"A-Z" INDEX & SEARCH

The Statistical Analysis of Morphometric Data From Honey Bee Colonies

Morphometric data is easy to collect, but requires sensible methods of analysing the data, this page shows some methods that may be used to refine the data and possibly eliminate data elements that have come from a different strain of bee.

The Raw Data from morphometric measurements is difficult to visualise if presented in simple tabular form.

In October 2000 I posed the following question to Jacob Kahn:-

I would like some guidance on the treatment of 'odd' entries on scattergrams.
Assuming that some drifting is inevitable we may inadvertently sample a 'wrong un' when doing wing morphometry. We may have a basically tight group with an odd dot that is some distance from the main group.
Do we ignore this 'off target' dot or is that cheating?

Jacob's Reply was as follows:-

Unfortunately the scattergram is not really very informative unless you can demonstrate a correlation between discoidal-shift and cubital-index; in most cases there is no correlation. Assuming that in the case you mentioned there was no correlation you can proceed as follows:

determine whether the extreme position of the outlying point is due to DS or CI or both;

calculate either for DS CI or both (depending on your answer in (a)) in a one tail test either the 5% or 95% quantile (depending at which end of the distribution your outlying point is positioned), if your point falls outside one of these values you are allowed to reject it as not belonging to the population you have measured at a 5% significance level.

Even if you exclude your point from your scattergram keep a record of it. Again I am assuming:-

Suppose you measured 30 wings from a population of say 30,000 bees then you have measured only 0.1% of that population and a second sample of 30 wings may give a different profile of measurements.

A mixture of different groups in a single population will be due to genetical as well as environmental factors. We need to know a lot more before we can embark on a partitioning exercise of this nature, and of course we would need the appropriate software, this kind of calculation cannot be done by hand.
A bivariate analysis of variance could be a good start. JK

Written... 24 May 2002, Upgraded... 13 June 2006,