4

I have made two linear regressions to estimate y and I get this results: One:

Residual standard error: 1.021 on 276 degrees of freedom
Multiple R-squared:  0.2347,    Adjusted R-squared:  0.2059 
F-statistic: 8.362 on 10 and 276 DF,  p-value: 6.878e-12

Second:

Residual standard error: 1.025 on 273 degrees of freedom
Multiple R-squared:  0.2312,    Adjusted R-squared:  0.1945 
F-statistic: 6.314 on 13 and 273 DF,  p-value: 2.085e-10

I know from $R^2$ that these models are not good, but which one is better from the other one? Can someone can explain the other factors beside R-squared? Maybe use ANOVA to compare?

usεr11852
  • 44,125
econmajorr
  • 155
  • 1
  • 5
  • 1
    This is a very broad question. I tried to go through every item in the output of an ANOVA comparing two models here. I wonder if this is what you need. – Antoni Parellada Feb 04 '16 at 21:32
  • The $R^2$ for these models are $\sim 0.20$, ie. decent unless you have good reasons to believe that your explanatory variables have strong linear relations with your dependent variable and you do not omit any other variables that have strong influence. Also given you do not tell us if these are nested models the easiest thing to do will be to just check their AIC before starting some cross-validation procedure. – usεr11852 Feb 04 '16 at 21:42
  • 2
    Are these nested models? If so, you can conduct an $F$-test – linksys Feb 05 '16 at 05:13
  • Just from glancing at this, I'm guessing that you added a few additional regressors to "Second" model, and that these additional regressors don't contribute much in terms of fit. I could be completely wrong though.

    Something reasonable is to take @linksys advice and test whether the coefficients on the additional regressors jointly are different from zero at a statistically significant level.

    In general though, what's good/bad/better/worse etc... unfortunately involves a lot of context specific knowledge and can be more impressionistic art than hard science.

     – Matthew Gunn Feb 05 '16 at 05:54

1 Answers1

6

As you correctly notice your $R^2$ associated with each model is very similar ($0.235$ vs $0.231$). In general none of the two though is "not good". $R^2$ being bad or good is entirely up to the actual application. As mentioned in my comment, unless you have good reasons to believe that your explanatory variables have strong linear relations with your dependent variable and you do not omit any other variables that have strong influence, an $R^2 \approx 0.23$ is far from catastrophic.

The obvious thing to suggest is to look at some kind of information criteria (Akaike Information Criterion, or Bayesian Information Criterion) to see if any of the two models is obviously better. Let me point out that these are not silver bullets. They make their own assumptions that have to be met. (eg. for the BIC you need your models to be nested.)

The factor "next to $R^2$" is the adjusted $R^2$: This is essentially the coefficient of determination but penalized so it accounts for the fact that you have a given number of explanatory variables in your model. Short of cross-validating or bootstrapping your model I would suggest using the AIC with a correction for finite sample sizes, dubbed AICc. You can use it by employing the AICc function available in package AICcmodavg to do that. Cross-validated has some excellent thread on the perils of automatic model selection (eg. here and here); I highly recommend reading them. To paraphrase the late George E. P.Box: Your model is certainly wrong, you just want to see if it is any useful. :)

Community
  • 1
usεr11852
  • 44,125