Doing a linear mixed effects model.

[#1]

Hi guys,

I am currently working on a project, where I have three independent values and a dependent value:

What I have attempted to do, is a three-way ANOVA, and here's what my project supervisor has written back to me: 

What you have done is a linear model with three independent variables (you can see that the results include a three-way interaction).

You could do a linear mixed effects model, where respondents could be included as a random effect. That would be more appropriate, as you have repeated measures.

You can read about one-way-ANOVA here: https://statistics.laerd.com/statistical-guides/one-way-anova-statistical-guide.php

Now I do have a question, and perhaps, more context needs to be added in. But let say for instance, I do my analysis like this: 

One-way ANOVA for the dependent variable and independent variable1(most important one)

Two way for: the dependent variable and independent variable1, 2

Three-way for: the dependent variable and independent variable 1, 2, 3.

And the reason why I wanna do this is to see whether the independent variables affect the dependent variable and each other.

However, seemingly, my project leader claims that  a linear mixed effects model, would be more appropriate, as because I have repeated measures. Could someone perhaps explain to me why that is? 

I am sorry for asking, I am relatively new to JMP etc.

[#2]

The first thing you need to do, is to go to the JMP Documentation Library and open the Fitting Linear Models document, and read the section on Mixed and Random Effect Model Reports and Options and the section on Mixed Models.

The JMP Documentation Library is available under the Help pull down menu.

[#3]

What you have done assume that your observations are independent. What your supervisor suggests implies that your observations are NOT independent. If "repeated measure" is the correct hint, then ask yourself: your dependent variable is repeatedly measured on what? E.g. multiple measurements on a lab rat, multiple measurements on a piece of metal, multiple measurements on a piece of land, etc. If your observations are not independent, but you analyze them as if they are, you may reach a wrong conclusion.

[#4]

Hi Peng

Thanks for the reply,

My dependent variable is a response to stimuli, that the participant's show after being exposed to a stimuli. Each participant has 4 different stimulis exposed to them, and I calculated their responses.

My primary independent variable is the effect of price, but there is also two another independent variables in scent category and gender, I want to examine.

I have tried to copy the steps of the video, and this is what I got.

[#5]

Looks like you found how to specify the analysis. If you are using JMP Pro, "Mixed Models" personality is the other platform that dedicates to this subject.

One thing I would like to point out, since you did not include in your screenshot. For such analysis, an important part is to interpret the Variance Component Estimates result.

I am not going to comment on your analysis and result, you should discuss with your supervisor.

[#6]

Hi Peng,

First and foremost, the results + interpretation is on me, correct.

What I actually found was an add-in that did the full-factorial analysis, and I got out it.

So is there anything other than the Variance Component Estimates result that is important to comment on?

[#7]

This is a warning sign. It means you don't have enough data to estimate the high order interaction. You may want to remove that interaction term from your model.

[#8]

Hi Peng,

I seem to have figured out why my observations are dependent, because I measure the participant's reaction before the stimuli and after each participant.

So that mean's that an ANOVA can't be conducted. In experiments similar to my own, people conduct one-way anova to find the Prob > F, which under the One-way ANOVA is referred to as: analysis of variance. Whereas, under the linear mixed effects model the table that is must similar to the analysis of variance, is the Fixed Effect Tests. So I assume, I should use this analysis table (or perhaps, if there's a significant difference in those two tables, you could enligthen me?)

[#9]

It is the time to discuss with your supervisor.

[#10]

In addition to the question, I had: Can one make an bivariate analysis in dependent observations? I am asking, because of it has analysis of variance in the data table, and is it that value you use when you comment whether there's a correlation between the two variables?