Date: Fri, 18 Jun 1999 06:10:52 +0200
Reply-To: z@gallup.com
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: Zenel Batagelj <z@CATI.SI>
Subject: Re: SPSSX-L Digest - 16 Jun 1999 to 17 Jun 1999
In-Reply-To: <19990618040053.3E1FC1F003@planja.arnes.si>
Content-Type: text/plain; charset="iso-8859-2"
pri faktorskih nima screeeja
z
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:SPSSX-L@UGA.CC.UGA.EDU]On Behalf Of
> Automatic digest processor
> Sent: Friday, June 18, 1999 6:00 AM
> To: Recipients of SPSSX-L digests
> Subject: SPSSX-L Digest - 16 Jun 1999 to 17 Jun 1999
>
>
> There are 23 messages totalling 1206 lines in this issue.
>
> Topics of the day:
>
> 1. Deleting Rows (3)
> 2. sample size in paired t-test for means
> 3. Numbering rows (2)
> 4. SPSS 9.0 GLM Post Hoc (2)
> 5. Interpolated median (7)
> 6. FW: Numbering rows
> 7. help with interactions
> 8. Clustering criteria
> 9. Numbering rows, $CASENUM & other system variables
> 10. massed data to individual records
> 11. Another higher level interaction question
> 12. Conjoint Analysis
> 13. <No subject given>
>
> ----------------------------------------------------------------------
>
> Date: Thu, 17 Jun 1999 08:40:44 -0500
> From: PEDERSEN BARRY <BPEDERSE@SMTP.ISBE.STATE.IL.US>
> Subject: Deleting Rows
>
> I'm using SPSS for Windows 7.5.1. I'm removing 30,000 rows of data from a
> file of about 100,000. The data has been cleared, but the rows are still
> there with all the cells marked as missing data. How can I
> select all those
> rows for deletion without scrolling down with the mouse?
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 09:15:36 -0500
> From: Tim Dunsworth <Tim.Dunsworth@METROSTATE.EDU>
> Subject: Re: sample size in paired t-test for means
>
> The tabled df's stop at 100 simply because above that N the t
> distribution =
> stops changing appreciably with N and is asymptotically identical to the =
> normal distribution. The issue of required sample size is a
> separate one, =
> and samples larger than 100 may well be necessary if your
> expected effect =
> is small relative to the data variability and/or if you want to
> achieve a =
> high degree of confidence (ie. alpha=3D .01 or .001 instead of the usual =
> .05).
>
> >>> Shailendra Dobson <s.dobson@MARKETPROBE.CA> 06/16 10:47 AM >>>
> Is there any maximum sample size requirement for dependent samples =
> t-tests?
>
>
> The degrees of freedom in student*s t-distribution goes up to 100 (and =
> beyond t values don*t change), that makes me think that sample
> size don*t =
> have to up to 30.=20
>
>
> Any input is appreciated. Thank you.
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 09:34:05 -0500
> From: Tim Dunsworth <Tim.Dunsworth@METROSTATE.EDU>
> Subject: Re: Deleting Rows
>
> First, I am curious to know what technique you used to achieve this =
> result, since I don't think I have seen a situation quite like this =
> before. But more importantly, what I would do now is to identify
> a column =
> that will be complete for all the good cases and missing for all the bad =
> cases, then do a Data, Select operation with ~ MISSING(VarX) as the =
> criterion and with the delete unselected cases option checked instead of =
> the filter out unselected cases option, then resave the resulting
> file of =
> good cases under a new filename. HTH.
>
> >>> PEDERSEN BARRY <BPEDERSE@SMTP.ISBE.STATE.IL.US> 06/17 8:40 AM >>>
> I'm using SPSS for Windows 7.5.1. I'm removing 30,000 rows of data from a
> file of about 100,000. The data has been cleared, but the rows are still
> there with all the cells marked as missing data. How can I select all =
> those
> rows for deletion without scrolling down with the mouse?
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 10:47:44 -0500
> From: PEDERSEN BARRY <BPEDERSE@SMTP.ISBE.STATE.IL.US>
> Subject: Numbering rows
>
> Thanks to everyone who made suggestions regarding methods for
> deleting large
> numbers of rows. There are several good ways.
>
> I have one more request for today.
>
> Now, I'd like to create a variable with case numbers. In the past, I've
> created a blank variable and pasted in a number series created from a
> spreadsheet. Surely, there's a better way. Each case has only
> one row. No
> two cases share a row. I just need SPSS to insert a number
> series from the
> top of the file to the bottom, with an increment of 1.
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 09:14:05 -0700
> From: Dale Glaser <glaser@PACIFIC-SCIENCE.COM>
> Subject: Re: Numbering rows
>
> The following will do the trick:
>
> Compute id=$casenum.
> Execute.
>
> This system variable will create a permanent case sequence number
> (see p.31
> in SPSS Base 7.5 syntax reference guide)
>
>
> Dale Glaser
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:SPSSX-L@UGA.CC.UGA.EDU] On Behalf Of
> PEDERSEN BARRY
> Sent: Thursday, June 17, 1999 8:48 AM
> To: SPSSX-L@UGA.CC.UGA.EDU
> Subject: Numbering rows
>
> Thanks to everyone who made suggestions regarding methods for
> deleting large
> numbers of rows. There are several good ways.
>
> I have one more request for today.
>
> Now, I'd like to create a variable with case numbers. In the past, I've
> created a blank variable and pasted in a number series created from a
> spreadsheet. Surely, there's a better way. Each case has only
> one row. No
> two cases share a row. I just need SPSS to insert a number
> series from the
> top of the file to the bottom, with an increment of 1.
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:18:48 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: SPSS 9.0 GLM Post Hoc
>
> Whether a predictor should be treated as a covariate or a factor
> depends on
> whether it's a "continuous" or categorical predictor. Note that
> by default,
> all interactions among factors are included in the model, but only main
> effects of covariates. The POSTHOC option is not available with covariates
> because many of the methods available there are not valid with covariates.
> POSTHOC is really for the simple one way situation, as even without
> covariates, unless the data are balanced, the unadjusted marginal
> means that
> get compared with POSTHOC aren't what you want to compare anyway, at least
> in most cases. Comparisons among levels of factors in more general designs
> is handled using EMMEANS COMPARE. In the dialog boxes, if you click on the
> Options button, you can specify estimated marginal means for one or more
> factors, and check a checkbox to compare main effects (which are averaged
> over the levels of other factors if those "main effects" are contained in
> interactions in the model). You can also specify Bonferroni or Sidak
> adjustments for maintaining Type I error levels for each factor's set of
> comparisons.
>
> First, figure out what model you want to fit, then look to see how the
> software will handle it.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Michael Mallen [SMTP:mjmallen@boombox.micro.umn.edu]
> > Sent: Wednesday, June 16, 1999 2:58 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: SPSS 9.0 GLM Post Hoc
> >
> > Hello,
> >
> > I'm trying to plug in many factors into the Univariate analysis window
> > and
> > I'm running into some problems. I have a large amount of factors that I
> > would like to place into the Fixed Factor but if I put too many factors
> > in,
> > my degrees of freedom bottom out at zero because of my small
> sample size.
> > I can place the factors in Covariates but this "greys" the Post Hoc
> > options,
> > which I need for my analysis. Is there a way to place factors in the
> > Covariate window and still perform the Post Hock tests? Or is there
> > another
> > way around this type of problem? Thank you in advance.
> >
> > Michael
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 12:09:58 -0700
> From: Libin Xu <libinxy@PGCPS.ORG>
> Subject: Interpolated median
>
> Dear SPSSX-Lers,
> I need to report the interpolated median by school. The following
> command is used to generate the score:
> select school#='12114'.
> freq var=readnp/grouped=readnp/stat=median.
> I would like to have the median for each school listed with other
> scores, such as mean scale score and mean nce score. Aggregate command
> will let me generate mean nce and scale score, but not the
> interpolated median. That means I have to run the above command many
> times to get median and manually type the result in. I tried the syntax
> manual, but did not find a better solution. Does anyone on the list have
> a more efficient way of doing this?
> Thank you for your attention.
>
> Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:20:45 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: SPSS 9.0 GLM Post Hoc
>
> In the middle of that, it should read "Comparisons among levels of factors
> in more general designs are handled," not "is handled."
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Nichols, David [SMTP:nichols@spss.com]
> > Sent: Thursday, June 17, 1999 11:19 AM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Re: SPSS 9.0 GLM Post Hoc
> >
> > Whether a predictor should be treated as a covariate or a factor depends
> > on
> > whether it's a "continuous" or categorical predictor. Note that by
> > default,
> > all interactions among factors are included in the model, but only main
> > effects of covariates. The POSTHOC option is not available with
> covariates
> > because many of the methods available there are not valid with
> covariates.
> > POSTHOC is really for the simple one way situation, as even without
> > covariates, unless the data are balanced, the unadjusted marginal means
> > that
> > get compared with POSTHOC aren't what you want to compare
> anyway, at least
> > in most cases. Comparisons among levels of factors in more
> general designs
> > is handled using EMMEANS COMPARE. In the dialog boxes, if you
> click on the
> > Options button, you can specify estimated marginal means for one or more
> > factors, and check a checkbox to compare main effects (which
> are averaged
> > over the levels of other factors if those "main effects" are
> contained in
> > interactions in the model). You can also specify Bonferroni or Sidak
> > adjustments for maintaining Type I error levels for each factor's set of
> > comparisons.
> >
> > First, figure out what model you want to fit, then look to see how the
> > software will handle it.
> >
> > David Nichols
> > Principal Support Statistician and
> > Manager of Statistical Support
> > SPSS Inc.
> >
> > > -----Original Message-----
> > > From: Michael Mallen [SMTP:mjmallen@boombox.micro.umn.edu]
> > > Sent: Wednesday, June 16, 1999 2:58 PM
> > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > Subject: SPSS 9.0 GLM Post Hoc
> > >
> > > Hello,
> > >
> > > I'm trying to plug in many factors into the Univariate
> analysis window
> > > and
> > > I'm running into some problems. I have a large amount of
> factors that I
> > > would like to place into the Fixed Factor but if I put too
> many factors
> > > in,
> > > my degrees of freedom bottom out at zero because of my small sample
> > size.
> > > I can place the factors in Covariates but this "greys" the Post Hoc
> > > options,
> > > which I need for my analysis. Is there a way to place factors in the
> > > Covariate window and still perform the Post Hock tests? Or is there
> > > another
> > > way around this type of problem? Thank you in advance.
> > >
> > > Michael
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 17:22:49 +0100
> From: "R. Allan Reese" <R.A.Reese@GRI.HULL.AC.UK>
> Subject: Re: Deleting Rows
>
> On Thu, 17 Jun 1999, PEDERSEN BARRY wrote:
>
> > I'm using SPSS for Windows 7.5.1. I'm removing 30,000 rows of
> data from a
> > file of about 100,000. The data has been cleared, but the rows
> are still
> > there with all the cells marked as missing data. How can I
> select all those
> > rows for deletion without scrolling down with the mouse?
>
> Intriguing. How did you "remove" the data? Surely not by scrolling?
> A recode could set all those values missing.
>
> Solution is to use Data/Select and click the radio button for dropping
> (rather than filtering) cases. Select those that are "NOT MISSING" on
> the variable(s) of interest.
>
> In my experience it's quite common for users who enter data through the
> data editor to end up with extra, spurious, cases due to moving the
> cursor down before checking there's another case to write. This leads to
> messy reports that look "correct" to the user but mutter about having
> used N cases and "dropped M with missing values."
>
>
> R. Allan Reese Email: r.a.reese@gri.hull.ac.uk
> Associate Manager Direct voice: +44 1482 466845
> Graduate Research Institute Voice messages: +44 1482 466844
> Hull University, Hull HU6 7RX, UK. Fax: +44 1482 466846
> ====================================================================
> This place is the Devil, or at least his principal residence,
> they call it the University. (Byron, letter 1805)
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 19:43:48 +0300
> From: Sorin Sion <Sorin.Sion@CONNEX.RO>
> Subject: FW: Numbering rows
>
> Try this, in the syntax window:
> compute v_temp1=1.
> execute.
> create
> v_temp2=csum(v_temp1).
> execute.
>
> I have SPSS8. I don't know if it works on prior versions.
>
> Sorin SION
>
> > -----Original Message-----
> > From: PEDERSEN BARRY [SMTP:BPEDERSE@SMTP.ISBE.STATE.IL.US]
> > Sent: Thursday, June 17, 1999 6:48
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Numbering rows
> >
> > Thanks to everyone who made suggestions regarding methods for deleting
> > large
> > numbers of rows. There are several good ways.
> >
> > I have one more request for today.
> >
> > Now, I'd like to create a variable with case numbers. In the past, I've
> > created a blank variable and pasted in a number series created from a
> > spreadsheet. Surely, there's a better way. Each case has only one row.
> > No
> > two cases share a row. I just need SPSS to insert a number series from
> > the
> > top of the file to the bottom, with an increment of 1.
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:49:10 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: help with interactions
>
> The two way interaction A*B is nested within the levels of C. From what
> you're saying, I'm assuming this is all within subjects or repeated
> measures. You would use the MMATRIX subcommand to specify the desired
> comparisons among the means. For simplicity, I'll assume that it's a 2x2x2
> and you want to look at the A*B effect within each level of C. You'd have
> something along the lines of
>
> GLM Y1 TO Y8
> /WSFACTORS A 2 B 2 C 2
> /MMATRIX 'A*B WITHIN C(1)' ALL 1 0 -1 0 -1 0 1 0
> 'A*B WITHIN C(2)' ALL 0 1 0 -1 0 -1 0 1.
>
> This will give results for each individual contrast, as well as an omnibus
> multivariate test for the set. If you have multiple degrees of freedom for
> the A*B interaction within each level of C, you might have to run multiple
> GLM commands, because you can only use one MMATRIX subcommand, and this
> wouldn't produce an omnibus test for A*B within C(1) or within C(2)
> separately. If you wanted that, you'd have to run GLM twice.
>
> Simple main effects or simple, simple main effects (A within the
> levels of B
> or B*C, for example) are more easily handled using EMMEANS COMPARE.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Michael Dennis [SMTP:awkmrd@GTE.NET]
> > Sent: Wednesday, June 16, 1999 3:39 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: help with interactions
> >
> > Can anybody out there please tell me how to, using GLM Repeated Measures
> > on
> > SPSS 8.0 for Windows, test for a 2-way interaction (A*B) AS
> NESTED WITHIN
> > a
> > 3-way interaction (A*B*C)?
> >
> > Thanks, Mike Dennis (Purdue University)
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:50:55 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: Clustering criteria
>
> I'm not aware of any SPSS macros for computing indices of validity for
> clustering solutions.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: professionale [SMTP:flebus@TIN.IT]
> > Sent: Wednesday, June 16, 1999 4:32 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: R:Clustering criteria
> >
> > In a1985 article in Psychometrika Millingan and Cooper published a
> > comprehensive review of thirty criteria to assess the validity of
> > clustering solutions. Could anybody let me know if there are
> SPSS macros
> > to
> > compute those clustering methods, possibly the best ones, like
> the cubic
> > clustering criteria ?
> > Or else, can anybody let me know what criteria he/she uses in
> clustering
> > subjects using answers to mental tests or questionnaires?
> >
> > Many thanks.
> >
> > Gian Battista Flebus
> > Istituto di Psicologia
> > Universit` di URBINO
> > <flebus@tin.it>
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:51:56 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: Interpolated median
>
> I think you'd have to get into scripting or custom programming computation
> of medians to get something more efficient.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > Sent: Thursday, June 17, 1999 2:10 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Interpolated median
> >
> > Dear SPSSX-Lers,
> > I need to report the interpolated median by school. The
> following
> > command is used to generate the score:
> > select school#='12114'.
> > freq var=readnp/grouped=readnp/stat=median.
> > I would like to have the median for each school listed
> with other
> > scores, such as mean scale score and mean nce score. Aggregate command
> > will let me generate mean nce and scale score, but not the
> > interpolated median. That means I have to run the above command many
> > times to get median and manually type the result in. I tried the syntax
> > manual, but did not find a better solution. Does anyone on the list have
> > a more efficient way of doing this?
> > Thank you for your attention.
> >
> > Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 11:53:18 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: Interpolated median
>
> What I meant by using scripting was reading the median values from the
> tables rather than taking them out "by hand."
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Nichols, David [SMTP:nichols@spss.com]
> > Sent: Thursday, June 17, 1999 11:52 AM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Re: Interpolated median
> >
> > I think you'd have to get into scripting or custom programming
> computation
> > of medians to get something more efficient.
> >
> > David Nichols
> > Principal Support Statistician and
> > Manager of Statistical Support
> > SPSS Inc.
> >
> > > -----Original Message-----
> > > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > > Sent: Thursday, June 17, 1999 2:10 PM
> > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > Subject: Interpolated median
> > >
> > > Dear SPSSX-Lers,
> > > I need to report the interpolated median by school. The
> > following
> > > command is used to generate the score:
> > > select school#='12114'.
> > > freq var=readnp/grouped=readnp/stat=median.
> > > I would like to have the median for each school listed with
> > other
> > > scores, such as mean scale score and mean nce score. Aggregate command
> > > will let me generate mean nce and scale score, but not the
> > > interpolated median. That means I have to run the above command many
> > > times to get median and manually type the result in. I tried
> the syntax
> > > manual, but did not find a better solution. Does anyone on
> the list have
> > > a more efficient way of doing this?
> > > Thank you for your attention.
> > >
> > > Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 20:02:58 +0300
> From: Sorin Sion <Sorin.Sion@CONNEX.RO>
> Subject: Re: Numbering rows, $CASENUM & other system variables
>
> Thanks. I forgot this.
>
> Btw, does anybody know other system variables than
> $CASENUM, $SYSMIS, $JDATE, $DATE, $TIME, $LENGTH, $WIDTH ?
>
>
>
> > -----Original Message-----
> > From: stevenp@greatorganizations.com
> [SMTP:stevenp@greatorganizations.com]
> > Sent: Thursday, June 17, 1999 7:53
> > To: 'Sorin Sion'
> > Subject: RE: Numbering rows
> >
> > Sorin: there is a much easier way to assign case numbers.
> >
> > Compute ID = $CASENUM.
> > Formats ID (f8.0).
> > Execute.
> >
> > $CASENUM is a reserved variable in SPSS. It will assign the
> row number as
> > an ID.
> >
> >
> > Steven M. Palmer
> > Partner
> > Center for Organization Effectiveness
> > StevenP@GreatOrganizations.com
> > www.GreatOrganizations.com
> > 608-833-3332 x:28
> >
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 13:20:24 -0700
> From: Libin Xu <libinxy@PGCPS.ORG>
> Subject: Re: Interpolated median
>
> Thank you for the suggestion.
> Unfortunately, I don't know how to write a script for this kind of
> problem. I have no idea if there is a syntax guide for script writing. I
> did see a message a few days ago using script to copy data from a pivot
> table and paste the data to word. But I don't see the possibility of
> modifying the script for my problem.
>
> Libin Xu
>
>
> "Nichols, David" wrote:
> >
> > What I meant by using scripting was reading the median values from the
> > tables rather than taking them out "by hand."
> >
> > David Nichols
> > Principal Support Statistician and
> > Manager of Statistical Support
> > SPSS Inc.
> >
> > > -----Original Message-----
> > > From: Nichols, David [SMTP:nichols@spss.com]
> > > Sent: Thursday, June 17, 1999 11:52 AM
> > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > Subject: Re: Interpolated median
> > >
> > > I think you'd have to get into scripting or custom
> programming computation
> > > of medians to get something more efficient.
> > >
> > > David Nichols
> > > Principal Support Statistician and
> > > Manager of Statistical Support
> > > SPSS Inc.
> > >
> > > > -----Original Message-----
> > > > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > > > Sent: Thursday, June 17, 1999 2:10 PM
> > > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > > Subject: Interpolated median
> > > >
> > > > Dear SPSSX-Lers,
> > > > I need to report the interpolated median by school. The
> > > following
> > > > command is used to generate the score:
> > > > select school#='12114'.
> > > > freq var=readnp/grouped=readnp/stat=median.
> > > > I would like to have the median for each school listed with
> > > other
> > > > scores, such as mean scale score and mean nce score.
> Aggregate command
> > > > will let me generate mean nce and scale score, but not the
> > > > interpolated median. That means I have to run the above command many
> > > > times to get median and manually type the result in. I
> tried the syntax
> > > > manual, but did not find a better solution. Does anyone on
> the list have
> > > > a more efficient way of doing this?
> > > > Thank you for your attention.
> > > >
> > > > Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 12:39:01 -0500
> From: "Nichols, David" <nichols@SPSS.COM>
> Subject: Re: Interpolated median
>
> You're welcome. I'd suggest checking the Script eXchange to see if there
> isn't something designed to do what you need to do, that's at
> least only in
> need of very slight modification. The URL is
> http://www.spss.com/software/spss/scriptexchange/.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > Sent: Thursday, June 17, 1999 3:20 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Re: Interpolated median
> >
> > Thank you for the suggestion.
> > Unfortunately, I don't know how to write a script for
> this kind of
> > problem. I have no idea if there is a syntax guide for script writing. I
> > did see a message a few days ago using script to copy data from a pivot
> > table and paste the data to word. But I don't see the possibility of
> > modifying the script for my problem.
> >
> > Libin Xu
> >
> >
> > "Nichols, David" wrote:
> > >
> > > What I meant by using scripting was reading the median values from the
> > > tables rather than taking them out "by hand."
> > >
> > > David Nichols
> > > Principal Support Statistician and
> > > Manager of Statistical Support
> > > SPSS Inc.
> > >
> > > > -----Original Message-----
> > > > From: Nichols, David [SMTP:nichols@spss.com]
> > > > Sent: Thursday, June 17, 1999 11:52 AM
> > > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > > Subject: Re: Interpolated median
> > > >
> > > > I think you'd have to get into scripting or custom programming
> > computation
> > > > of medians to get something more efficient.
> > > >
> > > > David Nichols
> > > > Principal Support Statistician and
> > > > Manager of Statistical Support
> > > > SPSS Inc.
> > > >
> > > > > -----Original Message-----
> > > > > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > > > > Sent: Thursday, June 17, 1999 2:10 PM
> > > > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > > > Subject: Interpolated median
> > > > >
> > > > > Dear SPSSX-Lers,
> > > > > I need to report the interpolated median by school. The
> > > > following
> > > > > command is used to generate the score:
> > > > > select school#='12114'.
> > > > > freq var=readnp/grouped=readnp/stat=median.
> > > > > I would like to have the median for each school
> listed with
> > > > other
> > > > > scores, such as mean scale score and mean nce score. Aggregate
> > command
> > > > > will let me generate mean nce and scale score, but not the
> > > > > interpolated median. That means I have to run the above
> command many
> > > > > times to get median and manually type the result in. I tried the
> > syntax
> > > > > manual, but did not find a better solution. Does anyone
> on the list
> > have
> > > > > a more efficient way of doing this?
> > > > > Thank you for your attention.
> > > > >
> > > > > Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 12:40:02 -0500
> From: "Matheson, David" <davidm@SPSS.COM>
> Subject: Re: Interpolated median
>
> Libin,
> I've pasted a solution below that was written by a coworker here at SPSS.
> As David Nichols noted, in SPSS 7.5 or later versions you could
> also write a
> script that would extract the medians from the output.
> You can see some script examples at
> http://www.spss.com/software/spss/scriptexchange/ .
> If you decide to go with scripting, you can avoid the multiple runs of
> select-frequencies command sets by using the split file command. Rather
> than:
>
> select if school#='12114'.
> freq var=readnp/grouped=readnp/stat=median.
>
> Try:
>
> sort cases by school#.
> split file by school#.
> freq var=readnp/grouped=readnp/stat=median.
> split file off.
>
> With this setup, you only need to run the single FREQUENCIES
> command and you
> will get a separate output display for each value of school#. If
> your file
> is already sorted by school#, you can skip the SORT CASES command.
>
> David Matheson
> SPSS Technical Support
>
>
> Here is the approach using SPSS command syntax:
>
>
> Q.
> The AGGREGATE procedure doesn't seem to have a summary function
> which will return medians within values of the BREAK group. Is there
> no way for me to do this in SPSS?
>
>
> A.
> SPSS's AGGREGATE procedure indeed does not have a median function, but
> there are ways to determine subgroup medians within SPSS. Note that the
> RMV (Replace Missing Values) procedure has a MEDIAN function which
> calculates medians across non-missing values within a variable and
> substitutes this value for any of its missing values. Coupled with the
> SPLIT FILE procedure, which can direct SPSS to repeat this process
> separately for each value within your intended BREAK group, you can
> affect the type of results an AGGREGATE median function might perform.
> The tricky part is forcing a known case with a missing value on the target
> variable into each category of the grouping variable. This is actually
> quite easily done with a combination of the AGGREGATE procedure itself and
> ADD FILES. Below is an annotated example illustrating the entire process,
> which you can copy and paste to an SPSS syntax window. It's instructive
> to run it in the sections outlined, so you can observe the changes to the
> active file and better apply the technique to your own data.
>
>
> ** First we read in some data values -- three "id" groups
> ** with readily identifiable medians on "score".
>
> DATA LIST FREE /id score.
> BEGIN DATA
> 1 1 1 2 1 3 1 4 1 5
> 2 1 2 3 2 5 2 7 2 9
> 3 2 3 4 3 6 3 8 3 10
> END DATA.
>
>
> ** Next we aggregate the file, breaking on the id variable.
>
> AGGREGATE OUTFILE 'aggmedian.sav'/
> PRESORTED/ BREAK=id/ nscore=N(score).
>
>
> ** We then add this aggregated outfile back to the original
> ** data file, interlacing the agreggated cases (1 per id)
> ** among their associated id subgroups, and creating a new
> ** variable (aggfile) to flag the cases coming from this
> ** aggregated file.
>
> ADD FILES FILE 'aggmedian.sav'/IN=aggfile/FILE */BY id.
> exe.
>
>
> ** Next we split the file by id, so that our next procedure
> ** is repeated for each id group.
>
> SPLIT FILE BY id.
>
>
> ** At this point, we know that "score" is missing for (at
> ** least) one case within each split file group; we've put this case
> ** there with the aggregation and re-merging of the file. We can now
> ** use the RMV procedure to substitute for that missing value
> ** the median across the remaining values for "score" (within
> ** each id subgroup, since SPLIT FILE is in effect).
>
> RMV medscore=MEDIAN(score, ALL).
>
>
> ** Only the flagged, "aggfile" cases hold the median values at
> ** this point; we still need to substitute the subgroup medians
> ** in "medscore" for each case within each id subgroup.
>
> IF id=LAG(id) medscore=LAG(medscore).
> EXE.
>
>
> ** There's no need to retain the cases that were generated
> ** by the AGGREGATE and ADD FILES and used solely to hold the
> ** id subgroup medians. We can select everything but them for
> ** the active file.
>
> SELECT IF NOT(aggfile).
> EXE.
>
> ** At this stage, each case in the file has a value for the
> ** "medscore" variable appropriate to its id subgroup. If this
> ** is sufficient, you can drop the "nscore" and "aggfile"
> ** variables the next time the file is saved. If your original
> ** aim was indeed an aggregated file with one case per id and
> ** the median within id subgroups on "score," you now have
> ** everything you need for this. Several of the AGGREGATE
> ** functions will capture the "medscore" values, which are
> ** effectively constants within each break group.
>
> AGGREGATE OUTFILE */BREAK=id/aggmed=MAX(medscore).
>
>
> -----Original Message-----
> From: Libin Xu [mailto:libinxy@PGCPS.ORG]
> Sent: Thursday, June 17, 1999 2:10 PM
> To: SPSSX-L@UGA.CC.UGA.EDU
> Subject: Interpolated median
>
>
> Dear SPSSX-Lers,
> I need to report the interpolated median by school. The following
> command is used to generate the score:
> select school#='12114'.
> freq var=readnp/grouped=readnp/stat=median.
> I would like to have the median for each school listed with other
> scores, such as mean scale score and mean nce score. Aggregate command
> will let me generate mean nce and scale score, but not the
> interpolated median. That means I have to run the above command many
> times to get median and manually type the result in. I tried the syntax
> manual, but did not find a better solution. Does anyone on the list have
> a more efficient way of doing this?
> Thank you for your attention.
>
> Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 12:45:31 -0500
> From: "Matheson, David" <davidm@SPSS.COM>
> Subject: Re: Interpolated median
>
> Also, scripts are described in the recent SPSS Base User's Guides.
>
> -----Original Message-----
> From: Nichols, David [mailto:nichols@spss.com]
> Sent: Thursday, June 17, 1999 12:39 PM
> To: SPSSX-L@UGA.CC.UGA.EDU
> Subject: Re: Interpolated median
>
>
> You're welcome. I'd suggest checking the Script eXchange to see if there
> isn't something designed to do what you need to do, that's at
> least only in
> need of very slight modification. The URL is
> http://www.spss.com/software/spss/scriptexchange/.
>
> David Nichols
> Principal Support Statistician and
> Manager of Statistical Support
> SPSS Inc.
>
> > -----Original Message-----
> > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > Sent: Thursday, June 17, 1999 3:20 PM
> > To: SPSSX-L@UGA.CC.UGA.EDU
> > Subject: Re: Interpolated median
> >
> > Thank you for the suggestion.
> > Unfortunately, I don't know how to write a script for
> this kind of
> > problem. I have no idea if there is a syntax guide for script writing. I
> > did see a message a few days ago using script to copy data from a pivot
> > table and paste the data to word. But I don't see the possibility of
> > modifying the script for my problem.
> >
> > Libin Xu
> >
> >
> > "Nichols, David" wrote:
> > >
> > > What I meant by using scripting was reading the median values from the
> > > tables rather than taking them out "by hand."
> > >
> > > David Nichols
> > > Principal Support Statistician and
> > > Manager of Statistical Support
> > > SPSS Inc.
> > >
> > > > -----Original Message-----
> > > > From: Nichols, David [SMTP:nichols@spss.com]
> > > > Sent: Thursday, June 17, 1999 11:52 AM
> > > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > > Subject: Re: Interpolated median
> > > >
> > > > I think you'd have to get into scripting or custom programming
> > computation
> > > > of medians to get something more efficient.
> > > >
> > > > David Nichols
> > > > Principal Support Statistician and
> > > > Manager of Statistical Support
> > > > SPSS Inc.
> > > >
> > > > > -----Original Message-----
> > > > > From: Libin Xu [SMTP:libinxy@PGCPS.ORG]
> > > > > Sent: Thursday, June 17, 1999 2:10 PM
> > > > > To: SPSSX-L@UGA.CC.UGA.EDU
> > > > > Subject: Interpolated median
> > > > >
> > > > > Dear SPSSX-Lers,
> > > > > I need to report the interpolated median by school. The
> > > > following
> > > > > command is used to generate the score:
> > > > > select school#='12114'.
> > > > > freq var=readnp/grouped=readnp/stat=median.
> > > > > I would like to have the median for each school
> listed with
> > > > other
> > > > > scores, such as mean scale score and mean nce score. Aggregate
> > command
> > > > > will let me generate mean nce and scale score, but not the
> > > > > interpolated median. That means I have to run the above
> command many
> > > > > times to get median and manually type the result in. I tried the
> > syntax
> > > > > manual, but did not find a better solution. Does anyone
> on the list
> > have
> > > > > a more efficient way of doing this?
> > > > > Thank you for your attention.
> > > > >
> > > > > Libin Xu
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 14:25:41 -0700
> From: Dale Glaser <glaser@PACIFIC-SCIENCE.COM>
> Subject: massed data to individual records
>
> Possibly there is some easy algorithm to get at the following, but my
> perusal of Defining Complex Files in the Base Manual did not quite get at
> what I need....any help greatly appreciated:
>
> I have received data as follows:
>
> Subject sup4 notsup4 ambig4 sup2 notsup2 ambig2
> 1 0 0 4 2 0
> 0
> 2 0 1 3 1 1
> 0
> 3 1 1 2 2 0
> 0
>
> sup=support
> notsup=not support
> amibig=ambiguous
>
> The numbers following the ascription in the variable name are the
> number of
> scenarios presented....I won't delve into the experimental
> context, but in a
> nutshell subject #1 receives four descriptions of a scenario, of
> which they
> rate each description as either being supporting, not supporting, or
> ambiguous; this is a completely within-subject design, so in this
> sample of
> data, subject #1 was presented with four situations and rated
> four as being
> ambiguous; subsequently they are given two situations (related to the same
> scenario) and rate them again (in this instance s#1 rates them
> both as being
> supportive); the above datapoints are for one scenario, so they
> actually had
> 4 scenarios each........we were interested at looking at the type of
> scenario x rating interaction (e.g., 4 x 3 chi-square), however,
> beyond the
> experimental logistics that may advise against this type of
> analysis (i.e.,
> nonindependence of data), this analysis was mainly a check after
> the fact to
> see if any descriptions are particularly problematic (i.e., ambiguous);
> well, unfortunately the data came to me in the above fashion
> meaning it was
> aggregated with each cell representing a sum (e.g., subject #3 said two
> descriptions were ambiguous for the '4' condition [ambig4]), instead of
> being true count data; I know how to use syntax to convert
> multiple records
> into one (e.g., vector outside a loop structure), but I am having troubles
> using some type of 'repeating data' command to create multiple records for
> each subject, and be able to convert the massed count for each cell into a
> RATING type variable, as described below......so, ideally my data should
> look like the following (for one subject and two contiguous scenarios):
>
> RATING
> Support=1
> Not support=2
> Ambiguous=3
>
> SCENARIO
> 4 descriptions=1
> 2 descriptions=2
>
>
> Subject Rating Scenario
> 1 3 1
> 1 3 1
> 1 3 1
> 1 3 1
> 1 1 2
> 1 1 2
>
> Are there any aggregate or looped commands that may remediate my
> perpelexity?
>
>
> Thank you for any assistance
>
> Dale Glaser, Ph.D.
> Pacific Science & Engineering/SDSU
> San Diego, CA
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 18:31:25 -0400
> From: Angela Ebreo <aebreo2@POP.UKY.EDU>
> Subject: Another higher level interaction question
>
> I have a 4-way interaction that is sig., along with a 2 way that is also
> sig. Can someone tell me different ways of testing whether the 2-way
> interaction differs across the other 2 conditions? All factors are
> between-ss, 2-levels.
>
> Thank you,
> Angela Ebreo
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 17:13:13 -0700
> From: Margaret Wells <meg_wells@YAHOO.COM>
> Subject: Conjoint Analysis
>
> I only have access to SPSS 6.1 professional statistics package and
> would like to run a conjoint analysis. Is it possible with this
> software? If not, what do I need to be updated to?
> Thanks in advance,
> Meg Wells
>
>
> _________________________________________________________
> Do You Yahoo!?
> Get your free @yahoo.com address at http://mail.yahoo.com
>
> ------------------------------
>
> Date: Thu, 17 Jun 1999 20:59:40 -0400
> From: Raynald Levesque <rlevesque@VIDEOTRON.CA>
> Subject: <No subject given>
>
> Hi Dale and listmembers!
>
> Using the variable names given in your example, the following syntax does
> the job:
>
> VECTOR v_rate(6) v_scen(6)v_data=sup4 TO ambig2.
> COMPUTE #done=0.
> LOOP #A=1 TO 6.
> DO IF v_data(#A)>0.
> LOOP #B=1 TO v_data(#A).
> COMPUTE v_rate(#done+#B)=#A - 3*(#A>3).
> END LOOP.
> COMPUTE #done=#done + v_data(#A).
> END IF.
> END LOOP.
>
> * now transform data from 1 record per subject to 6 records per subject.
> LOOP #cnt=1 TO 6.
> COMPUTE rating=v_rate(#cnt).
> COMPUTE scenario=1+(#cnt>4).
> XSAVE OUTFILE='temp.sav' /KEEP subject rating scenario.
> END LOOP.
> EXECUTE.
> GET
> FILE='temp.sav'.
> FORMATS subject TO scenario(F8.0).
> EXECUTE.
>
> Raynald Levesque
>
>
> ----- Original Message -----
> From: Dale Glaser <glaser@PACIFIC-SCIENCE.COM>
> Newsgroups: bit.listserv.spssx-l
> To: <SPSSX-L@UGA.CC.UGA.EDU>
> Sent: Thursday, June 17, 1999 5:25 PM
> Subject: massed data to individual records
>
>
> > Possibly there is some easy algorithm to get at the following, but my
> > perusal of Defining Complex Files in the Base Manual did not
> quite get at
> > what I need....any help greatly appreciated:
> >
> > I have received data as follows:
> >
> > Subject sup4 notsup4 ambig4 sup2 notsup2 ambig2
> > 1 0 0 4 2 0
> > 0
> > 2 0 1 3 1 1
> > 0
> > 3 1 1 2 2 0
> > 0
> >
> > sup=support
> > notsup=not support
> > amibig=ambiguous
> >
> > The numbers following the ascription in the variable name are the number
> of
> > scenarios presented....I won't delve into the experimental
> context, but in
> a
> > nutshell subject #1 receives four descriptions of a scenario, of which
> they
> > rate each description as either being supporting, not supporting, or
> > ambiguous; this is a completely within-subject design, so in this sample
> of
> > data, subject #1 was presented with four situations and rated four as
> being
> > ambiguous; subsequently they are given two situations (related
> to the same
> > scenario) and rate them again (in this instance s#1 rates them both as
> being
> > supportive); the above datapoints are for one scenario, so they actually
> had
> > 4 scenarios each........we were interested at looking at the type of
> > scenario x rating interaction (e.g., 4 x 3 chi-square), however, beyond
> the
> > experimental logistics that may advise against this type of analysis
> (i.e.,
> > nonindependence of data), this analysis was mainly a check
> after the fact
> to
> > see if any descriptions are particularly problematic (i.e., ambiguous);
> > well, unfortunately the data came to me in the above fashion meaning it
> was
> > aggregated with each cell representing a sum (e.g., subject #3 said two
> > descriptions were ambiguous for the '4' condition [ambig4]), instead of
> > being true count data; I know how to use syntax to convert multiple
> records
> > into one (e.g., vector outside a loop structure), but I am
> having troubles
> > using some type of 'repeating data' command to create multiple
> records for
> > each subject, and be able to convert the massed count for each
> cell into a
> > RATING type variable, as described below......so, ideally my data should
> > look like the following (for one subject and two contiguous scenarios):
> >
> > RATING
> > Support=1
> > Not support=2
> > Ambiguous=3
> >
> > SCENARIO
> > 4 descriptions=1
> > 2 descriptions=2
> >
> >
> > Subject Rating Scenario
> > 1 3 1
> > 1 3 1
> > 1 3 1
> > 1 3 1
> > 1 1 2
> > 1 1 2
> >
> > Are there any aggregate or looped commands that may remediate my
> > perpelexity?
> >
> >
> > Thank you for any assistance
> >
> > Dale Glaser, Ph.D.
> > Pacific Science & Engineering/SDSU
> > San Diego, CA
>
> ------------------------------
>
> End of SPSSX-L Digest - 16 Jun 1999 to 17 Jun 1999
> **************************************************
>
| 
