3
$\begingroup$

I have 1 categorical factor (3 treatments) and 1 continuous factor (weight) and then I have 5 continuous response variables.

From what I have read, I should not use a two way ANOVA as one of the factors is continuous. Is this correct? Should I be using a Multiple Regression instead?

I was advised that I can use ANOVA, but I'm not sure if this is correct based on what I have read. I could convert the continuous factor to categorical, but I have also read on this site that this is not the preferred option.

My aim with the data is to see if there is a significant difference between the 3 treatments in regards to the response variables, which would be a standard one-way ANOVA, but I also want to see if weight effects the response variables.

My analysis will be with R.

$\endgroup$
  • 5
    $\begingroup$ This question has come up before. ANOVA and so called "multiple regression" are equivalent; in R just use lm() in either case. The more tricky problem is how to deal with your five responses (rather than the more traditional one response variable). There are ways to do this though. $\endgroup$ – Peter Ellis Jan 24 '13 at 19:14
  • 3
    $\begingroup$ @PeterEllis, is right. An analysis w/ both categorical & continuous variables is often called ANCOVA (note the "C"). It may be that this is what you were advised, & you mis-heard or they accidentally mis-spoke. Typically w/ ANCOVA you are primarily concerned with the categorical vars, & the continuous are considered nuisance vars; also traditionally ANCOVA means you don't have interactions b/t the continuous & categorical. But all of this is subsumed by multiple reg, so you can do as you like. What is the deal w/ your 5 DVs? Are they all measures of the same idea, or independent questions? $\endgroup$ – gung - Reinstate Monica Jan 24 '13 at 21:18
  • $\begingroup$ Thank you for your advice. @gung Actually I was advised ANOVA, but maybe they didn't understand the part about having a continuous factor. My DVs are all measures of the same idea, for example, duration or frequency (all relating to sound measurements). $\endgroup$ – user20045 Jan 24 '13 at 22:30
  • $\begingroup$ @gung Just to add, am I correct in saying MANCOVA should be more suited to my data than ANCOVA as it also allows for more than one DV, and that MANCOVA & a Multivariate Linear Model are the same thing? Thanks again for all the help. $\endgroup$ – user20045 Jan 25 '13 at 0:11
  • 2
    $\begingroup$ My guess is that MANCOVA is a reasonable option for your situation. $\endgroup$ – gung - Reinstate Monica Jan 25 '13 at 0:51
3
$\begingroup$

This appendix (Multivariate Linear Models by J. Fox) should help you to deal with your multiple responses (see section 2 with the Iris data illustration).

$\endgroup$
  • 6
    $\begingroup$ +1, Welcome to the site, @EpiFunky, that's a nice contribution; Fox is really great. One of our goals here on CV is to create a permanent repository of actionable statistical info in the form of Q's & A's. As such, we worry about link rot. Would you mind giving a precis of the pdf in case the link goes dead in the future? (BTW, info like this is given in our about page & our FAQ, which you may want to read since you're new here.) $\endgroup$ – gung - Reinstate Monica Jan 24 '13 at 21:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.