ANCOVA with moderators

sverm

New Member
#1
I am currently doing research with an independent variable (four treatment groups), a continuous dependent variables, 5 control variables, and 2 continuous moderators. I figured out that I needed to do an ANCOVA analysis with the 5 control variables as covariates and the treatment groups as factor. I just cannot figure out how I can include the moderating variables into the analysis. Do I put them into the analysis as covariates (moderator1 & moderator1*treatment & moderator2 & moderator2*treatment) or should I do this differently, for instance not including the interaction variables?
 
#2
If you don't include the interaction variables you won't be able to test for moderation. So, if you don't want to use moderating varaibles as covariates (which to be honest, I am not sure if you should do or not), I can think of 2 options.:
1. Use a regression analysis. Problem is, you will have to turn your IV into dummy variables and use 15 different interaction variables.
2. Transform each continuous variable into an ordinal variable, with only a few possible values. Problem is, you will lose resolution. Plus, you need a big sample. However, this comes with a benefit - you will be able to detect non-linear patterns, if they exist.
 

sverm

New Member
#3
Thank you for your response. I will probably do a regression analysis with dummy variables then. I already did turn my IV into dummy variables. I don't fully understand how you came up with 15 different interaction variables. I should multiply my moderator with each dummy variable, right? So then I'll have 2(moderators)*4(treatment groups) = 8 interaction variables, or did I understand it wrong? Thanks in advance!!
 

Karabiner

TS Contributor
#4
4 treatment groups require 3 binary dummies, the last dummy would be redundant. So you'd get 3*2 interactions .

With kind regards

Karabiner
 
#5
Thank you! So you mean that I need to include 3 dummy variables for the independent variables and 3 interaction terms per moderator in the analysis? Or do I need to include 3 dummy variables for the independent variables but 4 interaction terms (for each treatment group) per moderator?
 
#6
You will have 3 dummy variables and 2 moderators. That is 6 2-way interactions. However, to truly understand the relationship between the variables you will have to also calculate another 3 3-way interactions. 9 in total (not 15, which for some reason seemed logical to me as I wrote my previous message)
 
#7
Thanks again for your response. I have conducted the linear regression with the 6 2-way interactions. All interaction terms are not significant. Can I conclude then that there is no moderating effect?