Kaplan-Meier Survival Analysis - advice wanted

#1
Hi

It's hard to find plain-talk on-line instruction in Survival Analysis, so I turn to you all for guidance.

Attached is a Survival Analysis table that I gleaned from techwalla.com, and adapted to my fictitious oak forest.

Will you please take a look at it and advise me if all is fine, or needs work and corrections.

Thanks.

Note: the forum software is not letting me attach the excel 2010 file worksheet (it indicates no files found) so I have to take a snapshot and enter it as a picture. Hope you can still offer me advice.
km.JPG
 

hlsmith

Omega Contributor
#2
So it has been about 10 years since I have done one of these by hand. Without any censoring things look too easy. I know I have mentioned this to you before, but in a lot of stats it is assumed observations are independent. But I wonder if the death of a tree affects other trees, say same culprit kills them or now they have greater resources. Also, are there different causes of tree death, so competing risks (natural death per se versus human or animal intervention)?

You also seem to be doing interval analyses, which seems fine just for getting survival probabilities, but if trying to get median survival or graphing - I would imagine that a tree could have died on the first, middle, or last day of the year - though they get classified just as during year. Keep this in mind when going forward. As mentioned, I don't delve into this area much, so I just wanted to get you some feedback but may not have been a tremendous help.
 
#3
Hey I take anything you have to say as important.

The techwalla chart just seems a little too simple compared to all the maths I've come across. Maybe in lieu of "Number Died" I should have multiple columns listing causes of death. With that, maybe multiple regression would be a better tool - or could you suggest one?

To keep the observations as independent as possible sample trees would be well-scattered, although insect pests can cover extremely vast areas. As far as trees are concerned, the death of one should not per se cause the death of a neighbor although the cause could be the same.

Thanks for your help and feel free to comment more if the spirit moves.
 

Miner

TS Contributor
#4
I use Minitab for reliability (survival) analysis. Because the data are left censored (no idea when the trees sprouted), right censored and interval censored (no idea when during a given year the tree died, I used a Turnbull estimate rather than Kaplan -Meier, but still duplicated your numerical results. See the attached PDF for the analysis.
 

Attachments

#5
Would it not be natural in a Cox regression or a Weibull regression to include explanatory variables like years with strong insects attacks or very dry years (or what ever that can harm trees)?
 
#6
GretaGarbo: you talkin' to me? I'm still trying to get a handle on survival analysis. I'll consult you about Messrs. Cox and Weibull later.
 
#9
I'm sure it would suit me if I knew anything about Cox and Weibull. I'm still very wet behind the ears in this science. I'll look for your advice when I tackle those two later.