01.06.2013 Views

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

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.

example discussed above we take a sample of 20 factories and second-stage<br />

samples of 50 men <strong>in</strong> each of the 20 factories. If there is systematic variation<br />

between the factories, due perhaps to variation <strong>in</strong> health conditions <strong>in</strong> different<br />

parts of the country or to differ<strong>in</strong>g occupational hazards, this variation will be<br />

represented by a sample of only 20 first-stage units. A random sample of 1000<br />

men, on the other hand, would represent 1000 random choices of first-stage units<br />

(some of which may, of course, be chosen more than once) and would consequently<br />

provide a better estimate of the national mean.<br />

A useful device <strong>in</strong> two-stage sampl<strong>in</strong>g is called self-weight<strong>in</strong>g. Each first-stage<br />

unit is given a probability of selection which is proportional to the number of<br />

second-stage units it conta<strong>in</strong>s. Second-stage samples are then chosen to have<br />

equal size. It follows that each second-stage unit <strong>in</strong> the whole population has an<br />

equal chance of be<strong>in</strong>g selected and the formulae needed for estimation are<br />

somewhat simplified.<br />

Cluster sampl<strong>in</strong>g<br />

Sometimes, <strong>in</strong> the f<strong>in</strong>al stage of multistage sampl<strong>in</strong>g, complete enumeration of<br />

the available units is undertaken. In the <strong>in</strong>dustrial example, once a survey team<br />

has <strong>in</strong>stalled itself <strong>in</strong> a factory, it may cost little extra to exam<strong>in</strong>e all the men <strong>in</strong><br />

the factory; it may <strong>in</strong>deed be useful to avoid the embarrassment that might be<br />

caused by <strong>in</strong>vit<strong>in</strong>g some men but not others to participate. When there is no<br />

sampl<strong>in</strong>g at the f<strong>in</strong>al stage, the method is referred to as cluster sampl<strong>in</strong>g. The<br />

<strong>in</strong>vestigator has no control over the number of sampl<strong>in</strong>g units <strong>in</strong> the clusters and<br />

this means that the loss of precision, compared with simple random sampl<strong>in</strong>g, is<br />

even greater than that <strong>in</strong> multistage sampl<strong>in</strong>g.<br />

Design effect<br />

19.2 The plann<strong>in</strong>g of surveys 655<br />

The ratio of the variance of an estimator from a sampl<strong>in</strong>g scheme to the variance<br />

of the estimator from simple random sampl<strong>in</strong>g with the same total number of<br />

sampl<strong>in</strong>g units is known as the design effect, often abbreviated to Deff. I n<br />

Example 19.1 the Deff for allocation C is 0 000739=0 000803 ˆ 0 92. Another<br />

way of look<strong>in</strong>g at the Deff is <strong>in</strong> terms of sample size. The same precision could<br />

have been achieved with a stratified sample of 92 people with allocation C as for<br />

a simple random sample of 100 people. The stratified sample is more efficient<br />

(Deff < 1) than simple random sampl<strong>in</strong>g and this will occur generally provided<br />

that there is a component of variation between strata. The efficiency of stratified<br />

sampl<strong>in</strong>g <strong>in</strong>creases with the <strong>in</strong>creas<strong>in</strong>g heterogeneity between strata and the<br />

consequent greater homogeneity with<strong>in</strong> strata.<br />

In contrast, multistage and cluster sampl<strong>in</strong>g will usually have a Deff > 1.<br />

That is, a larger sample size will be required than with simple random sampl<strong>in</strong>g.

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

Saved successfully!

Ooh no, something went wrong!