GFHi - practical meaning?

[scubadada]

...If I am reading Table 3 above correctly,
Median(StDev) GF among DCS victims was 79, versus
Median(StDev) GF in whole database was 66, with a P<.0001
Ergo, there is a statistical association between higher GF and DCS, notwithstanding the distribution of cases within the database.

Probably the Mean (Range), not that important

The article is very interesting, I thank @dmaziuk for the citation, but beg to differ on the interpretation

 

[rsingler]

You're looking at the posterior distribution and you're positing that the likelihood function is linear increase in surfacing GF. It does not follow. It's just as likely that all of the cases were diving at their computers/tables default conservatism and got bent well within the predicted -- using Craig's numbers above for illustration -- 1 in 323 or 1 in 769. It just so happens that in 75% of the cases that default was in the range of GF 0.7..0.9. Correlation is not causation, with a P of "absolutely certain".

If you run from the priors, like SAUL, you'll have to cut your bottom time in half and your GF: by 100%, to shift your chances from 1/500 to 1/5000. And still some 30 minutes after you surface you'll either be 100% bent or 100% not.

You're far more on top of the math than I am, but even I'm not suggesting a linear increase in risk.
Let's just say that it's "multifactorial", and that a soup-bowl full of dive reports is not a controlled study.
It's always fun debating you @dmaziuk , even if your GFLo might be too low

 

[dmaziuk]

You're far more on top of the math than I am, but even I'm not suggesting a linear increase in risk.

It's not that, it's your own x'ed peds example: you take 16 x'ed peds, their chances of getting x'ed are, after the fact, 100%. Now you run "statistics" on 'em and conclude, with very high probablity, that they had 15/16 chance of getting x'ed on the sidewalk. That's pretty much how GF interpretation is done in the cited study.

If a diver's a priori chances of getting bent were 1 in 5000 and she got bent on her 5001st dive while running GF 0.8, which do you think is more statistically relevant: her GF or her dive count?

How much do you want to bet that 100% of divers "undeservedly" bent while using Shearwater Teric in default OC Rec mode got bent at or below GF 40/85?

It's not the math part, it's the cause and effect part that I'm not buying.

 

[Jay]

The values of A and B make sense to me – beginning and end of deco obligations would mean that the SurfGF ~= GFHi (as said in the OP). However, I do not understand C or D.

- Why did Perdix have a deco obligation end at C /SurfGF(71) instead of B/SurfGF(85)?
- Why are Subsurface and Shearwater GFs so different upon surfacing (69% vs 77%)?

Maybe I am misinterpreting or not understanding what I am really looking at, but any insight would be appreciated

I *tthink* I might know ... re your q1/ in my earlier post @atdotde explained the NDL differences. That in turn would lead to the SurfGF computation being affected (at those points of NDL differences). My dives uploaded were quite 'jumpy' (must have been a swell) but even on dives that were stable (square wreck profiles), I found NDL and (surfacing) GF differences. Out of 8 dives I looked at, only one didn't have a differences (NDL & GF at surface) yet most dives appeared relatively stable, so I assume that even small differences in depth/timing create large enough differences in NDL and SurfGF, however stability did generally resulted in a smaller differences.

I guess more might be gleaned when SWCloud has SurfGF displayed graphically.

re q2/ well, that's a hard one to explain. At the surface, hopefully depth = 0 and is stable ... so I'm puzzled as well.

 

[biggpa]

I *tthink* I might know ... re your q1/ in my earlier post @atdotde explained the NDL differences. That in turn would lead to the SurfGF computation being affected (at those points of NDL differences). My dives uploaded were quite 'jumpy' (must have been a swell) but even on dives that were stable (square wreck profiles), I found NDL and (surfacing) GF differences. Out of 8 dives I looked at, only one didn't have a differences (NDL & GF at surface) yet most dives appeared relatively stable, so I assume that even small differences in depth/timing create large enough differences in NDL and SurfGF, however stability did generally resulted in a smaller differences.

I guess more might be gleaned when SWCloud has SurfGF displayed graphically.

re q2/ well, that's a hard one to explain. At the surface, hopefully depth = 0 and is stable ... so I'm puzzled as well.

Thanks for the reply. I looked at your other thread and it makes sense. I get that there are valid reasons for Shearwater vs Subsurface, but Shearwater claims I surfaced at 77%. It was a short amount of time between the end of the deco obligation and surfacing (less than 2 mins). Based on previous @rsingler posts about accent rate and safety stops it doesn't seem like there should have been a drop from 85% to 77%. Is my assumption that (SurfGF@NDL=0) ~= (GFHi) ~= (SurfGF@EndOfDeco) incorrect?

 

[scubadada]

I find this conversation very interesting. I find a 3 min SS and an ascent at 15 ft/min gives me about -5-6%. A 5 min SS gives me a about 8-9%.

 

[Jay]

At the moment, I think it's small, but perhaps not as small as I thought previously because .... playing around with SubSurface, on a 30m dive for 15m on air, surfacing at 9m/min versus 99m/min (max permitted) produces a difference of 9GFs. (using 16C).

The above quote is from the other thread when @rsingler asked about if the "instantaneous" SurfGF calc input used by SW had been more accurately calc'd than my prior guesstimation of "trivial" (which in turn was based on other member's estimations/reasoning earlier in the thread). Turns out it's not so trivial.

Thanks for the reply. I looked at your other thread and it makes sense. I get that there are valid reasons for Shearwater vs Subsurface, but Shearwater claims I surfaced at 77%. It was a short amount of time between the end of the deco obligation and surfacing (less than 2 mins). Based on previous @rsingler posts about accent rate and safety stops it doesn't seem like there should have been a drop from 85% to 77%. Is my assumption that (SurfGF@NDL=0) ~= (GFHi) ~= (SurfGF@EndOfDeco) incorrect?

The ceilings from SW and SubS ended at different times? (I've quoted your post below, and added that into your graphic), so I think it's a bit of an oranges/apples plus the above NDL/data-point timing issue to compare 85 & 77? (assuming I'm reading it correctly).

SubS shows a 2GF difference between end of deco obligation and (71-69) which seems sensible.

I think saying that (SurfGF@NDL=0) ~= (GFHi) ~= (SurfGF@EndOfDeco) is correct.

A: Start of deco obligation. SurfGF(87), GF99(On-Gassing)​

B: End of Ceiling from Subsurface. SurfGF(85), GF99(26)​

C: End of Ceiling from Sheatwater: SurfGF(71), GF99(8)​

D: Surface. SurfGF(69), GF99(77)​

 

[biggpa]

The above quote is from the other thread when @rsingler asked about if the "instantaneous" SurfGF calc input used by SW had been more accurately calc'd than my prior guesstimation of "trivial" (which in turn was based on other member's estimations/reasoning earlier in the thread). Turns out it's not so trivial.



The ceilings from SW and SubS ended at different times? (I've quoted your post below, and added that into your graphic), so I think it's a bit of an oranges/apples plus the above NDL/data-point timing issue to compare 85 & 77? (assuming I'm reading it correctly).

SubS shows a 2GF difference between end of deco obligation and (71-69) which seems sensible.

I think saying that (SurfGF@NDL=0) ~= (GFHi) ~= (SurfGF@EndOfDeco) is correct.

View attachment 553558​

Thanks for the info - I *think* I am understanding. I am finding it very interesting what/how each software does the "same data". Not nearly as clear as I thought it would be.

I have added SurfGF to my main screen and will start to watch it more carefully in the context of the total conversation in this thread. To me, the actual computer would be the gold standard for SurfGF, GF99, etc so that I can compare it to SW Desktop and SubS for some post dive analysis.

 

[Jay]

Thanks for the info - I *think* I am understanding. I am finding it very interesting what/how each software does the "same data". Not nearly as clear as I thought it would be..

On the topic of Buhlmann differences, this article is quote interesting Why is Bühlmann not like Bühlmann – The Theoretical Diver and there's a neat discussion about anchoring GFLow here ceiling/GF (SW posted #127, and SubS in #82). Buhlmann computers are compared here (showing some decent NDL variations) : Help with simple deco algorithm project


The latter thread was very interesting as there were even NDL differences on the same computer at different times due to variations in barometric pressure.