SPSSX Discussion

Missing subcommand in Mixed model

Classic

List

Threaded

3 messages Options

didier

Missing subcommand in Mixed model

Hi there
I just have one question regarding "missing" subcommand in Mixed Model.
It says:

MISSING Subcommand (MIXED command)
The MISSING subcommand specifies the way to handle cases with user-missing values.
• If this subcommand is not specified, the default is EXCLUDE.
• Cases, which contain system-missing values in one of the variables, are always deleted.
• The keywords EXCLUDE and INCLUDE are mutually exclusive. Only one of them can be specified at once.

I have a simple repeated design (time, 3 levels) with one independent variable (Var) and one dependent variable group (4 levels)
the data is organized in "stacked" form. The advantage of using mixed model is that I have missing data and would like the model to use all available data rather than doing list wise deletion.

If I am understanding the above statement properly, I need to define and state to include missing data in order to run the model and include subjects with missing data? However when doing so using -99 as the missing value code (see script below), the model actually considers the value to be -99 whenever the data is missing in the calculation (rather than considering that value to be missing).
Am I missing something :-)?

RECODE Var (SYSMIS=-99).
MISSING VALUES Var (-99).
MIXED Var BY Group Time
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=Group Time Group*Time | SSTYPE(3)
/METHOD=REML
/missing=INCLUDE
/REPEATED=Time | SUBJECT(SID) COVTYPE(AR1)
/EMMEANS=TABLES(Group*Time) COMPARE(Time) refcat(first) ADJ(lsd).

Alex Reutter

Re: Missing subcommand in Mixed model

Hi,

MIXED Var BY Group Time /FIXED=Group Time Group*Time | SSTYPE(3) /MISSING=INCLUDE /REPEATED=Time | SUBJECT(SID) COVTYPE(AR1).
is essentially a shortcut to

MISSING VALUES ALL ().
MIXED Var BY Group Time /FIXED=Group Time Group*Time | SSTYPE(3) /MISSING=INCLUDE /REPEATED=Time | SUBJECT(SID) COVTYPE(AR1).
MISSING VALUES [syntax setting all your user-missing values back to the way they were]
The MIXED procedure uses more of the available data than GLM in the situation where values are missing for some values of TIME, but not others. Recall that the data structure for GLM makes each case a separate subject, while MIXED spreads subjects across cases. For example, data for running GLM repeated measures might look like:

GROUP VAR.TIME1 VAR.TIME2 VAR.TIME3
3 . 10.3 8.4
2 14.1 7.6 .
1 10.0 . 9.2
...

And the same data for MIXED would be structured:

SUBJ GROUP TIME VAR
1 3 1 .
1 3 2 10.3
1 3 3 8.4
2 2 1 14.1
2 2 2 7.6
2 2 3 .
3 1 1 10.0
3 1 2 .
3 1 3 9.2
...

MIXED has 6 cases to work with, while GLM has none. If you want to use the cases with missing values for VAR, you need to impute values; using, for example, MULTIPLE IMPUTATION (see http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/mi_analysis.htm for the list of procedures that support it).

Cheers,
Alex

From: didier <[hidden email]>
To: [hidden email],
Date: 06/04/2013 01:58 PM
Subject: Missing subcommand in Mixed model
Sent by: "SPSSX(r) Discussion" <[hidden email]>

Hi there I just have one question regarding "missing" subcommand in Mixed Model. It says: MISSING Subcommand (MIXED command) The MISSING subcommand specifies the way to handle cases with user-missing values. • If this subcommand is not specified, the *default is EXCLUDE*. • Cases, which contain system-missing values in one of the variables, are *always* deleted. • The keywords EXCLUDE and INCLUDE are mutually exclusive. Only one of them can be specified at once. I have a simple repeated design (time, 3 levels) with one independent variable (Var) and one dependent variable group (4 levels) the data is organized in "stacked" form. The advantage of using mixed model is that I have missing data and would like the model to use all available data rather than doing list wise deletion. If I am understanding the above statement properly, I need to define and state to include missing data in order to run the model and include subjects with missing data? However when doing so using -99 as the missing value code (see script below), the model actually considers the value to be -99 whenever the data is missing in the calculation (rather than considering that value to be missing). Am I missing something :-)? RECODE Var (SYSMIS=-99). MISSING VALUES Var (-99). MIXED Var BY Group Time /CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=Group Time Group*Time | SSTYPE(3) /METHOD=REML /missing=INCLUDE /REPEATED=Time | SUBJECT(SID) COVTYPE(AR1) /EMMEANS=TABLES(Group*Time) COMPARE(Time) refcat(first) ADJ(lsd). -- View this message in context:http://spssx-discussion.1045642.n5.nabble.com/Missing-subcommand-in-Mixed-model-tp5720554.htmlSent from the SPSSX Discussion mailing list archive at 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

didier

Re: Missing subcommand in Mixed model

Thanks a lot Alex for taking the time to respond.

That answers my question. I was actually able to talk also to SPSS customer support who confirmed that the the Mixed model by default takes into account all available data. As you described, only cases with missing data at a particular timepoint is excluded. So in most cases, one should ignore the missing subcommand.

They mentioned that one may want consider choosing "include" option for the Missing Subcommand for special cases such as for categorical data when you want to run statistics on all data including "missing" data which will then be considered as a categorical value in its own right (to include for example % of missing data in the statistics).