GLM: discrepancy between ANOVA table and Parameters table

[Kirill Orlov]

data list list /y x1 x2.
begin data
1    1    1
3    1    1
3    1    1
3    1    1
6    1    1
2    1    2
2    1    2
4    1    2
2    1    2
3    1    2
4    1    2
5    1    2
4    2    1
6    2    1
6    2    1
4    2    1
3    2    1
6    2    1
3    2    1
1    2    1
1    2    1
5    2    2
5    2    2
7    2    2
5    2    2
4    2    2
4    2    2
5    2    2
2    2    2
4    2    2
end data.

UNIANOVA y BY x1 x2
  /METHOD=SSTYPE(3)
  /INTERCEPT=INCLUDE
  /CRITERIA=ALPHA(0.05)
  /PRINT=PARAMETER
  /DESIGN=x1 x2 x1*x2.

P-values of effects in ANOVA table is different (except for the interaction) from those in regressional parameters table.
This is because GLM outputs ANOVA table as corresponding to linear regression with factors recoded into deviation contrast predictors, whereas it outputs Parameters table as corresponding to linear regression with factors recoded into indicator contrast predictors (i.e. dummy variables). OK.

My question is: Is it possible somehow through syntax to order the command to output both tables with the same p-values, that is, corresponding to the same type of contast variables? It's interesting to note that old good MANOVA command allowed it.

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

[Ryan]

Kirill,

Here you go:

--

RECODE x1 x2 (2=0).
EXECUTE.

UNIANOVA y with x1 x2
  /METHOD=SSTYPE(3)
  /INTERCEPT=INCLUDE
  /CRITERIA=ALPHA(0.05)
   /PRINT=PARAMETER
  /DESIGN=x1 x2 x1*x2.

--

Ryan

On Wed, Nov 5, 2014 at 5:06 AM, Kirill Orlov <[hidden email]> wrote:

data list list /y x1 x2.
begin data
1    1    1
3    1    1
3    1    1
3    1    1
6    1    1
2    1    2
2    1    2
4    1    2
2    1    2
3    1    2
4    1    2
5    1    2
4    2    1
6    2    1
6    2    1
4    2    1
3    2    1
6    2    1
3    2    1
1    2    1
1    2    1
5    2    2
5    2    2
7    2    2
5    2    2
4    2    2
4    2    2
5    2    2
2    2    2
4    2    2
end data.

UNIANOVA y BY x1 x2
  /METHOD=SSTYPE(3)
  /INTERCEPT=INCLUDE
  /CRITERIA=ALPHA(0.05)
  /PRINT=PARAMETER
  /DESIGN=x1 x2 x1*x2.

P-values of effects in ANOVA table is different (except for the interaction) from those in regressional parameters table.
This is because GLM outputs ANOVA table as corresponding to linear regression with factors recoded into deviation contrast predictors, whereas it outputs Parameters table as corresponding to linear regression with factors recoded into indicator contrast predictors (i.e. dummy variables). OK.

My question is: Is it possible somehow through syntax to order the command to output both tables with the same p-values, that is, corresponding to the same type of contast variables? It's interesting to note that old good MANOVA command allowed it.

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

[Kirill Orlov]

Ryan :-) Thanks, but... are you kidding me?
I meant, of course, are there way do what my question posed by syntax of GLM/UNIANOVA comand itself, not by any external operations on the data.

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

[Ryan]

Kirill,

I changed x1 and x2 to covariates rather than factors (WITH instead of BY). That, by itself, will make both Tables produce identical results. 

I then decided to recode the variables so that they are dummy coded in a standard way.

Ryan

On Wed, Nov 5, 2014 at 2:21 PM, Kirill Orlov <[hidden email]> wrote:

Ryan :-) Thanks, but... are you kidding me?
I meant, of course, are there way do what my question posed by syntax of GLM/UNIANOVA comand itself, not by any external operations on the data.

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

[Kirill Orlov]

I know about that opportunity. Using factors as covariates won't help if factors are not dichotomous.

The relevant answer to my question might be "use LMATRIX subcommand accordingly". However, such answer won't suit me, either, because it won't affect the ANOVA and the Parameters tables. Rather, it will produce additional tables.

05.11.2014 22:27, Ryan Black пишет:

Kirill,

I changed x1 and x2 to covariates rather than factors (WITH instead of BY). That, by itself, will make both Tables produce identical results. 

I then decided to recode the variables so that they are dummy coded in a standard way.

Ryan

On Wed, Nov 5, 2014 at 2:21 PM, Kirill Orlov <[hidden email]> wrote:

Ryan :-) Thanks, but... are you kidding me?
I meant, of course, are there way do what my question posed by syntax of GLM/UNIANOVA comand itself, not by any external operations on the data.

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