Perhaps it's just my crabby nature, but the rules I've been thinking of
are all "don'ts".
(a) Don't confuse statistical significance with strength of
association. When most people talk about strength of association
between two variables, they mean that changes in one variable have an
appreciably large impact on the other, not that there's a good
statistical fit. It's the slope of the regression line that really
matters, rather than the correlation coefficient.
(b) Don't assume that apparently remote statistical associations won't
arise by chance. For example, if you correlate a variable with 42000
observations at a test of p=.001, there just might be 42 random
associations. (As if anyone would do that ... !)
(c) Don't be surprised if larger numbers lead to more observations. In
Utopia on Trial, Alice Coleman found an association between high
density housing and the presence of litter. All that means that where
there are more people, someone is more likely to drop something on the
pavement.
(d) Don't confuse indicators with measurement. An indicator is a
signpost, not the thing itself. Some figures are good indicators even
if they are bad measures: for example, mortality is a good indicator of
health, and income is a good indicator of poverty. Some figures are
good measures but bad indicators, like recorded crime or the claimant
count for JSA.
Paul Spicker
******************************************************
Please note that if you press the 'Reply' button your
message will go only to the sender of this message.
If you want to reply to the whole list, use your mailer's
'Reply-to-All' button to send your message automatically
to [log in to unmask]
*******************************************************
|