Crosstabulation statistics for a nominal and ordinal variable
Locked
Crosstabulation statistics for a nominal and ordinal variable
 
|
Dear list! I have a crosstabulation problem. I have 2
treatment groups (independent) as the column variable and an ordinal variable
(often, seldom, never) as the row variable. I get a highly significant chi
square using exact test (I got some expected frequencies less than 5). I,
however would like to pinpoint where the main differences lie (within the
ordinal variable) by pairwise comparisons. Besides doing this by creating new
dichotomized variables and performing new analyses on these, is there a simpler
and more economical way of doing this? best Staffan Lindberg Sweden |
Re: Crosstabulation statistics for a nominal and ordinal variable
Administrator
|
How about using TEMPORARY-SELECT IF? E.g., temp. select if (rowvar NE 3). /* 1 vs 2 . crosstabs rowvar by colvar / stat = chisqr. temp. select if (rowvar NE 2). /* 1 vs 3 . crosstabs rowvar by colvar / stat = chisqr. temp. select if (rowvar NE 1). /* 2 vs 3 . crosstabs rowvar by colvar / stat = chisqr.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by Staffan Lindberg
You might consider the column proportions test in CTABLES.
From: SPSSX(r) Discussion <[hidden email]> To: [hidden email] <[hidden email]> Sent: Thu Jun 25 04:50:29 2009 Subject: [SPSSX-L] Crosstabulation statistics for a nominal and ordinal variable Dear list! I have a crosstabulation problem. I have 2
treatment groups (independent) as the column variable and an ordinal variable
(often, seldom, never) as the row variable. I get a highly significant chi
square using exact test (I got some expected frequencies less than 5). I,
however would like to pinpoint where the main differences lie (within the
ordinal variable) by pairwise comparisons. Besides doing this by creating new
dichotomized variables and performing new analyses on these, is there a simpler
and more economical way of doing this? best Staffan Lindberg Sweden |
Re: Crosstabulation statistics for a nominal and ordinal variable
|
|
In reply to this post by Staffan Lindberg
Also recall, however, that such
tests suffer from inflated type I error rates and so one needs to be careful
about that +/- 1.96 criterion. Dr. Paul R. Swank, Professor and Director of
Research Children's Learning Institute University of Texas Health
Science Center-Houston From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Staffan Lindberg Dear list! I got an interesting solution
from Carol Paris. Never thought of “sresid” that way. Could maybe be of
interest to other members of the list. best Staffan Lindberg Sweden Från: Parise,
Carol A. [mailto:[hidden email]] Staffan, in the /cells subcommand, add "sresid". This gives you a
standardizsed residual for each cell of your crosstabs table. If a standardized
residual is <-1.96 or >1.96, the frequency of the cell is statistically
greater than or less than expected. It's sometimes helpful to included the /expected
value in the cell also so you can see what the expected frequency is versus the
observed frequency. Carol From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Staffan Lindberg Dear list! I have a crosstabulation problem. I have 2 treatment groups
(independent) as the column variable and an ordinal variable (often, seldom,
never) as the row variable. I get a highly significant chi square using
exact test (I got some expected frequencies less than 5). I, however
would like to pinpoint where the main differences lie (within the ordinal
variable) by pairwise comparisons. Besides doing this by creating new
dichotomized variables and performing new analyses on these, is there a simpler
and more economical way of doing this? best Staffan Lindberg Sweden |
|
|
Paul has a very good point about Type 1 error rate.
But i think they serve the purpose of finding patterns in
data.
I think of the standarized
residuals in analogous to the post-hoc test following a statistically
significant AVOVA with more than 2 levels of an IV where the post-hoc tests show
you where the differences occur. With a
statistically significant chisquare in a crosstab table, the sresids point
you to the cells that are contributing to the significant chisquare.
While
patterns tend to emerge when you see the residuals, It's not always straight
forward. I've had crosstabs with a statistically significant chiquare and none
of the sresids are above or below 1.96. In this case, there tends to be a
pattern among the cells with negative versus positive standardized residuals and
it helps you hone in on specific levels of categories.
Whether the result is straight forward or not, the residuals are mostly a
means of establishing patterns between levels of variables rather than a test of
statistical signficance.
Carol
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R Sent: Friday, June 26, 2009 8:35 AM To: [hidden email] Subject: Re: Crosstabulation statistics for a nominal and ordinal variable Also recall, however, that such
tests suffer from inflated type I error rates and so one needs to be careful
about that +/- 1.96 criterion. Dr. Paul R. Swank,
Professor and Director of
Research Children's Learning
Institute University of Texas Health
Science Center-Houston From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Staffan
Lindberg Dear
list! I got an interesting solution
from Carol Paris. Never thought of sresid that way. Could maybe be of interest
to other members of the list. best Staffan
Lindberg Sweden Från: Parise,
Carol A. [mailto:[hidden email]] Staffan, in the
/cells subcommand, add "sresid". This gives you a standardizsed residual for
each cell of your crosstabs table. If a standardized residual is <-1.96 or
>1.96, the frequency of the cell is statistically greater than or less than
expected. It's sometimes helpful to included the /expected value in the cell
also so you can see what the expected frequency is versus the observed
frequency. Carol From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Staffan
Lindberg Dear list! I have a crosstabulation problem. I have 2 treatment groups
(independent) as the column variable and an ordinal variable (often, seldom,
never) as the row variable. I get a highly significant chi square using
exact test (I got some expected frequencies less than 5). I, however would
like to pinpoint where the main differences lie (within the ordinal variable) by
pairwise comparisons. Besides doing this by creating new dichotomized variables
and performing new analyses on these, is there a simpler and more economical way
of doing this? best Staffan Lindberg Sweden |
| Free forum by Nabble | Edit this page |