[R] Package survey

Thomas Lumley tlumley at u.washington.edu
Wed Oct 22 22:32:43 CEST 2008


I think this has already been fixed in version 3.9-1.

Also, you probably want to add na.rm=TRUE to the call

  svyby(~p_igov,~div_a,desenho_nps,svytotal,drop.empty.groups=TRUE,vartype
  =c("se","var","cvpct"), na.rm=TRUE)

otherwise the estimated total will be NA whenever any observation is NA.

 	-thomas

On Wed, 22 Oct 2008, Rita Sousa wrote:

> Hi,
>
> Iÿÿm using the svyby for total statistics, for example:
>
>
>
> svyby(~p_igov,~div_a,desenho_nps,svytotal,drop.empty.groups=TRUE,vartype
> =c("se","var","cvpct"))
>
>
>
> In the numerical variable p_igov (and others) I have many non responses
> but if I maintain the NA it doesnÿÿt work.
>
>
>
> summary(base_nps$p_igov)
>
>   Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's
>
>    1.0     9.0    51.0   208.7   185.0 19470.0  1687.0
>
>
>
> desenho_nps <-
> svydesign(id=~npc,strata=~estr_final,weights=~pond_nps,fpc=~nps_univ,nes
> t=T,data=base_nps)
>
> svyby(~p_igov,~div_a,desenho_nps,svytotal,drop.empty.groups=TRUE,vartype
> =c("se","var","cvpct"))
>
> Error in if (any(coef(object) < 0)) warning("CV may not be useful for
> negative statistics") :
>
>  missing value where TRUE/FALSE needed
>
>
>
> If I replace the NA by zero it works without problem, but the doubt
> still remains because the cv estimated can be not real. Isnÿÿt it?
>
>
>
> base_nps[is.na(base_nps)]<-0
>
> svyby(~p_igov,~div_a,desenho_nps,svytotal,drop.empty.groups=TRUE,vartype
> =c("se","var","cvpct"))
>
> div_a statistics.p_igov se.p_igov var.p_igov cv%.p_igov
>
> 1     1          261791.2  9306.518   86611275   3.554940
>
> 2     2         1634590.5 45330.366 2054842079   2.773194
>
>
>
> I beg for a reply.
>
> Thanks.
>
> ---------------------------------------------------
> Rita Sousa
> Departamento de Metodologia e Sistemas de Informaÿÿÿÿo
>
> INE - DP: Instituto Nacional de Estatÿÿstica - Delegaÿÿÿÿo do Porto
>
> Tel.: 22 6072016 (Extensÿÿo: 4116)
> ---------------------------------------------------
>
>
>
>
>

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle


More information about the R-help mailing list