Hello everyone,
For chi-squared (based) statistics, p values based on the chi-square
distribution can be inaccurate if we have small expected frequencies in the
contingency table. Exact p-values (generated using the true distribution of
the statistic) are used to maintain accuracy in such situations.......my
question is related to the use of exact p values for 2 other tests:
1) I want to use a Mann Whitney Test. I have 2 independent groups - one
containing 16 cases, the other containing 120 cases.
2) I want to use the Wilsoxon Matched Pairs Signed Ranks Test. I have 2
related groups, both containing 7 cases.
Will exact p values maintain accuracy when we have such imbalance (as in
case 1 above) and such small groups (as in case 2)?
On another (different) point - if I have one nominal and one ordinal
variable and I wish to establish the *degree* of association between the 2
variables, am I correct in thinking that the Cramer's V statistic could be
used? I know that this is usually recommended when both variables are
nominal. [ I've read too that two oridnal variables can be examined using
nominal measures e.g. Cramer's V but other measures (e.g. Gamma) reflect the
additional information available from ranking and are therefore
recommended.]
Many thanks for you help on these points,
Kim.
|