Cell sizes Orthoplan

[student09]

Dear All,

I would be very glad if somebody might help me with the following question -

I need to conduct a factorial survey (a.k.a vignette study).

- There are 2 factors A and B.

- Factor A has 4 conditions, B has 5 conditions.

- It is a between-subjects ANOVA design.

I now want to gauge the minimum number of person needed for this study.
This is of course a question of test power, but my problem relates to a
rather specific aspect, namely:

The total number of cells would be 4*5=20.

However, should I consider the number of people per condition within each
factor (that would be 4 respectively 5)OR the number of people across all
possible combinations (the 4*5 = 20 cells) to gauge the sample size needed?

Obviously, even if one follows the rule of thumb of "use 25 obersvations
per cell", the conclusions would be quite different, because then I would
either had to use 4*25 = 100 respectivley 5*25 =125 people, or 20*25= 500
people

What would be the appropriate view here?

Many thanks for your ideas!!!!
Lea

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Mike]

If you're going to treat this as a 2-way independent groups ANOVA
then you should be asking what is the minimum level of power do you
need to detect a significant 2-way interaction.  The main effects will
have higher levels of power relative to the interaction.  So, one focus
to have is what are the effect sizes you're trying to detect, especially
in the 2-way interaction?  If the effect sizes are large, then a small number
of subjects are needed.  If the effect sizes are small, the a large number
of subjects will be needed.  It is best if you use one of the statistical
power packages to examine what happens in your situation when you
vary (a) sample size, (b) effect size, and (c) level of statistical power
(assuming you're using an alpha = 0.05).

-Mike Palij
New York University
[hidden email]

----- Original Message -----
From: "Lea" <[hidden email]>
To: <[hidden email]>
Sent: Friday, March 18, 2011 5:32 AM
Subject: Cell sizes Orthoplan

> Dear All,
>
> I would be very glad if somebody might help me with the following question -
>
> I need to conduct a factorial survey (a.k.a vignette study).
>
> - There are 2 factors A and B.
>
> - Factor A has 4 conditions, B has 5 conditions.
>
> - It is a between-subjects ANOVA design.
>
> I now want to gauge the minimum number of person needed for this study.
> This is of course a question of test power, but my problem relates to a
> rather specific aspect, namely:
>
> The total number of cells would be 4*5=20.
>
> However, should I consider the number of people per condition within each
> factor (that would be 4 respectively 5)OR the number of people across all
> possible combinations (the 4*5 = 20 cells) to gauge the sample size needed?
>
> Obviously, even if one follows the rule of thumb of "use 25 obersvations
> per cell", the conclusions would be quite different, because then I would
> either had to use 4*25 = 100 respectivley 5*25 =125 people, or 20*25= 500
> people
>
> What would be the appropriate view here?
>
> Many thanks for your ideas!!!!
> Lea
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Rich Ulrich]

If you are willing to focus on the main effects, to the
exclusion of worrying about the two-way interactions, then
your power question reduces to the two separate questions,
for 4 groups and for 5 groups.  You require an N that is
the larger of what is needed to detect the two effects that
are of interest.  (And, "power" is a good reason that the vast
majority of experimental designs contrast two groups, not
more.)

To examine power for more than two groups, you should also
be aware of the sort of *pattern*  of difference that you
expect, or want to be able to detect.  It can be easier to
detect one *big* divergent mean than to detect the difference
among a spread of results. And, certainly, easier to interpret.

I still recommend J. Cohen's book, since it discusses some
of the issues (instead of merely providing formulas).

--
Rich Ulrich



=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[MacGillivary Heather L]



Apologies, I am out of the office on vacation and will not return until  March 23.  I will be checking email periodically.

Thanks,

Heather