react properly, specially in this situation. It's actually the other way around - this crash is giving us a data point how a certain crew of certified pilots reacted in under specific circumstances. Actually we have a few more data points: penultimate Lion Air crash, ultimate Lion Air crash and ET crash. All are certified pilots. They reacted differently. Was the system designed to accommodate any certified crew? And I'll go further and ask: isn't it the fact that the crew can not be responsible. Either they should not have been cleared to fly (e.g. stripped of their license, fail the MAX rating). Either they meet the minimums or they don't. If they don't and they still got licensed, it's not their fault. It's the regulator's. If they crashed because the plane was not flyable by any certified crew, it's boeing's fault. The crew could only be at fault if they were under influence, if they cheated to pass the tests or hide any medical conditions etc... |
Originally Posted by hans brinker
(Post 10478156)
Q: If the average applicant scores 100 points on the pilot qualification test, made up of theoretical questions, and a simulator check, you say they should probably normally distributed. Lets say 60 % score between 90 and a 110 points, and the airlines only hire the pilots with over 90 points. With the 20% percent who didn't cut it missing, wouldn't the resulting curve be non-normally distributed? (and seeing hiring rates around 50% at interviews makes me think the skewing would be worse)
Had statistics class in college about 3 decades ago..... You could also think about how consistent such a cutoff would be implemented, probably the hiring decision also depends on non-piloting factors. In the end, the reason a pilot is given a license is because it is thought he is able to fly a plane. Obviously this needs to be consistent with what is required to actually do it, or there is a problem. |
Originally Posted by TehDehZeh
(Post 10478358)
In the end, the reason a pilot is given a license is because it is thought he is able to fly a plane. Obviously this needs to be consistent with what is required to actually do it, or there is a problem.
|
Originally Posted by hans brinker
(Post 10478156)
Q: If the average applicant scores 100 points on the pilot qualification test, made up of theoretical questions, and a simulator check, you say they should probably normally distributed. Lets say 60 % score between 90 and a 110 points, and the airlines only hire the pilots with over 90 points. With the 20% percent who didn't cut it missing, wouldn't the resulting curve be non-normally distributed? (and seeing hiring rates around 50% at interviews makes me think the skewing would be worse)
Anyway, average is probably the wrong measure of central tendency. Better is almost certainly the median: by definition, half the population will be at or below the median value. Clearly, aircraft have to be readily manageable by the half of qualified pilots who fall below the median--especially aircraft like narrow-body twins, which are likely to fly the bread-and-butter routes that don't attract the profoundly gifted. The issue, surely, is not to do with statistics, but with what the cut-off level is for getting, and maintaining, qualification (and it would be good if there were some measure of variability--you don't really want to have in charge of an airliner someone whose flying is sublime some days, but a total mare on others.) |
Average etc.
In past years in my airline pilots were qualified after each check as Above (later changed to High) average, Average, Below average (possibly requires extra training) and Failed (out if happens twice). The average was supposed to be the average for that particular airline, which was supposed to be pretty good. However this changed to Qualified and Not qualified (which required remedial action). I think that is what we have to look at - not whether someone has a licence - we all have a licence - but whether one is qualified to the professional standard demanded by the airline and stipulated by the manufacturer. I for one would not like to be a pax in an airliner piloted by two guys who had a licence but had not passed stringent company requirements. And by the way this qualification was applied on check rides and sim checks but there were also training (refresher) flights where all hell could be let loose to train combination failures. These were not qualified sessions but if someone screwed up further action (training) was given. |
Sim limitations
Originally Posted by GarageYears
(Post 10475680)
The crux of this discussion revolves around the simulator accurately representing the aircraft.
Was adequate data provided for the trim wheel loading? My guess is the data for the trim loading at high speed was NOT provided at low altitude at very high speed... I suspect that is true of all simulators, whether 737MAX. NG or whatever. - GY similar to demonstrated cross wind capability that becomes the cross wind limit. It is the limit because they had no incentive to go higher and risk an airframe. |
| All times are GMT. The time now is 17:33. |
Copyright © 2024 MH Sub I, LLC dba Internet Brands. All rights reserved. Use of this site indicates your consent to the Terms of Use.