I'm struggling with multiple linear regression where all independent variables are factors. My data set contain 6 columns: Day; (Levels: Friday Monday Saturday Sunday Thursday Tuesday Wednesday) TVChannel; (Levels: BFMTV C8 CStar France 2 France 3 France 5 HD1 RMC decouverte TF1 TMC) DayPart; (Levels: Access Day Night Peak) Spot; (Levels: Bien Etre Silence) Format; (Levels: 10 24) Visits dependent variable

I'm using lm() in r to built the linear regression:

fit<-lm(Visits ~ Day * TVChannel * DayPart * Spot * Format, data=source)

the summary of the model is quite long, as there is a lot of permutations. I can ask to show me only those of them that have significant impact on the visits, but how can I interpret all of my baselines?

My intercept is

(Intercept) estimate=1.548e+02   Pr(>|t|)=0.830

but this intercept contains all different baselines (one for every variable and one for every level of combinations of this variables).

So first of all how can I see which levels and combinations the model took as baselines and then how can I understand if some of this baselines have or not (significant or not) an impact on my dependent variable?

Thank you for your help!

  • $\begingroup$ What do you mean with baseline? $\endgroup$ Apr 12, 2017 at 14:31
  • $\begingroup$ @Tommaso the reference or a baseline of the dummy variables $\endgroup$ Apr 12, 2017 at 14:37

1 Answer 1


The short and quick answer (also late I supose..) is that the "baselines" that you mention are built, by default in R, as the first level of each predictor factor. So on your example would be:

  • Day: Friday
  • TVchanel: BFMTV
  • etc

Each day of the week will be estimated as an effect compared with Friday and each TVchanel with BFMTV. So what happens If you include the interaction between Day and TVchanel? Let´s take Monday and Cstar. The "baseline" now could be understood as the sum of the effects of Day:Monday and TVChanel:Cstar, and what you are aiming to compare is the interaction effect, I mean, Is there any extra effect on my dependent variable when we have both levels of the variables? If significant there is, otherwise is not, the interaction effect is or not different from 0.

Conclussion, the "baselines" could be built as a sum of all the effects that participate in the interaction you aim to check.

Good luck!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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