Thanks. I got it. It's a regular parallel analysis, bootstrapped.
(It's paralleled, the bootstrapped, rather than bootstrapped then
paralleled*.) [*Made up word]
Parallel analysis uses variables with the same skew, kurtosis, etc,
but with a zero population correlation matrix. If you find factors
with that data, it's because of the distributions, not because there
are really factors there. So you do that, and you see what the
biggest eigenvalues are - you know by definition that these factors
are not real, so if you find factors that big in your data, they're
not real.
The marginal bootstrap goes one step further to get 95% confidence
intervals of those pretend eigenvalues.
Jeremy
2009/10/10 Ioanna Vrouva <[log in to unmask]>:
> Hi Jeremy,
> Thanks for your response. This is the link.
> http://psico.fcep.urv.es/utilitats/factor/
> If you come to a conclusion please let me know.
> best wishes
> Ioanna
>
>
>
>
> Sunday, October 11, 2009 12:33 AM
> From:
Add
> sender to Contacts
> To:
> "Ioanna Vrouva" <[log in to unmask]>
> Cc:
> [log in to unmask]
> Hi Ionna,
>
> I don't know that software. Can you provide a link.
>
> Did it do a regular parallel analysis (which is sometimes called a
> univariate bootstrap) or did it do something more sophisticated, using
> a Markov Chain Monte Carlo (MCMC) approach. Sometimes marginal
> bootstrap refers to the MCMC.
>
> Jeremy
>
>
>
> 2009/10/10 Ioanna Vrouva
> <<[log in to unmask]" target="_blank">http:[log in to unmask]>[log in to unmask]>:
>> Dear all,
>> I am writing to request help with the following.
>> I have used the "Factor" software (By Lorenzo and Ferrando) to determine
>> the
>> number of factors to be retained from a questionnaire I have developed.
>> I described the procedure as follows:
>> " In order to decide the right number of factors to retain, we performed a
>> parallel analysis using marginally bootstrapped samples (PA-MBS; Lattin,
>> Carroll, & Green, 2003),
>> one of the most accurate factor retention techniques that involves the
>> generation of many correlation matrices of random variables based on the
>> same sample size and number
>> of variables in the actual data set using marginally bootstrapped samples.
>> In such samples, the number of variables, the sample size and the marginal
>> kurtosis are kept
>> at their original level, which is important because these factors
>> determine
>> the distribution of eigenvalues".
>> One of the reviewers asked me to " explain the concept of marginally
>> bootstrapped samples a bit more clearly".
>> I would be very grateful if you had any ideas that might help me do this,
>> or
>> perhaps if you might be able to suggest any relevant literature?
>> with many thanks for your time,
>> Ioanna
>>
>
>
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|