Date:         Wed, 19 Apr 2006 09:53:49 -0300
Reply-To:     Hector Maletta <hmaletta@fibertel.com.ar>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Hector Maletta <hmaletta@fibertel.com.ar>
Subject:      Re: Minimum number of cases for computing correlation
Comments: To: Rita Clivio <rclivio@tradelab.it>
In-Reply-To:  <4tkn68$9s6t43@ibs01.interbusiness.it>
Content-Type: text/plain; charset="us-ascii"

Rita,
There is no simple answer to your 'simple' question. Consider the following
factors:
1. Regression may involve any number of variables. An old rule of thumb, not
very well justified, says you need at least ten cases per variable. That is
an absolute minimum: better results need more.
2. As you suspect, finite populations may help. The standard error of an
estimate (or the size of the sample needed for a given standard error) is
affected by the so called finite population correction which is (N-n)/(N-1)
where N=population and n=sample size. This you should look up in a
statistics text for more details.
3. A regression with a given sample size may be statistically significant if
all cases lie near the regression line, or not if cases are more scattered
around. If your data closely conform to a linear equation you may need a
smaller sample.
Hope this helps.
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] En nombre de Rita
Clivio
Enviado el: Wednesday, April 19, 2006 7:25 AM
Para: SPSSX-L@LISTSERV.UGA.EDU
Asunto: Minimum number of cases for computing correlation

Dear List



I have a simple (?) statistics question.



When you run a correlation, which is the minimum number of case you have to
consider ?

For regression maybe is 30 cases ? Under 30 cases regression is not
recommended.

Is always referred to my population, i.e. 7 cases from a population of 50
units and 7 cases from a population of 1000 units are different ?



Thanks in advance



Rita