QUB |
Archaeology and Palaeoecology |
The 14Chrono Centre
Manual for psimpoll and pscomb
Menu
Md.3 produces the submenu for data transformations:
this is identical to that described for zonation
(menu Mc.5).
Square-root transformation is routinely used with covariance matrices for
the standard pollen data PCA. For proportional data,
centred log ratio) is probably the
most desirable statistically, using a method recommended by
Aitchison (1986) specifically for
compostional data such as pollen percentages. It works by transforming the
data values to log(value / geometric mean of all values in the sample),
and then calculating covariances in the usual way to derive the basic matrix. I
know of no use of this in the palynological literature (yet!). Because this
method
takes log values, zero data entries cannot be used. For proportional data, I
treat zeroes as
`trace zeroes': values that are not absolute zero, but only seen as zero at
the detection level of the count. I use a value of 0.001 as the rounding
error (0.5 × 1 / 500),
assuming that 500 is a typical pollen count.
For other data types, the centred log ratio method calculates
log(x + 1).
I have
not yet tried PCAs with the other two transformation methods (taken from
CONISS [Grimm 1987]: see under zonation).
Return to menu Md.
Back to contents page
Copyright © 1995-2007 K.D. Bennett
Archaeology and Palaeoecology | 42 Fitzwilliam St | Belfast BT9 6AX | Northern Ireland | tel +44 28 90 97 5136
Archaeology and Palaeoecology | The 14Chrono Centre | URL http://www.qub.ac.uk/arcpal/ |
WebMaster