[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
"Maureen Paul" <M.Paul@warwick.ac.uk> |

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

Subject |
st: RE: A question on quadchk |

Date |
Mon, 18 Nov 2002 15:10:40 +0000 |

According to the example in the FAQ, when the quad point was increases from 12 to 24, there was an improvement. In my case as well, increasing the quad points also led to an improvement. Also, the manual (v6) states that "When the number of points is increased, the range spanned by the points is also increased, and, on average, the points are only slightly closer together". Somewhere else mentioned that precision is increased sightly. It seems that these bits taken together suggest that increasing the quad points may help somewhat. Still, what should I make of things if I increase or reduce the quad points and still have one or two variables with a relative difference greater than 1%? Thanks >>> n.j.cox@durham.ac.uk 11/18/02 01:00PM >>> Maureen Paul > I am testing the reliability of the quadrature approximation after > running a random effects logit model. However the relative > difference on > 3 of my variables exceed 1% (they are 2 job sector dummies > with relative > differences of 1.3% and 48% and 1 industry dummy with a relative > difference of 1.3%) . I have increased the quad points (to > 20) but this > does not change the results. All other coefficients are fine. > > Looking at the example in the book, it appears that if one > coefficient > is problematic, the reliability of all the estimates are > questionable > whilst reading the FAQ suggest that if all other > coefficients are fine > while one may be problematic, I should be reasonably > confident with the > results. > > Since there does not appear to be much gain in increasing the quad > points, should I be confident with my results even though 3 > coefficients > exceeds 1% with one have a relative difference of 48%? > I am no expert on this, but a small but possibly important detail here is that sometimes you are better advised to decrease the # of quad points. Think of this as like retreating from approximating with a polynomial which has too many terms. This point is elaborated in the manual. Nick n.j.cox@durham.ac.uk * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: from variable to macro** - Next by Date:
**Re: st: -cluster- dissimilarity measures** - Previous by thread:
**st: dstdize: still no labels** - Next by thread:
**st: RE: survey median estimator** - Index(es):

© Copyright 1996–2017 StataCorp LLC | Terms of use | Privacy | Contact us | What's new | Site index |