ANOVA vs. T-Test - which one to use?
Susanne
Im trying to find about if there are any stat. sign. differences betw.
my two groups.
Categories I'm supposed to analyze are e.g. marrital status, religion
etc.. so what I did up to now is
take one column named "groups" and subdivide the values (naming the
whole population group '1' and the withdrawal group '2').
The 2nd column I named "status" and subdivided its values into '1' >
married, '2'> single, '3' divorced etc...
Then I ran a independent samples t-test, as well as an one-way ANOVA.
Both of them look pretty "ok", the Sig.value is higher than 0.5, so
there seems to be no stat.significant difference, which is what I
expected anyway.
Im just absolutely not sure which one of these tests is the right one
now?
Which one am I supposed to use if I have two groups with one variable
and diff. answer options to it? Thanks for your help already! :)
Art Kendall
Usually one want to compare 2 parts of the whole group, e.g., stayers
and withdrawals.
Also at least for marital status and religion, the independent
variables are strictly nominal level.
A 2 group t test and a 2 group one way ANOVA are the same thing. t**2 = F.
It would never be appropriate to use religion, marital or other nominal
variables as a DV.
Treating withdrawal as a DV would be plausible if you recode it to
zero/one so that what you are testing are proportions (percents).
For 2 variable analyses I suggest you try CROSSTABS. If you can deal
with it CATREG (categorical regression) would be more informative. With
CATREG you could include nominal, ordinal, and continuous variables as
predictors. See the archives for this list. IIRC there is a PYTHON
approach to differences of proportions especially if the number
withdrawing is small in cells.
Art Kendall
Social Research Consultants
Susanne
> It would be extremely unusual to compare the whole group to a part.
> Usually one want to compare 2 parts of the whole group, e.g., stayers
> and withdrawals.
>
> Also at least for marital status and religion, the independent
> variables are strictly nominal level.
>
Thanks for your answer.
"A 2 groupt testand a 2 group one wayANOVAare the same thing." > By 2
groups you mean two variables (in my case the variable "group" and say
"marital status" ? So it would be no difference to use a one way anova
or a t-test, if I do have only these two variables right?
If I divide my variable "marital status" into values such as 1
married, 2 single, 3 divorced etc. , would that be still nominal then?
"Treating withdrawal as a DV would be plausible if you recode it to
zero/one so that what you are testing are proportions (percents). "
> To see if there's any stat.sign.difference betw. these two groups would mean to take the 'group' variable as the DV I assume then? What do you mean by recode it to zero/one?
Thanks, Susanne
Rich Ulrich
wrote:
>On Aug 30, 7:55 pm, Art Kendall <A...@DrKendall.org> wrote:
[snip, previous]
>Thanks for your answer.
>
>"A 2 groupt testand a 2 group one wayANOVAare the same thing."
> By 2
>groups you mean two variables (in my case the variable "group" and say
>"marital status" ? So it would be no difference to use a one way anova
>or a t-test, if I do have only these two variables right?
>
>If I divide my variable "marital status" into values such as 1
>married, 2 single, 3 divorced etc. , would that be still nominal then?
>
"nominal" denotes unordered categories. Because they
are unordered, they could as easily be labeled in any other
other, assuming that the labels are numbers. Taking an
average of these arbitrarily assigned labels gives a result that
is generally useless -- and ANOVA compares averages on
meaningful *scores* (descriptors like age; or outcomes),
between multiple groups.
The t-test is nothing more or less than a simplified instance
of ANOVA -- It usually compares averages of two groups,
or occasionally it compares one average to a fixed value
(most often, zero).
- Read, or at least browse, a few textbooks on statistics.
Learning statistics from using a stat-pack is not something
that statisticians recommend -- You miss a lot that you should
know about "assumptions" and background of even minimal
theory. In your questions, you are demonstrating that you
also, so far, do not have the basic vocabulary for describing
possible analyses.
>"Treating withdrawal as a DV would be plausible if you recode it to
>zero/one so that what you are testing are proportions (percents). "
>> To see if there's any stat.sign.difference betw. these two groups would mean to take the 'group' variable as the DV I assume then? What do you mean by recode it to zero/one?
>
>Thanks, Susanne
--
Rich Ulrich