CODEWITHSUNDEEP
rlinear-regressionlm
Josep Ramos
Josep Ramos

Reputation: 11

Error conducting regression (stepwise) AIC is -infinity for this model, so 'step' cannot proceed

This is my first consult, happy to share the community :) (You would notice I'm not a native english speaker and also I'm not very good using R, actually these are my firsts steps).

I'm trying to conduct a regression (stepwise). I have one continuous dependent variable (General assessment of seminar quality) and I'm trying to know how well other questions I asked my students (quality of the materials, the level of participation, etc.) predict a good global assessment.

All variables are continuous (1-7)

Here is the code:

Regressió_Total_Forward <- lm(Val_general ~ Val_tema+Val_material+Val_interacció+Val_participa+Val_oral+Val_feina+Val_fatiga+Val_temps_pràctica+Val_ajust_temps, data=Sem_1)
step(Regressió_Total_Forward,direction="backward")

And R returns this:

Error in step(Regressió_Total_Forward, diretion = "backward") : AIC is -infinity for this model, so 'step' cannot proceed

What should I do??

Thank you <3

If I try to do a "forward" or "both" regression, the same error appears.

Upvotes: 1

Views: 34

Answers (1)

Roland
Roland

Reputation: 132969

I can reproduce the issue like this:

DF <- data.frame(x1 = 1, x2 = 1, y = 1)
 
fit <- lm(y ~ x1 + x2, data =DF)

step(fit, diretion = "backward")
#Error in step(fit, diretion = "backward") : 
#  AIC is -infinity for this model, so 'step' cannot proceed

An AIC of -Inf can only happen with a log-likelihood of Inf, which means a perfect fit. However, due to floating-point inaccuracies, you usually don't get that but get instead just a large negative AIC. step then gives a warning "attempting model selection on an essentially perfect fit is nonsense". The error only seems to occur if lm returns some NA coefficients. You should check your predictors for collinearity.

Upvotes: 1

Related Questions