Repeated measures ANOVA with missing data

[Steve-192]

Hey all,

Brief info about my dataset: Cell counts from histological samples of
liver tissue were taken for case and control patients.  Each patient was
supposed to get 3 readings done by a pathologist who counted the cells
(but some patients only had 2 readings done). Need to find out if patients
on the treatment had a higher/lower cell count than control patients.

I was deciding between running a repeated measures ANOVA. However, I hit a
snag with the latter model.  Since I have missing data (2 instead of 3
slides read) for half my patients, the model delets those cases from the
analysis.

Is there a way to take the missing data into consideration in the repeated
measures ANOVA?  Syntax is fine but I'm not well versed in Python.  Should
I use some other model (mixed model, perhaps)?  Or would it be fine to run
a t-test using the average counts of each patient (would avoid the whole 2
readings vs 3 reading issue)?

Thanks in advance.
-Steve

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Swank, Paul R]

You need to use a mixed model instead. This will make use of what data
is there (using maximum likelihood) and is much better with missing data
than GLM, which uses listwise deletion in repeated measures.

Paul R. Swank, Ph.D.
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center - Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:SPSSX-L@...] On Behalf Of
Steve
Sent: Thursday, June 19, 2008 6:35 PM
To: SPSSX-L@...
Subject: Repeated measures ANOVA with missing data

Hey all,

Brief info about my dataset: Cell counts from histological samples of
liver tissue were taken for case and control patients.  Each patient was
supposed to get 3 readings done by a pathologist who counted the cells
(but some patients only had 2 readings done). Need to find out if
patients
on the treatment had a higher/lower cell count than control patients.

I was deciding between running a repeated measures ANOVA. However, I hit
a
snag with the latter model.  Since I have missing data (2 instead of 3
slides read) for half my patients, the model delets those cases from the
analysis.

Is there a way to take the missing data into consideration in the
repeated
measures ANOVA?  Syntax is fine but I'm not well versed in Python.
Should
I use some other model (mixed model, perhaps)?  Or would it be fine to
run
a t-test using the average counts of each patient (would avoid the whole
2
readings vs 3 reading issue)?

Thanks in advance.
-Steve

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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
LISTSERV@... (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

[Reutter, Alex]

Also note that in order to use MIXED, you'll need to restructure your dataset.  If you look at Help > Case Studies..., then Advanced > Linear Mixed Models > Using Linear Mixed Models to Analyze Repeated Measurements, there are instructions on how to restructure a file from variables to cases (i.e., originally set up for use in GLM Repeated Measures).

Cheers,
Alex

-----Original Message-----
From: SPSSX(r) Discussion [mailto:SPSSX-L@...] On Behalf Of Swank, Paul R
Sent: Friday, June 20, 2008 10:29 AM
To: SPSSX-L@...
Subject: Re: Repeated measures ANOVA with missing data

You need to use a mixed model instead. This will make use of what data
is there (using maximum likelihood) and is much better with missing data
than GLM, which uses listwise deletion in repeated measures.

Paul R. Swank, Ph.D.
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center - Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:SPSSX-L@...] On Behalf Of
Steve
Sent: Thursday, June 19, 2008 6:35 PM
To: SPSSX-L@...
Subject: Repeated measures ANOVA with missing data

Hey all,

Brief info about my dataset: Cell counts from histological samples of
liver tissue were taken for case and control patients.  Each patient was
supposed to get 3 readings done by a pathologist who counted the cells
(but some patients only had 2 readings done). Need to find out if
patients
on the treatment had a higher/lower cell count than control patients.

I was deciding between running a repeated measures ANOVA. However, I hit
a
snag with the latter model.  Since I have missing data (2 instead of 3
slides read) for half my patients, the model delets those cases from the
analysis.

Is there a way to take the missing data into consideration in the
repeated
measures ANOVA?  Syntax is fine but I'm not well versed in Python.
Should
I use some other model (mixed model, perhaps)?  Or would it be fine to
run
a t-test using the average counts of each patient (would avoid the whole
2
readings vs 3 reading issue)?

Thanks in advance.
-Steve

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Steve-192]

Thanks Paul, Alex and Sara,

I was afraid that I might have to use mixed models.  Also, thanks to both
Alex and Sara (whose emails didn't go to the list and I'm posting below)
for giving me the heads up on restructuring the dataset.  It was
originally structured for a linear analysis and I had restructured it for
the ANOVA... so it's back to drawing board. Thanks again for the tips.

-Steve


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Sara House wrote on Fri, Jun 20, 2008 at 9:13AM

Steve -

If you use a mixed model, you wouldn't lose all your cases with missing
values.  Instead, you would be able to use all cases that have data at a
particular time.  At the risk of sounding like Don Hedeker's personal
cheerleader (I've recommended his site on this list before), if you go to
http://tigger.uic.edu/~hedeker/ you'll find SPSS and SAS syntax to do
mixed-model analyses with continuous data, and SAS syntax for dichotomous
data.

BTW, does anyone know if there is a way to do mixed model analyses with
dichotomous data in SPSS?  I don't have access to SAS at work and I'm more
familiar with SPSS anyway.

Sara

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Sara House wrote on Fri, Jun 20, 2008 at 11:43 AM

Steve -

Oop, sorry, didn't realize my first email went only to you and not the
list.  You can ignore the question in my email.

Also, Alex is right that you'll want to restructure your file.  So instead
of:

ID      T1Meas      T2Meas     T3Meas
1       x           y          z
2 ...

You'll want a single measurement variable and then a time variable:

ID         Meas       Time
1            x             1
1            y             2
1            z             3
2    ...

Hope this helps!


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Roberts, Michael]

Hi Listers,

I would appreciate some help with a way to create a counter of the
number of occurrences within records that have two variables - id and
date, that are the same.  In other words, when the id changes, or the
date changes, I want to begin counting again.

There are solutions but they all to require a numeric value for the id.
I seem to remember doing something like this relatively easily a long
while back, but cannot find the code.

Any help would be much appreciated.

TIA

Mike

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Richard Ristow]

At 05:20 PM 6/23/2008, Roberts, Michael wrote:

>I would like to create a counter of the number of occurrences within
>records that have two variables - id and date. When the id changes,
>or the date changes, I want to begin counting again.

There's a number of ways. This one (code not tested) is fairly easy.
It requires that the file be sorted by ID and Date; and it leaves
variable NewGroup in the file, which you may not want.

>There are solutions but they all to require a numeric value for the id.

This will work with any datatypes for ID and Date.

ADD FILES
   /FILE=*
   /BY ID Date
   /FIRST = NewGroup.

NUMERIC SeqInGrp (F4).
DO IF   NewGroup.
.  COMPUTE SeqInGrp = 1.
ELSE.
.  COMPUTE SeqInGrp = LAG(SeqInGrp) + 1.
END IF.

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Roberts, Michael]

Richard,

Thank you for the solution.  I did finally manage this by creating a
counter, setting its default value to 1, then testing both values (date
and ID)lagged by 1, and incrementing the counter by 1; seems to work.

Your solution, however looks very interesting, and I am going to check
it out in the morning. In my search for a solution, I did see some code
you posted a while back which includes 'Before' and 'After' commands.
My search for information on these two turned up very little.  How can I
find out more about them?

Thanking You
Sincerely

Mike




-----Original Message-----
From: Richard Ristow [mailto:wrristow@...]
Sent: Monday, June 23, 2008 6:29 PM
To: Roberts, Michael; SPSSX-L@...
Subject: Re: Creating a counter of ids within similar records

At 05:20 PM 6/23/2008, Roberts, Michael wrote:

>I would like to create a counter of the number of occurrences within
>records that have two variables - id and date. When the id changes,
>or the date changes, I want to begin counting again.

There's a number of ways. This one (code not tested) is fairly easy.
It requires that the file be sorted by ID and Date; and it leaves
variable NewGroup in the file, which you may not want.

>There are solutions but they all to require a numeric value for the id.

This will work with any datatypes for ID and Date.

ADD FILES
   /FILE=*
   /BY ID Date
   /FIRST = NewGroup.

NUMERIC SeqInGrp (F4).
DO IF   NewGroup.
.  COMPUTE SeqInGrp = 1.
ELSE.
.  COMPUTE SeqInGrp = LAG(SeqInGrp) + 1.
END IF.

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Richard Ristow]

At 06:40 PM 6/23/2008, Roberts, Michael wrote:

>Thank you for the solution.

You're welcome!

>I [managed] this by creating a counter, setting its default value to
>1, then testing both values (date and ID) lagged by 1, and
>incrementing the counter by 1 [when the lagged values both matched
>the current values?]; seems to work.

Yes; that's fundamentally very similar logic. Did you specify LEAVE
for the counter, or use LAG (as I did) to increment the counter from
the previous case?

>Your solution, however looks very interesting, and I am going to
>check it out in the morning.

It's the easiest to program, since it doesn't need separate checking
of the two key variables, and it doesn't need special-case logic for
the very first record in the file (as LAG logic sometimes does).
Apart from leaving the extra variable in the file, it has a lot going for it

>I did see some code you posted a while back which includes 'Before'
>and 'After' commands. My search for information on these two turned
>up very little. How can I find out more about them?

I don't recall that; could you refer me to the posting you're writing
about? Neither 'Before' nor 'After' is an SPSS command; I'd need to
see the context in which I mentioned them.

-Best wishes, and onward,
  Richard

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Valerie Guille-2]

Hey all,

I am hoping that someone will be able to help me here.

1st a brief info about my dataset: I have 2 groups of independent
subjects (1st group receive Placebo and 2nd group receive a treatment)
in which 3  response from 3 different types of stimuli are measured.

I believed I need to run a repeated measures ANOVA with the 3 response
the RM. I run a the following analysis in SPSS:

GLM
  PPI4 PPI8 PPI16 BY TTT_Code
  /WSFACTOR = PPI 3 Polynomial
  /CONTRAST (TTT_Code)=Simple(1)
  /METHOD = SSTYPE(3)
  /POSTHOC = TTT_Code ( LSD BONFERRONI )
  /EMMEANS = TABLES(TTT_Code) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(PPI) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(TTT_Code*PPI)
  /PRINT = DESCRIPTIVE ETASQ OPOWER HOMOGENEITY
  /CRITERIA = ALPHA(.05)
  /WSDESIGN = PPI
  /DESIGN = TTT_Code .

SPSS does not run the posthoc analysis on these data because I only have
2 groups. I would like to be able to compare my groups within each of my
responses (i.e. PPI4, PPI8 and PPI16).
I would greatly appreciate, if someone could help me with the commande I
need to add in the above syntax in order to do this.


Thanks in advance.
Valerie

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

[Zdaniuk, Bozena-2]

Hello, you need to change your syntax a bit. Change your
  /EMMEANS = TABLES(TTT_Code) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(PPI) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(TTT_Code*PPI)

to
  /EMMEANS = TABLES(TTT_Code) COMPARE (TTT_Code) ADJ(BONFERRONI)
  /EMMEANS = TABLES(PPI) COMPARE (PPI)  ADJ(BONFERRONI)
  /EMMEANS = TABLES(TTT_Code*PPI) COMPARE (TTT_Code)
  /EMMEANS = TABLES(TTT_Code*PPI) COMPARE (PPI)

This will give you all the comparisons you want. And I would not bother with ADJ subcommand - it does not work in SPSS (at least I never saw it making any adjustments).
Also, you may want to think whether you really have a RM design or just a multivariate ANOVA. Repeated measures analysis usually means that you measured people on the same thing multiple times. But arent' you saing that you have three measures of responses to three DIFFERENT stimuli?
Just a thought.
Bozena Zdaniuk



________________________________________
From: SPSSX(r) Discussion [SPSSX-L@...] On Behalf Of Valerie Guille [valerie.guille@...]
Sent: Sunday, June 29, 2008 10:07 PM
To: SPSSX-L@...
Subject: 2-WAY Repeated measures ANOVA-POST HOC ANALYSIS IN SPSS

Hey all,

I am hoping that someone will be able to help me here.

1st a brief info about my dataset: I have 2 groups of independent
subjects (1st group receive Placebo and 2nd group receive a treatment)
in which 3  response from 3 different types of stimuli are measured.

I believed I need to run a repeated measures ANOVA with the 3 response
the RM. I run a the following analysis in SPSS:

GLM
  PPI4 PPI8 PPI16 BY TTT_Code
  /WSFACTOR = PPI 3 Polynomial
  /CONTRAST (TTT_Code)=Simple(1)
  /METHOD = SSTYPE(3)
  /POSTHOC = TTT_Code ( LSD BONFERRONI )
  /EMMEANS = TABLES(TTT_Code) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(PPI) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(TTT_Code*PPI)
  /PRINT = DESCRIPTIVE ETASQ OPOWER HOMOGENEITY
  /CRITERIA = ALPHA(.05)
  /WSDESIGN = PPI
  /DESIGN = TTT_Code .

SPSS does not run the posthoc analysis on these data because I only have
2 groups. I would like to be able to compare my groups within each of my
responses (i.e. PPI4, PPI8 and PPI16).
I would greatly appreciate, if someone could help me with the commande I
need to add in the above syntax in order to do this.


Thanks in advance.
Valerie

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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
LISTSERV@... (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