Logistic regression and with interaction

[Charlotte-9]

Dear all,

I am trying to build a logistic regression model to investigate the
influences on uptake of a screening test (outcome 'screened or not').  My
independent variables are ethnicity (6 categories), gender (M/F), age
(continuous) and deprivation (continuous).  I have two main questions:

1.  An initial model including just the main effects suggests that
ethnicity, age and deprivation are all highly significant. However, when I
then run the model adding two interaction terms ethnicity*age and
ethnicity*deprivation, the interaction terms are significant but the
ethnicity main effect is not.  What does this mean?  Does it imply that
the effects of ethnicity are significantly affected by a person's age and
deprivation status?  I should point out that my overall analysis is very
much focused on whether ethnicity has an effect on screening uptake.

2.  As a general question, if I consider interactions of two categorical
variables (say ethnicity and age where ages have been grouped), how do I
interpret the overall model if some of the interactions are significant
and some are not, given that the overall interaction is significant?

Many thanks in advance,

Charlotte

[Hector Maletta]

         Charlotte:
         To your first question, your finding means that ethnicity has an
effect that is amplified (or attenuated, depending on sign) by age and
deprivation. You have also found that there is no other significant
ethnicity effect, apart from the ethnicity effect that is affected by age
and deprivation.
         To your second question: it would mean that certain combinations of
age-group and ethnicity categories alter significantly the odds of the
outcome, whilst other combinations don't. In balance, age-group and
ethnicity do have some effect on the odds of the outcome, and that's the
reason why the overall interaction is significant.

         Hector

         -----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de
Charlotte
Enviado el: 27 May 2007 11:34
Para: [hidden email]
Asunto: Logistic regression and with interaction

         Dear all,

         I am trying to build a logistic regression model to investigate the
         influences on uptake of a screening test (outcome 'screened or
not').  My
         independent variables are ethnicity (6 categories), gender (M/F),
age
         (continuous) and deprivation (continuous).  I have two main
questions:

         1.  An initial model including just the main effects suggests that
         ethnicity, age and deprivation are all highly significant. However,
when I
         then run the model adding two interaction terms ethnicity*age and
         ethnicity*deprivation, the interaction terms are significant but
the
         ethnicity main effect is not.  What does this mean?  Does it imply
that
         the effects of ethnicity are significantly affected by a person's
age and
         deprivation status?  I should point out that my overall analysis is
very
         much focused on whether ethnicity has an effect on screening
uptake.

         2.  As a general question, if I consider interactions of two
categorical
         variables (say ethnicity and age where ages have been grouped), how
do I
         interpret the overall model if some of the interactions are
significant
         and some are not, given that the overall interaction is
significant?

         Many thanks in advance,

         Charlotte

[Marta García-Granero]

Hi everybody

I'm interested in creating a "centroids" plot (created with a veeeery
old program - it used to run in Windows 3.1 - called Grafit). Since my
original posting (with a bitmap of the graph) was rejected (due to the
attachment), here it goes a pathetic ASCII-art imitation (should be
viewed with courier font):

Triglic.               -
|                      |
|                      |
|       -       |------*------|
|       |              |
|       |              |
|       |              -
|  |----O----|
|       |
|       |
|       |
|       -
|------------------------------
          Glucose levels

The circle&asterisc represent the mean of both variables (glucose and
triglycerids) in two groups: treated (O) and control (*). The error
bars are the SD of both variables in each group.

Right now, the process I have to follow is this one:

- I get mean&standard deviation (using any descriptive procedure or
  aggregate) for both variables&groups
- I copy the results and paste then in Grafit data windows
- I create the graph, add manually the error bars, modify  the axis
  and some other settings...
- I copy the graph and paste it in Powerpoint (or Word), convert it to
  Microsoft graph, ungroup all the elments, modify some details (like
  making one of the circles solid and the other hollow...) and regroup
  everything together again.

Then I start all over again with the next couple of variables I want
to graph (sssslow and booooring task).

Obvious question: can this task be done (either with raw or summarized
data) using SPSS graphs (legacy, Igraphs or SPSS 15 GPL)?

I have already taken a look at the GPL Reference Guide, but I found no
inspiration.

This is  syntax to generate the datasets (both with raw or summarized
data).

* Raw data *.
DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

* Summarized *.
AGGREGATE
  /OUTFILE=*
  /BREAK=Group
  /Gluc_mean = MEAN(Gluc)
  /Gluc_sd = SD(Gluc)
  /Triglic_mean = MEAN(Triglic)
  /Triglic_sd = SD(Triglic).

If it can be done I'd appreciate any hint you could provide (I don't
need the full polished-flawless syntax, I don't mind working a bit on
the problem, it's burning my eyelashes for nothing that bothers me).

Thanks a lot.

Surviving monday arrival as best as I can,
Marta

[Peck, Jon]

Of course you can do this with GPL.

DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc" LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW" HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: MEAN_Gluc=col(source(s), name("MEAN_Gluc"))
  DATA: LOW_Gluc=col(source(s), name("MEAN_Gluc_LOW"))
  DATA: HIGH_Gluc=col(source(s), name("MEAN_Gluc_HIGH"))
  DATA: MEAN_Triglic=col(source(s), name("MEAN_Triglic"))
  DATA: LOW_Triglic=col(source(s), name("MEAN_Triglic_LOW"))
  DATA: HIGH_Triglic=col(source(s), name("MEAN_Triglic_HIGH"))
  DATA: Group=col(source(s), name("Group"), unit.category())
  GUIDE: axis(dim(1), label("Glucose levels"))
  GUIDE: axis(dim(2), label("Triglic"))
  GUIDE: legend(aesthetic(aesthetic.color.exterior))
  ELEMENT: point(position(MEAN_Gluc*MEAN_Triglic), color.exterior(Group))
  ELEMENT: edge(position(MEAN_Gluc*(LOW_Triglic+HIGH_Triglic)), color.interior(Group))
  ELEMENT: edge(position((LOW_Gluc+HIGH_Gluc)*MEAN_Triglic),  color.interior(Group)) END GPL.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Marta García-Granero
Sent: Monday, May 28, 2007 10:12 AM
To: [hidden email]
Subject: [SPSSX-L] Bidimensional HILO graph with SPSS?

Hi everybody

I'm interested in creating a "centroids" plot (created with a veeeery
old program - it used to run in Windows 3.1 - called Grafit). Since my
original posting (with a bitmap of the graph) was rejected (due to the
attachment), here it goes a pathetic ASCII-art imitation (should be
viewed with courier font):

Triglic.               -
|                      |
|                      |
|       -       |------*------|
|       |              |
|       |              |
|       |              -
|  |----O----|
|       |
|       |
|       |
|       -
|------------------------------
          Glucose levels

The circle&asterisc represent the mean of both variables (glucose and
triglycerids) in two groups: treated (O) and control (*). The error
bars are the SD of both variables in each group.

Right now, the process I have to follow is this one:

- I get mean&standard deviation (using any descriptive procedure or
  aggregate) for both variables&groups
- I copy the results and paste then in Grafit data windows
- I create the graph, add manually the error bars, modify  the axis
  and some other settings...
- I copy the graph and paste it in Powerpoint (or Word), convert it to
  Microsoft graph, ungroup all the elments, modify some details (like
  making one of the circles solid and the other hollow...) and regroup
  everything together again.

Then I start all over again with the next couple of variables I want
to graph (sssslow and booooring task).

Obvious question: can this task be done (either with raw or summarized
data) using SPSS graphs (legacy, Igraphs or SPSS 15 GPL)?

I have already taken a look at the GPL Reference Guide, but I found no
inspiration.

This is  syntax to generate the datasets (both with raw or summarized
data).

* Raw data *.
DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

* Summarized *.
AGGREGATE
  /OUTFILE=*
  /BREAK=Group
  /Gluc_mean = MEAN(Gluc)
  /Gluc_sd = SD(Gluc)
  /Triglic_mean = MEAN(Triglic)
  /Triglic_sd = SD(Triglic).

If it can be done I'd appreciate any hint you could provide (I don't
need the full polished-flawless syntax, I don't mind working a bit on
the problem, it's burning my eyelashes for nothing that bothers me).

Thanks a lot.

Surviving monday arrival as best as I can,
Marta

[Peck, Jon]

This version gets rid of the two legends.

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc" LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW" HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: MEAN_Gluc=col(source(s), name("MEAN_Gluc"))
  DATA: LOW_Gluc=col(source(s), name("MEAN_Gluc_LOW"))
  DATA: HIGH_Gluc=col(source(s), name("MEAN_Gluc_HIGH"))
  DATA: MEAN_Triglic=col(source(s), name("MEAN_Triglic"))
  DATA: LOW_Triglic=col(source(s), name("MEAN_Triglic_LOW"))
  DATA: HIGH_Triglic=col(source(s), name("MEAN_Triglic_HIGH"))
  DATA: Group=col(source(s), name("Group"), unit.category())
  GUIDE: axis(dim(1), label("Glucose levels"))
  GUIDE: axis(dim(2), label("Triglic"))
  GUIDE: legend(aesthetic(aesthetic.color.interior), null())
  ELEMENT: point(position(MEAN_Gluc*MEAN_Triglic), color.exterior(Group))
  ELEMENT: edge(position(MEAN_Gluc*(LOW_Triglic+HIGH_Triglic)), color.interior(Group))
  ELEMENT: edge(position((LOW_Gluc+HIGH_Gluc)*MEAN_Triglic),  color.interior(Group)) END GPL.

-----Original Message-----
From: Peck, Jon
Sent: Tuesday, May 29, 2007 1:37 PM
To: 'Marta García-Granero'; [hidden email]
Subject: RE: [SPSSX-L] Bidimensional HILO graph with SPSS?

Of course you can do this with GPL.

DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc" LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW" HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: MEAN_Gluc=col(source(s), name("MEAN_Gluc"))
  DATA: LOW_Gluc=col(source(s), name("MEAN_Gluc_LOW"))
  DATA: HIGH_Gluc=col(source(s), name("MEAN_Gluc_HIGH"))
  DATA: MEAN_Triglic=col(source(s), name("MEAN_Triglic"))
  DATA: LOW_Triglic=col(source(s), name("MEAN_Triglic_LOW"))
  DATA: HIGH_Triglic=col(source(s), name("MEAN_Triglic_HIGH"))
  DATA: Group=col(source(s), name("Group"), unit.category())
  GUIDE: axis(dim(1), label("Glucose levels"))
  GUIDE: axis(dim(2), label("Triglic"))
  GUIDE: legend(aesthetic(aesthetic.color.exterior))
  ELEMENT: point(position(MEAN_Gluc*MEAN_Triglic), color.exterior(Group))
  ELEMENT: edge(position(MEAN_Gluc*(LOW_Triglic+HIGH_Triglic)), color.interior(Group))
  ELEMENT: edge(position((LOW_Gluc+HIGH_Gluc)*MEAN_Triglic),  color.interior(Group)) END GPL.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Marta García-Granero
Sent: Monday, May 28, 2007 10:12 AM
To: [hidden email]
Subject: [SPSSX-L] Bidimensional HILO graph with SPSS?

Hi everybody

I'm interested in creating a "centroids" plot (created with a veeeery
old program - it used to run in Windows 3.1 - called Grafit). Since my
original posting (with a bitmap of the graph) was rejected (due to the
attachment), here it goes a pathetic ASCII-art imitation (should be
viewed with courier font):

Triglic.               -
|                      |
|                      |
|       -       |------*------|
|       |              |
|       |              |
|       |              -
|  |----O----|
|       |
|       |
|       |
|       -
|------------------------------
          Glucose levels

The circle&asterisc represent the mean of both variables (glucose and
triglycerids) in two groups: treated (O) and control (*). The error
bars are the SD of both variables in each group.

Right now, the process I have to follow is this one:

- I get mean&standard deviation (using any descriptive procedure or
  aggregate) for both variables&groups
- I copy the results and paste then in Grafit data windows
- I create the graph, add manually the error bars, modify  the axis
  and some other settings...
- I copy the graph and paste it in Powerpoint (or Word), convert it to
  Microsoft graph, ungroup all the elments, modify some details (like
  making one of the circles solid and the other hollow...) and regroup
  everything together again.

Then I start all over again with the next couple of variables I want
to graph (sssslow and booooring task).

Obvious question: can this task be done (either with raw or summarized
data) using SPSS graphs (legacy, Igraphs or SPSS 15 GPL)?

I have already taken a look at the GPL Reference Guide, but I found no
inspiration.

This is  syntax to generate the datasets (both with raw or summarized
data).

* Raw data *.
DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

* Summarized *.
AGGREGATE
  /OUTFILE=*
  /BREAK=Group
  /Gluc_mean = MEAN(Gluc)
  /Gluc_sd = SD(Gluc)
  /Triglic_mean = MEAN(Triglic)
  /Triglic_sd = SD(Triglic).

If it can be done I'd appreciate any hint you could provide (I don't
need the full polished-flawless syntax, I don't mind working a bit on
the problem, it's burning my eyelashes for nothing that bothers me).

Thanks a lot.

Surviving monday arrival as best as I can,
Marta

[Bethany Cockburn]

Hi!  Does anyone have a recommendation for good web-based data entry
software?  Also, I've heard a little about SPSS Dimensions and would be
interested if anyone has an opinion or experience with it.  We will use data
entry for a wide variety of projects, from small (5 item) surveys to large
(4000 item) surveys.

Thanks for your advice!
Bethany




Bethany S. Cockburn, M.A.
Data Specialist
Child Assessor
University of Notre Dame
Data Management Center
Psychology Department
944 Flanner Hall
Notre Dame, IN 46556
574-631-0943

[Jim Moffitt]

If you are doing web-based surveys you don't need data entry software;
the respondents enter the data for you as they complete the
questionnaire. Take a look at Snap Surveys at snapsurveys.com

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Bethany Cockburn
Sent: Tuesday, May 29, 2007 3:23 PM
To: [hidden email]
Subject: Web-based Data entry recommendations

Hi!  Does anyone have a recommendation for good web-based data entry
software?  Also, I've heard a little about SPSS Dimensions and would be
interested if anyone has an opinion or experience with it.  We will use
data entry for a wide variety of projects, from small (5 item) surveys
to large (4000 item) surveys.

Thanks for your advice!
Bethany




Bethany S. Cockburn, M.A.
Data Specialist
Child Assessor
University of Notre Dame
Data Management Center
Psychology Department
944 Flanner Hall
Notre Dame, IN 46556
574-631-0943

[Stevan Nielsen]

We have had a great and long-lasting experience with Zoomerang --
inexpensive, flexible, easy to use.  Over the years we have gathered
more than 20,000 separate survey responses for many different surveys
(we can have as many unique surveys as we want) for about $250 per year.

www.zoomerang.com

Stevan Lars Nielsen, Ph.D.
Clinical Professor
Clinical Psychologist
2518 WSC, BYU
Provo, UT 84602

801-422-3035; fax [phone]

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Jim Moffitt
Sent: Tuesday, May 29, 2007 3:05 PM
To: [hidden email]
Subject: Re: Web-based Data entry recommendations

If you are doing web-based surveys you don't need data entry software;
the respondents enter the data for you as they complete the
questionnaire. Take a look at Snap Surveys at snapsurveys.com

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Bethany Cockburn
Sent: Tuesday, May 29, 2007 3:23 PM
To: [hidden email]
Subject: Web-based Data entry recommendations

Hi!  Does anyone have a recommendation for good web-based data entry
software?  Also, I've heard a little about SPSS Dimensions and would be
interested if anyone has an opinion or experience with it.  We will use
data entry for a wide variety of projects, from small (5 item) surveys
to large (4000 item) surveys.

Thanks for your advice!
Bethany




Bethany S. Cockburn, M.A.
Data Specialist
Child Assessor
University of Notre Dame
Data Management Center
Psychology Department
944 Flanner Hall
Notre Dame, IN 46556
574-631-0943

[la volta statistics]

Again a wonderful solution, Jon. Thanks
Marta, if you only want one SD (instead of 2 SD) you just change

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc"
LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW"
HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.

to:

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=
    MEANSD(Gluc, 1)[NAME="MEAN_Gluc" LOW="MEAN_Gluc_LOW"
HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic, 1)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW"
HIGH="MEAN_Triglic_HIGH"]
    Group
   MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.


Christian

-----Ursprüngliche Nachricht-----
Von: SPSSX(r) Discussion [mailto:[hidden email]]Im Auftrag von
Peck, Jon
Gesendet: Dienstag, 29. Mai 2007 20:39
An: [hidden email]
Betreff: Re: Bidimensional HILO graph with SPSS?


This version gets rid of the two legends.

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc"
LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW"
HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: MEAN_Gluc=col(source(s), name("MEAN_Gluc"))
  DATA: LOW_Gluc=col(source(s), name("MEAN_Gluc_LOW"))
  DATA: HIGH_Gluc=col(source(s), name("MEAN_Gluc_HIGH"))
  DATA: MEAN_Triglic=col(source(s), name("MEAN_Triglic"))
  DATA: LOW_Triglic=col(source(s), name("MEAN_Triglic_LOW"))
  DATA: HIGH_Triglic=col(source(s), name("MEAN_Triglic_HIGH"))
  DATA: Group=col(source(s), name("Group"), unit.category())
  GUIDE: axis(dim(1), label("Glucose levels"))
  GUIDE: axis(dim(2), label("Triglic"))
  GUIDE: legend(aesthetic(aesthetic.color.interior), null())
  ELEMENT: point(position(MEAN_Gluc*MEAN_Triglic), color.exterior(Group))
  ELEMENT: edge(position(MEAN_Gluc*(LOW_Triglic+HIGH_Triglic)),
color.interior(Group))
  ELEMENT: edge(position((LOW_Gluc+HIGH_Gluc)*MEAN_Triglic),
color.interior(Group)) END GPL.

-----Original Message-----
From: Peck, Jon
Sent: Tuesday, May 29, 2007 1:37 PM
To: 'Marta Garca-Granero'; [hidden email]
Subject: RE: [SPSSX-L] Bidimensional HILO graph with SPSS?

Of course you can do this with GPL.

DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=MEANSD(Gluc)[NAME="MEAN_Gluc"
LOW="MEAN_Gluc_LOW" HIGH="MEAN_Gluc_HIGH"]
    MEANSD(Triglic)[NAME="MEAN_Triglic" LOW="MEAN_Triglic_LOW"
HIGH="MEAN_Triglic_HIGH"]
    Group MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: MEAN_Gluc=col(source(s), name("MEAN_Gluc"))
  DATA: LOW_Gluc=col(source(s), name("MEAN_Gluc_LOW"))
  DATA: HIGH_Gluc=col(source(s), name("MEAN_Gluc_HIGH"))
  DATA: MEAN_Triglic=col(source(s), name("MEAN_Triglic"))
  DATA: LOW_Triglic=col(source(s), name("MEAN_Triglic_LOW"))
  DATA: HIGH_Triglic=col(source(s), name("MEAN_Triglic_HIGH"))
  DATA: Group=col(source(s), name("Group"), unit.category())
  GUIDE: axis(dim(1), label("Glucose levels"))
  GUIDE: axis(dim(2), label("Triglic"))
  GUIDE: legend(aesthetic(aesthetic.color.exterior))
  ELEMENT: point(position(MEAN_Gluc*MEAN_Triglic), color.exterior(Group))
  ELEMENT: edge(position(MEAN_Gluc*(LOW_Triglic+HIGH_Triglic)),
color.interior(Group))
  ELEMENT: edge(position((LOW_Gluc+HIGH_Gluc)*MEAN_Triglic),
color.interior(Group)) END GPL.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Marta Garc a-Granero
Sent: Monday, May 28, 2007 10:12 AM
To: [hidden email]
Subject: [SPSSX-L] Bidimensional HILO graph with SPSS?

Hi everybody

I'm interested in creating a "centroids" plot (created with a veeeery
old program - it used to run in Windows 3.1 - called Grafit). Since my
original posting (with a bitmap of the graph) was rejected (due to the
attachment), here it goes a pathetic ASCII-art imitation (should be
viewed with courier font):

Triglic.               -
|                      |
|                      |
|       -       |------*------|
|       |              |
|       |              |
|       |              -
|  |----O----|
|       |
|       |
|       |
|       -
|------------------------------
          Glucose levels

The circle&asterisc represent the mean of both variables (glucose and
triglycerids) in two groups: treated (O) and control (*). The error
bars are the SD of both variables in each group.

Right now, the process I have to follow is this one:

- I get mean&standard deviation (using any descriptive procedure or
  aggregate) for both variables&groups
- I copy the results and paste then in Grafit data windows
- I create the graph, add manually the error bars, modify  the axis
  and some other settings...
- I copy the graph and paste it in Powerpoint (or Word), convert it to
  Microsoft graph, ungroup all the elments, modify some details (like
  making one of the circles solid and the other hollow...) and regroup
  everything together again.

Then I start all over again with the next couple of variables I want
to graph (sssslow and booooring task).

Obvious question: can this task be done (either with raw or summarized
data) using SPSS graphs (legacy, Igraphs or SPSS 15 GPL)?

I have already taken a look at the GPL Reference Guide, but I found no
inspiration.

This is  syntax to generate the datasets (both with raw or summarized
data).

* Raw data *.
DATA LIST FREE/Group Gluc Triglic.
BEGIN DATA
1 5.17 1.51 1 5.25 1.64 1 5.31 1.59 1 4.58 1.52 1 5.56 1.64
1 5.15 1.35 1 5.33 1.68 1 5.29 1.30 1 5.01 1.91 1 5.88 1.30
2 4.92 1.32 2 4.71 1.41 2 4.79 1.24 2 4.04 1.24 2 4.82 1.39
2 4.41 1.32 2 4.89 1.36 2 4.60 1.33 2 4.75 1.20 2 5.33 1.41
2 4.67 1.18 2 4.08 1.37 2 5.19 1.28 2 4.77 1.05 2 4.94 0.99
2 4.65 1.31 2 4.76 1.29 2 5.33 1.20 2 4.70 1.18 2 5.02 1.13
END DATA.

VALUE LABEL Group 1 'Control' 2' Treated'.

* Summarized *.
AGGREGATE
  /OUTFILE=*
  /BREAK=Group
  /Gluc_mean = MEAN(Gluc)
  /Gluc_sd = SD(Gluc)
  /Triglic_mean = MEAN(Triglic)
  /Triglic_sd = SD(Triglic).

If it can be done I'd appreciate any hint you could provide (I don't
need the full polished-flawless syntax, I don't mind working a bit on
the problem, it's burning my eyelashes for nothing that bothers me).

Thanks a lot.

Surviving monday arrival as best as I can,
Marta

[Natalie Patterson]

Beth,
I have used SPSS dimensions for a few years and found it very easy to
adminster web based surveys. You can load questions items from ms word
and activating the job was very easy. The only part that I found
confusing was using quota and sample management. Would be happy to
address any specific questions you have.
 
Thanks,
Natalie
 
Natalie M. Patterson
Analyst | Access Management Measurement & Evaluation | Spectrum Health
Hospitals
Phone: [phone] | Fax: [phone]