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

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

From |
Nick Cox <njcoxstata@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: How to test heterokedasticity for 120 periods |

Date |
Wed, 20 Feb 2013 11:38:50 +0000 |

Just to add that I know that *scedasticity is in principle a matter of how the errors behave. In practice, however, if you choose a good scale on which to handle the response, the errors will usually be better behaved. My own sermonising take is that less focus on what the errors are, or might be, doing or more focus on how the response is expected to behave would improve many data analyses. On Wed, Feb 20, 2013 at 10:02 AM, Nick Cox <njcoxstata@gmail.com> wrote: > Like Roberto, I don't know why you are doing this, but I do wonder > what the strategy is here. > > Roberto sketches out code for what you ask, but what are you going to > with all the test results? > > 1. With 120 tests, some fraction is going to show up as significant at > conventional levels, even if the null is always right. This problem is > widely discussed under various headings, one of which is multiplicity. > > 2. If some tests come out significant and the rest not, what do you > intend to do? > > 3. Even if cross-sectional regressions make sense, the pattern of > results over time has to make sense too. > > 4. Heteroscedasticity could mean different things, most obviously > outliers as well as more systematic inconstancy of variance. The test > results are going to be silent on why it occurs. > > I'd -collapse- the data to period, mean of response and variance of > response, look at that as time series and as a scatter plot. It might > well be that such exploration exposes the need for some transformation > or link function and that side-steps all that *scedasticity stuff. > Most simply, it may even be that working on a logarithmic scale makes > more sense. Of course, you could be in a field that regards looking at > the data as strange or suspect. > > Nick > > On Wed, Feb 20, 2013 at 7:57 AM, Roberto Liebscher > <roberto.liebscher@ku.de> wrote: > >> although I do not know exactly but why are you testing for >> heteroskedasticity for each period separately (Is it appropriate in your >> case to estimate your model for each period subsample separately?) you could >> loop over the regressions using a command like this >> >> forval i = num1/num2 { >> quietly reg depvar indepvar if period == `i' >> quietly estat hettest >> display "Period " "`i'" >> display `r(p)' >> } >> >> If you want to conduct an ANOVA for the period groups this link might be >> helpful: >> http://www.stata.com/capabilities/anova-ancova/ > > Xixi Lin [edited] > >>> I am trying to test test heteroskedasticity for 120 periods. I randomly >>> tested for 3 periods, using White tests, and the result shows it is >>> homoskedasticity. But I want to make sure that all 120 periods show >>> homoskedasticity. Does anyone knows how to write a loop to test it? As >>> a result, I hope it tells me which period or if any periods are not >>> homoskedastic. * * 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: How to test heterokedasticity for 120 periods***From:*Xixi Lin <winnielxx@gmail.com>

**Re: st: How to test heterokedasticity for 120 periods***From:*Roberto Liebscher <roberto.liebscher@ku.de>

**Re: st: How to test heterokedasticity for 120 periods***From:*Nick Cox <njcoxstata@gmail.com>

- Prev by Date:
**st: Placement of legend above "note" in -grc1leg-** - Next by Date:
**Re: st: comparing equality of coefficients from two subsamples** - Previous by thread:
**Re: st: How to test heterokedasticity for 120 periods** - Next by thread:
**Re: st: How to test heterokedasticity for 120 periods** - Index(es):