Re: se and 1/se

From: Duncan Murdoch (murdoch_at_stats.uwo.ca)
Date: 09/26/04 

Date: Sun, 26 Sep 2004 13:33:21 -0400

On Sun, 26 Sep 2004 18:46:20 +0200, "Gonapse halden"
<l.******daren*******@yyyibelgique.com> wrote:

>Hi,
>
>If b = 1 /a may we say that
>se(b) = 1 / se(a) or var(b) = 1 /var(a)

No, definitely not. Take for example se(a) = 0; then se(b) = 0 too.

The standard error of 1/a is not determined from se(a), but for small
se(a) it can be approximated using the "delta method": a 1st order
Taylor expansion of the function 1/a around E(a).

Duncan Murdoch