[EMAIL PROTECTED] (Chen G) wrote in
news:[EMAIL PROTECTED]:
> Hello,
> My experience with statistics is very limited, and so I apologize in
> advance.
> I have 3 groups of usertypes marked "a", "b", and "c" that represent
> different psychological personalities that learn in the same class.
> In each group I have a different number of students ("a" - 13, "b" -
> 7, "c" - 7) and each have a grade.
> I would like to be able to tell whether there is a correlation (in the
> literal sense) between the different usertypes, and their grades. How
> do I do that statistically?
By grade, do you mean an ordinary course grade of the A-F variety? If
that's the case, or if it's some other numerical measure with approximately
equal spacing between units, then your problem becomes that of testing and
measuring the association between a nominal predictor variable (group type)
and a continuous outcome variable (grade). That's usually done by an
analysis of variance (which is actually a special type of linear regression
model). It looks like you want a one-way ANOVA with group as "treatment."
You'd test whether there is an association by looking at the F-ratio from
the test, and measure its strength by looking at the eta-squared value
(SSB/SST); the square root of that could be interpreted as a "correlation"
though almost nobody does so (eta-squared can be interpreted similarly to
R-squared in regression, as the proportional reduction in variance
resulting from taking the grouping into account).
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
