don't necessarily like, called principle components analysis. Without
going into detail it makes some assumptions about how the individual
items are related to each other, frequently overestimating this
communality. Its good for a first estimation of the underlying
factors, and for the initial development of a questionnaire, but for
determining the actual structure of the factors involved, I think its
inadequate.
I'll be downloading the data today and using some other analysis on
it later today and see what I come up with.
larry
>I think that this survey is much more valid than the other ones we've tried.
>
><http://politics.beasts.org/>http://politics.beasts.org/
>
>FWIW, here are my scores:
><http://politics.beasts.org/scripts/results?surveyid=838428544>http://politics.beasts.org/scripts/results?surveyid=838428544
>
>Axis Position
>1 left/right -7.2199 (-0.4346)
>2 pragmatism +2.2182 (+0.1335)
>
>Anyhow the following is part of the rationale of the survey given by
>its authors.
>--
>politicalcompass.org is a web site which asks a number of opinion
>questions of its visitors, and then places them in a two-dimensional
>space which is supposed to characterize their political views.
>Unfortunately, politicalcompass.org has a poor reputation; in
>particular, there is a suspicion that its questions are designed to
>make respondents lean towards an economically right-wing, socially
>liberal ("right libertarian") position, and the two axes of variation
>on which results are plotted are opaque in their derivation and may
>not be tremendously relevant.
>
>These suspicions are compounded by the problem that
>politicalcompass.org's methods are not open and, therefore, it is not
>possible to determine whether their selection of questions carries a
>bias which its operators are using to further their own ends.
>
>The purpose of this site is to do a survey of this type properly and
>openly, so that the methods and data in use are open to inspection.
>More detail
>
>The proper way to do this is to collect a bunch of questions and a
>bunch of answers to them, then take the space defined by all the
>answers to the questions, and construct a spanning basis for it. The
>natural way to do this is with principal components analysis, though
>as a non-statistician I can't comment on whether this is actually the
>best approach. We should then be able to discover -- in terms defined
>by the answers to the questions set -- the significant axes of
>variation in the data.
>
>This means that all the results we get are defined by the data: we do
>not measure anyone's views according to criteria we set out, but
>according to endogenous criteria. The only points at which our
>judgment enters the method are
>
> * when choosing questions (or, rather propositions); and
> * when we give context to the results.
>
>The first of those shouldn't matter, if the questions are reasonably
>unbiased and cover a wide enough range of subject materials. The
>second doesn't matter, since it's just a presentational issue.
>--
>
>So far I'm going over their analysis, and looking at how they did the
>factor analysis, it looks pretty good so far. I'm going download
>their data over the weekend and run it through a few of my stats
>programs (SPSS for the factor analysis and AMOS for the causal
>modelling/path analysis) and see if it holds. but my first impression
>by looking at their published eigenvectors, is that it looks legit.
>
>larry
>
>[
[Todays Threads] [This Message] [Subscription] [Fast Unsubscribe] [User Settings]
