Open smishr opened 1 year ago
We need to do something about tests. If we can't match answers with R, or any other language, it won't be good to merge this.
:exclamation: No coverage uploaded for pull request base (
v0.1.1@4fac910
). Click here to learn what that means. The diff coverage isn/a
.
@@ Coverage Diff @@
## v0.1.1 #247 +/- ##
=========================================
Coverage ? 92.66%
=========================================
Files ? 14
Lines ? 218
Branches ? 0
=========================================
Hits ? 202
Misses ? 16
Partials ? 0
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.
We need to do something about tests. If we can't match answers with R, or any other language, it won't be good to merge this.
svyrecvar
is a beast, but Im confident that a Julia native solution will be much faster, with less overheads.
Also Hartley Rao is good starting point. Better than returning a variance at all
@ayushpatnaikgit Should we
SurveyDesign
. So that the four summary statistics return some variance (instead of nothing)Discussing and adding the Taylor series / recvar
might take several months, as well as #277 and other big structural changes will take several months. Until then SurveyDesign
will return HR by default? After that we can change default
add Survey
total
withSurveyDesign
deff
and other srs formulaeI couldnt get R to match Hartley Rao results like for like. R isnt using Hartley Rao algorithm everywhere, only for certain PPS cases.
svyrecvar
isnt the same as HR.