Not really.
Strictly, bimodality means that there are two modes. That is if you
found the frequency of each value, two of those frequencies would be
equally high. However, if you're looking at a histogram, whether it
is bimodal depends on where you (or your computer program) decides to
put the bins.
Imagine you had people's age. Whilst the modes might be 10 people
aged 18, and 10 people aged 23, with no other ages having more than
10, you could say that your data were bimodal. However, what if the
modal ages were 20 and 21? Then, strictly, the data would be bimodal,
but it wouldn't have a bimodal shape.
However, if you measured age in months, rather than years, your modes
might be completely different - the 10 18 year olds might have been
born in 10 different months. But you might have 2 people who were
aged 31 years, 2 months, and 2 people aged 26 years, 2 months. Your
data are bimodal again, but but the modes are different.
Anyway, to conclude, bimodal has a technical meaning, but it's fairly
useless. Just as you might describe a road as straight (when it's
not) or an orange as round (when it's not), a distribution being
bimodal is a description of its approximate shape. You don't care if
your data are truly bimodal, if a variable has two humps in its
distribution, it's approximately bimodal.
(On a similar vein, I get annoyed when people write that their
variable was normally distributed. It almost certainly isn't, but it
might be close.).
One more point, now I think of it. If you really, really care, you
can specify a distribution type, and do a significance test to compare
your distribution to that pre-specified distribution. Some members of
the family of beta distributions (see
http://en.wikipedia.org/wiki/Beta_distribution ) are bimodal, for e.g.
I can't imagine why you'd ever want to do that though. (People do it
with normal distributions, and that's usually nonsensical enough.)
J
On 31/03/2008, Yuanyuan Zhao <[log in to unmask]> wrote:
> Hello, everyone,
> I would be grateful if anyone could help me with this?
> i'd like to know whether my data are bimodal distributed. i looked at the graph, it looks like that but i am not sure. is there any test i could do in the spss to confirm the bimodal distribution? and how to do that?
> Thank you very much.
>
> Yuanyuan
>
--
Jeremy Miles
Learning statistics blog: www.jeremymiles.co.uk/learningstats
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|