12.01.2015 Views

RESEARCH METHOD COHEN ok

RESEARCH METHOD COHEN ok

RESEARCH METHOD COHEN ok

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.

STATISTICAL SIGNIFICANCE 515<br />

alongtailatthepositiveendandthemajorityof<br />

the data at the negative end, and a negative skew<br />

has a long tail at the negative end and the majority<br />

of the data at the positive end.<br />

Statistical significance<br />

Much statistical analysis hinges on the notion of<br />

statistical significance. Kirk (1999: 337) indicates<br />

that ‘a statistically significant result is one<br />

for which chance is an unlikely explanation’.<br />

Research in a hypothetico-deductive mode often<br />

commences with one or more hypotheses. This is<br />

the essence of hypothesis testing in quantitative<br />

research. Typically hypotheses fall into two types.<br />

The null hypothesis, amajortypeofhypothesis<br />

states that, for example, there is no relationship<br />

between two variables, or that there has been<br />

no change in participants between a pretest<br />

and a post-test, or that there is no difference<br />

between three school districts in respect of<br />

their examination results, or that there is no<br />

difference between the voting of males and females<br />

on such-and-such a factor. The null hypothesis<br />

sits comfortably with the Popperian notion of<br />

the hallmark of a science being its essential<br />

falsifiability.<br />

The point here is that by casting the hypothesis<br />

in a null form the burden of proof is placed<br />

on the researcher not to confirm that null hypothesis.<br />

The task is akin to a jury starting with a<br />

presumption of innocence and having to prove<br />

guilt beyond reasonable doubt. Not only is it often<br />

easier simply to support a straightforward positive<br />

hypothesis, but also, more seriously, even if that<br />

positive hypothesis is supported, there may be insufficient<br />

grounds for accepting that hypothesis,<br />

as the finding may be consistent with other hypotheses.<br />

For example, let us imagine that our<br />

hypothesis is that a coin is weighted and, therefore,<br />

unfair. We flip the coin 100 times, and find<br />

that 60 times out of 100 it comes out as heads.<br />

It would be easy to jump to the conclusion that<br />

the coin is weighted, but, equally easily, other reasons<br />

may account for the result. Of course, if the<br />

coin were to come out as heads 99 times out of<br />

100 then perhaps there would be greater truth in<br />

the hypothesis. The null hypothesis is a stronger<br />

version of evidence, not only requiring that the<br />

negative hypothesis be ‘not supported’, but also indicating<br />

a cut-off point only above which the null<br />

hypothesis is ‘not supported’, and below which the<br />

null hypothesis is supported. In our coin example<br />

it may be required to find that heads comes up<br />

95 times out of 100 or 99 times out of 100, or<br />

even 999 times out of 1,000, to say, with increasing<br />

confidence in respect of these three sets<br />

of figures, that the null hypothesis is not supported<br />

(see http://www.routledge.com/textbo<strong>ok</strong>s/<br />

9780415368780 – Chapter 24, file 24.8.ppt).<br />

We use terminology carefully here. Some<br />

researchers state that the null hypothesis is<br />

‘rejected’; others say that it is ‘confirmed’ or ‘not<br />

confirmed’; other say that it is ‘accepted’ or ‘not<br />

accepted’. We prefer the terminology of ‘supported’<br />

or ‘not supported’. This is not mere semantics or<br />

pedantry; rather it signals caution. Rejecting a<br />

null hypothesis is not the same as ‘not confirming’<br />

or ‘not supporting’ that null hypothesis, rejection<br />

implying an absolute and universal state which the<br />

research will probably not be able to demonstrate,<br />

being bounded within strict parameters and not<br />

being applicable to all cases. Further, ‘confirming’<br />

and ‘not confirming’, like ‘rejecting’, is too strong,<br />

absolute and universal a set of terms for what is,<br />

after all, research that is bounded and within<br />

delineated boundaries. Similarly, one cannot<br />

‘accept’ a null hypothesis as a null hypothesis<br />

can never be proved unequivocally.<br />

A second type of hypothesis is termed the<br />

alternative hypothesis. Whereasthenullhypothesis<br />

states that there is no such-and-such (e.g.<br />

change, relationship, difference), the alternative<br />

hypothesis state that there is such-and-such, for<br />

example: there is achangeinbehaviourofthe<br />

school students; there is a difference between<br />

students’ scores on mathematics and science;<br />

there is a difference between the examination<br />

results of five school districts; there is adifference<br />

between the pretest and post-test results of suchand-such<br />

a class. This weaker form of hypothesis<br />

is often supported when the null hypothesis is<br />

‘not supported’, i.e. if the null hypothesis is not<br />

supported then the alternative hypothesis is.<br />

Chapter 24

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

Saved successfully!

Ooh no, something went wrong!