Timeline for Do all interactions terms need their individual terms in regression model?
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S May 21, 2018 at 15:28 | history | edited | Michael R. Chernick | CC BY-SA 4.0 |
fixed typos and made a bit easier to read
|
| S May 21, 2018 at 15:28 | history | suggested | theforestecologist | CC BY-SA 4.0 |
fixed typos and made a bit easier to read
|
| May 21, 2018 at 15:19 | review | Suggested edits | |||
| S May 21, 2018 at 15:28 | |||||
| Sep 25, 2014 at 21:47 | comment | added | Greg Snow | @Ben, I would suggest doing a full and reduced model test. Fit the model with the term(s) of interest and without, then compare them. If you want to compare models where the one model is not nested in the other then you can use AIC (or BIC) for that comparison (but you don't get a p-value). | |
| Sep 25, 2014 at 6:06 | comment | added | Ben | Yes right, but I mean to know if Z has interaction effect on X in 2008, I have to compare the t and p-value of B1X with B3XZ*2008.. is that right or I must create BX*2008? in other words, how can i know whether B3XZ*2008 improve the relationship? thanks a lot @Greg | |
| Sep 24, 2014 at 15:42 | comment | added | Greg Snow | @Ben, I am not sure that I understand the last question, but basically when making predictions you would use B1X for non 2008 and the combination of B1X and B3XZ for 2008 (B3 does use information from other than 2008 because of other assumptions made). | |
| Sep 24, 2014 at 12:57 | comment | added | Ben | thanks a lot that was helpful. Is it ok to compare the output of the interaction with $$B1X$$ although $$B1X$$ is for the whole period and the interaction is only for the year 2008 | |
| Sep 8, 2014 at 16:00 | comment | added | Greg Snow | @Ben, It is certainly possible to fit the above model which basically says you think there is (or may be) an interaction in 2008, but not in any other year. If you have a justification for this, then I think the model is fine. But it is an unusual enough assumption that you will probably need to justify this to any audience. | |
| Sep 7, 2014 at 15:26 | comment | added | Ben | thanks for your reply. I have two ways interaction and want to interact that with a dummy year. I only want to assess the year 2008 (regulation year) on the two ways interaction. so is it ok to use $$ Y= B_0+B_1 X+B_2 Z+B_3X*Z*2008+yeardummies $$ X & Z are continuous variables, Z is the regulation rating. the year 2008 is scored 1 and 0 for other years. so it is like if i only take the observations of the year 2008 without interaction. I read about the weak and strong heredity principle, but didn't clearly understand | |
| Sep 4, 2014 at 17:33 | comment | added | Greg Snow | @Ben, it can depend on both how you parameterize your variables (in my example switching the 0/1 to 1/0 for either variable would change the interpretation) and what questions you are trying to answer and what assumptions you are willing to make. | |
| Sep 4, 2014 at 12:44 | comment | added | Ben | So do you mean it all depends on the objectives of your study OR based on your parameters? | |
| May 8, 2012 at 5:04 | history | answered | Greg Snow | CC BY-SA 3.0 |