Stats! Stats! Stats! Another year of AH2!

[Lusche]

No I just added kills & deaths.

And I don't see the need to remove "kills vs itself", because I don't see it as double counting at all. Why? Because there are two players involved, both flying the same plane type. I see every kill as two "events", one kill and one death. One player wins, one player dies, regardless if it was Spit 16 vs P51D or Spit 16 vs Spit 16. If I would remove one "event" from a Spit 16 vs Spit 16 kill, I would basically remove one player, one "sortie" from the data pool. It would also distort the numbers again, because some planes with frequent usage and sufficient range (P-51) will fight more frequently vs each other, while rarer planes with short range have less opportunity to kill each other (Me 163, La-7)

And I'm just addign up the number of "events" to get a rough feel for how much a plane is used in the MA, in lieu of better data like sorties flown.

[dtango]

No I just added kills & deaths.

And I don't see the need to remove "kills vs itself", because I don't see it as double counting at all. Why? Because there are two players involved, both flying the same plane type. I see every kill as two "events", one kill and one death. One player wins, one player dies, regardless if it was Spit 16 vs P51D or Spit 16 vs Spit 16.

And I'm just addign up the number of "events" to get a rough feel for how much a plane is used in the MA, in lieu of better data like sorties flown.

It's matrix math.  You are double counting some of the events.  Take a look at the example above.  If we look at Model C:

Kills = 5 + 9 + 13
Deaths = 1 + 1 + 13

We are counting 13 twice.  The 13 is 13 events, not 26 events.

This number also affects the total aggregate because you have to remove all the double counts from the total %.

Tango, XO
412th FS Braunco Mustangs

[Lusche]

Let's take an example:

Fighter A) A very frequent and long ranged fighter, thus quite often fighting his own kind. Let's say a P-51

Fighter B) Somewhat less frequent and very short range, so it naturally happens to kill it's own kind less frequent. Let's say: La-7

Premise:
The P-51 kills 30 P-51's and 20 La-7's
The La-7 kills 30 P-51's and 5 La-7s

My version (simple kills & deaths):
P-51 has 50 kills and 60 deaths = 110
La-7 has 35 kills and 25 deaths = 60

Result: P-51 has usage number 65%, La-7 35%


Your version (removing all kills on "itself")
P-51 has 20 kills and 30 deaths = 50
La-7 has 30 kills and 20 deaths = 50

Result both planes have usage number 50%. Which would lead to the conclusion that both planes are having about the same commonness, which is not the case when you look at the raw data.

[dtango]

No I just added kills & deaths.

And I don't see the need to remove "kills vs itself", because I don't see it as double counting at all. Why? Because there are two players involved, both flying the same plane type. I see every kill as two "events", one kill and one death.

The problem is that the Kills+Deaths by model approach doesn't treat every kill as two events.  It only treats same model vs. same model situation as two events.

Tango, XO
412th FS Braunco Mustangs

[dtango]

Ships passing in the night.  Let me look at your latest post and consider .

Tango, XO
412th FS Braunco Mustangs

[Lusche]

Ships passing in the night. 

 

[Murdr]

		A Kills	B Kills	C Kills		Total Deaths
A Deaths		5	5	5		15
B Deaths		4	2	9		15
C Deaths		1	1	13		15
Total Kills		10	8	27		45

If we us the Kills+Deaths method if we are examining Plane C for instance we would get the following:
15 Kills + 27 Deaths = 42

I'm just trying to follow this discussion.  I understand what you are saying about the aggregate total, but on the other hand, the results are not raw totals but percentages. 

On your grid.  Via Lusche's method C=15+27=42   42/90=46.667% usage
HMM on the grid C is responisble for 33.3% of all deaths, and 60% of all kills.  The average of those 2 percentages gives me Lusche's number.
C=33.3% of deaths and C=27/45=60% of kills.  (.33+.6)/2=46.667% average
B=33.3% of deaths and 22.2% of kills (.333+.222)/2=27.778%
A=33.3% of deaths and 17.7% of kills (.333+.1778)/2=25.555%
100%=All kills and deaths.  While we do not have true data points for all true usage, an average between the percentage of total kills and percentage of total deaths seems like a resonable metric.

Via what you've said
C=29 / Ajusted total=70   29/70=41.4%
B=21/70=33.3%
A=20/70=28.5%

I look at this and percentage wise I am a bit perplexed. 100%=deaths+kills minus same plane kills?  C is still responisble for 33.3% of all deaths, and 60% of all kills.  How do we get to 41.4% usage from those 2 numbers?  Since we are looking at percentage results, what are we gaining by removing planeocide  events?

[dtango]

hehe .

OK my statement still stands.  The kills+deaths by model approach doesn't treat every kill as a double event.  It only treats the same model vs. same model as a double event therefore you need to remove the double count in this scenario.  I'm not saying remove all kills for same model on model.  If just have to remove the double count.

A clarification using your example:

	P51K	La7K	Total D
P51D	30	20	50
La7D	30	5	35
Total K	60	25

Let's look at the P-51.  The kills+deaths by model math only counts 20 deaths of the P-51 by the La7 and 30 kills of the La7 by the P-51 both once but it counts the 30 kills of the P-51 on itself twice.  

The kills+deaths by model approach doesn't treat every kill as two events.  That's why you have to adjust it one way or the other to count it correctly.  You either have to double the other kills and deaths to count them twice or remove the double count for the same model.

2ndly to clarify, my fix is not to remove all the kills for a given model of itself, just the double count so that you're counting it only once.

So the math works like this:

P-51: kills(30+20) + deaths(30) = 80
30 p-51 deaths left off because it's counted once in the kills already

La7: kills(30+5) + deaths(20) = 55
5 la-7 deaths left off because it's counted once in the kills already

Tango, XO
412th FS Braunco Mustangs

[dtango]

Murdr:

I'll try to explain with a pictoral example.

Tango, XO
412th FS Braunco Mustangs

[bongaroo]

Yea, I proved the, "All countries are exactly identical theory", was a complete load of festering, maggot ridden crap years ago. Anyone who says they're the same in terms of either pervasive behavior or cumulative end results is simply spewing a torrential maelstrom of stinky propaganda diarrhea that would make Joseph Gerbils proud.

I did that country analysis to confirm my suspicion that my direct observations were supported by hard data back then. Your data just reconfirms is wasn't just a two-year trend, but a permanent pattern with very little, if any, indication of significant change in five years, which in itself is quite remarkable considering a lot of individuals have left AH, joined AH or changed countries since then either permanently or on a rotation basis.

Birds of a feather flock together, always have, always will. AH's country demographics was, is and always will be a reflection of that most basic human psychosocial dynamic.

Everytime I read one of your posts I have trouble imagining how any single human body could hold so much ego.

 

[Lusche]

 The kills+deaths by model approach doesn't treat every kill as a double event.  It only treats the same model vs. same model as a double event therefore you need to remove the double count in this scenario.  

It's simply not true. It's not counting double. Removing model vs model kills would greatly skew the numbers, which are to reflect usage. Another, last, example:

4 Players taking off. One Spit, one LA, two P-51. Four sorties. Two of them (=50%) made in P-51
The Spit kills the La. One Pony kills another Pony.

Result:
Spit (1 kill) Usage=1 = 25%
La-7 (1 death) Usage=1 = 25%
P-51  (1 kill and 1 death) Usage=2 = 50%

When removing same model kills, it would be
Spit Usage=1 = 50%
La7- Usage=1 = 50%
P-51 Usage=0 = 0%

Which one does reflect reality better?

[dtango]

It's simply not true. It's not counte double. Removing model vs model kills would greatly skew the numbers, which are to reflect usage. Another, last, example:

4 Players taking off. One Spit, one LA, two P-51. Four sorties. Two of them (=50%) made in P-51
The Spit kills the La. One Pony kills another Pony.
Result:
Spit Usage=1
La-7 Usage=1
P-51 Usage=2

When removing same model kills, it would be
Spit Usage=1
La7- Usage=1
P-51 Usage=0

Well, technically using what I proposed it would be:

Spit Usage=1
La7 Usage=1
P-51 Usage=1

But your example is a good one because it convinces me that I'm wrong .  Sorry for the fuss guys! 

I think I've spotted my trouble with double counting.  Kills and deaths being both sides of the same coin yet counting them as individual events trips me up for some reason!   

However it also points out some of the pitfalls in using this approach.  What happens if the surviving P-51 goes on to kill the Spit?

Spit Usage=2
La7 Usage=1
P51 Usage=3

The P-51 sure gets used more than the other planes.  But technically we've only had the actual usage of:

Spit=1
La7=1
P-51=2

But I'm being melodramatic here.  I humbly support the use of kills+deaths as a proxy for usage with a bit of handwaving to explain away the pitfalls!

Tango, XO
412th FS Braunco Mustangs

[Lusche]

Well, technically using what I proposed it would be:

Spit Usage=1
La7 Usage=1
P-51 Usage=1

Uhm yes, I removed one too many ...


but hey, math had always been my worst subject in school. 

[Bruv119]

math geek purse fight 

Interesting stats lusche thanks for the time and research 

[Murdr]

Murdr:

I'll try to explain with a pictoral example.

But your example is a good one because it convinces me that I'm wrong .  Sorry for the fuss guys! 

Does this mean I'm not getting a pretty picture now?