svyarpr text says numbers match where they don't

[ajdamico]

if i'm understanding correctly, the new numbers in this changeset make the statement that vardpoor matches convey for linearized designs no longer true ?

https://github.com/guilhermejacob/context/pull/23/files#diff-91dbc690af363b1d2d087e78fec174a76a0437df7150b2d5808dbc4fcdac09f9

[guilhermejacob]

Ah, yes. The linearization-based variance estimators are different, so they no longer match.

[ajdamico]

i can fix but just checking: for the same reason as mentioned here?

https://github.com/guilhermejacob/context/blob/master/01-poverty.Rmd#L313

On Sun, Aug 13, 2023 at 11:45 AM Guilherme Jacob @.***> wrote:

Ah, yes. The linearization-based variance estimators are different, so they no longer match.

— Reply to this email directly, view it on GitHub https://github.com/ajdamico/convey/issues/386#issuecomment-1676396319, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABU5ZYCAJ2ZYMXIRVQKHMLXVDZCFANCNFSM6AAAAAA3OJ4ZGA . You are receiving this because you authored the thread.Message ID: @.***>

[guilhermejacob]

I see... I think I built the book with my "patched" vardpoor. I'll rebuild it again now with the CRAN version.

[guilhermejacob]

Or you could do it to avoid a merge mess, @ajdamico ;)

[ajdamico]

for sure! could you just give me one sentence explaining why the linearization-based variance estimators should differ from vardpoor? i had previously thought the estimates only differed on replication designs..

[guilhermejacob]

Discussed at length here, but possible one liners:

• Because Fprime is the derivative wrt to the estimated threshold, not the estimated quantile (point to Osier);
• Because our linearization results differ (point to issue).

For the usual reader, the second should do, but the first is more helpful, I think.