Date:         Mon, 3 Dec 2001 10:22:58 -0600
Reply-To:     "Straus, Ian" <Ian.Straus@VIAINFO.NET>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Straus, Ian" <Ian.Straus@VIAINFO.NET>
Subject:      Re: Calculating Sample Sizes
Comments: To: Robert Chatfield <robert@BERGENT.COM.AU>
Content-Type: text/plain

You CAN do that (400 &50), but your most sensitive method would be to use
two equal groups.

The big question is: What's your desired confidence level for detecting a
significant difference?  [How badly will this affect you if the real-world
action after your test goes wrong and you get burned?]
Subsidiary questions:
How much of a difference is strategically significant to you? [ If this
score is on a scale of 1-100 and you get a mean difference of 0.05 will you
care?]
Do you have any data now which would indicate the standard deviations of
scores?


Ian Straus
Market Research Specialist
VIA Metropolitan Transit
San Antonio, Texas

> -----Original Message-----
> From: Robert Chatfield [SMTP:robert@BERGENT.COM.AU]
> Sent: Sunday, December 02, 2001 11:18 PM
> To:   SPSSX-L@LISTSERV.UGA.EDU
> Subject:      Calculating Sample Sizes
>
> We want to be able to test whether a change of wording in a
> question creates a statistically significant difference in
> the mean scores of the responses.
>
> We currently have funds to test 450 people and are wondering
> how *few* people we need to administer the new wording to to
> be able to see whether the new wording makes a difference
> (it would be preferable if the new wording made no
> significant difference).
>
> Can anyone suggest a way to estimate how many people we need
> to administer the second wording of the question to in order
> to get a meaningful result?
>
> That is, can we give 400 people the original wording and 50
> people the new wording and simply perform a t-test on the
> resulting means?
>
> If this is a basic question, please forgive me as a humble
> market researcher :-)
>
> Thanks
>
> Robert Chatfield