Date: Thu, 6 Nov 2008 20:59:56 +0100
Reply-To: Marta García-Granero <email@example.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Marta García-Granero <firstname.lastname@example.org>
Subject: Re: nested effects?
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Zdaniuk, Bozena wrote:
> Ok, I get it (or so I hope:)). Now, the reason I asked was because I was trying to get around dyad problem.
> I have caregiver-care recipient dyads that were randomly assigned to treatment or control and I measure them on five outcomes. Pearson correlation tells me that on two outcomes correlation between CG and CR score is significant. According to Kenny, I have no choice but to analyze at the level of dyad. But I was still hoping to analyze on the level of individuals to retain more power. So, I thought I would include a dyad id as random factor nested in the treatment and this way I can say that I remove the impact of dyad from the treatment effect. But it looks like I cannot do it because my individuals are not randomly assigned to dyads, right?
Your design is not nested at all, but paired. Against your idea, if the
correlation between CG&CR score is significant, analyzing your data
using dyad as unit retains more power than treating your individuals
independently. Paired designs are very efficient since they remove
inter-pair heterogeneity. Just as a small proof: if you compute the
sample size you need to detect a standardized effect d=0.8 (Cohen's
threshold for large effects) as significant with a power of 80% in 2
independent samples design, you end up with 25 subjects per group. The
same calculation in a paired design will give you a figure close to 15
pairs. Therefore, a paired design is more efficient.
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
For a list of commands to manage subscriptions, send the command