Difficulty with 3-way ANOVA

[#1]

Hi everyone,

I am having some trouble running a 3-way ANOVA test. I think I am inputting it correctly based on what I have see others do (fit model, dependent variable in the y, full factorial macros). But my analysis does not generate an F ratio or Prob > F. 

Another issue I see is that my data do not have all the same N (there are only 10 550 lm + guide in the Working Channel column, but 20 empty and 20 550 lm).  If anyone has a work around for these problems, I would appreciate it!

Sincerely,

Kevin

[#2 · Accepted solution]

Hi @BizMarkie : There are a few problems that I can see. See my plot below,  Notice that not all combinations (of Working Channel, Irrigation System, and Scope) are present so that will cause some issues/biases if you include interactions in your model. Also, notice there are no replicates/repeats (there is only one observation per combination) so you have no degrees of freedom for error (and consequently no F ratios and p-values) if you include the 3-way interaction. So, in short, you are trying to fit a model with terms that can't be estimated, via an unbalanced data set without replication. I see your response is a mean from N=2.  Why not analyze the raw data directly? That said, if they are repeats, rather than replicates (pretty good descriptions here https://sixsigmadsi.com/repetition-vs-replication-key-differences/ ), then analyzing the means as you have may be appropriate.

[#3]

Thank you for your response! That is a great idea. I just ran the raw data. In my analysis, I have the "Lost DFs" prompt. Is this because my data is unbalanced? I have no data for Wolf 24.5 and Wolf 26 scopes with 550 lm + guide Working Channel. 

Also, my DFs go to 0 for Irrigation System. Why is that? Should I eliminate all of my 550 lm + guide data to balance everything out?

Thank you again!

Kevin

[#4]

Hi @BizMarkie : Yeah, having data at all combinations is important if you want all the interactions. There is a good discussion here.

 https://community.jmp.com/t5/Discussions/I-am-receiving-Lost-DFs-Biased-and-Zeroed-parameter-estimat...

I don't often recommend removing data to balance things out; but, as you say, you may consider that or just look at main effects only to get a sense for where the big movement is in the data. If you do that, it looks like Scope and Irrigation System are the big players. Then build a more complete model (with the interaction) with just those two factors. There is no real "right thing to do" here...though we could probably come up with all kinds of wrong things!  

[#5]

Got it. I think it would make sense for the purpose of our analysis to take out the lm 550 + guide  data to compare apples to apples. I can always do a separate analysis without 24.5 Wolf and 26 Wolf to include the lm 550 + guide. 

I have attached my data here. Thanks!

[#6]

Certainly is an option @BizMarkie !