Close Home Forum Sign up / Log in

Interpreting regression / anova output in R (categorical IVs with more than 2 levels)

T

Hi there

Can anyone suggest a detailed tutorial on how to successfully interpret regression output in R for a model that has an IV that has 3 levels (e.g., conditions A, B, and C)? I am very familiar with running linear models in R but up to now my categorical IVs have only had two levels (e.g., male/female, low/high). It's simple to interpret the output when this is the case. But when there are three levels to the IV it is rather confusing.

I have had a search and read about doing different types of contrast coding to help you interpret the output, but I haven't managed to find a tutorial that is really clear and that I can actually follow and use as a guide to interpret my output. If anyone could suggest one, or provide steps on how to do it, that would be great. Thank you!

Ps. I realise this isn't a stats forum. I'll post on one eventually if I need to.

E

Have you thought to ask the question in researchgate? You are more likely to get an answer.

T

I have and may do. I don't really like posting questions on there for some reason.

E

Quote From Tudor_Queen:
I have and may do. I don't really like posting questions on there for some reason.

Yes for privacy reasons. It is not nice when a colleague or line manager knows you are asking a question. I had to create a new account in a technical forum bcause my former account was my first name :)

T

Yes, come to think of it that is probably it! :D I ask lots of questions in person, and have no problem with it. But I think on the net I prefer to have a degree of anonymity and use a nickname.

A

You are really a nice person. I always like to read your replies to people. Always helpful. I wish I could help you but I'm really rubbish with stats. I hope you find the help you need.

I have done a fair bit of ANOVA and multi-variable statistics. I don't know if my experience is applicable to social sciences but I think you should be looking at lack of fit tests. With multi variable modelling over fitting the data is super easy and biases most regression values. So stuff like an F-test can tell you if your model is over fitted or use the predicted r^2. Also once you have a working model you can remove variables and see if it improves/deteriorates. I think ANOVA is less about interpreting the model but seeing how easily it breaks.

Unfortunately, I can't recommend any videos as I inherited some code from my supervisor and she was rather good at explaining the stats side. Are you just using R or another piece of software?

T

Thank you AmITAA! :)

Thanks Rewt - I appreciate your response. It's literally the interpretation of the output I need for categorical variables with more two levels. It's dead simple when there's two levels (eg. treatment vs control), as the first can be interpreted by the value of the intercept, and the second is the value of the predictor (in the output table I mean), and if it's significant then there was a sig diff between the two conditions. But when there's three levels you have to mess around with coding and somehow interpret it differently. I'll figure it out. Just being lazy and not trying hard enough to follow the online tutorials. It's R I'm using and I shouldn't have said anova in the title of my post as it's mixed regression (using the lme4 package) - although I thought maybe it might be similar to interpreting the output of anova so I put that too.

Hi Tudorqueen, I think interpreting the statistical data should be the same across both ANOVA and mixed regression, it is just requires coding. The only thing I can recommend is write out the multiple levels and coding on pen and paper beforehand, so you understand everything before translating it to R. I hate programming stats so make sure I have a clear plan before writing anything on the computer. Goodluck!

T

Yes exactly... and I am being too lazy to understand the coding... still haven't done it :D

59720