Timeline for AIC and its degrees of freedom for linear regression models
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 24, 2019 at 18:09 | vote | accept | Rodvi | ||
| Dec 24, 2019 at 18:06 | comment | added | usεr11852 | 1a. Yes. 1b. Yes, but do notice that for BIC we cannot compare models where M1 is not nested within M2. 2. Yes. Example: we want to model $y$, we did an experiment measuring $X_{Oc}$ and $y_{Oc}$ last October, got $N$ samples, we did the same experiment again in November, and recorded $X_{No}$ and $y_{No}$ got another $N$ samples; the fact we have two samples of equal size $N$ measuring the same variable $y$ does not mean we can compare the models $M_{Oc}$ and $M_{No}$ AIC. (The example is a bit simplistic but it can happen if we have access to the same $y$ from different data sources). | |
| Dec 24, 2019 at 17:54 | comment | added | Rodvi | Thanks! So, as I understood, the answer on my first question should be "yes" (we can compare AIC's values of all three models and choose the best one with the lowest AIC). And this is true even for BIC. But I want to ask – what do you mean by "models that potentially have different rows of data"? "Different rows" mean samples from completely different probabilistic distribution? | |
| Dec 24, 2019 at 17:06 | history | answered | usεr11852 | CC BY-SA 4.0 |