Comparing ORs

[Chandler, Helena K]

 Hi group -

This isn't necessarily an SPSS-specific question, but there seems to be a lot of general stats knowledge among this group so I hope you'll indulge me in my question about comparing odds ratios.

I have two models to predict health care use:

Model A includes controls and PTSD
Model B includes controls, PTSD and Medical Illness

Eyeballing these models, it's pretty clear that the OR for PTSD is far smaller once Medical Utilization is included in the model.  The log-likelihood ratios indicate that the 2nd model is a better fit, but what I want to be able to say is that the change in the Odds ratio for PTSD is significant.  I had thought that if the Confidence Intervals around the two ORs for PTSD were non-overlapping then they were statistically distinct.  I was told that this is incorrect.  So, how do you make a statement about the difference between the odds ratio for PTSD in Model A (which is about 3.5) vs. the odds ratio for PTSD in Model B (which is about 1.2)?  [BTW, event though the OR in Model B is so close to 1, it is a statistically significant predictor since our sample size is huge (20,000)]

Thanks for your help.

Helena Chandler, PhD


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Automatic digest processor
Sent: Friday, August 31, 2007 12:02 AM
To: Recipients of SPSSX-L digests
Subject: SPSSX-L Digest - 29 Aug 2007 to 30 Aug 2007 (#2007-243)

There are 15 messages totalling 1007 lines in this issue.

Topics of the day:

  1. Run regression analysis without excluding variables.
  2. Python Sorting (2)
  3. Assigning Random Numbers (Open) (2)
  4. unsubscribe (2)
  5. Vista and SPSS
  6. Assigning Number (2)
  7. Obtaining analysis results via command syntax
  8. List rejecting "duplicate" messages (2)
  9. Using LMATRIX to fix between-subject levels  10. van Westendorp extension

----------------------------------------------------------------------

Date:    Wed, 29 Aug 2007 23:28:28 -0700
From:    Abdus Salam <[hidden email]>
Subject: Run regression analysis without excluding variables.

Dear Experts;

  I am facing a problem in one of my regression model; To developed that model I have one dependent variable which shows proportion of initiate on medicine and 23 attributes variable as an independent variables; when I run simple linear regression then some of the attributes excluded so to get rid of that excluded variable I run the regression in Enter method and take the attributes block wise. Now my question: is there any other way to get rid of that type of excluded variables. I can run also first time factor analysis with in that attributes and then
  Run also regression with that factor solution; but client needs 23 attributes derived Importance. So is there any other option to run the regression analysis without using block.

  Thanks for any of your's quick response.

  Mousumi.




---------------------------------
Be a better Heartthrob. Get better relationship answers from someone who knows.
Yahoo! Answers - Check it out.

------------------------------

Date:    Thu, 30 Aug 2007 09:44:37 +0200
From:    Georg Maubach <[hidden email]>
Subject: Python Sorting

Dear Listers,

we resend our question to this list because the mail system rejected at the first try saying we have asked this question before. We checked our files and found we did not. Maybe that the mail system had a similar mail in the records.

----------------

Dear Listers,

we have begun to use Python for one of our research projects. We need to create sorted variables lists. The variable have indexes in the variable name at 3 positions:

The generic format of the variable is q411_Index1_Index2_Index3_07.

Example:=20
q411_101_101_101_07
q411_101_102_101_07
q411_101_103_101_07

q411_101_101_102_07
q411_101_102_102_07
q411_101_103_102_07

q411_101_101_103_07
q411_101_102_103_07
q411_101_103_103_07

q411_102_101_101_07
q411_102_102_101_07
q411_102_103_101_07

q411_102_101_102_07
q411_102_102_102_07
q411_102_103_102_07

q411_102_101_103_07
q411_102_102_103_07
q411_102_103_103_07
and so on ....

We need to sort the variables

first by index1
second by index3
third by index2

The variables are stored in a list like [q411_101_101_101_07, q411_101_102_101_07, q411_101_103_101_07, q411_101_101_102_07, q411_101_102_102_07, q411_101_103_102_07, .....].

How can we sort that list using Python sorting functionality?

Best regards

Georg Maubach
Research Manager

------------------------------

Date:    Thu, 30 Aug 2007 11:39:26 +0200
From:    =?iso-8859-2?Q?B=F3dai_Zolt=E1n?= <[hidden email]>
Subject: Re: Assigning Random Numbers (Open)

Hi Richard,

I am afraid your calculation in the birthday problem is a little simpler than it should be. You don't take into account conditional probability when deriving your formula.

So, in my opinion, the result would be *much* more complex.

It is easier to calculate the probability that all the k birthdays are on *different* days.

P= 1*(1-1/365)*(1-2/365)*...*(1-(k-1)/365) = 365!/(365^k*(365-k)!) if I am not mistaken.

Or, more generally, P= n!/(n^k*(n-k)!)

So the probability we wanted to find is 1-P. It is the probability that there will be at least 1 duplication.

If k=3E5 and n=9E15, this results in a very similar(*almost* the same) vale to Richard's: 5E-6, but not exactly the same.

But if ratio k/n was larger, the result would be much different. Just consider k=30, n=365: your formula would give a value larger than 1 for a probability!

Just couldn't miss replying to your note 'simpler in closed form than it looks at first'. I hope you don't mind my correcting the formula.

Zoltan Bodai
Hungary



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Richard Ristow
Sent: Thursday, August 30, 2007 4:45 AM
To: [hidden email]
Subject: Re: Assigning Random Numbers

The birthday problem -

At 06:55 PM 8/29/2007, Fry, Jonathan B. wrote:

>One might reasonably suppose that the possibility of duplication could
>be ignored [when drawing 300,000 numbers] with the Mersenne twister,
>but I think not.  Even if you use all 53 bits, I think there's a
>substantial possibility of duplication in 300,000 cases.  It's a
>birthday problem with 2^53 possible birthdays and 300,000 tries.  I
>have not worked it out.

The birthday problem is simpler in closed form than it looks at first.
Let the population size be N (N=365, for the birthday problem). Draw 1, and there's 0 chance of a duplicate. Draw a 2nd, and there's one chance for a duplicate, or probability 1/N. If that draw isn't a duplicate, draw a 3rd and *at that draw* there are 2 chances for a duplicate, probability 2/N, etc.

Generally, the probability that the *first* duplicate occurs on the kth draw is (k-1)/N. So the probability that *any* duplicate occurs
*through* the kth draw is 1/N times

   Sum(i=1,k)(i-1)
= Sum(i=0,k-1)(i)
= k*(k-1)/2    -- exactly square-law, in k

For k=3E5 and N=2^53, that is

4.5E10/9.0E15 = 5E-6 -- pretty small.

You'd need to draw about 9.5 million numbers, for a 1% chance of a duplicate (ignoring any non-random properties of the series).

------------------------------

Date:    Thu, 30 Aug 2007 07:29:22 -0500
From:    "Peck, Jon" <[hidden email]>
Subject: Re: Python Sorting

This can be done very compactly in Python by making use of its special support for customizing the sorting of lists.  When you use the sort method of a list object, you can specify a key function that prepares the list items for comparison.  The following will do what you want.

def keyf(x):
  xs = x.split("_")
  return "_".join([xs[1], xs[3], xs[2]])

That function extracts the three indexes, rearranges the order, and then joins them back together with the "_".  Sorting that rearranged string in the standard way then gives you the sorted list.

To use it, if lis is your list of names, do this.
lis.sort(key=keyf)

In your example the first part of a name is always q411 and the last part is always 07.  If these can vary, just include xs[0] and xs[4] in the join list above.

Regards,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Georg Maubach
Sent: Thursday, August 30, 2007 2:45 AM
To: [hidden email]
Subject: [SPSSX-L] Python Sorting

Dear Listers,

we resend our question to this list because the mail system rejected at the first try saying we have asked this question before. We checked our files and found we did not. Maybe that the mail system had a similar mail in the records.

----------------

Dear Listers,

we have begun to use Python for one of our research projects. We need to create sorted variables lists. The variable have indexes in the variable name at 3 positions:

The generic format of the variable is q411_Index1_Index2_Index3_07.

Example:=20
q411_101_101_101_07
q411_101_102_101_07
q411_101_103_101_07

q411_101_101_102_07
q411_101_102_102_07
q411_101_103_102_07

q411_101_101_103_07
q411_101_102_103_07
q411_101_103_103_07

q411_102_101_101_07
q411_102_102_101_07
q411_102_103_101_07

q411_102_101_102_07
q411_102_102_102_07
q411_102_103_102_07

q411_102_101_103_07
q411_102_102_103_07
q411_102_103_103_07
and so on ....

We need to sort the variables

first by index1
second by index3
third by index2

The variables are stored in a list like [q411_101_101_101_07, q411_101_102_101_07, q411_101_103_101_07, q411_101_101_102_07, q411_101_102_102_07, q411_101_103_102_07, .....].

How can we sort that list using Python sorting functionality?

Best regards

Georg Maubach
Research Manager

------------------------------

Date:    Thu, 30 Aug 2007 08:19:19 -0500
From:    "Marks, Jim" <[hidden email]>
Subject: Re: Assigning Random Numbers (Open)

A better calculation of the birthday probability is 1 - P(not matching).

With two people the odds of not matching are    364/365.
With three people the odds of not matching are  363/365 ...

The probability of not matching is the product of the series:
364/365 * 363/365 * ...

With 30 people the odds of not matching are about 30%, so the odds of a match are 1-30% = 70%. You can amaze your friends at parties (or make money betting) with the prediction of a match in a relatively small group

--jim

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bódai Zoltán
Sent: Thursday, August 30, 2007 4:39 AM
To: [hidden email]
Subject: Re: Assigning Random Numbers (Open)

Hi Richard,

I am afraid your calculation in the birthday problem is a little simpler than it should be. You don't take into account conditional probability when deriving your formula.

So, in my opinion, the result would be *much* more complex.

It is easier to calculate the probability that all the k birthdays are on *different* days.

P= 1*(1-1/365)*(1-2/365)*...*(1-(k-1)/365) = 365!/(365^k*(365-k)!) if I am not mistaken.

Or, more generally, P= n!/(n^k*(n-k)!)

So the probability we wanted to find is 1-P. It is the probability that there will be at least 1 duplication.

If k=3E5 and n=9E15, this results in a very similar(*almost* the same) vale to Richard's: 5E-6, but not exactly the same.

But if ratio k/n was larger, the result would be much different. Just consider k=30, n=365: your formula would give a value larger than 1 for a probability!

Just couldn't miss replying to your note 'simpler in closed form than it looks at first'. I hope you don't mind my correcting the formula.

Zoltan Bodai
Hungary



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Richard Ristow
Sent: Thursday, August 30, 2007 4:45 AM
To: [hidden email]
Subject: Re: Assigning Random Numbers

The birthday problem -

At 06:55 PM 8/29/2007, Fry, Jonathan B. wrote:

>One might reasonably suppose that the possibility of duplication could
>be ignored [when drawing 300,000 numbers] with the Mersenne twister,
>but I think not.  Even if you use all 53 bits, I think there's a
>substantial possibility of duplication in 300,000 cases.  It's a
>birthday problem with 2^53 possible birthdays and 300,000 tries.  I
>have not worked it out.

The birthday problem is simpler in closed form than it looks at first.
Let the population size be N (N=365, for the birthday problem). Draw 1, and there's 0 chance of a duplicate. Draw a 2nd, and there's one chance for a duplicate, or probability 1/N. If that draw isn't a duplicate, draw a 3rd and *at that draw* there are 2 chances for a duplicate, probability 2/N, etc.

Generally, the probability that the *first* duplicate occurs on the kth draw is (k-1)/N. So the probability that *any* duplicate occurs
*through* the kth draw is 1/N times

   Sum(i=1,k)(i-1)
= Sum(i=0,k-1)(i)
= k*(k-1)/2    -- exactly square-law, in k

For k=3E5 and N=2^53, that is

4.5E10/9.0E15 = 5E-6 -- pretty small.

You'd need to draw about 9.5 million numbers, for a 1% chance of a duplicate (ignoring any non-random properties of the series).

------------------------------

Date:    Wed, 29 Aug 2007 09:01:21 -0400
From:    Richard Ristow <[hidden email]>
Subject: Re: unsubscribe

FAQ: How to unsubscribe, or leave list SPSSX-L:

Requests to unsubscribe that are posted to the list, will never be acted on.

You must send the request to [hidden email].

 From the E-mail address from which you're subscribed to the list, send a message to [hidden email] with the following words in the body of the message:

SIGNOFF SPSSX-L

Don't put anything else (your name, etc.) in the body of the message.

It should work. If it doesn't, go to the following Web page:

http://www.listserv.uga.edu/cgi-bin/wa?SUBED1=spssx-l&A=1

and unsubscribe from there.

..........................
More information:

When you subscribed to the list, you received a welcome message (I'm copying it below) with instructions (including asking you to save it).

 From the welcome message:

>Your  subscription to  the SPSSX-L  list (SPSSX(r)  Discussion) has
>been accepted.
>
>Please save this message for future  reference, [...]
>
>You may leave the list at any time by sending a "SIGNOFF SPSSX-L"
>command to [hidden email].

There are many other commands that can be sent to the same address, to manage your subscription. If you send mail to [hidden email] with the text

INFO REFCARD

and no other text, you will be mailed a file describing those commands.

------------------------------

Date:    Thu, 30 Aug 2007 11:24:11 -0400
From:    Mark A Davenport MADAVENP <[hidden email]>
Subject: Re: Vista and SPSS

I don't recall ever receiving this notice.  And I am a beta tester.

***************************************************************************************************************************************************************
Mark A. Davenport Ph.D.
Senior Research Analyst
Office of Institutional Research
The University of North Carolina at Greensboro
336.256.0395
[hidden email]

'An approximate answer to the right question is worth a good deal more than an exact answer to an approximate question.' --a paraphrase of J. W.
Tukey (1962)






Kylie Lange <[hidden email]>
Sent by: "SPSSX(r) Discussion" <[hidden email]>
08/30/2007 10:56 AM
Please respond to
[hidden email]


To
[hidden email]
cc

Subject
Re: Vista and SPSS






Hi David,

I am the SPSS contact for my university and on June 27th I received one of their 'Proactive Notification' emails warning about the issue you raised. The text of the message was:

====================================
Subject:        SPSS Proactive Notification - Deleting variables from
dataset may
generate subsequent incorrect results with SPSS 15.0.1 dated November 22,
2006

NOTE: If you are not running SPSS 15.0.1 dated November 22, 2006 - please disregard this email.

SPSS Technical Support recently detected a problem with SPSS 15.0.1 dated November 22, 2006.  Deleting variables in a dataset may cause data to be incorrect upon subsequent transformations.  This only occurs when variables are deleted interactively from the user interface or invoking the DELETE VARIABLES syntax command.  This would be apparent when a new variable is generated in the active dataset.

SPSS is evaluating the most appropriate solution but until it is available, the workaround would be to cache your dataset after deleting a variable.  From the data editor go to File->Cache Data and choose Cache Now.  Via command syntax, please add the CACHE command following every DELETE VARIABLES command.

If you have any questions or problems, please contact your local SPSS support team (http://support.spss.com/contactus.asp ).

SPSS apologizes for any inconvenience this has caused. We are continually striving to provide more proactive communication to our clients.  However, if you do not wish to receive further proactive e-mails from SPSS Support, just reply to this message with REMOVE on the subject line.

Thank you,
SPSS Technical Support
====================================


Then, on July 12th another Proactive Notification was sent announcing the
15.0.1.1 patch that fixes the problem:


====================================
Subject:        SPSS Proactive Notification - SPSS 15.0.1.1 Patch
availability
notice addressing the delete variables issue with SPSS 15.0.1 dated November 22, 2006

SPSS Technical Support announces the availability of the SPSS 15.0.1.1 Patch for Windows users. This patch will work only for English Windows versions of SPSS 15.0.1.

This patch addresses the issue that was previously announced on June 26,
2007
relating to potential data transformation errors after deleting variables from your dataset in SPSS 15.0.1. Please see Resolution 70501 for more details.
This
patch does not correct previously saved SPSS data files.

For detailed information on the description of this patch, installation instructions, and downloading go to: http://support.spss.com  and click on "Login to Online Tech Support" then enter your user name and password.
Select
the "SPSS" menu link then on "Patches" followed by "SPSS for Windows". If you have forgotten your user id or password you can use the "Find Password"
feature
on our Web site. This can be found at:
http://support.spss.com/password.html

We are continually striving to improve our communication to our clients.
However, if you do not wish to receive further proactive e-mails from SPSS Support, please reply to this message with REMOVE on the subject line.

Thank you,
SPSS Technical Support
====================================


It sounds like these notifications don't get distributed to as many users as they should, but presumably if people aren't receiving them they can contact Technical Support and request to be added to the distribution list for the future.

In the meantime, the 15.0.1.1 patch can be downloaded from http://support.spss.com.

Hope this helps.

Cheers,
Kylie.


Quoting "Laflamme, David" <[hidden email]>:

> Mark,
>
> I have been using SPSS 15.0.1 on Vista since the beginning of June.
> Other than the issue below which is not specific to Vista, I have had
> no problems.  I originally thought the issue below was a Vista
> problem, but later learned from SPSS that it is not operating system specific (e.g.
> affects WinXP too).
>
> I found what I consider to be a serious bug, however SPSS has told me
> that they do not intend to fix it in the current version.  Their term
> below is "inconvenience".  I believe version 16 isn't due out until
> the 3rd quarter of this year.  I was very disappointed, of course.
> Even more disappointing is that the bug appears not to have been publicized.
> The SPSS tech I spoke with told me that one other person had reported
> the problem -- so they were aware of it already.  His message to me is
> below.  I did express my disappointment to the tech on the phone.
>
> I sent a message (first message pasted below) to the person who
> manages the SPSS license at my university, who then sent it out to all
> of our SPSS users.  Fortunately, there is a workaround --
> unfortunately, it is not a workaround that can be implemented
> automatically.  Rather, it relies on the user to remember to take action frequently.
>
> **Here is the issue and workaround:  After deleting a variable, you
> must save/cache the dataset.  If you don't, then when another variable
> is created, any missing values in the new variable are populated with
> whatever values used to be in the now deleted variable.  Not always
> easy to detect unless you're looking for it, so it is very possible
> that many SPSS users have unknowingly used faulty data.  I had to go
> back and check a couple weeks worth of work.
>
> I am considering moving to SAS because of the handling of this issue.
>
> I was not a member of this list when I discovered this, or would have
> shared it at that time.  Also, it is possible that SPSS has since
> fixed the bug but not notified me -- though again, their stated
> intention was not to fix it in the current version.  I know there are
> several SPSS staffers on this list, so perhaps they will chime in.  I
> would be happy to learn that the info I received was incorrect!
>
> -David
>
> >>>>6/20/07
> I identified what I think is a pretty serious bug in SPSS yesterday.
> The SPSS tech support person was able to replicate it while I was on
> the phone with him.  He later sent me the email below.  It looks like
> the bottom line is that SPSS has no intention of issuing a patch.
> Instead, they will wait until version 16 is released.  At the very
> least, I would expect them to notify their customers.  The
> ramifications are not insignificant.  I will need to go back and
> recheck several analyses I've done using version 15.0.1.  I think it
> may be worth sending out a note to all SPSS users.  There is a
> workaround, but if you don't know about the problem/solution you could
> end up with faulty data.  That is what happened to me. Fortunately, I
> did a little detective work and got lucky and found the bug.
>
> The summary below is from another SPSS client who originally found the
> bug.  I can describe how you can replicate it on your machine if you
> want.
>
> Anyway, I just thought I should let you know in case you want to let
> the other UNH SPSS users know about the problem.  Someone who intends
> to publish results compiled using SPSS version 15.0.1 would probably
> want to recheck their work.
>
> >>>>>
> From: SPSS Support [mailto:[hidden email]]
> Sent: Tuesday, June 19, 2007 5:27 PM
> To: Laflamme, David
> Subject: SPSS CaseID : DED/501861 - Strange behaviour with duplicating
> values
>
> David,
>     A quick update here - before submitting this to development, I ran
> this issue by a colleauge, who informed me that he's already submitted
> it!  It's fixed in 16.0, and I ran your data and syntax through a 16
> beta, to confirm that I can't replicate the problem there.  Looks like
> someone else got to this one before we did.
>
> Problem Summary:
> Deleting a variable may cause subsequent transformations to be
> incorrect
>
> Problem Description:
> I'm using SPSS for Windows version 15.0.1 and I compute a new variable
> in a dataset, for instance:
>
> COMPUTE z = x + y
>
> where 'x' and 'y' are existing variables. I then delete 'z' by either
> right-clicking on it and selecting Clear or via a DELETE VARIABLES
> command. I then run something like this:
>
> IF (gender = 1) a = 1.
>
> The variable 'a' now holds the same position in the dataset that 'z'
> used to. And any cell of 'a' not populated with a 1 is populated with
> data that had been in 'z' before it was "deleted".
>
> It appears 'z' wasn't really deleted.
>
> This problem did not occur in 15.0.0 or earlier. What is wrong?
>
>
> Resolution Description:
> This appears to be a defect and has been reported to SPSS development
> for remediation in an upcoming release of SPSS. As a workaround until
> the behavior is corrected, please cache your dataset after Clearing a
> variable or issuing a DELETE VARIABLES command.
>
> This error has been fixed in SPSS 16.0 for Windows.
>
> We apologize for the inconvenience.
>
> Devon Dejno - SPSS Technical Support
> [hidden email]
> Tel: [phone]    Fax: [phone]
> Have a question for SPSS Technical Support?
> Try our resolution search at: http://support.spss.com
>

------------------------------

Date:    Thu, 30 Aug 2007 12:22:32 -0400
From:    "<R. Abraham>" <[hidden email]>
Subject: Assigning Number

I have a slightly different question on randomly assigning a number.

I want to assign a random number between the ranges 19 and 35, if a case meets a specific criteria.

e.g.
If Inferred Age = '1' and Tenure <19,

Then Assign a Random Number between 19 and 35 for the value of 'Age', such that there is a difference of atleast 16 years between the values of 'Tenure' and 'Age'.

Note:
Inferred Age '1' = 19-35 years
'Age' Value is Missing

The SAS code for this is:
if InferredAge=1 and Tenure<19 then Age=16*ranuni(0)+19;

Can somebody help with a similar syntax in SPSS?

Thanks.
Renji Abraham

------------------------------

Date:    Thu, 30 Aug 2007 12:55:44 -0500
From:    Barth Riley <[hidden email]>
Subject: Obtaining analysis results via command syntax

Hello



I have written a program that repeatedly generates a dataset and performs a statistical analysis on that data for purposes of estimating power and sample size. I would like to know if there is a way to find out whether the result from an analysis was significant for a particular effect in the model. I could then calculate the number of significant tests relative to the total number of tests (iterations) conducted. Can anything like this be done in command syntax?



Thanks



Barth Riley, Ph.D.

Res. Asst. Professor & Associate Program Director

Dept. of Disability and Human Development M/C 626

University of Illinois-Chicago

1640 W. Roosevelt Rd.

Chicago, IL 60608

Voice [phone]

Fax:   [phone]

Email:  <mailto:[hidden email]> [hidden email]

------------------------------

Date:    Thu, 30 Aug 2007 12:53:08 -0400
From:    Richard Ristow <[hidden email]>
Subject: List rejecting "duplicate" messages

At 03:44 AM 8/30/2007, Georg Maubach wrote:

>we resend our question to this list because the mail system rejected at
>the first try saying we have asked this question before. We checked our
>files and found we did not. Maybe that the mail system had a similar
>mail in the records.

Actually, the first try was distributed properly. You'll see it, if your settings for this list have it send you copies of your own postings.

Those "rejected duplicate posting" messages happen from time to time, referring to a posting you made that was distributed correctly.

It took me a while not to worry about them. They appear to result from certain kinds of mail bounce, that send the bounce message, with the original message text, back to the *list*, but appearing to come from you. It's a good thing the LISTSERV software does recognize and reject them; otherwise, at best your posting would be duplicated, and at worst there'd be a cascade of new bounce messages posted to the list.

Fortunately, these don't happen very often. The LISTSERV software may also report these duplicates to the list administrator, who'll unsubscribe the offending account; the software may even do it automatically.

The correct response is to make sure your original posting was distributed (look for it on the list, or the acknowledgement from "L-Soft list server at UGA (1.8d)" <[hidden email]>); and if it was, ignore and discard the 'duplicate' message.

------------------------------

Date:    Thu, 30 Aug 2007 19:48:11 +0100
From:    Bonnie Poon Zahl <[hidden email]>
Subject: Using LMATRIX to fix between-subject levels

Hi,

I was wondering if anyone knew the syntax to test for simple interaction effects of two within-subject variables at four levels of a between-subject variable? I have gotten all the interactions down except this one! Would be grateful for any help.

Best wishes,
Bonnie Poon Zahl

------------------------------

Date:    Thu, 30 Aug 2007 14:52:34 -0400
From:    Richard Ristow <[hidden email]>
Subject: Re: unsubscribe

At 01:52 PM 8/30/2007, Chang,Victor wrote to me, off-list:

>SIGNOFF SPSSX-L


I don't like to nag, but since you're posting this as a reply, you must have seen the very recent postings about unsubscribing, including a complete set of the following instructions. There is a great deal to be said for reading available relevant information.

Anyway, here you go:

FAQ: How to unsubscribe, or leave list SPSSX-L:

Requests to unsubscribe that are posted to the list, will never be acted on.
....................
ADDENDUM: Unsubscribing users isn't my job, either. In fact, I don't have listowner privileges to do it (and don't want them!). There is a very competent and helpful person who administers the list, but *she* shouldn't be asked to handle tasks, like subscription and un-subscription, that the list server handles easily.
....................


You must send the request to [hidden email].

 From the E-mail address from which you're subscribed to the list, send a message to [hidden email] with the following words in the body of the message:

SIGNOFF SPSSX-L

Don't put anything else (your name, etc.) in the body of the message.

It should work. If it doesn't, go to the following Web page:

http://www.listserv.uga.edu/cgi-bin/wa?SUBED1=spssx-l&A=1

and unsubscribe from there.

...........................
More information:

When you subscribed to the list, you received a welcome message (I'm copying it below) with instructions (including asking you to save it).

 From the welcome message:

Your  subscription to  the SPSSX-L  list (SPSSX(r)  Discussion) has  been accepted.

Please save this message for future  reference, [...]

You may leave the list at any time by sending a "SIGNOFF SPSSX-L"
command to [hidden email].

There are many other commands that can be sent to the same address, to manage your subscription. If you send mail to [hidden email] with the text

INFO REFCARD

and no other text, you will be mailed a file describing those commands.

------------------------------

Date:    Thu, 30 Aug 2007 14:53:02 -0400
From:    UGA Listserv Admin <[hidden email]>
Subject: Re: List rejecting "duplicate" messages

On Thu, 30 Aug 2007, Richard Ristow wrote:

> At 03:44 AM 8/30/2007, Georg Maubach wrote:
>
>> we resend our question to this list because the mail system rejected
>> at the first try saying we have asked this question before. We
>> checked our files and found we did not. Maybe that the mail system
>> had a similar mail in the records.
>
> Actually, the first try was distributed properly. You'll see it, if
> your settings for this list have it send you copies of your own postings.
>
> Those "rejected duplicate posting" messages happen from time to time,
> referring to a posting you made that was distributed correctly.
>
> It took me a while not to worry about them. They appear to result from
> certain kinds of mail bounce, that send the bounce message, with the
> original message text, back to the *list*, but appearing to come from
> you. It's a good thing the LISTSERV software does recognize and reject
> them; otherwise, at best your posting would be duplicated, and at
> worst there'd be a cascade of new bounce messages posted to the list.
>
> Fortunately, these don't happen very often. The LISTSERV software may
> also report these duplicates to the list administrator, who'll
> unsubscribe the offending account; the software may even do it automatically.

Sadly, listserv doesn't inform the listowner about these (but I'll probably get one from posting this and be able to figure out who the culprit is), so if you are gettting these and want to stop it, send me a copy of the bounce.

>
> The correct response is to make sure your original posting was
> distributed (look for it on the list, or the acknowledgement from
> "L-Soft list server at UGA (1.8d)" <[hidden email]>); and
> if it was, ignore and discard the 'duplicate' message.

By the way, if you don't get a copy of your posts, and want to, set your subscription to REPRO  under 'Acknowledgements' at
http://listserv.uga.edu/cgi-bin/wa?SUBED1=spssx-l&A=1

--
***Jean Snow                                       [hidden email] ***
*** EITS Core Services Support/Postmaster/Listserv Admin            ***

------------------------------

Date:    Thu, 30 Aug 2007 17:12:29 -0400
From:    Richard Ristow <[hidden email]>
Subject: Re: Assigning Number

At 12:22 PM 8/30/2007, <R. Abraham> wrote:

>I want to assign a random number between the ranges 19 and 35, if a
>case meets a specific criteria.
>
>e.g.
>If Inferred Age = '1' and Tenure <19,
>
>Then Assign a Random Number between 19 and 35 for the value of 'Age',
>such that there is a difference of atleast 16 years between the values
>of 'Tenure' and 'Age'.
>
>The SAS code for this is:
>if InferredAge=1 and Tenure<19 then Age=16*ranuni(0)+19;

Well, that translates directly into SPSS; in fact, it can be made the corresponding SPSS with only a little editing. (In fact, I've made some changes not strictly necessary.) I don't have a SAS manual right here; I'm assuming that 'ranuni(0)' returns a number uniform-distributed on (0,1). This is not tested, but:

if InferredAge EQ 1 and Tenure LT 19
    Age=16*rv.uniform(0,1)+19.

BUT, I think the SAS code may be buggy, with the bugs carried over to the SPSS code:

- The result won't be an integer, in either SPSS or SAS, if that matters.

- You say you want "a difference of at least 16 years between the values of 'Tenure' and 'Age'". You won't get that. This code is triggered by any value of Tenure less than 19; it can generate a value as close to 19 as you like. The minimum difference is 0, though that can never be reached exactly; or 1, if 'Tenure' is always an integer.

AND, are you sure this is a good idea at all? It looks like you may be filling in reasonable but fictitious values for missing values. At very best, this loses statistical power; at a likely worst, it can give results that are plain wrong.

-Good luck and onward,
  Richard

------------------------------

Date:    Thu, 30 Aug 2007 20:56:18 -0400
From:    Scott Porter <[hidden email]>
Subject: van Westendorp extension

Has anyone played with the Newton, Miller and Smith (1993) extension of the van Westendorp model where they add two purchase probability questions (BARGAIN and EXPENSIVE price points) in order to model maximum trial and maximum revenue points.

I found a reference here:

http://www.sawtoothsoftware.com/download/techpap/2001Proceedings.pdf (page
113-114 or page 117-118 depending on which page numbers you use)

Before I try to figure out how to calculate the trial curve i wondered if there is any additional documentation anywhere or if someone's already written some code snippets, etc.

------------------------------

End of SPSSX-L Digest - 29 Aug 2007 to 30 Aug 2007 (#2007-243)
**************************************************************