Performing stats on developmenatal data

[Nickhimali at aol.com]

Dear all,
I have collected a sample of data (using an elicitation technique) on
children's acquisition of four case markers in Sinhala. The total sample size
is 37. 19 of the children came from a rural setting and the remaining 18 from
an urban setting. I want to find any effects of gender and age group (old and
young)  on acquisition. I used two way ANOVA to check significance and
interaction. Although I got the results, I came across the problem of data
not being 'normally distributed' - owing to the type of data, i.e. being of
developmental in nature the number of case markers children had acquired
increased with age and reached a maximum around 2;6 years. Thus, I have a
high proportion of children who have acquired all the case markers, with only
a few who hadn't got any. I came across this problem when administering all
the parametric tests. I am able to use Mann-whitney instead of the T-test,
but what alternatives could I used for the ANOVA (both one and two way)? Is
it possible for me to use ANOVA despite not meeting all the assumptions? I do
not want to miss out information by performing multiple Mann-Whitney's
instead of the ANOVA. I realise there are ways to transform the
non-parametric data to parametric. Is this necessary?

Your comments would be greatly appreciated.
Himali Clarke