| Author |
Message |
< ASReml ~ Bootstrapping unequal samples in ASReml
|
| kabuck |
 Posted: Mon Apr 23, 2012 7:18 pm |
 |
|
|
Joined: 23 Feb 2012
Posts: 6
|
I am trying to estimate heritability for a repeated measures trait in captive birds.
A data set for one treatment considerably smaller than my other treatments, and I am worried that differences observed between the models will reflect the unequal sample sizes between the treatment groups.
Here's my idea: Can I use bootstrapping to reduce the size of the larger data set to be the same as the smaller data set in order to detangle effects of sample size from legitimate genetic effects between treatments?
Is it possible to automate this process in ASReml to prevent me from bootstrapping in excel and running many separate models?
Kenton |
|
|
| Back to top |
 
|
| Arthur |
| Posted: Tue Apr 24, 2012 11:46 pm |
|
|
|
Joined: 05 Aug 2008
Posts: 277
Location: Orange, NSW
|
Dear Kenton,
You could use a transformation to create a uniform random variable,
and then use it to discard records from the population you wanted to reduce
in size. (A ferew more trnsformations).
When you have that working, insert a
!CYCLE 1:1000
to do it 1000 times.
That should achieve your purpose. |
_________________
Arthur Gilmour
Retired Principal Research Scientist (Biometrics) |
|
| Back to top |
 
|
| kabuck |
| Posted: Tue May 08, 2012 6:05 pm |
|
|
|
Joined: 23 Feb 2012
Posts: 6
|
Dear Arthur,
Thank you very much for your advice; it has been extremely helpful.
I ended up using the !SETU qualifier to create a uniform random variable. I couldn’t figure out how to discard a specific number of samples, so I used !D to arrive at an approximation.
!DEBUG !LOGFILE
Heritability of ZEFI HG levels at 0.6 dietary dose
ANIMAL !P
DATE !A
ORIGIN !A
REAR !A
BLOODHG
RANDOM !SETU 5 807 !D<1.3625
ZEFI.ped !skip 1
ZEFI_BOOTSTRAP2.dat !skip1 !DDF 1 !FCON !PRINT 6
!CYCLE 1:1000
BLOODHG ~ mu DATE !r ANIMAL ide(ANIMAL)
VPREDICT ZEFI_BOOTSTRAP.pin
I know this is a weird operation to perform in ASReml. I will keep looking for a way to randomly discard a specific number of records. |
|
|
| Back to top |
|
| Arthur |
| Posted: Tue May 08, 2012 10:12 pm |
|
|
|
Joined: 05 Aug 2008
Posts: 277
Location: Orange, NSW
|
1) There is no way of using this mechanism to drop a specific number of records. Just a proportion which will be subject to random sampling.
2) If you want the record dropping to be conditional on family,
you will need to add transformations to identify the family and incorporate that in the test.
3) The whole concept of mixed models is that there is weighting according to family size so I expect you will discover your fears are unfounded. |
_________________
Arthur Gilmour
Retired Principal Research Scientist (Biometrics) |
|
| Back to top |
|
|
|
![[vsni.co.uk]](/common/images/address.gif){kind=small}
Contact us | Site map
News Feed
