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

From |
"Nick Cox" <n.j.cox@durham.ac.uk> |

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

Subject |
st: RE: Re: Weighted number of observations |

Date |
Tue, 3 Aug 2004 15:34:00 +0100 |

This isn't a -tabstat- issue as such. -tabstat- just passes the buck to -summarize-, which behaves in the same way. The issue is delicate, but hinges, I surmise, on this distinction. When -summarize- (e.g.) _uses_ the sum of the weights, it rescales first. As quoted, [U] 14.1.6 includes the expression "when it uses them", which thus appears not ornamental, but crucial. When -summarize- _displays_ the sum of the weights, it displays the unscaled sum. Nick n.j.cox@durham.ac.uk Friedrich Huebler > Dear Toyoto, > > The auto data has 22 observations with foreign=1, not 50950. In [R] > tabstat we read: "aweights and fweights are allowed; see [U] 14.1.6 > weight." [U] 14.1.6 states that most Stata commands rescale the > aweights to sum to N. However, -tabstat- does not rescale the > weights. The fact that -tabstat- treats aweights the same way as > fweights is not clearly documented. One could also argue that this > behavior is inconsistent. > > Friedrich Huebler > > --- Toyoto Iwata <iwata@med.akita-u.ac.jp> wrote: > > Dear Friedrich Huebler > > > > You wrote about > > > > > . tabstat foreign [aw=weight], stat(sum) > > > > > > variable | sum > > > -------------+---------- > > > foreign | 50950 > > > ------------------------ > > > > Perhaps I miss the point, but, > > > > .gen eachmean = foreign*weight > > > > .gen sumeachmean = sum(eachmean) /* I don't know the mean of this. > > */ > > > > . list sumeachmean in l > > > > +----------+ > > | sumeac~n | > > |----------| > > 74. | 50950 | > > +----------+ > > > > This seems to agree with the definition of the aweight. > > > > [Online help says,] > > > > aweights, or analytic weights, are weights that are inversely > > proportional to the variance of an observation; i.e., > > the variance of the j-th observation is assumed to be sigma^2/w_j, > > where w_j are the weights. > > Typically, the observations represent averages and the weights are > > the number of elements that gave rise to the average. > > > > > > > The Stata User's Guide states in section 14.1.6: "For most Stata > > > commands, the recorded scale of aweights is irrelevant; Stata > > > internally rescales them to sum to N, the number of observations > > in > > > your data, when it uses them." It would be useful if the Stata > > > documentation could make clear which commands don't use aweights > > in this manner. > > > * * 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/

**Follow-Ups**:**Re: st: RE: Re: Weighted number of observations***From:*Richard Williams <Richard.A.Williams.5@nd.edu>

**st: Testing OLS assumptions in panel data***From:*"Andrea Molinari" <A.Molinari@sussex.ac.uk>

- Prev by Date:
**st: Re: Weighted number of observations** - Next by Date:
**st: Suggestion for an added graph feature** - Previous by thread:
**st: RE: Marginal & Impact Effects in a Tobit** - Next by thread:
**st: Testing OLS assumptions in panel data** - Index(es):

© Copyright 1996–2014 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |