Notice: On March 31, it was **announced** that Statalist is moving from an email list to a **forum**. The old list will shut down on April 23, and its replacement, **statalist.org** is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Caroline Wilson <wilson_cj@hotmail.com> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
RE: st: high correlation interaction & main effect |

Date |
Tue, 5 Mar 2013 19:18:23 +0000 |

Thank you Maarten - this is very helpful!! ---------------------------------------- > Date: Tue, 5 Mar 2013 10:32:57 +0100 > Subject: Re: st: high correlation interaction & main effect > From: maartenlbuis@gmail.com > To: statalist@hsphsun2.harvard.edu > > On Tue, Mar 5, 2013 at 6:12 AM, Caroline Wilson wrote: > > I'm using Stata to run a multilevel model. I have a few interaction variables in my model, which are the product of a continuous variable (time) and a categorical variable. My variable for time was defined as years from the start of the study, to the nearest day, and was defined this way because it makes sense in the context of my study. The interaction variables are in some cases highly correlated (0.8) with the main effects. > > A couple of questions: > > 1. How much should I worry about the high correlation? > > 2. What, practically speaking, can I do about the high correlation? I'd like to keep the interaction terms in the model since they represent my key research question. > > You should know about it, and know about the consequences: low power, > i.e. you are less likely to find a effect when you should. This is not > nice but it is an accurate representation of the amount of information > available in the data. One way you can see this is by looking at the > minumum number of observations necessary to estimate an effect. The > larger that minimum, the less information is present in each > individual observation. > > Consider a main effect. The absolute minimum number of observations > necesary for estimating it is 2: one observation in group 1, the other > in group 2 and the difference in response is your effect. > > For an interaction effect the absolute minimum number of observation > necessary is 4: You need 2 observations in group a, one each in group > 1 and 2, in order to compute the effect of group 1 versus 2 in group > a. You need another 2 observations in group b, one each in group 1 and > 2, in order to compute the effect of group 1 versus 2 in group b. The > difference in these effects is your interaction effect. > > So striclty speaking there is nothing you can and should do about this > high correlation. > > -- Maarten > > --------------------------------- > Maarten L. Buis > WZB > Reichpietschufer 50 > 10785 Berlin > Germany > > http://www.maartenbuis.nl > --------------------------------- > > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/faqs/resources/statalist-faq/ > * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/

**References**:**st: high correlation interaction & main effect***From:*Caroline Wilson <wilson_cj@hotmail.com>

**Re: st: high correlation interaction & main effect***From:*Maarten Buis <maartenlbuis@gmail.com>

- Prev by Date:
**st: Need help programming for dates in Stata** - Next by Date:
**Re: st: odds ratios and confidence intervals after mlogit with a dummy X** - Previous by thread:
**Re: st: high correlation interaction & main effect** - Next by thread:
**st: fitted values in Poisson and NB** - Index(es):