Hello.
I have an outdoor experiment with 2 species, 5 salinity levels and 2 light treatments (50% and 100% daylight). To produce the light treatments we used shade cloth and for practical reasons we created one large plot with no shade cloth (100%) and another with shade cloth (50%). Within each light treatment each species were combined with all salinity levels and replicated 8 times. To even out random variation from the two light plots, shade cloth and all plants were switch over approximately half way through the experiment (day 36 of toal 90 days). Initially i have analysed the data with a three way anova but i have been told that the nesting is a problem and switching over does account for this. Therefor my statistical analysis is supposedly flawed. To my understanding im not able to conduct an nested analysis as i only have 1 replicate of each light tretament. I was adviced that an GLM is more appropriate due to the design!? Can anyone shed light on why an GLM should be more appropriate? Any alternative analyses?
Please provide references to support your answer and indicate you research field(biology, economics etc...) in you answer.
Thanks in advance :)
Dear Emil，sorry，I do not understand " replicated 8 times " and " i only have 1 replicate of each light tretament". What is the number of total samlpes? I guess the number of total samples may be 2*5*2*8=160.If so, why " only have 1 replicate of each light tretament " ?
Of course, If you have only one big shape cloth, some researchers may think it is a " simple pseudoreplication". After all, you may need 8 shape clothes and 8 light big blocks, but this is difficult.
I think linear regression is a good way. An example is shown in picture 1.
GLM may be Ok, because this model uses fitting but it also needs analysis of variance. If you want to use ANOVA, you can use two-way ANOVA in light and shade treatment to explore individual and interactive effects of species and salinity.
If you have more problems, you can send email to me ([email protected]) or Huijia Song.
thanks for the reply. regarding the 1 replicate of the light treatment (each block with shadecloth or without) this was in relation to a nested anova. you are correct in that i have 160 samples in total. i thought that linear regression was similiar to an ANOVA and therefor also assumes the same?
i think i have been able to transform my parameters successful but i still dont understand why/how a GLM is better than a ANOVA for my experimental design...?
In simple terms, linear model, including simple or mixed, allows no replicates. Becasue linear model is used to explore the relationshps between X and Y. But ANOVA must have replicates (n>=3), becasue of comparation the variance in this anylsis.
For example, "Biotic interactions drive ecosystem responses to exotic plant invaders" published in the Science, 2020. The replicates is n=2, in this study, authors used fit linear mixed effects models to explore the 3 kinds of treatments on the traits.
Fit linear mixed effects models is a very very useful analysis, which can solve almost all problems. It has minimal requirements for the data.
- Badges
- Science topic