Hi Colleagues,
For example. I have a data set which can be categorized based on sampling time and treatment. However, I only concern the main effect of treatment on parameters I measured. What I did is to conduct one-way ANOVA with treatment as the only factor to analyze data for each sampling time, respectively (e.g., day by day).
I want to know that is there any rules ask us to must conduct or report the two-way ANOVA before a one-way ANOVA even we don't want to discuss in the paper? I only want to enhance the readability and stress the effect of treatment.
Thanks.
Best,
Guijun
There is no such rule. However if you use an experimental design for a two way anova. It would be incorrect to analyze it as a 1 way ANOVA See for example the differences between a completely randomized Design and a factorial design in Montgomery at this link:
https://www.amazon.com/Design-Analysis-Experiments-Douglas-Montgomery/dp/1118146921/ref=pd_rhf_cr_p_img_5?_encoding=UTF8&psc=1&refRID=157G22TZZRKYQB2MJTXW
Best, David Booth
There is no such rule. However if you use an experimental design for a two way anova. It would be incorrect to analyze it as a 1 way ANOVA See for example the differences between a completely randomized Design and a factorial design in Montgomery at this link:
https://www.amazon.com/Design-Analysis-Experiments-Douglas-Montgomery/dp/1118146921/ref=pd_rhf_cr_p_img_5?_encoding=UTF8&psc=1&refRID=157G22TZZRKYQB2MJTXW
Best, David Booth
There is no rule, but one could analyze as two-way ANOVA and if the interaction term is not significant it is then a one-way ANOVA problem. However, I would be careful in this because it is assuming that a failure to reject the null hypothesis (that the interaction term is zero) is equivalent to proving that the interaction term is actually zero. That conclusion is not a proper use of statistics. In any case, you will have to present the two-way ANOVA first. After that I don't see a point in also presenting one-way ANOVAs where the outcome is either redundant or wrong.
Analysis depends on planning of experiment.If u have designed 2-way expt. then u should do it.
If is a 2way experiment, or interaction is not significant (and correctly Timothy A Ebert showed the risk in not considering interaction) or you have to present all the effect, starting from interaction. If interaction is significant this mind that the effect of one factor is not constant along the level of the other one factor and strictly speaking, main effect could be not considered at all and you have to discuss interaction only. I read about sampling in time... if there are repeated measures on the same subject you have to use an ANOVA model for repeated measure and not the standard ANOVA.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis