MxN variation of McNemar-Bowker test?

[rsijben]

Dear SPSS list,

I have several models which categorize my data. The first model categorizes
it in 3 conditions (0,1,2) while the 2nd (and later) model categorizes it in
4 conditions (0,1,2,3).

My research question is whether the proportions of the categories differ
significantly between the 2 models. Some research led to me to
McNemar-Bowker test which is, unfortunately, limited to NxN crosstabs.

I've read about some alternatives in the form of repeated measures
multinomial logistic regression (which don't seem to be supported by SPSS).

Is there something similar to a MxN variation of the McNemar-Bowker test
available in SPSS?



--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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

[Jon Peck]

Analyze > Mixed Models > Generalized Linear supports multinomial logistic with repeated measures.

On Wed, Jan 24, 2018 at 8:10 AM, rsijben <[hidden email]> wrote:

Dear SPSS list,

I have several models which categorize my data. The first model categorizes
it in 3 conditions (0,1,2) while the 2nd (and later) model categorizes it in
4 conditions (0,1,2,3).

My research question is whether the proportions of the categories differ
significantly between the 2 models. Some research led to me to
McNemar-Bowker test which is, unfortunately, limited to NxN crosstabs.

I've read about some alternatives in the form of repeated measures
multinomial logistic regression (which don't seem to be supported by SPSS).

Is there something similar to a MxN variation of the McNemar-Bowker test
available in SPSS?



--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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

--

===================== 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

[Kornbrot, Diana]

Can do this in procedure mixed, but nee data in LONG form

i.e. frequency for cat 2 beneath freq for cat 1 etc with 4 values underneath each other for each unit [subject]

best

Diana

On 24 Jan 2018, at 15:14, Jon Peck <[hidden email]> wrote:

Analyze > Mixed Models > Generalized Linear supports multinomial logistic with repeated measures.

On Wed, Jan 24, 2018 at 8:10 AM, rsijben <[hidden email]> wrote:

Dear SPSS list,

I have several models which categorize my data. The first model categorizes
it in 3 conditions (0,1,2) while the 2nd (and later) model categorizes it in
4 conditions (0,1,2,3).

My research question is whether the proportions of the categories differ
significantly between the 2 models. Some research led to me to
McNemar-Bowker test which is, unfortunately, limited to NxN crosstabs.

I've read about some alternatives in the form of repeated measures
multinomial logistic regression (which don't seem to be supported by SPSS).

Is there something similar to a MxN variation of the McNemar-Bowker test
available in SPSS?



--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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

--

===================== 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

_____________________________________
Professor Diana Kornbrot
Mobile
[phone]
Work
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
[phone]
[hidden email]
http://dianakornbrot.wordpress.com/
http://go.herts.ac.uk/Diana_Kornbrot
skype:  kornbrotme
Home
19 Elmhurst Avenue
London N2 0LT, UK
[phone]                                                   
 ------------------------------------------------------------                                    

===================== 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

[Bruce Weaver]

Maybe it's just too early in the morning, but that research question makes no
sense to me.  How can the proportions be the same if the number of
categories differ?  Thanks for clarifying.



rsijben wrote

> Dear SPSS list,
>
> I have several models which categorize my data. The first model
> categorizes
> it in 3 conditions (0,1,2) while the 2nd (and later) model categorizes it
> in
> 4 conditions (0,1,2,3).
>
> My research question is whether the proportions of the categories differ
> significantly between the 2 models. Some research led to me to
> McNemar-Bowker test which is, unfortunately, limited to NxN crosstabs.
>
> I've read about some alternatives in the form of repeated measures
> multinomial logistic regression (which don't seem to be supported by
> SPSS).
>
> Is there something similar to a MxN variation of the McNemar-Bowker test
> available in SPSS?
>
>
>
> --
> Sent from: http://spssx-discussion.1045642.n5.nabble.com/
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (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





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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

[rsijben]

Thank you for your suggestions,

Considering how each case is assigned to 1 category per model I created the
following long format:

1 Column for case/subject number, each case listed twice as we're comparing
2 models.
1 Column listing model number 1 or 2, used as repeated measure.
1 Column listing the assigned category for each case by the respective
model.


Inputting this to the multinomial logistic under generalized linear grants
the error: "Repeated measures analysis is not supported for the multinomial
probability distribution".


@Mr. Weaver,

The first model assigns each case to 1 of 3 categories while the 2nd model
assigns these cases to 4 categories. Model 1 could, for example, assign the
data evenly over the conditions yielding a proportion of 0.33 for each
category. Model 2, being more optimized, might recognize that half the cases
in category 3 actually belong in a new category 4, creating the proportions
0.33, 0.33, 0.1565, 0.1565 for categories 1, 2, 3, and 4, respectively. The
proportion of cases assigned to category 1 and 2 would remain identical.





--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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

[Bruce Weaver]

If you mean that all 4 categories were available for the first model, but the
4th category just was not used, fair enough.  In that case, I think you can
solve your problem by using INTEGER mode on your CROSSTABS command.  

* McNemar-Bowker test for marginal homogeneity with non-square table.
DATA LIST FREE / res1980 res1985 wt .
BEGIN DATA.
1 1 11607
1 2 100
1 3 366
1 4 124
2 1 87
2 2 13677
2 3 515
2 4 302
3 1 172
3 2 225
3 3 17819
3 4 270
4 1 63
4 2 176
4 3 286
4 4 10192
END DATA.
WEIGHT BY wt.

* Create the symmetry and quasi-symmetry indicator variables .
COMPUTE s1 = any(1,res1980,res1985).
COMPUTE s2 = any(2,res1980,res1985).
COMPUTE s3 = any(3,res1980,res1985).

COMPUTE qs12 = s1*s2 .
COMPUTE qs13 = s1*s3 .
COMPUTE qs14 = s1*any(4,res1980,res1985).
COMPUTE qs23 = s2*s3 .
COMPUTE qs24 = s2*any(4,res1980,res1985).
COMPUTE qs34 = s3*any(4,res1980,res1985).

FORMATS s1 to qs34 (F4).

* Symmetry model - reported in Agresti, page 357 .
GENLOG
res1980 res1985 WITH s1 s2 s3 qs12 qs13 qs14 qs23 qs24 qs34
/PRINT FREQ ESTIM
/PLOT NONE
/DESIGN qs12 qs13 qs14 qs23 qs24 qs34 s1 s2 s3 .

CROSSTABS RES1980 by RES1985 /STATISTICS=MCNEMAR.

* RECODE RES1980 so that all 4s become 3s.
RECODE RES1980 (4=3) (ELSE=COPY) into RES1980r.
* Now redo the CROSSTABS command--but note that we need
* to use INTEGER mode with a non-square table.
CROSSTABS VARIABLES = RES1980r (1,4) RES1985(1,4)
 /TABLES = RES1980r by RES1985 /STATISTICS=MCNEMAR.

* References.
*
http://spssx-discussion.1045642.n5.nabble.com/MxN-variation-of-McNemar-Bowker-test-td5735446.html#a5735452
* http://www-01.ibm.com/support/docview.wss?uid=swg21477019.



rsijben wrote

> Thank you for your suggestions,
>
> Considering how each case is assigned to 1 category per model I created
> the
> following long format:
>
> 1 Column for case/subject number, each case listed twice as we're
> comparing
> 2 models.
> 1 Column listing model number 1 or 2, used as repeated measure.
> 1 Column listing the assigned category for each case by the respective
> model.
>
>
> Inputting this to the multinomial logistic under generalized linear grants
> the error: "Repeated measures analysis is not supported for the
> multinomial
> probability distribution".
>
>
> @Mr. Weaver,
>
> The first model assigns each case to 1 of 3 categories while the 2nd model
> assigns these cases to 4 categories. Model 1 could, for example, assign
> the
> data evenly over the conditions yielding a proportion of 0.33 for each
> category. Model 2, being more optimized, might recognize that half the
> cases
> in category 3 actually belong in a new category 4, creating the
> proportions
> 0.33, 0.33, 0.1565, 0.1565 for categories 1, 2, 3, and 4, respectively.
> The
> proportion of cases assigned to category 1 and 2 would remain identical.
>
>
>
>
>
> --
> Sent from: http://spssx-discussion.1045642.n5.nabble.com/
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (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





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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