Statistical testing question from a novice

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Statistical testing question from a novice

Robert Morgan-3
I'm a stats & SPSS novice, and I'm trying to think through a statistical
testing question I have.

My company fielded a survey composed of both a local Web panel and
respondents gathered via pop-up solicitations on our Web site. I'm only
interested in analyzing from the Web panel universe.

For a large number of questions, I want to compare panel respondents that
are Women 25-54 with all panel respondents to see if there are significant
differences between Women 25-54 on the panel and all panel respondents. For
the discussion I'll assume I'm dealing with ordinal-level data
(i.e. "daily", "several times a week", "several times a month", "once a
month or less", "never").

My thinking is that this type of comparison would constitute a
nonparametric test of paired samples, so the best test to use would be a
Wilcoxon matched-pairs signed-ranks test.

In order to create my variable of Women 25-54 from the Web panel in SPSS, I
would need to use a Boolean operation to create the variable - would I
simply use "Transform," then "Recode into Different Variables"? Is that the
best way to create new groupings of the sample? I've always used "Select
Cases," but that doesn't permanently change the selection. I'm not sure how
I should create that new grouping to submit it for testing within SPSS.

Thanks in advance for any thoughts on how to approach this.
Reply | Threaded
Open this post in threaded view
|

Re: Statistical testing question from a novice

Marta García-Granero
Hi Robert

Perhaps I haven't understood your data layout, but I don't think you
have paired data at all, unless you artificially pair every "woman
25-54" with another respondent. Also, you can't test the group of
"women 25-54" with "all panel respondents" (that would include those
women in both groups). You must split your respondents in two groups
(·Women 25-54· vs "Other respondents") and use an independent samples
test, like Mann-Whitney's U test (since your data are ordinal level).

The best way of splitting the dataset in two would be a filter
variable (1=Women 25-54; 0=All other respondents).

RM> My company fielded a survey composed of both a local Web panel and
RM> respondents gathered via pop-up solicitations on our Web site. I'm only
RM> interested in analyzing from the Web panel universe.

RM> For a large number of questions, I want to compare panel respondents that
RM> are Women 25-54 with all panel respondents to see if there are significant
RM> differences between Women 25-54 on the panel and all panel respondents. For
RM> the discussion I'll assume I'm dealing with ordinal-level data
RM> (i.e. "daily", "several times a week", "several times a month", "once a
RM> month or less", "never").

RM> My thinking is that this type of comparison would constitute a
RM> nonparametric test of paired samples, so the best test to use would be a
RM> Wilcoxon matched-pairs signed-ranks test.

RM> In order to create my variable of Women 25-54 from the Web panel in SPSS, I
RM> would need to use a Boolean operation to create the variable - would I
RM> simply use "Transform," then "Recode into Different Variables"? Is that the
RM> best way to create new groupings of the sample? I've always used "Select
RM> Cases," but that doesn't permanently change the selection. I'm not sure how
RM> I should create that new grouping to submit it for testing within SPSS.

RM> Thanks in advance for any thoughts on how to approach this.

RM> __________ Información de NOD32, revisión 1.1713 (20060817) __________

RM> Este mensaje ha sido analizado con  NOD32 antivirus system
RM> http://www.nod32.com




--
Regards,
Dr. Marta García-Granero,PhD           mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)