One-Sample Test Options
Locked
 
|
We have a survey designed to ascertain
whether respondents feel worse off, about the same or better off
following an intervention in thirty different areas of functioning
(after the intervention, do you feel much worse off, about the same, or
much better off with respect to a, b, c....). The survey utilized a nine
point anchored scale that looked something like this: Area A Much Worse Off o o o o About the Same o o o o Much Better OffArea B Much Worse Off o o o o About the Same o o o o Much Better Off Area C Much Worse Off o o o o About the Same o o o o Much Better Off We
let SPSS Non Parametric do out thinking for us and it returned most of
the items as normally distributed (one sample Kolmogorov - Smirnov). We
then ran: NPTESTS /ONESAMPLE TEST (area of functioning) WILCOXON(TESTVALUE=5) /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE /CRITERIA ALPHA=0.05 CILEVEL=95.
Thanks MUCH for any assistance. I'm in the unenviable position of knowing just enough statistics to be dangerous. Eric |
|
|
Maybe you did this and I didn’t understand your posting. Don’t you want to work with a dichotomous variable (same/worse versus better)? Null hypothesis would
be 50/50. Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Eric We have a survey designed to ascertain whether respondents feel worse off, about the same or better off following an intervention in thirty different areas of functioning (after the intervention, do you feel
much worse off, about the same, or much better off with respect to a, b, c....). The survey utilized a nine point anchored scale that looked something like this: Area A Much Worse Off o o o o About the Same o o o o Much Better Off We are interested in whether people feel better off or not which appear as a response greater than 5, which is the midpoint of our scale. N was about 40 (it depends on the item, there is some missing data). We let SPSS Non Parametric do out thinking for us and it returned most of the items as normally distributed (one sample Kolmogorov - Smirnov). We then ran: in order to test whether the sample tended to feel better than "About the Same" which, as above, is represented by a value of "5" in the data set. I was very surprised to find that even with relatively small mean differences (they ranged from 0.67 to 1.85) all were statistically significant according the the test above.
Some Questions:
Thanks MUCH for any assistance. I'm in the unenviable position of knowing just enough statistics to be dangerous. Eric Eric Graig ===================== To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
|
|
|
In reply to this post by Eric Graig-2
Hi,
Not really my area but your results actually
sound consistent. That
being said, you may want to do the
following:
(1) Check out the following webpage on the IBM
website:
It explains how to do the one-sample Wilcoxon
test (but you already
know that) and that it is a special case of
the Wilcoxon matched-pairs
test where the second value is the null
hypothesized value (i.e., "5")
which you don't seem to
appreciate.
(2) The IBM website above cites Section 5.1 in
the following:
W. J. Conover's (1971) Practical Nonparametric Statistics (Wiley).
Conover's book is up to a 3rd edition, more info about that here:
You might want to check out either the earlier
or the later edition (I'd
go for the later edition because new
developments may have altered
some of the interpretation of the one-sample
Wilcoxon text). So,
I suggest that Ye Get Thee To A Library or a
bookstore where you
read the book for free (I couldn't find an
electronic copy online; you
might have better luck).
(3) A general Google search turns up about
3,800 hits for "one-sample
Wilcoxon test" and a search of
scholar.google.com turns up about 483
hits; for the latter see:
So, I don't understand how/why you can't find
references for the test.
If you have access to www.jstor.org, which has the Journal of
the
American Statistical Association (JASA), the
American Statistician,
and a shipload of other statistical-ish
journal, you'll find article there
as well. But you might want to start
just by looking at the above
mentioned Conover book first.
(4) One point that you seem to miss is that
you are focusing on the
difference between means (or, technically in
the one sample Wilcoxon
test, medians) but forget that this difference
has to be interpreted
relative to the amount of sampling error in
your sample statistics.
Cohen's d (if you use sample means and
standard deviations) is
the most representative difference to use
because it reflects the
difference between means relative to a measure
of their sampling
error. So, don't just look at the simple
difference between means.
I'm sure if I said anything crazy/absurd/wrong
above, someone will
correct me.
-Mike Palij
New York University
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
| Free forum by Nabble | Edit this page |