SPSSX Discussion

Emmeans and WITH(cov=MEAN) sets covariate to 1.00

Classic

List

Threaded

4 messages Options

Garry Anderson

Emmeans and WITH(cov=MEAN) sets covariate to 1.00

Dear List,

In Unianova I am using the subcommand
/emmeans = tables(treat) with(wt1=MEAN)

The output sets the wt1 covariate = 1.00, even though it is really 470.
I am using version 17. I repeat the same commands in version 16 and the
covariate is correctly set to 470. If I use a different covariate in
version 17 I also obtain that covariate being set to 1.00.

Any help would be appreciated as to why the covariate is always being
set to 1.00.

Cheers, Garry
[hidden email]

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

SPSS Support

Re: Emmeans and WITH(cov=MEAN) sets covariate to 1.00

I'm not able to replicate a problem like this with Release 17. I've asked Garry offline for further information, but in the meantime if anyone else should encounter such an issue, please let us know.

David Nichols

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Anderson
Sent: Wednesday, November 19, 2008 2:35 AM
To: [hidden email]
Subject: [SPSSX-L] Emmeans and WITH(cov=MEAN) sets covariate to 1.00

Dear List,

In Unianova I am using the subcommand
/emmeans = tables(treat) with(wt1=MEAN)

The output sets the wt1 covariate = 1.00, even though it is really 470.
I am using version 17. I repeat the same commands in version 16 and the covariate is correctly set to 470. If I use a different covariate in version 17 I also obtain that covariate being set to 1.00.

Any help would be appreciated as to why the covariate is always being set to 1.00.

Cheers, Garry
[hidden email]

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

Marta Garcia-Granero

Re: Emmeans and WITH(cov=MEAN) sets covariate to 1.00

SPSS Support wrote:
> I'm not able to replicate a problem like this with Release 17. I've asked Garry offline for further information, but in the meantime if anyone else should encounter such an issue, please let us know.
>

I couldn't either. I have tested it with this dataset:

data list list/treatment nplants yield (3 F8).
begin data
1 9 14
1 7 13
1 13 17
1 25 30
2 19 36
2 15 17
2 14 18
2 18 22
3 16 7
3 18 9
3 25 22
3 20 16
4 18 19
4 24 33
4 16 25
4 21 37
5 15 8
5 17 10
5 18 14
5 17 14
6 17 12
6 16 11
6 18 13
6 20 17
end data.

UNIANOVA yield BY treatment WITH nplants
/METHOD=SSTYPE(3)
/INTERCEPT=INCLUDE
/EMMEANS=TABLES(treatment) WITH(nplants=MEAN)
/CRITERIA=ALPHA(.05)
/DESIGN=nplants treatment.

Results are exactly the same in SPSS 15 and 17. Perhaps Garry can use it
to check his SPSS 17 behavior.

HTH,
Marta

> David Nichols
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Anderson
> Sent: Wednesday, November 19, 2008 2:35 AM
> To: [hidden email]
> Subject: [SPSSX-L] Emmeans and WITH(cov=MEAN) sets covariate to 1.00
>
> Dear List,
>
> In Unianova I am using the subcommand
> /emmeans = tables(treat) with(wt1=MEAN)
>
> The output sets the wt1 covariate = 1.00, even though it is really 470.
> I am using version 17. I repeat the same commands in version 16 and the covariate is correctly set to 470. If I use a different covariate in version 17 I also obtain that covariate being set to 1.00.
>
> Any help would be appreciated as to why the covariate is always being set to 1.00.
>
> Cheers, Garry
> [hidden email]
>
> =====================
> 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
>
>

SPSS Support

Re: Emmeans and WITH(cov=MEAN) sets covariate to 1.00

Marta, thank you for also trying this. The reason that neither you nor I could replicate the issue is that it occurs only when you have a factor variable that's a long string (width set to more than eight characters). I discovered this after Garry sent me his files. Many thanks to Garry for pointing out the problem and for sending in the information necessary to diagnose the specific conditions under which it occurs.

The problem affects both the GLM and UNIANOVA procedures (the two procedures accessed via Analyze>General Linear Model in the menus) in Release 17.0.0. If you name a variable as a factor on the main command line that is a long string, whether or not that factor is specified on EMMEANS or even used in the model, and you specify an EMMEANS subcommand with the default specification of fixing covariates at their mean values, either by specifying WITH(cov=MEAN) or by not specifying WITH at all, the procedures will fix covariates at incorrect values in computations of EMMEANS. If there is a single covariate, it will be set to 1 rather than its mean. If there are multiple covariates, the last one listed will be set to 1, and others will apparently be set to different incorrect values.

This should be fixed in the 17.0.1 patch release. Until then, there are two ways to work around the issue:

1) Specify the mean value of each covariate explicitly. This requires use of command syntax. If you're using the menus, click Paste rather than OK, and replace MEAN with the actual mean value of the covariate in the EMMEANS subcommand.

2) Create a new variable to use as a factor that is not a long string. The AUTORECODE procedure (Transform>Automatic Recode in the menus) provides an easy way to do this. It will create a numeric variable with the original variable's strings as value labels. The value labels will be shown in the output, so the levels of the factor will (with default settings) be labeled as they were with the string variable.

Again thanks to Garry for pointing out the issue and we apologize for any resulting inconvenience.

David Nichols

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Marta García-Granero
Sent: Wednesday, November 19, 2008 1:12 PM
To: [hidden email]
Subject: Re: [SPSSX-L] Emmeans and WITH(cov=MEAN) sets covariate to 1.00

SPSS Support wrote:
> I'm not able to replicate a problem like this with Release 17. I've asked Garry offline for further information, but in the meantime if anyone else should encounter such an issue, please let us know.
>

I couldn't either. I have tested it with this dataset:

data list list/treatment nplants yield (3 F8).
begin data
1 9 14
1 7 13
1 13 17
1 25 30
2 19 36
2 15 17
2 14 18
2 18 22
3 16 7
3 18 9
3 25 22
3 20 16
4 18 19
4 24 33
4 16 25
4 21 37
5 15 8
5 17 10
5 18 14
5 17 14
6 17 12
6 16 11
6 18 13
6 20 17
end data.

UNIANOVA yield BY treatment WITH nplants
/METHOD=SSTYPE(3)
/INTERCEPT=INCLUDE
/EMMEANS=TABLES(treatment) WITH(nplants=MEAN)
/CRITERIA=ALPHA(.05)
/DESIGN=nplants treatment.

Results are exactly the same in SPSS 15 and 17. Perhaps Garry can use it to check his SPSS 17 behavior.

HTH,
Marta

> David Nichols
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf
> Of Garry Anderson
> Sent: Wednesday, November 19, 2008 2:35 AM
> To: [hidden email]
> Subject: [SPSSX-L] Emmeans and WITH(cov=MEAN) sets covariate to 1.00
>
> Dear List,
>
> In Unianova I am using the subcommand
> /emmeans = tables(treat) with(wt1=MEAN)
>
> The output sets the wt1 covariate = 1.00, even though it is really 470.
> I am using version 17. I repeat the same commands in version 16 and the covariate is correctly set to 470. If I use a different covariate in version 17 I also obtain that covariate being set to 1.00.
>
> Any help would be appreciated as to why the covariate is always being set to 1.00.
>
> Cheers, Garry
> [hidden email]
>
> =====================
> 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
>
>

--
For miscellaneous statistical stuff, visit:
http://gjyp.nl/marta/

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