I'm still looking at these child welfare surveys based on questionnaires
distributed to child protection authorities. I'm wondering what sort of
conventional wisdom might exist as to setting alpha levels for a
contingency table relating each parameter to three measures of a
dichotomous outcome - whether the kid is the problem or abuse and neglect
are the problem, to put it simplistically. I'm using chi-square tests and
Cramer's V and contingency coefficient in SPSS to go about sifting these
variables. I would suppose each question is considered to be the unit of
analysis, but I'm not experienced in this area. My own inclination is to
use good ol' .05 as a bottom-line cut-off for the chi-squares and then to
use a measure of association to decide empirically where it would be useful
to cut off for further analysis. But is there any justification for setting
the alpha for the chi-squares to a stricter level?
David Klein
|