04.02.2015 Views

Dictionary of Evidence-based Medicine.pdf

Dictionary of Evidence-based Medicine.pdf

Dictionary of Evidence-based Medicine.pdf

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.

V<br />

Variance<br />

Variance is a measure <strong>of</strong> the spread <strong>of</strong> a distribution. If in a population <strong>of</strong><br />

N individuals the measurements are x { with i = 1 to N and the mean is ^i<br />

then the variance o 2 is given by:<br />

In practice, population variances are estimated using samples <strong>of</strong>, say, size<br />

. To reflect the reduced precision, the variance is then estimated by s 2 :<br />

Variance inflation factor (VIF)<br />

A measure <strong>of</strong> how much the variance associated with a coefficient <strong>of</strong> a statistical<br />

model is inflated by the lack <strong>of</strong> orthogonality (predictor variables<br />

correlated) in the design. The standard error <strong>of</strong> a model coefficient is<br />

increased by a factor equal to the square root <strong>of</strong> the VIF, when compared<br />

to the corresponding standard error in an orthogonal model. VIF values<br />

greater than 10 indicate that the coefficients are poorly estimated due to<br />

multicolinearity (presence <strong>of</strong> correlated predictor variables in the model).<br />

A coefficient with a VIF <strong>of</strong> 25 has a standard error which is five times<br />

higher than it would be if the design were orthogonal.<br />

VIF (see Variance inflation factor)

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

Saved successfully!

Ooh no, something went wrong!