SPSSX Discussion

Standardization question

Classic

List

Threaded

5 messages Options

KEVIN MANNING

Standardization question

Hi all,

I have 3 groups who perform on a wide variety of cognitive tests. I want to standardize the test variables in the hope of creating cognitive indices (I do not have nearly enough cases in my sample for a factor analysis) for comparison.

My question relates to how SPSS creates z scores from the descriptives command - if I create z scores based on the whole dataset, I think I will lose between-group performance differences. But is there a way to split the file into three groups, and then run descriptives to create the z scores??

Any help or other suggestions would be greatly appreciated! Thank you.
Kevin

ViAnn Beadle

Re: Standardization question

Did you try the split file command? This appears to produce zscores normed
to the split.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
KEVIN MANNING
Sent: Tuesday, June 05, 2007 8:52 AM
To: [hidden email]
Subject: Standardization question

Hi all,

I have 3 groups who perform on a wide variety of cognitive tests. I
want to standardize the test variables in the hope of creating cognitive
indices (I do not have nearly enough cases in my sample for a factor
analysis) for comparison.

My question relates to how SPSS creates z scores from the descriptives
command - if I create z scores based on the whole dataset, I think I will
lose between-group performance differences. But is there a way to split the
file into three groups, and then run descriptives to create the z scores??

Any help or other suggestions would be greatly appreciated! Thank you.
Kevin

Richard Ristow

Re: Standardization question

In reply to this post by KEVIN MANNING

At 10:51 AM 6/5/2007, KEVIN MANNING wrote:

> I have 3 groups who perform on a wide variety of cognitive
> tests. I want to standardize the test variables in the hope of
> creating cognitive indices for comparison.
>
> My question relates to how SPSS creates z scores from the
> descriptives command - if I create z scores based on the whole
> dataset, I will lose between-group performance differences. But is
> there a way to split the file into three groups, and then
> run descriptives to create the z scores??

ViAnn Beadle's suggestion to use SPLIT FILES should give you what
you're asking for; but I doubt it will give you what you need. If you
do it, your standardized variables will have mean 0 and S.D. 1,
exactly, within each group. If you're interested in difference of mean
score between groups, those won't do you much good. Standardizing over
the whole file sounds more like what you want.

Hector Maletta

Re: Standardization question

Richard,
I was thinking exactly what you say, though I opted for not going
into that in my response. If the z-scores are computed on the whole group,
the differences between the two groups will be easily gauged since each will
have a different mean and a different distribution.

Then I started thinking whether there could be some other untold
side of the question. Sometimes people talk about computing z-scores not in
relation to the observed distribution of the cases under analysis, but in
relation to a reference or normative distribution. For instance, observed
child weights by age can be measured in a z-score scale taken from a
reference growth curve for healthy children, and so you can learn whether a
particular child is above or below the normal weight for his age (the
difference being measured in standard deviations, and the mean and SD being
the mean and SD for each age). But even in this case, there is no point in
computing the z-scores separately or jointly since they would give the same
results.
Therefore, in conclusion, for "internal standardization" of a
variable called X, Kevin should use DESCRIPTIVES X/SAVE, thus creating a
variable called (by default) ZX that equals the z-score version of X for the
whole dataset, with overall zero mean and unit SD, and then analyze the new
variable by subgroup (mean, SD, distribution, median --the works).

For an external reference or normative z-score (such as normal
weight for age) Kevin would need an external conversion table giving the
normative z-score for any given weight-age combination. This table should be
a second SPSS file to be matched by weight and age with the original data
set.

Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Richard Ristow
Sent: 07 June 2007 15:48
To: [hidden email]
Subject: Re: Standardization question

At 10:51 AM 6/5/2007, KEVIN MANNING wrote:

> I have 3 groups who perform on a wide variety of cognitive
> tests. I want to standardize the test variables in the hope of
> creating cognitive indices for comparison.
>
> My question relates to how SPSS creates z scores from the
> descriptives command - if I create z scores based on the whole
> dataset, I will lose between-group performance differences. But
is
> there a way to split the file into three groups, and then
> run descriptives to create the z scores??

ViAnn Beadle's suggestion to use SPLIT FILES should give you what
you're asking for; but I doubt it will give you what you need. If
you
do it, your standardized variables will have mean 0 and S.D. 1,
exactly, within each group. If you're interested in difference of
mean
score between groups, those won't do you much good. Standardizing
over
the whole file sounds more like what you want.

statisticsdoc

Re: Standardization question

Hector,

You make a well stated distinction between internal and external
standardization. I think there may be another way to program SPSS to carry
out the external standardization. Using the example of standardizing weight
by age:

1.) Create a lookup table with a separate record for each age level of
interest. Each record contains the mean weight and the standard deviation
of weight at that age level (MWEIGHT and SDWEIGHT).

2.) Do a one to many match for age, so each case in the data is matched with
the relevant mean weight and standard deviation for each age level.

3.) Compute the standardized weight as (WEIGHT - MWEIGHT) / SDWEIGHT .
Individual weights will be standardized around the mean for a specific age
level (using the standard deviation for that age group).

HTH,

Steve

For personalized and professional consultation in statistics and research
design, visit
www.statisticsdoc.com

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]]On Behalf Of
Hector Maletta
Sent: Thursday, June 07, 2007 5:10 PM
To: [hidden email]
Subject: Re: Standardization question

Richard,
I was thinking exactly what you say, though I opted for not going
into that in my response. If the z-scores are computed on the whole group,
the differences between the two groups will be easily gauged since each will
have a different mean and a different distribution.

Then I started thinking whether there could be some other untold
side of the question. Sometimes people talk about computing z-scores not in
relation to the observed distribution of the cases under analysis, but in
relation to a reference or normative distribution. For instance, observed
child weights by age can be measured in a z-score scale taken from a
reference growth curve for healthy children, and so you can learn whether a
particular child is above or below the normal weight for his age (the
difference being measured in standard deviations, and the mean and SD being
the mean and SD for each age). But even in this case, there is no point in
computing the z-scores separately or jointly since they would give the same
results.
Therefore, in conclusion, for "internal standardization" of a
variable called X, Kevin should use DESCRIPTIVES X/SAVE, thus creating a
variable called (by default) ZX that equals the z-score version of X for the
whole dataset, with overall zero mean and unit SD, and then analyze the new
variable by subgroup (mean, SD, distribution, median --the works).

For an external reference or normative z-score (such as normal
weight for age) Kevin would need an external conversion table giving the
normative z-score for any given weight-age combination. This table should be
a second SPSS file to be matched by weight and age with the original data
set.

Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Richard Ristow
Sent: 07 June 2007 15:48
To: [hidden email]
Subject: Re: Standardization question

At 10:51 AM 6/5/2007, KEVIN MANNING wrote:

> I have 3 groups who perform on a wide variety of cognitive
> tests. I want to standardize the test variables in the hope of
> creating cognitive indices for comparison.
>
> My question relates to how SPSS creates z scores from the
> descriptives command - if I create z scores based on the whole
> dataset, I will lose between-group performance differences. But
is
> there a way to split the file into three groups, and then
> run descriptives to create the z scores??

ViAnn Beadle's suggestion to use SPLIT FILES should give you what
you're asking for; but I doubt it will give you what you need. If
you
do it, your standardized variables will have mean 0 and S.D. 1,
exactly, within each group. If you're interested in difference of
mean
score between groups, those won't do you much good. Standardizing
over
the whole file sounds more like what you want.