Data Censoring and Normality Tests

From: Berton Gunter

No Andy, they're (left) censored alright: their values are 
known to be below
2.4, but not how far below. Truncated data "arise when 
observations are
actually observed only when they take on values in a 
particular range (above
2.4, say). For observations outside that range, their EXISTENCE is not
known." -- Meeker and Escobar, "Statistical Methods for 
Reliability Data."

Cheers,
Bert
 
 

-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch 
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Liaw, Andy
Sent: Tuesday, November 09, 2004 3:12 PM
To: 'Kenneth Frost'; r-help at stat.math.ethz.ch
Subject: RE: [R] Data Censoring and Normality Tests

I don't know much about censored data, but what you described 
doesn't sound
like censored data, but rather truncated data.  

I think normality tests are over-(ab)used, and would suggest 
inspection of
normal QQ plots instead (as you already do).  For formal 
tests, you might be
able to cook up something that test for truncated normality.

Cheers,
Andy

From: Kenneth Frost

Hello,

I would like to know if there is a function in R that 

will test for 

normality and handle censored data sets.  Currently, I 

evaluate each 

censored data set by the extent to which a normal scores plot 
approximate a straight line.  For complete data sets I use 
shapiro.test().

Below is an example of a censored data set.

data1<-c(0.00, 0.00, 0.00, 5.86, 5.17,  8.17,  5.12, 4.92, 
7.08, 5.73, 
5.44, 6.61, 6.34, 6.23, 5.97, 5.86, 5.15, 7.98, 6.72, 5.15, 
3.58, 6.86, 
6.12, 4.58, 6.07, 5.38, 5.21, 3.78)

The zero values occur because I cannot detect a pathogen 

with a value 

below 2.4.  Using shapiro.test(), it seems that the data are 
different 
from a normal distribution.  However, the normal scores 

plot suggest 

otherwise.

Using R version 1.9.1

Thanks in advance.

Ken