I acquired a set of data with larvae dealing with pesticide. To describe the toxicity of pesticide, I record the behavior responses under different treatments. For example, I recorded the "move ability" after pesticide treatments. I recorded the number of larvae with 6 different classes: "behave normal and swim quickly, behave normal and response slow, body convulsion & response slow & can not swim, move after several touches, alive and little response after touching, dead". I am wandering how to analyze this kind of data. Can I apply 0-5 values to each class (from zero influence to sever affected) and do Anova analyses?
Thank you very much for answering : )
You should use non-parametric statistics for ordinal data. Look at the following web site for a free package of such tests: http://vassarstats.net/
A chi-square test of association may be appropriate. ... Do you consider the measured variable MoveMobilty to be ordinal (ordered categorical) or nominal categorical ?
Please explain your coding scheme. It appears you have several variables (convolution, movement without prompt, movement after prompt). How did you come up with your six value scale (what were your inter-rater reliability statistics for this?). It seems you have one binary variable with presumably little measurement error (dead or not) and then some variable based on the amount of movement (and presumably can ignore the movement with nudge since that looks just to classify dead or not). So the first question is whether you have some metric for amount of movement? Second, how many/what proportion dead? Would a hurdle model make more sense?
Thank you Daniel Wright . We did not count the amount of movement during a period of time. I just recorded different behavior responses at a certain time and ranked the behaviors. I attached the figure of my data.
You are right that I can try with dead or not (binary data). However, this will lose some details of low concentration pesticide impacting the larvae.
Bin Jiang , what does this: "I just recorded different behavior responses at a certain time and ranked the behaviors" mean? What did you record? Ranked according to what? How many raters did you use and what statistics do you have to checking the reliability of this scheme? The point I was making about dead or not likely being binary, is that your model would be something like a hurdle model, so the variable after the hurdle can be metric or otherwise.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation