History, Politics And Current Affairs
http://www.tboverse.us/HPCAFORUM/phpBB3/

RAF Fighters V6.2
http://www.tboverse.us/HPCAFORUM/phpBB3/viewtopic.php?f=12&t=18112
Page 2 of 5

Author:  Johnnie Lyle [ Wed Sep 23, 2015 10:20 pm ]
Post subject:  Re: RAF Fighters V6.2

KDahm wrote:
Johnnie Lyle wrote:
KDahm wrote:
To really complete this, we need to have some kind of idea what % of flying hours training and the other phases of pilotage use. If, for example, training is about 10% of a typical pilot's experience, then 18% loss is bad. If it's 24%, then it's not so dire.


Sorta.

The oft cited figure is 90:8:2 - 90% of casualties are from pilot deficiencies (which roughly looks to be 60% physical and 30% skill), 8% aircraft deficiencies, and 2% enemy action. That is apparently based on an analysis of British 1914-1915 losses - where it came from is unknown. Using the stats from Hobson (Airmen died in the Great War, 1914 - 1918) on total RFC and RNAS killed, it's 59% KIA and 41% DWF (died while flying, which covers pretty much anything that killed you other than the Germans killing you while you were in the cockpit). Hobson reports 2266 KIA and 1554 killed while flying not from enemy action, and based on the RFC struck off lists, my WAG is 2195 killed in France, of which 313 were due to training ops. We can probably cross-reference the two lists and identify which of the not noted as killed really died, and which noted as a WAG kill survived (need more money and time for that). But, the error should not differ by aircraft type, assuming we have enough of an N.

Certainly, if our outcome of interest is "risk of death due to training" as an absolute measure, yes, we need a flight hour estimate. But if our interest is whether the Camel was deadlier to experienced pilots than other fighters, no, not really.

If the aircraft was more likely to kill an experienced pilot than others, the key value is the % of pilots/crewman killed per training accident that wrote off the aircraft. There, the Camel is solidly in the middle of the pack - the D.H.2 was equal, the Bristol F2B, the SPAD S.VII, and the Pup all had hgher relative fatality rates in training ops than the Camel among experienced personnel, while the Nieuport 11/16, D.H.5, Nieuport 17/23, B.E.2/B.E.12, F.E.2., Strutter, S.E.5/S.E.5a, Dolphin and Snipe were safer. We can't really conclude about the Triplane, the SPAD S.XIII, the F.B.5, F.E.8, the Morane-Saulniers, Nieuport 24 or 27, or the Bristol Scout because of insufficient numbers of crashes (probably due to lack of earlier records).

Of those, the only really important diferences (ie, greater than 5%) with the Camel are the Nieuport 11/16, Strutter, Snipe and S.E.5a.

Of the fighter planes with sufficient N, it looks like the SE5/SE5a was the safest - ie, the plane least likely to kill you. So it's pretty sound to conclude that the Camel's fearsome reputation as a man killer is really due to pilots who were still learning to fly, and possibly because it's contemporary (the S.E.5a) was a whole lot safer. Certainly the Camel wasn't that much more dangerous to an experienced pilot than the B.E.2, which was what most learned how to fly on.

So, we must amend our conclusion a little - the Camel was no more dangerous to an experienced pilot than most of it's predicessors (the exception being the Nieuport), or most Spowith products (except the Snipe and Strutter), but was more dangerous than it's contemporary, the S.E.5a.

Except that pilots killed per training accident results in the assumption that a training accident occurred. To really see how likely the plane is to contribute to the pilot death, the numbers have to include some measure for how much flying occurred for the plane type, ideally split between training and non-training.

A plane which has a low accident rate, but has a high number of pilot deaths per accident, may be a very safe plane to fly but faster and more fragile than others. A plane that has a high accident rate, but rarely kills it's pilot may have some nasty flight characteristics but is heavy enough to protect the pilot. Flight hours training/non-training would be one way of getting an idea, whether tracked by plane type or by pilot. Numbers of plane produced/shot down/crashed for other reasons would be another way. There may be insufficient data for any of those.

This is not to say anything bad about the work you've done on this, nor the wonderful work at http://www.airhistory.org.uk in collecting the data. I'm just trying to get a more complete picture. You know engineers - when one says that more data would be worthless, it's time to check his pulse.


I'm not sure we can easily get flight hour data. That would require going through all the pilot log books, and the routine orders books. Lots of pilot log books were destroyed, and the routine order books are dirty, dirty data with lots of duplicates. Cleaning that is hundreds of man hours, probably, and even then, we'll still be missing a lot of data.

We already have the training vs total aircraft written off - that is the column % Of Aircraft Struck Off In Training. That gives us a figure for training losses as part of overall losses. So we can compare that column, and, again, the Camel is in the middle of the pack. That indicates the plane wasn't any more likely to be struck off on a training op than any other aircraft.

The Snipe data is interesting - roughly half the Snipe's lost were lost in training, but the Snipe wasn't in combat service that long. So that looks like we're seeing the expected losses for a conversion/early combat period there.

We can probably refine the data by defining what a conversion period is, and restrict the analysis to only aircraft struck off during a squadron's conversion period. Effectively, that's normalizing everybody to the Snipe. That would require data on when each squadron converted to what aircraft, which is out there but tedious to get into a readily analyzable format. Even then, we'd be excluding replacement fliers. So it could go either way, depending on how high a unit's casualties were.

More would refine the risk estimate, yeah, but we'd probably get roughly the same answer to the question asked - was the Camel disproportionately deadly to experienced fliers?. If it is, then the Camel should have a disproportionately higher number of killed pilots, pilots killed in training, aircraft struck off due to training, or pilots killed in training to overall pilots. Instead, it's generally in the middle.

So, I think we can say that it's not that much deadlier than other WWI aircraft. If anything, it's comparable to the B.E.2 in terms of likelyhood to be destroyed in a training accident, and likelihood to kill somebody in a training accident - and the B.E.2 was the standard trainer by the end of the war. The interesting bit is comparing the Camel to the S.E.5a - and the Camel has higher number of fatalities per lost aircraft, a higher accident rate, and much higher fatality rates in training accidents, and training deaths to overall deaths.

In light of this, I think it's time to start digging into the Camel's lethal reputation, and see if that is overblown. Or if it grew up being compared to the S.E.5a, and not the RFC/RNAS/RAF as a whole. That would require a deeper analysis of training records, airman records to see who trained on what, and whatnot, since they are not readily available at present.

KDahm wrote:
BTW - Will you be sending this to http://www.airhistory.org.uk?


I hadn't thought about it, really. They mostly seem to have raw data, not analysis.

Author:  Belushi TD [ Thu Sep 24, 2015 8:34 am ]
Post subject:  Re: RAF Fighters V6.2

Ok.

Here's my take on how the Camel got the reputation.

The percentages of dead Snoopys fall into the middle of the pack.

However, the sheer number of dead Snoopys is due to the huge absolute number of losses. I'm assuming that there is a corresponding huge number of airframes produced.

As a result, the reputation is due to the sheer number of dead Snoopys, which is close to an order of magnitude higher than for most other models. So lots of people heard about their old pal SoandSo, who snuffed it while flying a Camel, and they told their buddy, who had another buddy who died in the same manner, and it multiplied.

Since there were not a comparable number of dead that occurred in other types, it became an article of faith that the Camel tried to kill its pilots.

Belushi TD

Author:  KDahm [ Thu Sep 24, 2015 9:26 am ]
Post subject:  Re: RAF Fighters V6.2

Belushi TD wrote:
Ok.

Here's my take on how the Camel got the reputation.

The percentages of dead Snoopys fall into the middle of the pack.

However, the sheer number of dead Snoopys is due to the huge absolute number of losses. I'm assuming that there is a corresponding huge number of airframes produced.

As a result, the reputation is due to the sheer number of dead Snoopys, which is close to an order of magnitude higher than for most other models. So lots of people heard about their old pal SoandSo, who snuffed it while flying a Camel, and they told their buddy, who had another buddy who died in the same manner, and it multiplied.

Since there were not a comparable number of dead that occurred in other types, it became an article of faith that the Camel tried to kill its pilots.

Belushi TD

That sounds reasonable.

Here's another thought about how to get the information I was thinking about. If a good sampling of the number of days between assignment and disposal/crashing for each plane type can be easily extracted, we can multiply that by the number of airframes produced to get an accident rate/plane/flight day. It's a really crude approach, but better than nothing.

Author:  Craiglxviii [ Thu Sep 24, 2015 10:51 am ]
Post subject:  Re: RAF Fighters V6.2

According to Wiki (I know!) 5490 Camels of all types were built.

Se5a- 5265 built.
SPAD X111- 764
Albatros D.V- 900
Albatros D.Va- 1612

There were more Camels built than any of its contemporaries, by either a small (SE5a) or a significant margin (anything else).

Author:  Belushi TD [ Thu Sep 24, 2015 2:16 pm ]
Post subject:  Re: RAF Fighters V6.2

This reflects the % dead and % struck off, doesn't it, along with the raw numbers?

The Se5A had a similar number built, but a lower number struck off, so IIRC, it was lower on the "deadly" scale?

Belushi TD

Author:  The Bushranger [ Thu Sep 24, 2015 6:12 pm ]
Post subject:  Re: RAF Fighters V6.2

Belushi TD wrote:
Ok.

Here's my take on how the Camel got the reputation.

The percentages of dead Snoopys fall into the middle of the pack.

However, the sheer number of dead Snoopys is due to the huge absolute number of losses. I'm assuming that there is a corresponding huge number of airframes produced.

As a result, the reputation is due to the sheer number of dead Snoopys, which is close to an order of magnitude higher than for most other models. So lots of people heard about their old pal SoandSo, who snuffed it while flying a Camel, and they told their buddy, who had another buddy who died in the same manner, and it multiplied.

Since there were not a comparable number of dead that occurred in other types, it became an article of faith that the Camel tried to kill its pilots.


This is extremely logical and, I'd argue, quite likely - and in fact it's related to the "statistics" that the Environutzies pull out to try to "prove" that, for instance, the manatees here in Florida are still endangered and need more (and more and more and more) protection. Their argument is to point at the numbers of 'manatees struck by boats' and say 'see? See how many manatees are being hit by boats? The numbers aren't going down, so we need more no-motor/no-boating zones to protect the manatees!'.

...the devil in this particular detail, of course, is that the number of boat strikes as a percentage of the manatee population has (pardon the pun) crashed. It's just that the number of manatees has risen substantially - and, thus, the raw numbers remain the same.

Of course, the Environutzies' response? To reject the data that says manatee populations has risen as being fabricated... :roll:

But it's the same principle: there's a Big Number of dead pilots in Camel crashes, so obviously the Camel was unusually lethal - but (as Johnnie has shown) once you correct for the fact there's umpteen bazillion more Camels than (almost) anything else, the Camel shows itself as no more deadly than (almost) anything else.

And I'm sure the fact that one of the few aircraft the Camel was rather more deadly than was the almost-as-plentiful SE.5 doesn't help, because it lets people point at "Dead Camel Pilots" vs "Dead SE.5 Pilots" and go "AH-HA! PROOF!".

Author:  Johnnie Lyle [ Thu Sep 24, 2015 8:07 pm ]
Post subject:  Re: RAF Fighters V6.2

Here's the other bit to consider.

If the Camel is God of the Air, Death to the Hun, and better than shagging a dozen supermodels, you're more likely to die crashing it than being shot down. Because Ze Germans can't touch you, total combat deaths go down, automatically increasing the % of non-combat deaths to total deaths, and the number of buddies who you can name that copped it in a practice flight.

Author:  pdf27 [ Fri Sep 25, 2015 1:15 am ]
Post subject:  Re: RAF Fighters V6.2

Actually, I think the comparison to the SE.5 is more relevant - between the two they made up the overwhelming majority of single-seat fighters on the Western Front, and clearly from the statistics the SE.5 is the safer of the two. Now, the classical accident pyramid suggests that for every fatality there will be a substantial number of injuries and a very large number of near misses (given the aircraft of the time the injuries will probably be artificially compressed though since the scope for only being injured is quite small). That suggests that there were a very large number of Camel pilots out there who had experienced their aircraft trying to kill them, and a far smaller number of SE.5 pilots (with a substantial number who crossed one way or the other and as a result were convinced that the SE.5 was safer to fly).
Now it seems rather obvious to me that the influence of live pilots on the reputation of an aircraft is going to be far stronger than that of dead pilots - you're aren't going to have a dead guy bending your ear about how dangerous such-and-such an aircraft was compared to everything else (when actually he only flew the SE.5 for comparison, but as this is 20 years later and he's big-timing it in the mess he isn't going to admit that). Not only that, but the live pilots will stick around for an awfully long time with their experiences while the dead sort of fade away a bit - their names are remembered, but by and large their cause of death isn't.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 1:19 am ]
Post subject:  Re: RAF Fighters V6.2

KDahm wrote:
Belushi TD wrote:
Ok.

Here's my take on how the Camel got the reputation.

The percentages of dead Snoopys fall into the middle of the pack.

However, the sheer number of dead Snoopys is due to the huge absolute number of losses. I'm assuming that there is a corresponding huge number of airframes produced.

As a result, the reputation is due to the sheer number of dead Snoopys, which is close to an order of magnitude higher than for most other models. So lots of people heard about their old pal SoandSo, who snuffed it while flying a Camel, and they told their buddy, who had another buddy who died in the same manner, and it multiplied.

Since there were not a comparable number of dead that occurred in other types, it became an article of faith that the Camel tried to kill its pilots.

Belushi TD

That sounds reasonable.

Here's another thought about how to get the information I was thinking about. If a good sampling of the number of days between assignment and disposal/crashing for each plane type can be easily extracted, we can multiply that by the number of airframes produced to get an accident rate/plane/flight day. It's a really crude approach, but better than nothing.


That requires a LOT of work. We'd need the individual squadron war diaries to get when a plane was actually issued from the air parks to the squadrons, and they're not readily available. We do have combined fate files for most RFC aircraft, and combined movement files, but the data is dirty and would require lots of man hours (hundreds) to clean, since it's almost a line by line process.

On further discovery, there's a lot of duplicate data in the struck-off lists. So the original analyses are inflated. Cleaning that up will require work, since it's going to require a lot of hand examination. I also found possible undercounting of training losses.

We have month of despatch to various forces by aircraft serial for 8778 aircraft, but that's only when they were despatched to the air parks, not when they were issued to squadrons. They also We also have the individual RFC/RAF serial files for most, but not all aircraft. With 15,922 aircraft recorded as struck off, we will not have a complete picture, since a LOT of losses will not be in our data set with issuing data, even accounting for duplicate data. Similarly, since the despatch files include aircraft not going to the Expeditionary Force, we may be missing crucial losses. We also will have to assume a censoring date for aircraft not lost. I chose to use the last date delivered, since we have records for 1919.

With these caveats, we could make an estimate using the known aircraft delivered, knowing that it overestimates airplane days (our time at risk measure) and underestimates both aircraft despatched and aircraft lost. With aircraft days and losses, we could do a univariate Cox analysis on type, using the Camel as baseline, with practice accident or pilot killed in practice as our event of interest. It's backwards, but it will identify aircraft that were statistically safer than a Camel. For this analysis, I'll be super conservative and only use aircraft where we know a crewman was killed - otherwise, I've got to guess which of the missing and unknown to have the crew die, and which not, which I am not comfortable doing if we're using individual values for plane days.

So I did. We lost almost all the French purchased aircraft, because they were not on the dispatch sheets. We also lost a lot of the early stuff, since, again, they were not on the dispatch sheets. Same with naval aircraft. We also lost the Snipe, since we only have a dispatch record for 1 Snipe that was struck off. Finally, because of so few numbers for the R.E.7, I couldn't include it in the Cox Models.

So, here goes. For ease of identification, I have colored the Camel in red, and the S.E.5a (our presumed "safer" contemporary) in green.

Table I: Aircraft losses, mean time to loss, and crew fate - Only aircraft we know when they got shipped out of the UK, and were assigned to the RFC/RAF
Type Despatched Median Plane Days Total Aircraft Struck Off Not Struck Off No Crew All Crew OK At Least One Crewman Injured At Least One Crewman Killed At Least One Crewman Missing At Least One CrewmanPoW Crew Fate Not Specified
Airco DH.2 200 158.5 137 63 2 70 17 13 29 0 6
Airco DH.4 646 105.0 584 62 35 303 67 49 110 0 20
Airco DH.5 342 86.5 238 104 2 138 36 15 30 0 17
Airco DH.9 236 43.5 182 54 18 89 25 6 37 0 7
Armstrong Whitworth F.K.8 517 101.0 491 26 35 302 58 33 19 0 44
Bristol F2B 662 91.0 577 85 42 235 92 47 122 0 39
Martinsyde G.100/G.102 120 185.0 87 33 0 11 11 4 34 0 27
Royal Aircraft Factory B.E.12 138 93.0 94 44 6 40 11 5 19 0 13
Royal Aircraft Factory B.E.2 720 174.0 531 189 38 224 88 67 64 0 50
Royal Aircraft Factory F.E.2 845 106.0 742 103 63 271 147 66 137 0 58
Royal Aircraft Factory F.E.8 121 137.0 94 27 2 50 18 7 13 0 4
Royal Aircraft Factory R.E.7 32 408.5 15 17 3 6 1 0 2 0 3
Royal Aircraft Factory R.E.8 1694 121.0 1524 170 176 760 249 159 105 2 73
Royal Aircraft Factory S.E.5 1083 82.0 1037 46 72 606 100 36 176 0 47
Sopwith 11/2 strutter 408 89.0 346 62 8 193 33 25 72 0 15
Sopwith Camel 1569 91.0 1368 201 109 645 126 71 335 2 80
Sopwith Dolphin 171 99.0 159 12 17 80 26 9 22 0 5
Sopwith Pup 263 97.0 206 57 31 51 28 4 51 0 41
Total 9813 101.0 8413 1400 659 4075 1133 616 1377 4 549


Table II: Aircraft losses on a training mission, mean time to loss, and crew fate - Only aircraft we know when they got shipped out of the UK, and were assigned to the RFC/RAF
Type Despatched Total Aircraft Struck Off Aircraft Struck Off After Training Flight Median Time to Struck Off Training Mission (Days) No Crew All Crew OK At Least One Crewman Injured At Least One Crewman Killed At Least One Crewman Missing At Least One CrewmanPoW Crew Fate Not Specified
Airco DH.2 200 137 17 116.0 0 11 2 2 0 0 2
Airco DH.4 646 584 105 79.0 0 79 13 8 1 0 4
Airco DH.5 342 238 43 47.0 0 25 7 7 1 0 3
Airco DH.9 236 182 20 47.0 0 15 2 1 0 0 2
Armstrong Whitworth F.K.8 517 491 69 103.0 1 49 8 4 0 0 7
Bristol F2B 662 577 77 129.0 0 41 13 14 0 0 9
Martinsyde G.100/G.102 120 87 9 221.0 0 3 1 1 0 0 4
Royal Aircraft Factory B.E.12 138 94 19 63.0 0 13 3 2 0 0 1
Royal Aircraft Factory B.E.2 720 531 79 120.0 0 45 12 7 3 0 12
Royal Aircraft Factory F.E.2 845 742 67 88.0 0 40 10 8 0 0 9
Royal Aircraft Factory F.E.8 121 94 11 104.0 0 7 3 1 0 0 0
Royal Aircraft Factory R.E.7 32 15 3 57.0 0 2 0 0 0 0 1
Royal Aircraft Factory R.E.8 1694 1524 186 117.0 0 115 30 27 1 0 13
Royal Aircraft Factory S.E.5 1083 1037 176 88.0 0 131 24 12 0 0 9
Sopwith 11/2 Strutter 408 346 70 78.0 0 55 6 4 0 0 5
Sopwith Camel 1569 1368 253 90.0 0 164 31 31 2 0 25
Sopwith Dolphin 171 159 22 68.5 0 11 5 5 0 0 1
Sopwith Pup 263 206 27 61.0 0 11 8 1 0 0 7
Total 9813 8413 1253 91.0 1 817 178 135 8 0 114


Finally, here's the cleaned up table of % killed, and % killed on a practice mission, using our WAG factor for estimating the fate of those missing or not specified. I have given each plane a ranking for each of our % categories. Low is good for our pilots, high is bad for our pilots.

Table III: Summary of Aircraft despatched, struck off and struck off after a Training Mission, with rankings by four deadliness factors
Type Despatched Total Aircraft Struck Off Aircraft Struck Off After Training Flight WAG for KIA WAG for KIA after a training flight % Pilots Killed per Total Aircraft Struck Off % Pilots Killed per Total Aircraft Struck Off in Training % Of Aircraft Struck Off Due To Training Accident % Of all Pilots Killed In Training Accidents % Pilots Killed per Total Aircraft Struck Off (Rank Order) % Pilots Killed per Total Aircraft Struck Off in Training (Rank Order) % Of Aircraft Struck Off Due To Training Accident (Rank Order) % Of all Pilots Killed In Training Accidents (Rank Order)
Airco DH.2 200 137 17 31 3 22.6% 17.6% 12.4% 9.7% 14 8 6 6
Airco DH.4 646 584 105 114 11 19.5% 10.5% 18.0% 9.6% 8 5 13 5
Airco DH.5 342 238 43 39 10 16.4% 23.3% 18.1% 25.6% 6 14 14 16
Airco DH.9 236 182 20 29 2 15.9% 10.0% 11.0% 6.9% 4 3 3 2
Armstrong Whitworth F.K.8 517 491 69 65 8 13.2% 11.6% 14.1% 12.3% 1 6 10 11
Bristol F2B 662 577 77 128 19 22.2% 24.7% 13.3% 14.8% 13 15 8 14
Martinsyde G.100/G.102 120 87 9 35 3 40.2% 33.3% 10.3% 8.6% 18 18 2 4
Royal Aircraft Factory B.E.12 138 94 19 22 3 23.4% 15.8% 20.2% 13.6% 16 7 17 12
Royal Aircraft Factory B.E.2 720 531 79 124 15 23.4% 19.0% 14.9% 12.1% 15 12 11 10
Royal Aircraft Factory F.E.2 845 742 67 164 13 22.1% 19.4% 9.0% 7.9% 12 13 1 3
Royal Aircraft Factory F.E.8 121 94 11 16 1 17.0% 9.1% 11.7% 6.3% 7 1 4 1
Royal Aircraft Factory R.E.7 32 15 3 3 1 20.0% 33.3% 20.0% 33.3% 10 17 16 18
Royal Aircraft Factory R.E.8 1694 1524 186 249 35 16.3% 18.8% 12.2% 14.1% 5 11 5 13
Royal Aircraft Factory S.E.5 1083 1037 176 148 17 14.3% 9.7% 17.0% 11.5% 2 2 12 9
Sopwith 11/2 strutter 408 346 70 69 7 19.9% 10.0% 20.2% 10.1% 9 4 18 8
Sopwith Camel 1569 1368 253 279 45 20.4% 17.8% 18.5% 16.1% 11 9 15 15
Sopwith Dolphin 171 159 22 23 6 14.5% 27.3% 13.8% 26.1% 3 16 9 17
Sopwith Pup 263 206 27 51 5 24.8% 18.5% 13.1% 9.8% 17 10 7 7
Total 9813 8413 1253 1580 196 18.8% 15.6% 14.9% 12.4%


Ok, this essentially recreates the previous analysis, cleaning up the data, and only using aircraft that we know were issued, while adding time to struck-off, as requested by KDahm. As we can see, the Camel's median plane days to striking off are comparable to all it's contemporaries. It's got a shorter time to strike-off than many predicessors (except the Pup), but I suspect that is due to Bloody April - the D.H.2 and most British two-seaters had an easy life in 1916, only to come to a screeching halt in April, 1917. As such, they would have longer longevity while snacking on Eindeckers (yum, yum) and Halberstadts. The other interesting bit is the D.H.9, which had a median time from delivery to issueance of about a month and a half. That fits, since the D.H.9 was a deathtrap.

We can also see that the Camel's median time to a fatal training accident is the same as the median time to strike-off, and higher than that of any other Sopwith product, or any fighter than the D.H.2 or the Bristol F.2.B. Of note, again, are the D.H.5 and the D.H.9, which have very short median time to loss in a training accident.

As we can see in Table III, of the 18 aircraft we have despatch records for and sufficient numbers to analyze, the Camel is in the middle of the pack of total pilots killed (any way possible), and those specifically killed in a training accident that destroyed the aircraft. Interestingly, the Camel is high on the ranking of training accidents, and of the portion of pilots killed in training accidents to total killed.

Now, we go to the univariate Cox Proportional Hazards Models:

Table IV: Univariate Cox Proportional Hazards Model for risk of pilot killed, aircraft destroyed in training, and pilot killed in training - Super Conservative Case
Type Despatched Total Aircraft Struck Off Aircraft Struck Off After Training Flight Crew Killed Crew Killed on a Training Flight HR for Pilot Killed 95% CI Upper Limit 95% CI Lower Limit p-value HR for Struck Off on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value HR for Pilot Killed on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value
Airco DH.2 200 137 17 13 2 0.90 0.50 1.63 0.7366 0.33 0.20 0.53 <.0001 0.28 0.07 1.18 0.0833
Airco DH.4 646 584 105 49 8 1.46 1.01 2.09 0.0435 0.86 0.69 1.08 0.2036 0.53 0.25 1.16 0.1133
Airco DH.5 342 238 43 15 7 0.82 0.47 1.44 0.4974 0.66 0.48 0.91 0.0112 0.83 0.37 1.89 0.658
Airco DH.9 236 182 20 6 1 1.06 0.46 2.45 0.886 1.05 0.66 1.65 0.8486 0.43 0.06 3.18 0.4118
Armstrong Whitworth F.K.8 517 491 69 33 4 1.38 0.92 2.09 0.1231 0.81 0.62 1.06 0.1229 0.39 0.14 1.10 0.0759
Bristol F2B 662 577 77 47 14 1.65 1.14 2.39 0.0077 0.76 0.59 0.98 0.0321 1.13 0.60 2.12 0.7127
Martinsyde G.100/G.102 120 87 9 4 1 0.48 0.17 1.31 0.1504 0.29 0.15 0.57 0.0003 0.25 0.03 1.83 0.1725
Royal Aircraft Factory B.E.12 138 94 19 5 2 0.64 0.26 1.59 0.3391 0.68 0.43 1.09 0.1112 0.51 0.12 2.15 0.3583
Royal Aircraft Factory B.E.2 720 531 79 67 7 1.32 0.94 1.85 0.104 0.43 0.33 0.55 <.0001 0.28 0.12 0.65 0.0028
Royal Aircraft Factory F.E.2 845 742 67 66 8 1.51 1.08 2.12 0.0153 0.43 0.33 0.56 <.0001 0.40 0.19 0.88 0.0222
Royal Aircraft Factory F.E.8 121 94 11 7 1 0.93 0.43 2.01 0.8434 0.40 0.22 0.74 0.0031 0.27 0.04 2.01 0.2027
Royal Aircraft Factory R.E.8 1694 1524 186 159 27 1.73 1.31 2.29 0.0001 0.56 0.47 0.68 <.0001 0.66 0.39 1.10 0.1118
Royal Aircraft Factory S.E.5 1083 1037 176 36 12 0.82 0.55 1.22 0.3269 1.13 0.93 1.37 0.2086 0.64 0.33 1.26 0.1968
Sopwith 11/2 strutter 408 346 70 25 4 1.28 0.81 2.02 0.287 1.01 0.77 1.31 0.9506 0.44 0.15 1.25 0.1221
Sopwith Camel 1569 1368 253 71 31 1.00 - - - 1.00 - - - 1.00 - - -
Sopwith Dolphin 171 159 22 9 5 1.10 0.55 2.21 0.7814 0.76 0.49 1.17 0.213 1.46 0.57 3.75 0.4351
Sopwith Pup 263 206 27 4 1 0.29 0.11 0.80 0.0161 0.55 0.37 0.82 0.0032 0.16 0.02 1.16 0.0702


As you can see, we get a VERY different picture - one where the Camel looks a lot deadlier. However, remembering our tasty Eindecker theory, I think we need to add a covariate - German fighter opposition. Without access to German and British TROMs, I've got to come up with something quick and dirty. Here's where we can use the PM's aircraft model, and weight each German's fighter ranking by the estimated number of days our poor British pilots will have to face the Dastardly Hun(tm). I did this quick and dirty, assuming equal daily risk of a given German fighter if it was in frontline service.

Absolute German Fighter Risk=sum(RankingGerman Fighter x possible days exposedGerman Fighter)/(Allied Plane Days*number of fighter types in service)

Obviously, to do this, we also have to truncate the observation period to Armistice day. That takes us from 9813 to 9673 aircraft. We are also not weighting fighter risk by the # of German fighters in service on a given day, or the risk of encountering a given fighter type, or the aircraft's own fighter rating. Based on my Red Baron (my favorite WWI fighter rule set) Aircraft Availability, here are the parameters I used for estimating possible days exposed:
German Fighter Rating Service Start Date Service End Date
Fokker E.II 23.61 6/15/1915 6/31/1916
Fokker E.III 18.23 8/1/1915 6/31/1916
Halberstad D.II 26.2 6/1/1916 5/1/1917
Pfalz D.III 31.46 8/1/1917 11/11/1918
Albatross D.II 32.9 9/1/1916 11/31/1917
Albatross D.III 32.7 1/1/1917 3/1/1918
Albatross D.V 32.9 5/1/1917 11/11/1918
Fokker Dr.I 35.02 8/1/1917 9/31/1918
Roland D.VI 34.72 5/1/1918 11/11/1918
Fokker D.VII/MIII 33.28 4/1/1918 11/11/1918
Fokker D.VII/MIIIa 34.45 4/1/1918 11/11/1918
Fokker D.VII/BMW 38.39 4/1/1918 11/11/1918
Pfalz D.VIII 30.77 4/1/1918 11/11/1918
Pfalz D.XII 34.64 8/1/1918 11/11/1918
Siemens D.III 33.53 4/1/1918 11/11/1918
Siemens D.IV 30.35 8/1/1918 11/11/1918
Fokker D.VIII 24.37 8/1/1918 11/11/1918
Junkers D.1 29.54 10/1/1918 11/11/1918
I trimmed it a little, either for lack of data, or to remove two-seater escort fighters. I know, I argued for putting them in, but this should work well enough.

And here we go:
Table V: Univariate Cox Proportional Hazards Model for risk of pilot killed, aircraft destroyed in training, and pilot killed in training - Daily German Fighter Risk
Covariate HR for Pilot Killed 95% CI Upper Limit 95% CI Lower Limit p-value HR for Struck Off on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value HR for Pilot Killed on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value
German Fighter Risk 1.126 1.11 1.143 <.0001 1.132 1.12 1.144 <.0001 1.17 1.131 1.211 <.0001


Here's the Cox Model with both aircraft type and German Fighter Risk:
Table VI: Multivariate Cox Proportional Hazards Model for risk of pilot killed, aircraft destroyed in training, and pilot killed in training - incorporating Daily German Fighter Risk
Covariate HR for Pilot Killed 95% CI Upper Limit 95% CI Lower Limit p-value HR for Struck Off on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value HR for Pilot Killed on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value
Airco DH.2 2.22 1.21 4.06 0.0098 0.73 0.44 1.19 0.2064 0.65 0.15 2.83 0.5618
Airco DH.4 1.71 1.19 2.46 0.004 0.99 0.79 1.25 0.9257 0.64 0.29 1.38 0.2522
Airco DH.5 1.05 0.60 1.83 0.8725 0.85 0.61 1.17 0.3168 1.07 0.47 2.45 0.8649
Airco DH.9 0.93 0.41 2.15 0.873 0.87 0.55 1.39 0.5638 0.38 0.05 2.76 0.3354
Armstrong Whitworth F.K.8 1.63 1.08 2.47 0.0207 0.93 0.71 1.21 0.5741 0.47 0.17 1.33 0.1538
Bristol F2B 2.02 1.40 2.92 0.0002 0.88 0.68 1.14 0.3146 1.41 0.75 2.66 0.2882
Martinsyde G.100/G.102 1.19 0.43 3.28 0.7402 0.65 0.34 1.28 0.2152 0.63 0.08 4.70 0.6501
Royal Aircraft Factory B.E.12 1.78 0.71 4.45 0.2203 1.74 1.08 2.81 0.0228 1.27 0.28 5.70 0.7566
Royal Aircraft Factory B.E.2 2.98 2.10 4.23 <.0001 0.88 0.68 1.15 0.3594 0.64 0.27 1.53 0.3168
Royal Aircraft Factory F.E.2 2.54 1.81 3.57 <.0001 0.69 0.52 0.90 0.0066 0.70 0.32 1.53 0.3675
Royal Aircraft Factory F.E.8 2.04 0.93 4.48 0.0743 0.81 0.44 1.49 0.5047 0.60 0.08 4.52 0.6227
Royal Aircraft Factory R.E.8 2.27 1.71 3.01 <.0001 0.72 0.59 0.87 0.0007 0.88 0.53 1.49 0.6444
Royal Aircraft Factory S.E.5 0.88 0.59 1.32 0.5455 1.20 0.99 1.46 0.0677 0.70 0.36 1.37 0.2936
Sopwith 11/2 strutter 2.26 1.42 3.59 0.0006 1.67 1.27 2.19 0.0002 0.82 0.29 2.38 0.7193
Sopwith Camel 1.00 - - - 1.00 - - - 1.00 - - -
Sopwith Dolphin 1.29 0.65 2.59 0.4697 0.87 0.56 1.34 0.5256 1.72 0.67 4.44 0.2627
Sopwith Pup 0.55 0.20 1.52 0.2516 1.00 0.67 1.49 0.9872 0.32 0.04 2.38 0.2667
German Fighter Risk 1.15 1.13 1.17 <.0001 1.13 1.12 1.14 <.0001 1.16 1.12 1.20 <.0001

Now, isn't that interesting . . . the daily German fighter risk is not only significant at predicting whether an RFC/RAF pilot will be killed (which is logical), but also at predicting whether he'll crash his plane in a training op, or be killed crashing his plane in a training op. Essentially, the higher the German fighter risk (ie, the higher the PM would rate the average German fighter on Any Given Sunday), the increased likelyhood of a British training accident, and increased likelyhood of killing the pilot in said training accident! A 1 point increase in the value of a German WWI fighter increases the risk of killing an Allied pilot by 15%, a 13% increased risk of crashing the plane training, and a 16% chance of the pilot being killed training. When we incorporate the average German fighter plane rating into the model, the Camel looks damn good, being less likely to get you killed than any single seat fighter, except the Pup and the S.E.5a. It's also more likely to keep you alive than any two seater, except the D.H.9. I suspect that is the result of so many D.H.9 aborts and losses due to bad engines.

Even more interestingly, when we look at the risk of wrecking your plane in a training op, most aircraft are less likely to be written off in a training accident. When we look at the odds of being killed when you wreck your plane on a training op, incorporating the German fighter risk results in most Camel contemporaries (except the S.E.5a) being more likely to kill you. However, if we incorporate the German Fighter risk, what airplane you are flying is not a significant predictor of whether you'll die in said training op (overall p for aircraft type=0.7012) - just the German Fighter Risk!

What does that mean? I think it supports my earlier conclusion. Airplane ratings increased with the war, and so German fighter risk is a proxy for British fighter capability. As such, the better your fighter is overall, the less likely you are to be shot down, and thus the greater chance you have killing yourself other ways.

We can test this in a subset, by changing our method of rating the German fighter opposition to account for the Allied plane rating. Now, this will eliminate many of the two seaters, reducing us to just 5613 fighters, but we can try it. It does argue for rating them on the fighter model to estimate their effectiveness in defending themselves from a fighter, especially in WWI. Since I don't have data on which engine was used in the Camels, I'm going to assume the 130hp Clerget, since that was standard.

This results in the Relative German Fighter Risk variable:
Relative German Fighter Risk=sum((RankingGerman Fighter-RankingBritish Fighter) x possible days exposedGerman Fighter)/(Allied Plane Days*number of German fighter types in service)

Our variable isn't perfectly normally distributed, but it's close enough for HPCA work. Positive values = German Advantage, negative values = Allied Advantage. Overall median Relative German Fighter Risk is -0.39 and mean is 0.54. Shown below are the relative matchups on rating, and then median and mean Relative German Fighter Risk for each fighter type:
Albatross D.II Albatross D.III Albatross D.V Fokker D.VII/BMW Fokker D.VII/MIII Fokker D.VII/MIIIa Fokker D.VIII Fokker Dr.I Fokker E.II Fokker E.III Halberstad D.II Junkers D.I Pfalz D.III Pfalz D.VIII Pfalz D.XII Roland D.VI Siemens-Schuckert D.III Siemens-Schuckert D.IV
Allied Aircraft Allied Aircraft Rating 32.9 32.7 32.9 38.39 33.28 34.45 24.37 35.02 23.61 18.23 26.2 29.54 31.46 30.77 34.64 34.72 33.53 30.35 Median Relative German Fighter Risk Mean Relative German Fighter Risk Minimim Relative German Fighter Risk Maximum Relative German Fighter Risk
Airco DH.2 30.41 2.49 2.29 2.49 7.98 2.87 4.04 -6.04 4.61 -6.8 -12.18 -4.21 -0.87 1.05 0.36 4.23 4.31 3.12 -0.06 -0.39 -0.58 1.67 -7.52
Airco DH.5 30.23 2.67 2.47 2.67 8.16 3.05 4.22 -5.86 4.79 -6.62 -12 -4.03 -0.69 1.23 0.54 4.41 4.49 3.3 0.12 2.60 2.40 0.64 1.55
Bristol F2B 31.21 1.69 1.49 1.69 7.18 2.07 3.24 -6.84 3.81 -7.6 -12.98 -5.01 -1.67 0.25 -0.44 3.43 3.51 2.32 -0.86 1.62 1.65 0.61 -0.61
Royal Aircraft Factory F.E.2 23.45 9.45 9.25 9.45 14.94 9.83 11 0.92 11.57 0.16 -5.22 2.75 6.09 8.01 7.32 11.19 11.27 10.08 6.9 6.10 6.05 2.42 -0.77
Royal Aircraft Factory F.E.8 25.87 7.03 6.83 7.03 12.52 7.41 8.58 -1.5 9.15 -2.26 -7.64 0.33 3.67 5.59 4.9 8.77 8.85 7.66 4.48 3.60 3.36 1.74 -3.19
Royal Aircraft Factory S.E.5 35.37 -2.47 -2.67 -2.47 3.02 -2.09 -0.92 -11 -0.35 -11.76 -17.14 -9.17 -5.83 -3.91 -4.6 -0.73 -0.65 -1.84 -5.02 -1.53 -1.60 0.57 -4.77
Sopwith 11/2 strutter 29.73 3.17 2.97 3.17 8.66 3.55 4.72 -5.36 5.29 -6.12 -11.5 -3.53 -0.19 1.73 1.04 4.91 4.99 3.8 0.62 1.01 1.15 1.51 -3.53
Sopwith Camel 36.06 -3.16 -3.36 -3.16 2.33 -2.78 -1.61 -11.69 -1.04 -12.45 -17.83 -9.86 -6.52 -4.6 -5.29 -1.42 -1.34 -2.53 -5.71 -2.13 -2.03 0.74 -3.23
Sopwith Dolphin 33.89 -0.99 -1.19 -0.99 4.5 -0.61 0.56 -9.52 1.13 -10.28 -15.66 -7.69 -4.35 -2.43 -3.12 0.75 0.83 -0.36 -3.54 -0.17 -0.17 0.39 -0.76
Sopwith Pup 31.15 1.75 1.55 1.75 7.24 2.13 3.3 -6.78 3.87 -7.54 -12.92 -4.95 -1.61 0.31 -0.38 3.49 3.57 2.38 -0.8 1.32 1.04 0.71 -0.55

As you can see, we can immediately see both the qualitative superiority of the Camel (and the S.E.5a) over most German aircraft, and the relative inferiority of many contemporary British aircraft adjusted for the fighters each aircraft could have faced over each Allied aircraft's life. We also can pretty much explain Bloody April just with that table!

So, here is the univariate model for the Relative German Fighter Risk:
Table VII: Univariate Cox Proportional Hazards Model for risk of pilot killed, aircraft destroyed in training, and pilot killed in training using Relative Daily German Fighter Risk
Covariate HR for Pilot Killed 95% CI Upper Limit 95% CI Lower Limit p-value HR for Struck Off on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value HR for Pilot Killed on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value
Relative German Fighter Risk 1.011 0.97 1.053 0.6147 0.851 0.822 0.88 <.0001 0.866 0.786 0.956 0.0041


So, and the multivariate model for the Relative German Fighter Risk:
Table VIII: Univariate Cox Proportional Hazards Model for risk of pilot killed, aircraft destroyed in training, and pilot killed in training using Relative Daily German Fighter Risk
Covariate HR for Pilot Killed 95% CI Upper Limit 95% CI Lower Limit p-value HR for Struck Off on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value HR for Pilot Killed on Training Op 95% CI Upper Limit 95% CI Lower Limit p-value
Airco DH.2 1.322 0.719 2.432 0.3695 0.514 0.311 0.851 0.0097 0.45 0.101 2.012 0.2959
Airco DH.5 2.334 1.194 4.564 0.0132 2.173 1.402 3.368 0.0005 3.848 1.134 13.066 0.0307
Bristol F2B 3.687 2.277 5.971 <.0001 1.925 1.343 2.759 0.0004 3.718 1.381 10.008 0.0093
Royal Aircraft Factory F.E.2 8.226 4.152 16.297 <.0001 3.185 1.857 5.463 <.0001 4.173 0.871 19.999 0.074
Royal Aircraft Factory F.E.8 3.065 1.25 7.517 0.0144 1.697 0.841 3.427 0.14 1.455 0.15 14.141 0.7463
Royal Aircraft Factory S.E.5 0.844 0.564 1.261 0.4074 1.197 0.985 1.454 0.0708 0.673 0.344 1.317 0.2475
Sopwith 11/2 strutter 2.593 1.534 4.386 0.0004 2.35 1.677 3.295 <.0001 1.29 0.384 4.329 0.6804
Sopwith Camel 1.00 - - - 1.00 - - - 1.00 - - -
Sopwith Dolphin 1.638 0.804 3.337 0.1739 1.253 0.795 1.973 0.3312 2.619 0.939 7.305 0.0658
Sopwith Pup 0.595 0.21 1.686 0.3285 1.277 0.813 2.006 0.2892 0.462 0.057 3.766 0.4709
Relative German Fighter Risk 0.773 0.702 0.851 <.0001 0.73 0.674 0.79 <.0001 0.681 0.536 0.865 0.0016

So, what do we see? The first is a problem - Table VIII shows a switch in the risk of being killed, where increasing relative German fighter advantage becomes protective (as shown by the HR less than 1). Doing a quick check of fighter kills vs relative German fighter risk, we have as many pilots killed in the group with the greatest Allied advantage (71) as we do in the greatest Allied disadvantage (67). I have two theories to explain this.

Theory 1: we are overcorrecting for the German fighter advantage by including both aircraft type as a variable, and penalizing the German fighter risk for the relative rating. As you can see in the table, the hazard ratio for most aircraft goes up compared to the baseline (Camel). That suggests overcorrecting.

Theory 2: this goes back to the PM's original point, that the model only measures aircraft ability, not pilot ability and combat conditions. Essentially, lots of S.E.5a and Camel pilots (since the model predicts they should pawn the Dastardly Hun) should not have died, if it was a straight plane to plane comparison. Because they did, it skews the statistical model for the relative German advantage. To further test this, we'd need data about pilot skill and combat conditions. We could probably get flying time as a proxy for pilot skill, since we have crew names for each aircraft. That would be a hell of a task, but it is doable (the data exists). Or we could use an average, or the number of hours in training before being sent to the front. We could also rate all the British two-seaters as if they were fighters, and see if adding much easier targets to our data set (the Airco DH.4, Airco DH.9, Armstrong Whitworth F.K.8, Martinsyde G.100/G.102, Royal Aircraft Factory B.E.12, Royal Aircraft Factory B.E.2, Royal Aircraft Factory F.E.2, Royal Aircraft Factory F.E.8, Royal Aircraft Factory R.E.8) will offset the Camels and S.E.5as shot down.

So, is this analysis rubbish? Nope. Let's look at the other two conditions - aircraft wrecked in training, and pilot killed in an aircraft wrecked in training. In both of those situations, the relative German advantage is protective. Subject to the caveats above, this seems to fit our hypothesis - the greater German advantage, the less likely you and your trusty product of British aviation were to live long enough to crash training (bad for the plane) or die training (bad for the pilot, the plane might not mind much). When we incorporate this, we see that the risks of both your plane and you dying in training are higher if you are riding almost anything but a Camel.

Ignore the Pup snarking over there about it's little sister. The Pup has always been problematic in these analyses. 1,770 of them were built, but we only have records for 260-odd; I suspect many more went to the RNAS (and were missing in these analyses), or didn't go to France. We also don't have many confirmed killed in a Pup and, since our Cox model only uses killed, wrecked in practice, and killed while wrecked in practice, the small numbers are a problem for the model. This is reflected in the wide confidence intervals for the Pup. I suspect, with more data, the Pup will not compare so well with the Camel. The D.H.2 suffers from much the same data problem as the Pup, though with 453 built instead of 1,770 we have, proportionately, much less missing data. Given that you're anywhere from 30-220% more likely to get killed riding one, though, being marginally less likely to be in a training accident is not much comfort.

The S.E.5a, on the other hand, has more than earned the right to talk you out of buying a Camel. We also have sufficient numbers - 1000 odd S.E.5as and 1600 Camels - to be confident in the matchup. You're 12-16% less likely to be killed overall if your mount is an S.E.5a compared to a Camel, and if a training op goes bad, you're 30-33% less likely to die. But you are more likely to sit in your cockpit crying as they come to take your poor, mortally wounded mount away.

So, here's what I think we can conclude - the Camel was no deadlier to her pilots in the RFC than most RFC aircraft, especially other single seat products of the Sopwith company. She was also, by these analyses, the second best Allied fighter, after the S.E.5a. That suggests that being more maneuverable wasn't as great as it looks in the raw rating, especially at the cost of speed and performance at altitude. It also may indicate that the handling characteristics - easy and safer to fly - saved pilots who would have died were they in a Camel. It also might have encouraged folk to take risks they wouldn't have, leading to more crashes in training. We don't know.

So, where did the Camel's reputation come from? I think the key factor is the Camel's superiority over the German planes. That gives you more chances to die in a training operation, and automatically means your squadron mates will feel the training losses are more severe. Also, compared to the S.E.5a, Camel pilots did die more often in training accidents (absolutely and proportionately), so that adds to the reputation of the Camel as "dangerous."

With that, given the limits of the current data, I think this analysis has gone as far as it can go. We just don't have the training records, RNAS records or other theater records at the moment to easily crank the data. Should HM Government wish to change that, they know where to find me. The only edits I think we can make without massive data crunching are rating the missing aircraft as fighters and adding them back in.

And, now, I think this post is the one that deserves the single malt - both for answering the question, and testing the efficacy of the rating model.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 1:25 am ]
Post subject:  Re: RAF Fighters V6.2

pdf27 wrote:
Actually, I think the comparison to the SE.5 is more relevant - between the two they made up the overwhelming majority of single-seat fighters on the Western Front, and clearly from the statistics the SE.5 is the safer of the two. Now, the classical accident pyramid suggests that for every fatality there will be a substantial number of injuries and a very large number of near misses (given the aircraft of the time the injuries will probably be artificially compressed though since the scope for only being injured is quite small). That suggests that there were a very large number of Camel pilots out there who had experienced their aircraft trying to kill them, and a far smaller number of SE.5 pilots (with a substantial number who crossed one way or the other and as a result were convinced that the SE.5 was safer to fly).
Now it seems rather obvious to me that the influence of live pilots on the reputation of an aircraft is going to be far stronger than that of dead pilots - you're aren't going to have a dead guy bending your ear about how dangerous such-and-such an aircraft was compared to everything else (when actually he only flew the SE.5 for comparison, but as this is 20 years later and he's big-timing it in the mess he isn't going to admit that). Not only that, but the live pilots will stick around for an awfully long time with their experiences while the dead sort of fade away a bit - their names are remembered, but by and large their cause of death isn't.


Actually, there is a huge scope for a pilot not dying. From our struck-off and despatch records, 60% of S.E.5as had the pilot survive the aircraft's loss uninjured, while ~50% of Camel pilots survived uninjured the loss of their aircraft. About 10% of pilots for both aircraft were injured and so noted in the records, while about 4% of S.E.5a pilots and about 5% of Camel pilots were killed. The largest category of not-ok is missing, and not all of them were killed. We guesstimate about 50%, but that may well be high.

From my number crunching, it looks like the difference is due to the S.E.5a being safter, but also a lot more due to your Camel keeping the Germans from killing you, so it had more opportunity to take a swipe at you.

Author:  pdf27 [ Fri Sep 25, 2015 4:54 am ]
Post subject:  Re: RAF Fighters V6.2

I was more thinking of incidents that give you a bad scare but which the aircraft lands safely after - they're likely to be the most numerous and are probably in proportion to the crash rates.

The Relative German Fighter Risk table also flags up something I've been wondering about for some time, and that's the effect of prevailing winds on combat at the time. Both the major British designs have a clear advantage over just about every German design, yet AIUI loss rates don't represent this well tending to favour the Germans more than they should (is there a way to look at this with the stats?). Wind in Northern Europe tends to be from West to East, meaning that Entente fighter pilots trying to get away had a harder time than German ones, since for this type of aircraft wind speed is an appreciable fraction of maximum speed. That's going to cause a level of loss from people who simply got blown too far behind enemy lines and couldn't get back either through enemy action or fuel exhaustion.

Author:  Belushi TD [ Fri Sep 25, 2015 8:48 am ]
Post subject:  Re: RAF Fighters V6.2

Ho.... Ly..... SH!T!

My compliments. I know/remember just enough statistics to be able to follow your model. Furthermore, I'm just enough of a geek/nerd to find this analysis FACINATING.

Should we ever find ourselves in the same bar, I owe you a single malt.

Belushi TD

Author:  KDahm [ Fri Sep 25, 2015 10:16 am ]
Post subject:  Re: RAF Fighters V6.2

This goes far beyond what I was thinking. It definitely deserves a place of its own in the Essays section. May I suggest waiting a week, reproofing it, then putting it there.

I'll have to read it thoroughly over the weekend and spend some time cogitating before making an informed comment.

Quote:
That requires a LOT of work. We'd need the individual squadron war diaries to get when a plane was actually issued from the air parks to the squadrons, and they're not readily available. We do have combined fate files for most RFC aircraft, and combined movement files, but the data is dirty and would require lots of man hours (hundreds) to clean, since it's almost a line by line process.

On further discovery, there's a lot of duplicate data in the struck-off lists. So the original analyses are inflated. Cleaning that up will require work, since it's going to require a lot of hand examination. I also found possible undercounting of training losses.

We have month of despatch to various forces by aircraft serial for 8778 aircraft, but that's only when they were despatched to the air parks, not when they were issued to squadrons. They also We also have the individual RFC/RAF serial files for most, but not all aircraft. With 15,922 aircraft recorded as struck off, we will not have a complete picture, since a LOT of losses will not be in our data set with issuing data, even accounting for duplicate data. Similarly, since the despatch files include aircraft not going to the Expeditionary Force, we may be missing crucial losses. We also will have to assume a censoring date for aircraft not lost. I chose to use the last date delivered, since we have records for 1919.

With these caveats, we could make an estimate using the known aircraft delivered, knowing that it overestimates airplane days (our time at risk measure) and underestimates both aircraft despatched and aircraft lost. With aircraft days and losses, we could do a univariate Cox analysis on type, using the Camel as baseline, with practice accident or pilot killed in practice as our event of interest. It's backwards, but it will identify aircraft that were statistically safer than a Camel. For this analysis, I'll be super conservative and only use aircraft where we know a crewman was killed - otherwise, I've got to guess which of the missing and unknown to have the crew die, and which not, which I am not comfortable doing if we're using individual values for plane days.

What I was thinking about. A sampling of the data using a decent set of known dates. While there is a lot of fuzziness in the data (manufacture to dept, depot to squadron, squadron to repair/rebuild facility to squadron), it just has to be consistent across plane types. And the reader has to understand fuzziness exists.

Thank you.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 12:22 pm ]
Post subject:  Re: RAF Fighters V6.2

pdf27 wrote:
I was more thinking of incidents that give you a bad scare but which the aircraft lands safely after - they're likely to be the most numerous and are probably in proportion to the crash rates.

The Relative German Fighter Risk table also flags up something I've been wondering about for some time, and that's the effect of prevailing winds on combat at the time. Both the major British designs have a clear advantage over just about every German design, yet AIUI loss rates don't represent this well tending to favour the Germans more than they should (is there a way to look at this with the stats?). Wind in Northern Europe tends to be from West to East, meaning that Entente fighter pilots trying to get away had a harder time than German ones, since for this type of aircraft wind speed is an appreciable fraction of maximum speed. That's going to cause a level of loss from people who simply got blown too far behind enemy lines and couldn't get back either through enemy action or fuel exhaustion.


It's certainly a factor, though it may be counterbalanced by Allied pilots being helped by the tailwind when they are flying with the prevailing winds. Sun advantage is also to the Germans in the morning, on the British front. None of those factors, however, are as strong on the French fronts, where the line ran east-west instead of north-south. The consolidation of the front at the retreat to the Hindenberg Line in 1917 would also mitigate the wind advantage, taking it from a north south line in the Somme sector to a northwest-southeast line.

Another one that is complementary is line advantage - he who is operating behind friendly lines or over the front has a much better chance of recovering downed fliers. That was a major factor in Bloody April - the British had to operate behind German lines, while the Germans chose not to.

There's also training. Bloody April revealed that the Germans were a lot better at training their pilots, and resulted in a serious revamp of how the British ran their flight schools and how many hours a pilot had before they got into combat. The use of Jastas as flying fire brigades also helped - good pilots were concentrated and shuttled from trouble spot to trouble spot, giving the Germans a qualitative pilot advantage they wouldn't have had normally.

The one I would be interested in looking at is the parachute. German and Austrian aviators had them, starting in 1918. Allied ones didn't. It'd be interesting to see if the parachute gave a German pilot a significantly greater chance of surviving longer, and thus getting more experienced.

Of course, a lot of these would require having similar data for the Germans, and I don't think we do.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 12:30 pm ]
Post subject:  Re: RAF Fighters V6.2

KDahm wrote:
This goes far beyond what I was thinking. It definitely deserves a place of its own in the Essays section. May I suggest waiting a week, reproofing it, then putting it there.


Possibly. I think, right now, it would be best if we got an analysis section of the Aircraft Ranking library, and the discussion of the Camel was created as it's own thread. That way, it doesn't clutter up the RAF rankings, and can be easily found.

KDahm wrote:
I'll have to read it thoroughly over the weekend and spend some time cogitating before making an informed comment.


I've thought of a few other things I want to try, and will see how they work. The biggest criticisms I can see are these:
  1. We are missing key aircraft (French ones, RNAS ones, earlier model aircraft) for any of these analyses.
  2. The use of just pilots we know were killed in the Cox model is too conservative, and that we should expand it to include Missing as well.
  3. Pilot fate is the wrong variable in the Cox model, and that we should expand it to use all combat (as opposed to operational) losses as our outcome of interest to test effectiveness, if not lethality to pilots.
  4. German fighter risk and relative German fighter risk are treated as a constant for the duration of a plane's service life. Effectively, the risk as new planes arrive is diluted by averaging it over a wider period.
  5. The use of a Cox Proportional Hazards model is itself improper, and what we should really be using is an Accellerated Time to Failure model. The Cox model looks at the risk of an event hapening, and whether the covariates make it more or less likely to happen. The Accellerated Failure Time model looks at whether the covariates make the outcome happen sooner or later. Since most of the aircraft on our list were struck-off, we could look at an AFT model where being struck-off is the outcome, and see if plane type influences that.

Of these, items 2 and 5 are the easiest to do, followed by 4. Those just require new analysis code. Items 1 and 3 require lots of data cleaning. We have reconstructions of most of the serials, reconstructions of a lot of routine orders, and casualty reports, so it's possible, but time consuming. Adding more aircraft where we do have info requires the PM rating them. If that can be done, then it makes sense to stop here until we have ratings for them, then try the AFT model or a more liberal definition of "killed."

Author:  pdf27 [ Fri Sep 25, 2015 12:36 pm ]
Post subject:  Re: RAF Fighters V6.2

Johnnie Lyle wrote:
It's certainly a factor, though it may be counterbalanced by Allied pilots being helped by the tailwind when they are flying with the prevailing winds. Sun advantage is also to the Germans in the morning, on the British front. None of those factors, however, are as strong on the French fronts, where the line ran east-west instead of north-south. The consolidation of the front at the retreat to the Hindenberg Line in 1917 would also mitigate the wind advantage, taking it from a north south line in the Somme sector to a northwest-southeast line.

Not really - when it comes to performance relative to other aircraft then it's irrelevant since both aircraft are part of the same air mass. The issue is that when trying to base because they're low on fuel or trying to escape from an enemy force they're flying into the prevailing wind (or later on into a component of it), which cuts down their ground speed and increases the time they're exposed to danger/running out of fuel for the British and French while reducing it for the Germans. That can be very significant indeed for aircraft this slow - up to a 50% change in groundspeed when you change direction.

Johnnie Lyle wrote:
Another one that is complementary is line advantage - he who is operating behind friendly lines or over the front has a much better chance of recovering downed fliers. That was a major factor in Bloody April - the British had to operate behind German lines, while the Germans chose not to.

Ties in to wind too - getting home is further and into wind.

Johnnie Lyle wrote:
There's also training. Bloody April revealed that the Germans were a lot better at training their pilots, and resulted in a serious revamp of how the British ran their flight schools and how many hours a pilot had before they got into combat. The use of Jastas as flying fire brigades also helped - good pilots were concentrated and shuttled from trouble spot to trouble spot, giving the Germans a qualitative pilot advantage they wouldn't have had normally.

That should be amenable to the sort of stats you're doing, although you probably don't have the time to do it on that level of detail - mortality rates versus logged hours.

Johnnie Lyle wrote:
The one I would be interested in looking at is the parachute. German and Austrian aviators had them, starting in 1918. Allied ones didn't. It'd be interesting to see if the parachute gave a German pilot a significantly greater chance of surviving longer, and thus getting more experienced.

The PM's already looked at this one a bit - early parachutes were so heavy and fighters so light (420 kg empty for a Camel) that the aircraft take a major performance hit from carrying both the parachute and the pilot.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 1:19 pm ]
Post subject:  Re: RAF Fighters V6.2

pdf27 wrote:
Johnnie Lyle wrote:
It's certainly a factor, though it may be counterbalanced by Allied pilots being helped by the tailwind when they are flying with the prevailing winds. Sun advantage is also to the Germans in the morning, on the British front. None of those factors, however, are as strong on the French fronts, where the line ran east-west instead of north-south. The consolidation of the front at the retreat to the Hindenberg Line in 1917 would also mitigate the wind advantage, taking it from a north south line in the Somme sector to a northwest-southeast line.

Not really - when it comes to performance relative to other aircraft then it's irrelevant since both aircraft are part of the same air mass. The issue is that when trying to base because they're low on fuel or trying to escape from an enemy force they're flying into the prevailing wind (or later on into a component of it), which cuts down their ground speed and increases the time they're exposed to danger/running out of fuel for the British and French while reducing it for the Germans. That can be very significant indeed for aircraft this slow - up to a 50% change in groundspeed when you change direction.


Except prevailing wind is also a problem for the Germans, especially for bombing, ground attack, photo recon, artillery direction and fighter sweeps. It's just that you fly into the prevaling winds getting to the target, instead of getting home from it. So while a badly damaged or low-fuel Allied plane could be seriously hindered getting home, a German one could easily find itself in strife if the meterorological conditions were worse than they planned. As 1918 wore on, and the Germans had serious fuel issues, that is a not minor consideration.

pdf27 wrote:
Johnnie Lyle wrote:
Another one that is complementary is line advantage - he who is operating behind friendly lines or over the front has a much better chance of recovering downed fliers. That was a major factor in Bloody April - the British had to operate behind German lines, while the Germans chose not to.

Ties in to wind too - getting home is further and into wind.


Not really. If you're behind friendly lines when you get into trouble, all you need is a decently flat spot to put down.

Effectively, you are home - your plane may not be, but you are. You may have to walk to your aerodrome or spend the night with a friendly farmer, but you're not dodging enemy patrols looking for downed fliers. Also, if you're far enough back, nobody is shelling you, while if you're near the front, there's ample cover to get under with friendly infantrymen. Diving into an enemy trench cause your own side is shelling your plane to deny it to the enemy (especially the Lewis guns!) might lead to some bonding through shared misery, but is not exactly going to get you brownie points and make friends.

So wind is the least important of the beneficial factors of coming down behind your own lines by starting behind your own lines. Much more important is that the people are nicer, and, assuming you didn't hurt anything, you and maybe even your plane get a nice ride back to the aerodrome, not the less than fun PoW camp.

pdf27 wrote:
Johnnie Lyle wrote:
There's also training. Bloody April revealed that the Germans were a lot better at training their pilots, and resulted in a serious revamp of how the British ran their flight schools and how many hours a pilot had before they got into combat. The use of Jastas as flying fire brigades also helped - good pilots were concentrated and shuttled from trouble spot to trouble spot, giving the Germans a qualitative pilot advantage they wouldn't have had normally.

That should be amenable to the sort of stats you're doing, although you probably don't have the time to do it on that level of detail - mortality rates versus logged hours.


If we have a rough average of pilot hours over time, we could use it as a covariate - average pilot skill at time of despatch. We don't really have logged hours, just the time the plane was in service.

pdf27 wrote:
Johnnie Lyle wrote:
The one I would be interested in looking at is the parachute. German and Austrian aviators had them, starting in 1918. Allied ones didn't. It'd be interesting to see if the parachute gave a German pilot a significantly greater chance of surviving longer, and thus getting more experienced.

The PM's already looked at this one a bit - early parachutes were so heavy and fighters so light (420 kg empty for a Camel) that the aircraft take a major performance hit from carrying both the parachute and the pilot.

The question is whether the performance hit was worth it. Plenty of Allied pilots thought it would be (commentary is available here: http://ww1centenary.oucs.ox.ac.uk/machi ... parachute/) and the Germans and Austrians clearly did think it was, since they issued them starting in 1918. If we had German data, we could quantify this.

Anectdotally, Allied policy on parachutes was not a performance issue, but that the pilot would prematurely leave the aircraft. I've seen this reported lots of places, but can't tie it down to any official source.

Author:  Francis Urquhart [ Fri Sep 25, 2015 1:54 pm ]
Post subject:  Re: RAF Fighters V6.2

Johnnie Lyle wrote:
Anectdotally, Allied policy on parachutes was not a performance issue, but that the pilot would prematurely leave the aircraft. I've seen this reported lots of places, but can't tie it down to any official source.


I've never come across a reputable example of that opinion being stated. The only official report I've seen stated categorically that the available parachutes were too large and heavy to be carried in scout aircraft and the effects on performance and balance were dire.

The nearest I've come is in some (particularly American) accounts dating from the 1930s. They seem to be based on the "callous top brass who cared nothing for the men on the front" meme. There were inventors in WW1 who claimed they had designed parachutes that could be made small and light enough for use in scouts but the non-use may well have been simply a case of too many inventors. WW1 pilots also claimed post-war that the top brass didn't allow the use of chutes. All of this material was then picked up in the 1960s when there was a spate of "lions led by donkeys" stories inspired by "Oh what a lovely war" and based on the 30s material.

Parachutes were used by men in balloons and I do have a statement that the early parachutes only worked when used from reasonably stationary platforms. The early parachutes used with observation balloons were fixed release devices. The observer leapt from the balloon tethered to a line which jerked open the parachute. This method did not work well at all with fixed wing aircraft because when the time came that a parachute was needed, the plane was typically on fire or spinning out of control. It was in 1919, the year after the war ended, that the individually controlled release was invented, one which allowed the parachuting pilot to free fall and clear the plane before opening the chute. So, the major reason that fixed wing pilots didn't have parachutes was simple....they seldom worked.

It's worth noting that German aircraft had more power than RAF and French aircraft so the extra weight may have been acceptable. The parachute carried by the German aircraft worked like this

Attachment:
parachute.jpg
parachute.jpg [ 56.09 KiB | Viewed 907 times ]


The system weight about 30 pounds and it worked about one time in three. Thirty pounds was equivalent in weight to a second Spandau or an armored seat or self-sealing fuel tanks.

Author:  Johnnie Lyle [ Fri Sep 25, 2015 2:45 pm ]
Post subject:  Re: RAF Fighters V6.2

Which is why I want to model it. We have a ballpark estimate of a 33% effectiveness, and a measured penalty. Depending on the exact introduction date, we also have pre/post parachute exposures and losses for most German aircraft that could have used them.

So, if we can get a good mass of German aircraft data, we can see whether they were worth using, or if they were not.

Much like the "is the Camel deadlier" issue, we can see if perceptions are correct, nor if not backed up by data, explain why people thought that way.

Author:  Andy L [ Wed Jan 13, 2016 8:54 am ]
Post subject:  Re: RAF Fighters V6.2

Why does the Spitfire Mk IX score higher than the Spitfire VIII? The Mk IX was the older Mk V airframe with the Merlin 60, the MK VIII was a new airframe designed for the Merlin 60.

Andy

Page 2 of 5 All times are UTC - 5 hours
Powered by phpBB® Forum Software © phpBB Group
https://www.phpbb.com/