For which analysis? If you are using two measures at each time
instead of one, then you reduce measurement error, that reduces
variance which increases effect size. But not by a great deal
(probably - you can work it out but you need to make an assumption
about the reliability of the test). So I would leave the power
analysis the same. For the slope with a break, it's fairly complex,
so I'd do the power for the simple one, and say it's going to be more
than that.
That's a standard approach for a power analysis for a complex analysis
- say "I'm going to base the power analysis on a simple analysis, the
complex analysis is probably better, and therefore has more power.
That means, if anything, I'm underestimating power, which I would
rather do than overestimate power".
Also, keep in mind that the Devil's Dictionary of Statistics by
Stephen Senn defines power analysis as "Guessing, disguised as
mathematics".
Jeremy
On 27 October 2010 10:23, Davies, Nicola <[log in to unmask]> wrote:
> Thanks Jeremy.
>
> In regard to power analysis when choosing how many participants to recruit, how is this best achieved? I have used GPower to estimate number of participants needed based on the research design. Should I be basing power some how on the questionnaires I will be using? If so, how do I do this?
>
> Thanks for your help,
>
> Nicola
>
>
>
> ________________________________________
> From: Jeremy Miles [[log in to unmask]]
> Sent: 26 October 2010 23:24
> To: Davies, Nicola
> Cc: [log in to unmask]
> Subject: Re: Power Calculations and Statistical Tests
>
> if you don't expect a trend, I would average the four (or two)
> pre-test scores, and the four (two) post-test scores, and then do a
> repeated measures t-test.
>
> If you expect some sort of consistent trend that you need to remove
> then you need to do something like a multilevel model.
>
> Jeremy
>
>
> On 26 October 2010 14:44, Davies, Nicola <[log in to unmask]> wrote:
>> Hi,
>>
>> Does anyone know which statistical test I would need for the following:
>>
>> I have two independent groups: One group is before the service redesign; The second group is after the service redesign. Both groups will provide data at 4 points in time over a 12-month period.
>>
>> I need to know the statistical test I will use in order to calculate how many participants I need per group (I am using Gpower to do this).
>>
>> I will be trying to determine differences between the groups.
>>
>> Thank you for your help,
>>
>> Nicola
>>
>
>
>
> --
> Jeremy Miles
> Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|