regression on censored data

[ariklee]

Hi everyone. I am trying to do statistical analysis on censored data. I have paired data: for each stream flow measurement I have a bacteria count from a sample collected at the time the stream flow was measured.

The laboratory sometimes does not make enough dilutions on the bacteria sample, and perhaps one in twenty results is censored, for example, ">2000". How do I relate stream flow and bacteria count with data that is censored? The bacteria data is not always censored at the same value.

Here is a subset of the bacterial count data (the "ND" is non-detect, which means less than the detection limit but not necessarily zero, more like "<2". I don't have a problem with those. Only I cannot understand how to use the >2000 and >20000 and I even have one that is >2,000,000):

130
52
>2000
130
120
9500
180
11000
>20000
180
570
410
ND
500
370
7700
6500
12000
4300
4
2
ND
220
32
670


Thanks anyone for any help at all !!!

[bugman]

I suggests you look at TOBIT regression models.

Here is a good resource:

Statistical Methods
in Water Resources
By D.R. Helsel and R.M. Hirsch

Tobit models are covered on page 371.

Phil

[bugman]

Its available to download free through the United States Geological Survey