prentice.test anomaly

KnutWittkowski - 22 Feb 2010 - 15:48

Hi Roger,

Only your sixth block contains observations in both groups. Hence, as stated in the warning, 10 of your 12 observations are removed. With only two observations left (one per group), computing a standard deviation would be a challenge If you reverse the role of x and z, you get an informative (albeit still quite extreme) design and some results.

Thanks for helping us in improving error messages.

Best, Knut

RogerKoenker - 22 Feb 2010 - 15:08

I wondered if you could take a look at the code for prentice.test and see whether some better treatment of the following problem is possible:

load("xyz.Rda")

cbind(y,x,z)
# x y z
# 1 1 0.04960942 1
# 2 1 -0.13812130 1
# 3 3 -0.04611998 1
# 4 3 0.44127078 1
# 5 4 0.37898117 1
# 6 5 0.17372399 1
# 7 5 -0.23365750 1
# 8 6 0.37456275 1
# 9 7 0.42939303 1
#10 2 0.28364409 2
#11 2 0.19375529 2
#12 6 -0.08388159 2

require(muStat)
prentice.test(y,z,x)
Error in prentice.test(y, z, x) : object 'Mi' not found
In addition:
Warning message: In prentice.test(y, z, x) :
10 observation(s) in blocks with <2 of 'z' (groups) removed.

Admittedly the example is rather extreme, but the behavior does not seem to be what was intended.

Tia, Roger

Up