02.04.2013 Views

The psychopathology of everyday art: a quantitative Study - World ...

The psychopathology of everyday art: a quantitative Study - World ...

The psychopathology of everyday art: a quantitative Study - World ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>The</strong>re are few answers to this criticism which do not lead to possible inferential errors,<br />

and it is a question <strong>of</strong> judgement in using a procedure which is appropriate for the sample<br />

and reflects the classification and variety <strong>of</strong> variables encompassed.<br />

Wolf 222 describes a method using only the most significant results from each<br />

study, but this technique lends itself to type I error. Rosenthal's 223 adaption <strong>of</strong> the<br />

Stouffer method, i.e. averaging between significance levels transformed to z-scores to<br />

produce one variable per study, are clearly not appropriate here because the standard<br />

deviations and direction <strong>of</strong> significance <strong>of</strong> studies must be similar, and the variables non-<br />

correlated, so that they can be weighted equally 224 . Furthermore, although 4 from 11<br />

studies show a very high effect size aggregated in this way and 3 more show a small to<br />

medium effect, 3 studies did not actually report the large numbers <strong>of</strong> non-significant<br />

statistics, <strong>of</strong> which their study mostly consisted, and therefore estimates <strong>of</strong> p=1 would<br />

probably contribute to <strong>art</strong>ificially low averages resulting in a much too conservative<br />

estimate.<br />

Kulik, Rosenthal and others recommend using separate analyses for each variable,<br />

when each study uses several <strong>of</strong> the same type <strong>of</strong> variables. <strong>The</strong>re were two main<br />

problems which made this type <strong>of</strong> analysis difficult in this study: (1) <strong>The</strong>re were tests<br />

between patients and normal controls and between patients and patient controls; and (2)<br />

there was no way <strong>of</strong> knowing how similar variables with similar terms, which were<br />

tabulated together, actually were. Few studies contributed variables to most <strong>of</strong> the<br />

222 Wolf (1986) op.cit. p.46.<br />

223 Rosenthal (1984) op.cit. p.72.<br />

172

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!