You might find it enlightening to perform Tukey's "median polish" to
this two-way table, and display the residuals in the quasi-3-dimensional
plot there demonstrated.  (See Tukey, "Exploratory Data Analysis",
Addison-Wesley, 1977;  or it may be in Mosteller & Tukey, "Regression
and Data Analysis";  neither of which is conveniently at hand for me to
check.)

 For K-W to find a significant effect, as Thom and Rich have pointed
out, the median educational level needs to differ across marital status.

 For the chi-square analysis to be significant, all you need is that the
proportions of cases falling into the several educational levels differ
across marital status;  and as Thom's "simple example" showed, this can
happen, and even happen quite strongly, without inducing any differences
at all in the _median_ educational levels for each marital status.  I'd
guess that something like that example, if perhaps a little more
complex, is going on in your data.

 In fact, it sounds as though the data might be quite interesting in
these regards.  I don't have access to the book you mention;  would you
be kind enough to post, or send to me privately, the two-way table of
frequencies?

On Thu, 20 May 2004, G.S.Clarke wrote:

> I have asked some students to use SPSS to analyse some data from
> Pallant�s book �the SPSS survival manual� and I am working through the
> questions myself.  I�ve asked the students to examine whether there is
> any relationship between educational status (ranging from not much
> education � running through to higher degree level) which can be
> considered an ordinal variable, and marital status (married, single,
> divorced, widowed etc.) � which really seems to be a categorical
> variable.
>
> In doing the examples though, some thoughts/problems occur to me, which
> probably have a simple answer, though right now � I can�t see it!
>
> I initially decided to group educational status, using marital status as
> the factor and apply a Kruskal-Wallis test.  This test gives a �non
> significant� result.  Next I decided to apply a Chi-Square analysis
> treating both variables as categorical and this time both the pearson�s
> and likelihood ratio variants of the test say that there is a highly
> significant result.
>
> OK � two questions.
>
> First � is there any good reason why kruskal-wallis (which I would
> have taken as being the more powerful technique) doesn�t show
> significance when the chi-squared tests do (in other words, am I
> correct in my assumptions as to which is the more powerful technique).
>
> Secondly chi-squared provides a �symmetrical� analysis � it says (at
> least I think that it does) that having information about one of the
> variables provides information about the other variable � and vice
> versa.  However is Kruskal-Wallis �symmetrical� in this sense.  Does
> not KW say (had it been significant) that the marital status of the
> subjects (the factor) informs us somewhat on their educational status
> (the dependent variable). Now logically if this is true then the
> educational status should inform us somewhat on the marital status,
> however is this also the case �statistically� i.e. will any
> significance level generated by the KW using education (dependent) v
> marital status (independent)  also apply to phrasing the question the
> �other way around�?
>
> I hope that what I�ve written makes some kind of sense.  I appreciate
> that this might be a �simple� or even �stupid� question (I have these
> 'blips' every now and again)  but I�d like to get it clear in my own
> head before approaching the students with what I think is the �right
> answer�.
>
> Best Regards and Many Thanks
>
> Graham

 ------------------------------------------------------------
 Donald F. Burrill                              [EMAIL PROTECTED]
 56 Sebbins Pond Drive, Bedford, NH 03110      (603) 626-0816
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to