[問題] 如果比例的95% CI.上界超過1可以嗎?

看板Statistics作者時間17年前 (2009/01/22 11:17), 編輯推噓0(0018)
留言18則, 3人參與, 7年前最新討論串1/1
想請問一下 如果在算 p_hat 的95% C.I.時 得到的 p_hat=0.8235 及 95% C.I.=(0.642,1.005) 95% C.I.的公式為 p_hat +or- 1.96*sqrt[ p_hat*(1-p_hat)/n ] 上界的部份超過1 一般會直接照公式就將此組C.I.寫上? 還是會將上界用p的最大上界1 代入呢? 謝謝!!

01/22 11:26, , 1F
Replace it with 1; or use exact method to obatin CIs.
01/22 11:26, 1F

01/22 11:30, , 2F
若是以1取代的話 在哪本書或哪篇paper內有提到呢?
01/22 11:30, 2F

01/22 11:31, , 3F
因為我印象中有看過取代1的這種方式 不過一直找不到
01/22 11:31, 3F

01/22 11:31, , 4F
在哪裡看到的.. 因為要證明給別人看用1取代..
01/22 11:31, 4F

01/22 11:32, , 5F
所以希望能有本書或文章可以證明..謝謝!!
01/22 11:32, 5F

01/22 11:32, , 6F
Do you know how to interpret CI???
01/22 11:32, 6F

01/22 11:33, , 7F
100次有95次那個p_hat包含在此區間內 <-這樣對嗎?
01/22 11:33, 7F

01/22 11:33, , 8F
哈.. 是此區間包含p_hat...
01/22 11:33, 8F

01/22 11:34, , 9F
The problem is that the use of asymptotic variance
01/22 11:34, 9F

01/22 11:35, , 10F
to construct CIs may not yield good coverage probability.
01/22 11:35, 10F

01/22 11:36, , 11F
espically when \hat p is close to 0 or 1.
01/22 11:36, 11F

01/22 11:37, , 12F
That's why I recommend using exact CIs instead.
01/22 11:37, 12F

01/22 11:38, , 13F
Hey, your interpretation of CI is completely wrong.
01/22 11:38, 13F

01/22 11:38, , 14F
那可以請教一下exact CI要怎麼求得嗎?
01/22 11:38, 14F

01/22 11:39, , 15F
use binomial distribution
01/22 11:39, 15F
※ 編輯: goodafternoo 來自: 140.128.153.200 (01/22 11:48)

01/22 12:04, , 16F
oops, 11樓 should be p instead of \hat p
01/22 12:04, 16F

11/09 14:41, , 17F
100次有95次那個p https://daxiv.com
11/09 14:41, 17F

01/02 14:49, 7年前 , 18F
Do you know https://noxiv.com
01/02 14:49, 18F
文章代碼(AID): #19T-LQHJ (Statistics)