I am helping to analyze some bacteria count data and could use some guidance. Milk samples were collected before and after pasteurization on farms and various types of bacteria in the samples were plated and counted. We would like to report log reductions as a result of pasteurization. The sticking point is that some of the counts are zero both before and after pasteurization. Calculating a log reduction for these samples results in a zero value. Should we use the zero LR values or treat those samples as missing data? The concern is that by using the zero values, the effectiveness of pasteurization is biased when the log reductions for all samples are averaged. We also calculate percentage log reduction, and because that calculation requires dividing by the before count, any sample with a zero before ends up with a missing value for percentage log reduction.
0
Log reduction
Started by Coleen Jones, Jan 30 2013 08:36 AM
3 replies to this topic
#2
hobglobin
-
- Active Members
-
- 5,261 posts
Growing old is mandatory, growing up is optional...
60
Excellent
Excellent
Posted 30 January 2013 - 09:28 AM
Can't you use log+1 ?
One must presume that long and short arguments contribute to the same end. - Epicurus
...except casandra's that belong to the funniest, most interesting and imaginative (or over-imaginative?) ones, I suppose.
#3
Coleen Jones
-
- Members
-
- 2 posts
member
0
Neutral
Neutral
Posted 30 January 2013 - 09:54 AM
hobglobin,
We have added 1 to the raw counts and then taken the log. Is it acceptable to add one to the log values?
We have added 1 to the raw counts and then taken the log. Is it acceptable to add one to the log values?
#4
DRT
-
- Active Members
-
- 134 posts
Veteran
4
Neutral
Neutral
Posted 30 January 2013 - 11:23 AM
Just add 1 to the raw values not the logs.
There is an analytical technique called "zero inflated statistics" which might be useful. Ecologists use it but I'm not sure if it is only applicable to poisson (count) distributions, which may make it necessary to separate your data into divisions as if you were drawing a bar graph.
There is an analytical technique called "zero inflated statistics" which might be useful. Ecologists use it but I'm not sure if it is only applicable to poisson (count) distributions, which may make it necessary to separate your data into divisions as if you were drawing a bar graph.
