How to run ANOVA in this case with repeated measures?

Comments

• EJ

  September 2021

  Hi Friedrich,

  I think you first have to decide whether you want to use ANOVA or a linear mixed model.

  Cheers,

  E.J.
• Mientusz

  September 2021

  Hi E.J.,

  thanks for the information. I honestly don't know which option is preferable. I want to compare the average of each variant with the additional information from each dish and see if there is any effects.

  Best,

  Fritz
• EJ

  September 2021

  I'll ask someone from to team to assist (I am a bit swamped right now).
• JohnnyB

  September 2021

  Hi @Mientusz,

  Based on your last reply I think a linear mixed model is the way to go here. This model will account for each individual dish, while allowing you to compare the average (i.e., fixed) effect between the different groups.

  I looked at the csv you uploaded here, and it seems you some columns that have no header - did something go wrong there? If I have the correct data format, I can make you a jasp file to illustrate how to use jasp for mixed models.

  Kind regards

  Johnny
• Mientusz

  September 2021

  CRESS Blankdaten JASP.csv

  CRESS Blankdaten JASP.xlsx

  
  
• Mientusz

  September 2021

  Thanks Johnny for Your kind offer!

  I am wondering about what You refer to mentioning that some columns do not have a header? I uploaded both the csv and an Excel file and in both all columns have a header - same in JASP? However, for some root length there simply isn't any data because the seeds didn't germinate....

  Does this help?

  All best,

  Mientusz
• JohnnyB

  September 2021

  Hi @Mientusz,

  No problem, it turned out to be a problem with the delimiter character, but all good now!

  So I have some more questions:

  • You would like to test for differences between the substrates, and between the FerroSorp factors?
  • Is germination rate included, or is it not relevant for this analysis?
  • You have 10 columns for each seed, so you have basically 3x10 = 30 observations for each combination of FerroSorp and substrate? where each of those 30 observations is a separate "participant"?

  The tricky thing with these analyses is that the data need to be in a specific format (long format). While JASP does not offer the possibility to go from wide to long (it's on our todo list still), I can do this in R for you, but then I need the answers to the questions above..

  Cheers,

  Johnny
• Mientusz

  September 2021 edited September 2021

  Hi Johnny,

  thanks so much!

  • The germination rate can be avoided.
  • I would like to test if there are effects caused by the substrate and - more important - by FerroSorp. Substrate probably is pretty obvious but FerroSorp I have some difficulties.
  • Exactly, each combination was repeated with 10 seeds. The test should consider the variation within each repetition so I can't just use the average of a repetition because valuable information might be lost. I have never used the term participant but I guess that's what You refer to.

  All the best,

  Mientusz
• TarandeepKang

  September 2021

  I was just going to see if I can lend a hand, but it looks like you are already being taken care of. Just to clarify for you; it seems that most if not all of the developers of JASP are psychologists, as are many of the users. So yes we are in the habit of modelling the response to subjects or participants, in human behavioural experiments, but though your seeds are not participants, it is from them that you take your observations, so it's just a case of differences in terminology between disciplines!
• JohnnyB

  September 2021

  Hi @Mientusz,

  Thanks for the additional info (and thanks @TarandeepKang for the clarification!).

  What I ended up doing was to run a RM ANOVA, with each Seed column as a repeated measurement, to keep the design intact (so having 10 levels of a repeated factor I called Seedling). However, this variable is not of interest (unless you are interested in differences between the seeds), but we can still look at differences between the two between-subject factors Substrate and FerroSorp. You can take a look at the JASP file I uploaded here, where I ran both a Bayesian and frequentist RM ANOVA this way. You can see that there are no differences between the levels of FerroSorp (Inclusion Bayes factor of 0.1, p-value non-significant, raincloud plots indicate no difference). The Bayes factor here allows you to quantify evidence against the inclusion of FerroSorp (i.e., evidence for the null hypothesis, which would be 1/0.1 = 10). Please let me know if you have additional questions - hopefully I understood your design correctly.

  Cheers,

  Johnny