My experiment design contains two factors: instar of larvae (contain 3 levels: "2th","4th","6th") and days with chemical treatment (contain 5 levels:1,2,3,4,5(day)).
Then I measured a series of results, such as death or alive( 0 or 1), pupal weight,food utilization rate...blabla..
Some results were measured each after a day (all of them were integer data or binary data about the behavior of the larvae) and some results were got at the end of the experiment.
However,"6th" and "4th" larvae could complete their development in 3 days. So the treatment was designed like this:
X1 X2
2th: 1day, 2day, 3day, 4day, 5day
4th: 1day, 2day,
6th: 1day, 2day
I don't think it's a good method to regard "days with chemical treatment" as continuous variable.
Is any good method to deal with it? How to explain the interaction between two factors?
A thousand thanks!!!
Hello, if you design is unbalanced one option is do a factorial anova with type III sum of squares and LSM (least-squares mean). This two aproximations resolve the problem Read this document and search other papers with this two concepts. Bye bye
http://documentation.statsoft.com/STATISTICAHelp.aspx?path=GXX/Glm/Overviews/VisualGLMHypothesisTestingTypeIIISumsofSquares
The design is a little unclear... Are you measuring each after 1 day, and after 2 day, etc., or do you just care about the results at the _end_ of the treatment?
Thanks Karin!
Your adivice is constructive, I'll look into it.
Best wishes!
Hi,Mr Salvatore!
Thanks for your question.it's my fault
Some results were measured each after a day and some result were got at the end of the treatment.
I think these results measured by each days is not independent, because the result were not only effected by two factors of this experiment but also infulunced by some results which was measured before this result. For example, some results about the behavior of the larvae which occured in first day would have an effect on the latter result in second or third day. Nevertheless,all of them were effected by two initial factors.Do you have any good idea about it?
Best wishes
You are right about the need for repeated measures for the measurements where the same subject was measured on day 1, day 2, etc.
But let's ignore that for the moment.
Take a look at the following example and see it makes sense. You can do the same with your experiment, doing a one-way anova where each treatment is a combination of treatments from your experiment.
http://rcompanion.org/rcompanion/h_01.html
When you do get to the part of specifying the repeated measures, you can use glht for the contrasts with either nlme or lmer.
BTW, yes, I'm treating chemical treatment as a nominal (factor) variable.
I think you could present results with a plot like the following, with chemical treatment along the x-axis, and some chemical treatments would have just blue boxes (2nd instar).
http://rcompanion.org/rcompanion/images/d_08_01.jpg
Yes, ggplot handles this fine.
Thanks a lot, Salvatore!
I fully understand it. It‘s a perfect way to solve this problem.
best
Try to take a look at Bayesian models. They assume different probabilistic distributions and all are probabilistic models that take into account uncertainty. These models handle naturally unbalanced data, offer you the opportunity to make predictions and give naturally probabilistic interval values, instead of the ad hoc confidence intervals.
I know it, but it looks like a little obscure to me. I‘m trying to be a beginner on bayes statistics recently.
Thanks a lot ,Jhonathan
Jincheng Zhou,
I am sure if you take these video classes you will be able to handle it easily:
https://www.youtube.com/watch?v=pid0lUH467o
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology