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

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Significance tests<br />

4.1 <strong>Statistical</strong> <strong>in</strong>ference: tests and estimation 87<br />

Data are often collected to answer specified questions, such as: (i) do workers <strong>in</strong><br />

a particular <strong>in</strong>dustry have reduced lung function compared with a control group?<br />

or (ii) is a new treatment beneficial to those suffer<strong>in</strong>g from a certa<strong>in</strong> disease<br />

compared with the standard treatment? Such questions may be answered by<br />

sett<strong>in</strong>g up a hypothesis and then us<strong>in</strong>g the data to test this hypothesis. It is<br />

generally agreed that some caution should be exercised before claim<strong>in</strong>g that some<br />

effect, such as a reduced lung function or an improved cure rate, has been<br />

established. The way to proceed is to set up a null hypothesis, that there is no<br />

effect. So, <strong>in</strong> (ii) above the null hypothesis is that the new treatment and the<br />

standard treatment are equally beneficial. Then an effect is claimed only if the<br />

data are <strong>in</strong>consistent with this null hypothesis; that is, they are unlikely to have<br />

arisen if it were true.<br />

The formal way of proceed<strong>in</strong>g is one of the most important methods of<br />

statistical <strong>in</strong>ference, and is called a significance test. Suppose a series of observations<br />

is selected randomly from a population and we are <strong>in</strong>terested <strong>in</strong> a certa<strong>in</strong><br />

null hypothesis that specifies values for one or more parameters of the population.<br />

The question then arises: do the observations <strong>in</strong> the sample throw any light<br />

on the plausibility of the hypothesis? Some samples will have certa<strong>in</strong> features<br />

which would be unlikely to arise if the null hypothesis were true; if such a sample<br />

were observed, there would be reason to suspect that the null hypothesis was<br />

untrue.<br />

A very important question now is how we decide which sample values are<br />

`likely' and which are `unlikely'. In most situations, any set of sample values is<br />

peculiar <strong>in</strong> the sense that precisely the same values are unlikely ever to be chosen<br />

aga<strong>in</strong>. A random sample of 5 from a normal distribution with mean zero and<br />

unit variance might give the values (rounded to one decimal) 0 2, 1 1, 0 7, 0 8,<br />

0 6. There is noth<strong>in</strong>g very unusual about this set of values: its mean happens to<br />

be zero, and its sample variance is somewhat less than unity. Yet precisely those<br />

values are very unlikely to arise <strong>in</strong> any subsequent sample. But, if we did not<br />

know the population mean, and our null hypothesis specified that it was zero, we<br />

should have no reason at all for doubt<strong>in</strong>g its truth on the basis of this sample. On<br />

the other hand, a sample compris<strong>in</strong>g the values 2 2, 0 9, 2 7, 2 8, 1 4, the mean of<br />

which is 2 0, would give strong reason for doubt<strong>in</strong>g the null hypothesis. The<br />

reason for classify<strong>in</strong>g the first sample as `likely' and the second as `unlikely' is<br />

that the latter is proportionately very much more likely on an alternative<br />

hypothesis that the population mean is greater than zero, and we should like<br />

our test to be sensitive to possible departures from the null hypothesis of this<br />

form.<br />

The significance test is a rule for decid<strong>in</strong>g whether any particular sample is <strong>in</strong><br />

the `likely' or `unlikely' class, or, more usefully, for assess<strong>in</strong>g the strength of the

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

Saved successfully!

Ooh no, something went wrong!