I am using a number of word variables (word frequency; number of phonemes
etc.) to predict word reading difficulty for young children (measured as the
number of children in a class who can read/spell a word; I use the zscore
transformation as the dependent variable).
The words can be classified as having one-to-one correspondence for sound
and symbol (e.g. "and"; "long") or as containing complex symbols/graphemes
("though": 2 phonemes made up of a digraph[th] and a quadgraph[ough]).
Complex words are classified as having (1) or not (0) for 3 dichotomous
categories (di-, tri-, quad- graph) The presence of a di-, tri- or quad-
graph has a dramatic effect on word difficulty, but within the 150 words
that the pupils are reading there are relatively few tri- or quad- graphs.
4 questions:
1. does the number of cases (relative to the total number of cases, 150 in
this study) in a dichotomous category matter (I can't find any reference to
this in the stats books I use): there are 3 quadgraphs, 15 trigraphs, 40
digraphs, and the rest are 1-to-1.
2. I am using zscores for the dependent variable: does it make sense to
quote the unstandardised B values rather than standardised betas?
3. Do beta values make any sense at all for a dichotomous variable?
4. Is this the kind of question I can put to this list???
Thanks to those of you who have taken the trouble to read this.
Ken S.
Dr Ken Spencer
Centre for Educational Studies
Institute for Learning
Hull University
Cottingham Road
HULL HU6 7RX
Tel: +44 (0)1482 46 5954
|