GLM-repeated measures or Mixed Models

[Byrd Sellers, Johnita]

Hello everyone,

 

I am currently working on a study with a doctor in orthopedics.  They
have used a new method to test for compartment syndrome.  Here is some
quick background about the study.  Usually, when people are suspected of
having compartment syndrome, the only method is an invasive procedure.
The current method is a noninvasive way for physicians to evaluate a
suspected compartment syndrome.  This new method will be known as the
NIRS measurements.

 

There was a group of 20 patients evaluated.  At different levels, each
patient had their perfusion pressure (diastolic cuff pressure) and their
NIRS value read.  Also, each patient's BMI was measured.  There are
between 9-13 different pressure levels where a reading was taken for the
perfusion pressure and the NIRS value.  I would like to compare the two
values (perfusion pressure and NIRS value) to one another.  I would like
to find out the NIRS value when the perfusion pressure is 0 and things
like that.  I would also like to see if BMI will have an effect on these
measurements.

The problem is the pressure levels are somewhat continuous.  In my first
analysis, I was trying to break the problem up as follows:

I used the GLM-repeated measures method.

Within-Subjects factors were the 13 pressure levels. And for the
measure, I had assigned a value for the perfusion pressure readings and
a value for the NIRS readings.

Between-Subject factor was the BMI (since I wanted to know what type of
effect this may have on the results.)

 

I guess my question is this:

If I use repeated measures, will I be able to find the conclusions that
I am looking for.  Also, after a certain pressure level, there are some
patients who have no readings.  Really, once you pass the 9th pressure
level, there are a lot of patients with no readings.  When I evaluate
the problem using all methods, my results have a lot of missing values.
But when I stop at about the 9th level, there is a small amount of
missing values.

 

I have posed a similar question like this a couple of weeks ago (and
received great insight), however, I didn't take into account that the
pressure levels should be considered continuous.  I was trying to take
the average at each pressure level for all of the patients and then
perform a regression analysis on the results.  But, it didn't seem like
I would truly be answering my question.

 

Any help that anyone can give on the subject would be greatly
appreciated.  Thanks in advance.

 

Johnita

[Hashmi, Syed S]

Hello everyone,

I have some data where the cases have numerical codes assigned to them
in a variable called "code1" the format #.## and range from 1.10 to 9.79
(though it does not include every single number in that range).  Each of
the three characters in that code stands for some common characteristic,
eg. 1.10 is similar to 1.20 or 1.34 (all are in 1.## format).
Similarly, 1.10 shares some characteristics with 2.13 or 5.18 (all are
in #.1# format).

I wanted to make some computations based on the values of these codes.
It's easy when I wanted to do the computations for all cases with
code1=2.** or code=3.2* since it involved a simple DO IF statement:
        DO IF (code1 GE 1.00 & code1 LE 1.99).

My issue arises when I want to do computations on cases where the codes
are similar at the second character, eg. all cases with a #.2# code.
Does SPSS (I'm on v.15) have any wild card characters that I could use
for doing this?? If not, can I use the INDEX subcommand to specify the
character instead?

I would appreciate any help that anyone might be able to provide. Thanks
a heap in advance.

- Shahrukh Hashmi

[Melissa Ives]

Look into a calculation using a combination of MOD and TRUNC

MOD. MOD(numexpr,modulus). Numeric. Returns the remainder when numexpr
is divided by modulus. Both arguments must be numeric, and modulus must
not be 0.

TRUNC. TRUNC(numexpr). Numeric. Returns the value of numexpr truncated
to an integer (toward 0).

Something like this (totally untested)--where you want to divide code1
by 1 to keep only the portion following the decimal, then multiply by 10
and truncate the result so you get just the 1 in your 2.13/5.18 example
below.

Do if (trunc((mod(code1,1)/10)).

Melissa

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Hashmi, Syed S
Sent: Thursday, July 12, 2007 2:53 PM
To: [hidden email]
Subject: [SPSSX-L] Wild cards

Hello everyone,

I have some data where the cases have numerical codes assigned to them
in a variable called "code1" the format #.## and range from 1.10 to 9.79
(though it does not include every single number in that range).  Each of
the three characters in that code stands for some common characteristic,
eg. 1.10 is similar to 1.20 or 1.34 (all are in 1.## format).
Similarly, 1.10 shares some characteristics with 2.13 or 5.18 (all are
in #.1# format).

I wanted to make some computations based on the values of these codes.
It's easy when I wanted to do the computations for all cases with
code1=2.** or code=3.2* since it involved a simple DO IF statement:
        DO IF (code1 GE 1.00 & code1 LE 1.99).

My issue arises when I want to do computations on cases where the codes
are similar at the second character, eg. all cases with a #.2# code.
Does SPSS (I'm on v.15) have any wild card characters that I could use
for doing this?? If not, can I use the INDEX subcommand to specify the
character instead?

I would appreciate any help that anyone might be able to provide. Thanks
a heap in advance.

- Shahrukh Hashmi


PRIVILEGED AND CONFIDENTIAL INFORMATION
This transmittal and any attachments may contain PRIVILEGED AND
CONFIDENTIAL information and is intended only for the use of the
addressee. If you are not the designated recipient, or an employee
or agent authorized to deliver such transmittals to the designated
recipient, you are hereby notified that any dissemination,
copying or publication of this transmittal is strictly prohibited. If
you have received this transmittal in error, please notify us
immediately by replying to the sender and delete this copy from your
system. You may also call us at [phone] for assistance.

[Hashmi, Syed S]

Thanks Melissa,

I got the following error when I tried running your syntax:
An undefined variable name, or a scratch or system variable was specified in a variable list which accepts only standard variables.  Check spelling and verify the existence of this variable.¬This command not executed.¬

However, I broke it up into two separate commands and can use the final variable in the DO IF statements:

   COMPUTE MODstep = MOD(code1, 1).
   COMPUTE TRUNCstep = TRUNC(MODstep*10).
   DO IF TRUNCstep = 1.

It works great! Thanks again.

Shahrukh

> -----Original Message-----
> From: Melissa Ives [mailto:[hidden email]]
> Sent: Thursday, July 12, 2007 3:02 PM
> To: Hashmi, Syed S; [hidden email]
> Subject: RE: [SPSSX-L] Wild cards
>
> Look into a calculation using a combination of MOD and TRUNC
>
> MOD. MOD(numexpr,modulus). Numeric. Returns the remainder when numexpr
> is divided by modulus. Both arguments must be numeric, and modulus must
> not be 0.
>
> TRUNC. TRUNC(numexpr). Numeric. Returns the value of numexpr truncated
> to an integer (toward 0).
>
> Something like this (totally untested)--where you want to divide code1
> by 1 to keep only the portion following the decimal, then multiply by 10
> and truncate the result so you get just the 1 in your 2.13/5.18 example
> below.
>
> Do if (trunc((mod(code1,1)/10)).
>
> Melissa
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Hashmi, Syed S
> Sent: Thursday, July 12, 2007 2:53 PM
> To: [hidden email]
> Subject: [SPSSX-L] Wild cards
>
> Hello everyone,
>
> I have some data where the cases have numerical codes assigned to them
> in a variable called "code1" the format #.## and range from 1.10 to 9.79
> (though it does not include every single number in that range).  Each of
> the three characters in that code stands for some common characteristic,
> eg. 1.10 is similar to 1.20 or 1.34 (all are in 1.## format).
> Similarly, 1.10 shares some characteristics with 2.13 or 5.18 (all are
> in #.1# format).
>
> I wanted to make some computations based on the values of these codes.
> It's easy when I wanted to do the computations for all cases with
> code1=2.** or code=3.2* since it involved a simple DO IF statement:
>         DO IF (code1 GE 1.00 & code1 LE 1.99).
>
> My issue arises when I want to do computations on cases where the codes
> are similar at the second character, eg. all cases with a #.2# code.
> Does SPSS (I'm on v.15) have any wild card characters that I could use
> for doing this?? If not, can I use the INDEX subcommand to specify the
> character instead?
>
> I would appreciate any help that anyone might be able to provide. Thanks
> a heap in advance.
>
> - Shahrukh Hashmi
>
>
> PRIVILEGED AND CONFIDENTIAL INFORMATION
> This transmittal and any attachments may contain PRIVILEGED AND
> CONFIDENTIAL information and is intended only for the use of the
> addressee. If you are not the designated recipient, or an employee
> or agent authorized to deliver such transmittals to the designated
> recipient, you are hereby notified that any dissemination,
> copying or publication of this transmittal is strictly prohibited. If
> you have received this transmittal in error, please notify us
> immediately by replying to the sender and delete this copy from your
> system. You may also call us at [phone] for assistance.

[Peck, Jon]

If you can use programmability, you will find very powerful and general regular expressions available including, of course simple wildcards.  You can read about this in the Data Management Book downloadable as PDF from SPSS (see SPSS Developer Central for link, www.spss.com/devcentral), and make use of the trans and extendedTransforms modules on Developer Central.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Hashmi, Syed S
Sent: Thursday, July 12, 2007 2:53 PM
To: [hidden email]
Subject: [SPSSX-L] Wild cards

Hello everyone,

I have some data where the cases have numerical codes assigned to them
in a variable called "code1" the format #.## and range from 1.10 to 9.79
(though it does not include every single number in that range).  Each of
the three characters in that code stands for some common characteristic,
eg. 1.10 is similar to 1.20 or 1.34 (all are in 1.## format).
Similarly, 1.10 shares some characteristics with 2.13 or 5.18 (all are
in #.1# format).

I wanted to make some computations based on the values of these codes.
It's easy when I wanted to do the computations for all cases with
code1=2.** or code=3.2* since it involved a simple DO IF statement:
        DO IF (code1 GE 1.00 & code1 LE 1.99).

My issue arises when I want to do computations on cases where the codes
are similar at the second character, eg. all cases with a #.2# code.
Does SPSS (I'm on v.15) have any wild card characters that I could use
for doing this?? If not, can I use the INDEX subcommand to specify the
character instead?

I would appreciate any help that anyone might be able to provide. Thanks
a heap in advance.

- Shahrukh Hashmi

[Hashmi, Syed S]

Thanks Jon... and Melissa for the earlier reply.  I've used the TRANS
command and it worked fine.

Shahrukh


> -----Original Message-----
> From: Peck, Jon [mailto:[hidden email]]
> Sent: Thursday, July 12, 2007 5:03 PM
> To: Hashmi, Syed S; [hidden email]
> Subject: RE: [SPSSX-L] Wild cards
>
> If you can use programmability, you will find very powerful and
general
> regular expressions available including, of course simple wildcards.
You
> can read about this in the Data Management Book downloadable as PDF
from
> SPSS (see SPSS Developer Central for link, www.spss.com/devcentral),
and
> make use of the trans and extendedTransforms modules on Developer
Central.
>
> HTH,
> Jon Peck
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf
Of
> Hashmi, Syed S
> Sent: Thursday, July 12, 2007 2:53 PM
> To: [hidden email]
> Subject: [SPSSX-L] Wild cards
>
> Hello everyone,
>
> I have some data where the cases have numerical codes assigned to them
> in a variable called "code1" the format #.## and range from 1.10 to
9.79
> (though it does not include every single number in that range).  Each
of
> the three characters in that code stands for some common
characteristic,

> eg. 1.10 is similar to 1.20 or 1.34 (all are in 1.## format).
> Similarly, 1.10 shares some characteristics with 2.13 or 5.18 (all are
> in #.1# format).
>
> I wanted to make some computations based on the values of these codes.
> It's easy when I wanted to do the computations for all cases with
> code1=2.** or code=3.2* since it involved a simple DO IF statement:
>         DO IF (code1 GE 1.00 & code1 LE 1.99).
>
> My issue arises when I want to do computations on cases where the

codes
> are similar at the second character, eg. all cases with a #.2# code.
> Does SPSS (I'm on v.15) have any wild card characters that I could use
> for doing this?? If not, can I use the INDEX subcommand to specify the
> character instead?
>
> I would appreciate any help that anyone might be able to provide.
Thanks
> a heap in advance.
>
> - Shahrukh Hashmi

[Maguin, Eugene]

I'd like to check my understanding of an analysis problem. The data are
generated from a type of dating behavior study. For each of the
participants' dating partners, the participants make a rating for each of
several behaviors, if they engaged in that behavior. In addition
participants classify the relationship with each of their partners. Some
participants have one partner; others may have several.

The aspects of this problem that are confusing to me are the relationship
classification and the partners. Relationship type first. Partners are
classified in one and only one relationship type. Relationship type seems
like a between sujects variable but I don't think it can be considered one
because if a participant has multiple relationships, then the same
participant appears in multiple places.

Partners is the second problem. When there are multiple partners, there may
be multiple of relationships of the same type or one each of several
different relationship types.

Sample size will impose constraints but I'd first like to get an
understanding of how to thing of this problem. If someone can help, I'd
greatly appreciate it.

Thanks, Gene Maguin

[Jerabek Jindrich]

Hello,

only an idea.
There are some problems with classifying respondents, one respondent may have more than one partner, more than one relationship, if I understand you right.

Maybe you could look at relationships instead looking at respondents. I mean the observed unit, the row in the datafile, is a relationship, not a respondent. Then the relationship type and behaviour would be clearly defined and every relationship can be flaged as "one to one partner", "one to more", "many to many".

best regards
Jindra

> ------------ Původní zpráva ------------
> Od: Gene Maguin <[hidden email]>
> Předmět: Analysis problem
> Datum: 18.7.2007 18:03:17
> ----------------------------------------
> I'd like to check my understanding of an analysis problem. The data are
> generated from a type of dating behavior study. For each of the
> participants' dating partners, the participants make a rating for each of
> several behaviors, if they engaged in that behavior. In addition
> participants classify the relationship with each of their partners. Some
> participants have one partner; others may have several.
>
> The aspects of this problem that are confusing to me are the relationship
> classification and the partners. Relationship type first. Partners are
> classified in one and only one relationship type. Relationship type seems
> like a between sujects variable but I don't think it can be considered one
> because if a participant has multiple relationships, then the same
> participant appears in multiple places.
>
> Partners is the second problem. When there are multiple partners, there may
> be multiple of relationships of the same type or one each of several
> different relationship types.
>
> Sample size will impose constraints but I'd first like to get an
> understanding of how to thing of this problem. If someone can help, I'd
> greatly appreciate it.
>
> Thanks, Gene Maguin
>
>
>