I would like to know if CFU is considered a continuous or discrete variable , in order to determine the appropriate statistical tests to use.
Thanks in advance
Assuming you have the potential to get values from, say, 0 to 600 CFU/100ml, rounded to a whole number (e.g. 126 CFU/100ml) --- it's no more discrete than any other measurement variable that is rounded off to three significant figures. Compare: 12.7 cm, 19.5 °C .
A few other notes: I assume you are measuring CFU / 100 ml, or something similar. This kind of data is often log-normally distributed, which is why the geometric mean is often used. Your data has a hard left edge (0 or non-detect). The data may also be right-censored (at 300 or 600 or 1200). Gamma regression is probably appropriate for this kind of data. But most commonly, the log of the values is used, and resulting distribution approximates the normal distribution.
Thanks a lot , you mean that I should first transform the data to log ?
When I tried testing the normality of data I found it is not normally distributed.
I can't answer that. It depends on your data and what analysis you are going to conduct. But often, for bacterial counts in water, a log transformation is used and the resultant distribution is close enough to normal to use traditional parametric tests (like anova). In the U.S. benchmarks for bacteria in water are often expressed as geometric mean.
Actually the data was CFU in water assesed for 15 patients at zero time and after several laser treatment , so the data is for related samples. I used Non parametric friedman two way anova for related samples to test for significant difference in CFU between the groups and it was powerful enough to detect the difference
- Badges
- Science method