This is a cross post from the spss forum, but did not capture the
public's imagination!

I am doing some exploratory analysis to see if a task we are using
in a within subjects design is affected by age.

Participants have completed the task 5 times in close succession
giving T1, T2, T3, T4, T5.

Using repeated measures GLM in SPSS with no covariate the main
effect is significant.

if age is added into the analysis as a covariate this difference is
no longer significant which suggests that age plays some role.

Is there any subsequent analysis I can do to further explore this
role, or some option in SPSS which will show this?

Matt wrote:
This is a cross post from the spss forum, but did not capture the
public's imagination!

I am doing some exploratory analysis to see if a task we are using in
a within subjects design is affected by age.

Participants have completed the task 5 times in close succession
giving T1, T2, T3, T4, T5.

Using repeated measures GLM in SPSS with no covariate the main effect
is significant.

if age is added into the analysis as a covariate this difference is no
longer significant which suggests that age plays some role.

Is there any subsequent analysis I can do to further explore this
role, or some option in SPSS which will show this?

Thanks

IIRC, the repeated measures GLM in SPSS automatically includes the
covariate x RM factor interaction in the model you described, and there
is no way to exclude it.  Right?  Is that interaction significant?  Is
the main effect of age significant?

--
Bruce Weaver
bwea...@lakeheadu.cawww.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."

Matt <matt.lew...@gmail.com> wrote in news:516d811d-3181-4948-ac54-
bad19acd9...@c23g2000hsa.googlegroups.com:



It is a fairly restricted age range (11-18 and only 1 person each is
11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time is
the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to use
age as a between subjects factor in the repeated measures analysis
rather than a covariate?

What is important is that you accurately describe the relationships in
your data. You say that the "form of [trend to improvement over time]
is different". I think you would get more sensible replies if you were
both more expansive and more specific about what you are seeing.

--
David Winsemius

This is a cross post from the spss forum, but did not capture the
public's imagination!

I am doing some exploratory analysis to see if a task we are using in
a within subjects design is affected by age.

Participants have completed the task 5 times in close succession
giving T1, T2, T3, T4, T5.

Using repeated measures GLM in SPSS with no covariate the main effect
is significant.

if age is added into the analysis as a covariate this difference is no
longer significant which suggests that age plays some role.

Is there any subsequent analysis I can do to further explore this
role, or some option in SPSS which will show this?

Thanks

On Jan 30, 1:27 pm, Bruce Weaver <bwea...@lakeheadu.ca> wrote:
Matt wrote:
This is a cross post from the spss forum, but did not capture the
public's imagination!

I am doing some exploratory analysis to see if a task we are using in
a within subjects design is affected by age.

Participants have completed the task 5 times in close succession
giving T1, T2, T3, T4, T5.

Using repeated measures GLM in SPSS with no covariate the main effect
is significant.

if age is added into the analysis as a covariate this difference is no
longer significant which suggests that age plays some role.

Is there any subsequent analysis I can do to further explore this
role, or some option in SPSS which will show this?

Thanks

IIRC, the repeated measures GLM in SPSS automatically includes the
covariate x RM factor interaction in the model you described, and there
is no way to exclude it.  Right?  Is that interaction significant?  Is
the main effect of age significant?

--
Bruce Weaver
bwea...@lakeheadu.cawww.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."


output gives

rounds completion time

Then

rounds completion time by age

I have different variables that I am looking at the "time" effect for

For some the interaction with age is significant and in others it is
not significant

It is a fairly restricted age range (11-18 and only 1 person each is
11 and 1. Total n is 85

So a plot for age shows that the trend to improvement over time is
the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to use
age as a between subjects factor in the repeated measures analysis
rather than a covariate?

On Jan 30, 4:02 pm, David Winsemius wrote:

Matt <matt.lew...@gmail.com> wrote :


It is a fairly restricted age range (11-18 and only 1 person each
is 11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time
is the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to
use age as a between subjects factor in the repeated measures
analysis rather than a covariate?

What is important is that you accurately describe the relationships
in your data. You say that the "form of [trend to improvement over
time] is different". I think you would get more sensible replies if
you were both more expansive and more specific about what you are
seeing.

-- David Winsemius


the n refers to the number of subjects that completed the assessment
(which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n = 1,
and there are no 17 year olds) show an improvement in performance on
this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved performance
for each age. Slopes would differ however.

On Jan 30, 4:02 pm, David Winsemius <doe_s...@comcast.n0T> wrote:
Matt <matt.lew...@gmail.com> wrote in news:516d811d-3181-4948-ac54-
bad19acd9...@c23g2000hsa.googlegroups.com:

It is a fairly restricted age range (11-18 and only 1 person each is
11 and 1. Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time is
the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to use
age as a between subjects factor in the repeated measures analysis
rather than a covariate?

What is important is that you accurately describe the relationships in
your data. You say that the "form of [trend to improvement over time]
is different". I think you would get more sensible replies if you were
both more expansive and more specific about what you are seeing.

--
David Winsemius

the n refers to the number of subjects that completed the assessment
(which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n = 1,
and there are no 17 year olds) show an improvement in performance on
this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved performance
for each age. Slopes would differ however.

Hope this helps

AgeY Mean Std. Deviation
T1 12 1.84208 .174069
13 1.83516 .163855
14 1.92776 .221794
15 1.96213 .224959
16 1.85757 .141574
Total 1.88160 .192976
T2 12 1.73961 .174920
13 1.66242 .153462
14 1.73162 .166005
15 1.66133 .133866
16 1.59036 .065950
Total 1.69248 .158289
T3 12 1.66504 .198612
13 1.57135 .144364
14 1.61489 .173038
15 1.48859 .155504
16 1.52754 .151598
Total 1.58929 .173259
T4 12 1.62130 .117304
13 1.51471 .150148
14 1.52314 .141670
15 1.46118 .105629
16 1.43424 .082742
Total 1.52642 .140363
T5 12 1.60579 .160730
13 1.46109 .110084
14 1.53830 .170144
15 1.42927 .156137
16 1.38047 .115301
Total 1.50336 .160644

On Jan 29, 9:20 pm, Matt <matt.lew...@gmail.com> wrote:



On Jan 30, 4:02 pm, David Winsemius <doe_s...@comcast.n0T> wrote:
Matt <matt.lew...@gmail.com> wrote in news:516d811d-3181-4948-ac54-
bad19acd9...@c23g2000hsa.googlegroups.com:

It is a fairly restricted age range (11-18 and only 1 person each is
11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time is
the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to use
age as a between subjects factor in the repeated measures analysis
rather than a covariate?

What is important is that you accurately describe the relationships in
your data. You say that the "form of [trend to improvement over time]
is different". I think you would get more sensible replies if you were
both more expansive and more specific about what you are seeing.

--
David Winsemius

the n refers to the number of subjects that completed the assessment
(which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n = 1,
and there are no 17 year olds) show an improvement in performance on
this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved performance
for each age.  Slopes would differ however.

Hope this helps

        AgeY    Mean    Std. Deviation
T1      12      1.84208 .174069
        13      1.83516 .163855
        14      1.92776 .221794
        15      1.96213 .224959
        16      1.85757 .141574
        Total   1.88160 .192976
T2      12      1.73961 .174920
        13      1.66242 .153462
        14      1.73162 .166005
        15      1.66133 .133866
        16      1.59036 .065950
        Total   1.69248 .158289
T3      12      1.66504 .198612
        13      1.57135 .144364
        14      1.61489 .173038
        15      1.48859 .155504
        16      1.52754 .151598
        Total   1.58929 .173259
T4      12      1.62130 .117304
        13      1.51471 .150148
        14      1.52314 .141670
        15      1.46118 .105629
        16      1.43424 .082742
        Total   1.52642 .140363
T5      12      1.60579 .160730
        13      1.46109 .110084
        14      1.53830 .170144
        15      1.42927 .156137
        16      1.38047 .115301
        Total   1.50336 .160644

The means and s.d.s correlate .65 . What is your d.v.?
Should you transform it, or use a heteroscedastic model?

Matt <matt.lew...@gmail.com> wrote:
On Jan 30, 4:02 pm, David Winsemius  wrote:
Matt <matt.lew...@gmail.com> wrote :

It is a fairly restricted age range (11-18 and only 1 person each
is 11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time
is the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to
use age as a between subjects factor in the repeated measures
analysis rather than a covariate?

What is important is that you accurately describe the relationships
in your data. You say that the "form of [trend to improvement over
time] is different". I think you would get more sensible replies if
you were both more expansive and more specific about what you are
seeing.

-- David Winsemius

the n refers to the number of subjects that completed the assessment
(which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n = 1,
and there are no 17 year olds) show an improvement in performance on
this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved performance
for each age.  Slopes would differ however.

It makes clear what you were saying about the age*repeat-test effect.
Older individuals learn to do the test faster over 5 repetitions than
do younger individuals. Weren't you also dealing with another covariate
whose effect diminished when age was entered?

--
David Winsemius

On Feb 1, 12:28 pm, Ray Koopman <koop...@sfu.ca> wrote:
On Jan 29, 9:20 pm, Matt <matt.lew...@gmail.com> wrote:
On Jan 30, 4:02 pm, David Winsemius <doe_s...@comcast.n0T> wrote:
Matt <matt.lew...@gmail.com> wrote in news:516d811d-3181-4948-ac54-
bad19acd9...@c23g2000hsa.googlegroups.com:

It is a fairly restricted age range (11-18 and only 1 person each is
11 and 1. Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over time is
the same, but the form of this is different.

With the age range so curtailed like this is it more sensible to use
age as a between subjects factor in the repeated measures analysis
rather than a covariate?

What is important is that you accurately describe the relationships in
your data. You say that the "form of [trend to improvement over time]
is different". I think you would get more sensible replies if you were
both more expansive and more specific about what you are seeing.

--
David Winsemius

the n refers to the number of subjects that completed the assessment
(which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n = 1,
and there are no 17 year olds) show an improvement in performance on
this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved performance
for each age. Slopes would differ however.

Hope this helps

AgeY Mean Std. Deviation
T1 12 1.84208 .174069
13 1.83516 .163855
14 1.92776 .221794
15 1.96213 .224959
16 1.85757 .141574
Total 1.88160 .192976
T2 12 1.73961 .174920
13 1.66242 .153462
14 1.73162 .166005
15 1.66133 .133866
16 1.59036 .065950
Total 1.69248 .158289
T3 12 1.66504 .198612
13 1.57135 .144364
14 1.61489 .173038
15 1.48859 .155504
16 1.52754 .151598
Total 1.58929 .173259
T4 12 1.62130 .117304
13 1.51471 .150148
14 1.52314 .141670
15 1.46118 .105629
16 1.43424 .082742
Total 1.52642 .140363
T5 12 1.60579 .160730
13 1.46109 .110084
14 1.53830 .170144
15 1.42927 .156137
16 1.38047 .115301
Total 1.50336 .160644

The means and s.d.s correlate .65 . What is your d.v.?
Should you transform it, or use a heteroscedastic model?

Ray

I don't fully understand the idea of a heteroscedastic model.

Is this along the same lines as assessing for sphericity prior to
interpreting the analysis?

On Feb 1, 3:50 am, David Winsemius <doe_s...@comcast.n0T> wrote:
Matt <matt.lew...@gmail.com> wrote:
On Jan 30, 4:02 pm, David Winsemius  wrote:
Matt <matt.lew...@gmail.com> wrote :

It is a fairly restricted age range (11-18 and only 1 person
each is 11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over
time is the same, but the form of this is different.

With the age range so curtailed like this is it more sensible
to use age as a between subjects factor in the repeated
measures analysis rather than a covariate?

What is important is that you accurately describe the
relationships in your data. You say that the "form of [trend to
improvement over time] is different". I think you would get more
sensible replies if you were both more expansive and more
specific about what you are seeing.

-- David Winsemius

the n refers to the number of subjects that completed the
assessment (which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n =
1, and there are no 17 year olds) show an improvement in
performance on this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved
performance for each age.  Slopes would differ however.

It makes clear what you were saying about the age*repeat-test
effect. Older individuals learn to do the test faster over 5
repetitions than do younger individuals. Weren't you also dealing
with another covariate whose effect diminished when age was
entered?

--
David Winsemius

David

there has only been age used as a covariate. I'm sorry if i gave
the impression otherwise.

My query was more related to the fact that inclusion of age as a
covariate makes the main effect of time no longer significant.

I was curious to know if there is a way to examine this effect in
more depth, rather than provide a subjective description of the form
of the data.

Matt <matt.lew...@gmail.com> wrote innews:60d40feb-246a-4370-b041-b3d46fdd90ad@y5g2000hsf.googlegroups.com:



On Feb 1, 3:50 am, David Winsemius <doe_s...@comcast.n0T> wrote:
Matt <matt.lew...@gmail.com> wrote:
On Jan 30, 4:02 pm, David Winsemius  wrote:
Matt <matt.lew...@gmail.com> wrote :

It is a fairly restricted age range (11-18 and only 1 person
each is 11 and 1.  Total n is 85

Number of subjects or number of test points?

So a plot for age shows that the trend to improvement over
time is the same, but the form of this is different.

With the age range so curtailed like this is it more sensible
to use age as a between subjects factor in the repeated
measures analysis rather than a covariate?

What is important is that you accurately describe the
relationships in your data. You say that the "form of [trend to
improvement over time] is different". I think you would get more
sensible replies if you were both more expansive and more
specific about what you are seeing.

-- David Winsemius

the n refers to the number of subjects that completed the
assessment (which is actually 82 given exlusions)

all ages (have excluded the 11 and 18 year olds as they were n > >> > 1, and there are no 17 year olds) show an improvement in
performance on this particular task

The data file is below of one of the examples

I'd love to post the graph but cannot unfortunately.

If a line was fit for each age, there would be improved
performance for each age.  Slopes would differ however.

It makes clear what you were saying about the age*repeat-test
effect. Older individuals learn to do the test faster over 5
repetitions than do younger individuals. Weren't you also dealing
with another covariate whose effect diminished when age was
entered?

--
David Winsemius

David

there has only been age used as a covariate.  I'm sorry if i gave
the impression otherwise.

My query was more related to the fact that inclusion of age as a
covariate makes the main effect of time no longer significant.

I was curious to know if there is a way to examine this effect in
more depth, rather than provide a subjective description of the form
of the data.

I am not offering a "subjective description". I was talking about
observable patterns in your data. Your younger subjects only inmproved
from 1.84 to 1.61 while you oldest sujects improved from 1.86 to 1.38.
The question now comes up whether the measure variance and numbers of
sujects among the age groups (not yet specified) is sufficiently small
to allow a test of trend among the the youngest groups to give a
positive result when testing the hypothesis that the younger subjects
improved at all. That is probably why your sequence effect became non-
significant. Most of the variability I will bet that the p-value was
still suggestive though, perhaps around p=0.10. What was the trial
effect and p-value in your sequence*age model?  Or even better, give us
all the data for the age=12 subgroup.

And while you are at it, you should probably give us the model formula
that you used.

--
David Winsemius