+ Reply to Thread
Results 1 to 2 of 2

Thread: Removing interaction effect on 2-way ANOVA?

  1. #1
    Points: 119, Level: 2
    Level completed: 38%, Points required for next Level: 31

    Thanked 0 Times in 0 Posts

    Removing interaction effect on 2-way ANOVA?

    Hello, I've run a 2 way ANOVA of my research results and got these 3 p-values: A=0.001, B=0.069, and A*B=0.167. My professor now wants me to remove the interaction effect from my analysis. Does anyone know how I can do this? In the program I am using (SigmaPlot), there is no option for removing the interaction effect. Also, what exactly would this do to the data and why is it important to remove it? I want to have a clear understanding of this, but I do not know what my professor is asking. He wants me to figure all of this out myself, and I have tried, but I do need help. Thanks in advance.

  2. #2
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39
    Miner's Avatar
    Greater Milwaukee area
    Thanked 405 Times in 363 Posts

    Re: Removing interaction effect on 2-way ANOVA?

    This is one of those topics that will inevitably result in disagreement. A purist will say that you should only create an ANOVA table for those factors that you "A Priori" want to test for significance. Others will say that you include all factors and interactions first, then simplify the model by removing non-significant factors until you end up with a model consisting only of significant factors. As one that has applied ANOVA many, many times in industrial situations, I am firmly in the latter category.

    I cannot help you with the limitations of your software, but the methodology is really very simple. Okay, I confess, I had to calculate ANOVA by hand for 15 years BC (Before Computer). But really, all you do is add the SS for the AB interaction to the SS for the Error term. Add the df for the AB interaction to the df for the error term. Calculate the new MSE and use it to recalculate the new F-ratios and p-values.

    The reason that it is important is that it make for a more powerful test.

    Now, I am certain that there will be some vehement disagreement. In my defense, I have had 25 years of success in an industrial environment doing it this way. If it didn't work, I would be explaining a LOT of scrap and rework. Success in application beats theory any day.

+ Reply to Thread


Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts

Advertise on Talk Stats