SPSSX Discussion

percentiles

Classic

List

Threaded

11 messages Options

Rodrigo Briceño

percentiles

Dear SPSS users.
I have a database with some groups of data (groups based on categories
of patients). For each group of patients I must estimate the percentage
of outliers (using the rule of percentile 25-1.5 IQR and percentile 75+
1.5 IQR).
For that purpose I need to find how to construct the 25th percentile and
the 75th percentile in each group of patient, and get these values as
new variables. I found many references to percentiles in frequencies,
tables, explore and other SPSS commands, but nothing about what function
to use in order to COMPUTE these values as new variables.
Anyone could help me please?

Thanks for your time.
--

__________________________________________________________________

*Rodrigo Briceño*
Project Manager
Sanigest Internacional

+506 22-91-12-00 ext. 113 * Oficina* *Costa Rica*
+506 22-32-08-30 *Fax
*+506 88-86-11-77* Celular**
*[hidden email] <mailto:[hidden email]>
www.sanigest.com <http://www.sanigest.com/>

MSN: [hidden email] <mailto:[hidden email]>
SKYPE: rbriceno1087

_____________________

This communication contains legal information which is privileged and
confidential. It is for the exclusive use of the address and
distribution, dissemination, copying or use by others is strictly
prohibited. If you have received this communication by error, please
delete the original message and e-mail us.

Esta comunicación contiene información legal privilegiada y confidencial
para el uso exclusivo del destinatario. La distribución, diseminación,
copia u otro uso por terceras personas es estrictamente prohibida. Si
usted ha recibido esta comunicación por error, le rogamos borrar el
mensaje original y comunicárnoslo a esta misma dirección.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Marta Garcia-Granero

Re: percentiles

Rodrigo Briceño escribió:

> Dear SPSS users.
> I have a database with some groups of data (groups based on categories
> of patients). For each group of patients I must estimate the percentage
> of outliers (using the rule of percentile 25-1.5 IQR and percentile 75+
> 1.5 IQR).
> For that purpose I need to find how to construct the 25th percentile and
> the 75th percentile in each group of patient, and get these values as
> new variables. I found many references to percentiles in frequencies,
> tables, explore and other SPSS commands, but nothing about what function
> to use in order to COMPUTE these values as new variables.
> Anyone could help me please?
>
> Thanks for your time.

You will find a code that it's close to what you want (except that it
cleans the outliers, instead of counting them) if you search the
archived messages at

http://listserv.uga.edu/archives/spssx-l.html

And look for a message I sent in september 24th with the title "Re:
IQR&outliers"

At least, it will save P25&P75, and compute the so called inner fences
(+/- 1.5IQR). You can then modify it to flag the outliers (instead of
eliminating then).

I'm leaving right now, be back in a couple of hours. I'll help you then
if you are still stuck

Marta García-Granero

> --
>
> __________________________________________________________________
>
> *Rodrigo Briceño*
> Project Manager
> Sanigest Internacional
>
> +506 22-91-12-00 ext. 113 * Oficina* *Costa Rica*
> +506 22-32-08-30 *Fax
> *+506 88-86-11-77* Celular**
> *[hidden email] <mailto:[hidden email]>
> www.sanigest.com <http://www.sanigest.com/>
>
> MSN: [hidden email] <mailto:[hidden email]>
> SKYPE: rbriceno1087
>
> _____________________
>
> This communication contains legal information which is privileged and
> confidential. It is for the exclusive use of the address and
> distribution, dissemination, copying or use by others is strictly
> prohibited. If you have received this communication by error, please
> delete the original message and e-mail us.
>
>
> Esta comunicación contiene información legal privilegiada y confidencial
> para el uso exclusivo del destinatario. La distribución, diseminación,
> copia u otro uso por terceras personas es estrictamente prohibida. Si
> usted ha recibido esta comunicación por error, le rogamos borrar el
> mensaje original y comunicárnoslo a esta misma dirección.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

--
For miscellaneous statistical stuff, visit:
http://gjyp.nl/marta/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Oliver, Richard

Re: percentiles

In reply to this post by Rodrigo Briceño

One solution is to use OMS to extract the results from Frequencies, as this example from the Programming and Data Management book illustrates:

GET FILE='Employee data.sav'.
PRESERVE.
SET TVARS NAMES TNUMBERS VALUES.
DATASET DECLARE freq_table.
***split file by job category to get group percentiles.
SORT CASES BY jobcat.
SPLIT FILE LAYERED BY jobcat.
OMS
/SELECT TABLES
/IF COMMANDS=['Frequencies'] SUBTYPES=['Statistics']
/DESTINATION FORMAT=SAV
OUTFILE='freq_table'
/COLUMNS SEQUENCE=[L1 R2].
FREQUENCIES
VARIABLES=salary
/FORMAT=NOTABLE
/PERCENTILES= 25 50 75.
OMSEND.
***restore previous SET settings.
RESTORE.
MATCH FILES FILE=*
/TABLE='freq_table'
/rename (Var1=jobcat)
/BY jobcat
/DROP command_ TO salary_Missing.
EXECUTE.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Rodrigo Briceño
Sent: Wednesday, November 05, 2008 10:32 AM
To: [hidden email]
Subject: percentiles

Dear SPSS users.
I have a database with some groups of data (groups based on categories
of patients). For each group of patients I must estimate the percentage
of outliers (using the rule of percentile 25-1.5 IQR and percentile 75+
1.5 IQR).
For that purpose I need to find how to construct the 25th percentile and
the 75th percentile in each group of patient, and get these values as
new variables. I found many references to percentiles in frequencies,
tables, explore and other SPSS commands, but nothing about what function
to use in order to COMPUTE these values as new variables.
Anyone could help me please?

Thanks for your time.
--

__________________________________________________________________

*Rodrigo Briceño*
Project Manager
Sanigest Internacional

+506 22-91-12-00 ext. 113 * Oficina* *Costa Rica*
+506 22-32-08-30 *Fax
*+506 88-86-11-77* Celular**
*[hidden email] <mailto:[hidden email]>
www.sanigest.com <http://www.sanigest.com/>

MSN: [hidden email] <mailto:[hidden email]>
SKYPE: rbriceno1087

_____________________

This communication contains legal information which is privileged and
confidential. It is for the exclusive use of the address and
distribution, dissemination, copying or use by others is strictly
prohibited. If you have received this communication by error, please
delete the original message and e-mail us.

Esta comunicación contiene información legal privilegiada y confidencial
para el uso exclusivo del destinatario. La distribución, diseminación,
copia u otro uso por terceras personas es estrictamente prohibida. Si
usted ha recibido esta comunicación por error, le rogamos borrar el
mensaje original y comunicárnoslo a esta misma dirección.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Albert-Jan Roskam

Re: percentiles

Hi,

This is one way to calculate ntiles:
rank variables = salary / ntiles (100) into salary_percentile.

Cheers!!
Albert-Jan

--- On Wed, 11/5/08, Oliver, Richard <[hidden email]> wrote:

> From: Oliver, Richard <[hidden email]>
> Subject: Re: percentiles
> To: [hidden email]
> Date: Wednesday, November 5, 2008, 6:02 PM
> One solution is to use OMS to extract the results from
> Frequencies, as this example from the Programming and Data
> Management book illustrates:
>
> GET FILE='Employee data.sav'.
> PRESERVE.
> SET TVARS NAMES TNUMBERS VALUES.
> DATASET DECLARE freq_table.
> ***split file by job category to get group percentiles.
> SORT CASES BY jobcat.
> SPLIT FILE LAYERED BY jobcat.
> OMS
> /SELECT TABLES
> /IF COMMANDS=['Frequencies']
> SUBTYPES=['Statistics']
> /DESTINATION FORMAT=SAV
> OUTFILE='freq_table'
> /COLUMNS SEQUENCE=[L1 R2].
> FREQUENCIES
> VARIABLES=salary
> /FORMAT=NOTABLE
> /PERCENTILES= 25 50 75.
> OMSEND.
> ***restore previous SET settings.
> RESTORE.
> MATCH FILES FILE=*
> /TABLE='freq_table'
> /rename (Var1=jobcat)
> /BY jobcat
> /DROP command_ TO salary_Missing.
> EXECUTE.
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]]
> On Behalf Of Rodrigo Briceño
> Sent: Wednesday, November 05, 2008 10:32 AM
> To: [hidden email]
> Subject: percentiles
>
> Dear SPSS users.
> I have a database with some groups of data (groups based on
> categories
> of patients). For each group of patients I must estimate
> the percentage
> of outliers (using the rule of percentile 25-1.5 IQR and
> percentile 75+
> 1.5 IQR).
> For that purpose I need to find how to construct the 25th
> percentile and
> the 75th percentile in each group of patient, and get these
> values as
> new variables. I found many references to percentiles in
> frequencies,
> tables, explore and other SPSS commands, but nothing about
> what function
> to use in order to COMPUTE these values as new variables.
> Anyone could help me please?
>
> Thanks for your time.
> --
>
> __________________________________________________________________
>
> *Rodrigo Briceño*
> Project Manager
> Sanigest Internacional
>
> +506 22-91-12-00 ext. 113 * Oficina* *Costa Rica*
> +506 22-32-08-30 *Fax
> *+506 88-86-11-77* Celular**
> *[hidden email] <mailto:[hidden email]>
> www.sanigest.com <http://www.sanigest.com/>
>
> MSN: [hidden email] <mailto:[hidden email]>
> SKYPE: rbriceno1087
>
> _____________________
>
> This communication contains legal information which is
> privileged and
> confidential. It is for the exclusive use of the address
> and
> distribution, dissemination, copying or use by others is
> strictly
> prohibited. If you have received this communication by
> error, please
> delete the original message and e-mail us.
>
>
> Esta comunicación contiene información legal privilegiada
> y confidencial
> para el uso exclusivo del destinatario. La distribución,
> diseminación,
> copia u otro uso por terceras personas es estrictamente
> prohibida. Si
> usted ha recibido esta comunicación por error, le rogamos
> borrar el
> mensaje original y comunicárnoslo a esta misma dirección.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body
> text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the
> command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body
> text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the
> command
> INFO REFCARD

Oliver, Richard

Re: percentiles

Rank will generate a new variable that indicates which percentile group each case belongs in, but it will not give you the values that define the percentile cutpoints. And when you rank within categories of some grouping variable, cases in different groups may have the same percentile ranks, but those ranks are based on different cutpoints. The example provided from the programming and data management book generates new variables with the 25th, 50th, and 75th percentile values for each case, and those values for each case vary based on what group the case belongs in.

-----Original Message-----
From: Albert-jan Roskam [mailto:[hidden email]]
Sent: Thursday, November 06, 2008 11:32 AM
To: [hidden email]; Oliver, Richard
Subject: Re: percentiles

Hi,

This is one way to calculate ntiles:
rank variables = salary / ntiles (100) into salary_percentile.

Cheers!!
Albert-Jan

--- On Wed, 11/5/08, Oliver, Richard <[hidden email]> wrote:

mpirritano

evaluating differences in count data

In reply to this post by Albert-Jan Roskam

Hello listers!

I'm trying to find out if there is a way to do inferential tests on count data. I am comparing the number of individuals who have had one lipid assay in the year prior to an intervention with the number of individuals who have had one lipid assay and in the year after the intervention. So I have two numbers which are counts. Let's say 280 individuals with one lipid assay prior to the intervention and 365 individuals with an assay after the intervention.

I can't see how I could do any inferential stats with just two numbers, but I have come across some analyses focused on count data. Please advice!

I'd greatly appreciate any help, advice, etc!

Thanks!
Matt

Matthew Pirritano, Ph.D.
Research Analyst IV
Medical Services Initiative (MSI)
Orange County Health Care Agency
(714) 834-3566

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Albert-jan Roskam
Sent: Thursday, November 06, 2008 9:32 AM
To: [hidden email]
Subject: Re: percentiles

Hi,

This is one way to calculate ntiles:
rank variables = salary / ntiles (100) into salary_percentile.

Cheers!!
Albert-Jan

--- On Wed, 11/5/08, Oliver, Richard <[hidden email]> wrote:

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Maguin, Eugene

Re: evaluating differences in count data

Matthew,

>>I'm trying to find out if there is a way to do inferential tests on count
data. I am comparing the number of individuals who have had one lipid assay
in the year prior to an intervention with the number of individuals who have
had one lipid assay and in the year after the intervention. So I have two
numbers which are counts. Let's say 280 individuals with one lipid assay
prior to the intervention and 365 individuals with an assay after the
intervention.

I'm confused by your description. Are you saying that you had a pool of
people (say, N=500) who were 'assigned' to a pre-intervention lipid assay,
an intervention and a post-intervention lipid assay? 280 of 500 did the
pre-intervention assay and 365 of the 500 did the post-intervention assay?
And you want to compare the proportion who completed the pretest to the
proportion who completed the posttest?

Or, do I misunderstand and you are interested in something else?

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

David Greenberg

Re: evaluating differences in count data

In reply to this post by mpirritano

If you can assume that the counts were generated by a Poisson process you can do a test for the statistical significance of the difference. David Greenberg, Sociology Department, New York University

----- Original Message -----
From: "Pirritano, Matthew" <[hidden email]>
Date: Thursday, November 6, 2008 2:29 pm
Subject: evaluating differences in count data
To: [hidden email]

> Hello listers!
>
> I'm trying to find out if there is a way to do inferential tests on
> count data. I am comparing the number of individuals who have had one
> lipid assay in the year prior to an intervention with the number of
> individuals who have had one lipid assay and in the year after the
> intervention. So I have two numbers which are counts. Let's say 280
> individuals with one lipid assay prior to the intervention and 365
> individuals with an assay after the intervention.
>
> I can't see how I could do any inferential stats with just two
> numbers, but I have come across some analyses focused on count data.
> Please advice!
>
> I'd greatly appreciate any help, advice, etc!
>
> Thanks!
> Matt
>
> Matthew Pirritano, Ph.D.
> Research Analyst IV
> Medical Services Initiative (MSI)
> Orange County Health Care Agency
> (714) 834-3566
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf
> Of Albert-jan Roskam
> Sent: Thursday, November 06, 2008 9:32 AM
> To: [hidden email]
> Subject: Re: percentiles
>
> Hi,
>
> This is one way to calculate ntiles:
> rank variables = salary / ntiles (100) into salary_percentile.
>
> Cheers!!
> Albert-Jan
>
> --- On Wed, 11/5/08, Oliver, Richard <[hidden email]> wrote:
>
> > From: Oliver, Richard <[hidden email]>
> > Subject: Re: percentiles
> > To: [hidden email]
> > Date: Wednesday, November 5, 2008, 6:02 PM
> > One solution is to use OMS to extract the results from
> > Frequencies, as this example from the Programming and Data
> > Management book illustrates:
> >
> > GET FILE='Employee data.sav'.
> > PRESERVE.
> > SET TVARS NAMES TNUMBERS VALUES.
> > DATASET DECLARE freq_table.
> > ***split file by job category to get group percentiles.
> > SORT CASES BY jobcat.
> > SPLIT FILE LAYERED BY jobcat.
> > OMS
> > /SELECT TABLES
> > /IF COMMANDS=['Frequencies']
> > SUBTYPES=['Statistics']
> > /DESTINATION FORMAT=SAV
> > OUTFILE='freq_table'
> > /COLUMNS SEQUENCE=[L1 R2].
> > FREQUENCIES
> > VARIABLES=salary
> > /FORMAT=NOTABLE
> > /PERCENTILES= 25 50 75.
> > OMSEND.
> > ***restore previous SET settings.
> > RESTORE.
> > MATCH FILES FILE=*
> > /TABLE='freq_table'
> > /rename (Var1=jobcat)
> > /BY jobcat
> > /DROP command_ TO salary_Missing.
> > EXECUTE.
> >
> > -----Original Message-----
> > From: SPSSX(r) Discussion [mailto:[hidden email]]
> > On Behalf Of Rodrigo Briceño
> > Sent: Wednesday, November 05, 2008 10:32 AM
> > To: [hidden email]
> > Subject: percentiles
> >
> > Dear SPSS users.
> > I have a database with some groups of data (groups based on
> > categories
> > of patients). For each group of patients I must estimate
> > the percentage
> > of outliers (using the rule of percentile 25-1.5 IQR and
> > percentile 75+
> > 1.5 IQR).
> > For that purpose I need to find how to construct the 25th
> > percentile and
> > the 75th percentile in each group of patient, and get these
> > values as
> > new variables. I found many references to percentiles in
> > frequencies,
> > tables, explore and other SPSS commands, but nothing about
> > what function
> > to use in order to COMPUTE these values as new variables.
> > Anyone could help me please?
> >
> > Thanks for your time.
> > --
> >
> > __________________________________________________________________
> >
> > *Rodrigo Briceño*
> > Project Manager
> > Sanigest Internacional
> >
> > +506 22-91-12-00 ext. 113 * Oficina* *Costa Rica*
> > +506 22-32-08-30 *Fax
> > *+506 88-86-11-77* Celular**
> > *[hidden email] <
> > www.sanigest.com <
> >
> > MSN: [hidden email] <
> > SKYPE: rbriceno1087
> >
> > _____________________
> >
> > This communication contains legal information which is
> > privileged and
> > confidential. It is for the exclusive use of the address
> > and
> > distribution, dissemination, copying or use by others is
> > strictly
> > prohibited. If you have received this communication by
> > error, please
> > delete the original message and e-mail us.
> >
> >
> > Esta comunicación contiene información legal privilegiada
> > y confidencial
> > para el uso exclusivo del destinatario. La distribución,
> > diseminación,
> > copia u otro uso por terceras personas es estrictamente
> > prohibida. Si
> > usted ha recibido esta comunicación por error, le rogamos
> > borrar el
> > mensaje original y comunicárnoslo a esta misma dirección.
> >
> > =====================
> > To manage your subscription to SPSSX-L, send a message to
> > [hidden email] (not to SPSSX-L), with no body
> > text except the
> > command. To leave the list, send the command
> > SIGNOFF SPSSX-L
> > For a list of commands to manage subscriptions, send the
> > command
> > INFO REFCARD
> >
> > =====================
> > To manage your subscription to SPSSX-L, send a message to
> > [hidden email] (not to SPSSX-L), with no body
> > text except the
> > command. To leave the list, send the command
> > SIGNOFF SPSSX-L
> > For a list of commands to manage subscriptions, send the
> > command
> > INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

mpirritano

Re: evaluating differences in count data

In reply to this post by Maguin, Eugene

Sorry, I did leave a little info out. The intervention was intended to
increase the use of preventive medical procedures. The increased use was
encouraged by rewarding their use with monetary incentives. The
intervention also provided money to increase the population size
overall, which contains a diabetic population. Lipid assays are a
preventive measure for diabetics. So, we'd expect the number of lipid
assays to go up if doctors are taking advantage of the monetary
incentives. Hence the two hypothetical counts I gave below. The number
of lipid assays went up, but is increase greater than would be expected
if the intervention had no effect ( = the null hypothesis).

Thanks
matt

Matthew Pirritano, Ph.D.
Research Analyst IV
Medical Services Initiative (MSI)
Orange County Health Care Agency
(714) 834-3566
-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Gene Maguin
Sent: Thursday, November 06, 2008 12:19 PM
To: [hidden email]
Subject: Re: evaluating differences in count data

Matthew,

>>I'm trying to find out if there is a way to do inferential tests on
count
data. I am comparing the number of individuals who have had one lipid
assay
in the year prior to an intervention with the number of individuals who
have
had one lipid assay and in the year after the intervention. So I have
two
numbers which are counts. Let's say 280 individuals with one lipid assay
prior to the intervention and 365 individuals with an assay after the
intervention.

I'm confused by your description. Are you saying that you had a pool of
people (say, N=500) who were 'assigned' to a pre-intervention lipid
assay,
an intervention and a post-intervention lipid assay? 280 of 500 did the
pre-intervention assay and 365 of the 500 did the post-intervention
assay?
And you want to compare the proportion who completed the pretest to the
proportion who completed the posttest?

Or, do I misunderstand and you are interested in something else?

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Marta Garcia-Granero

Re: evaluating differences in count data

Pirritano, Matthew wrote:

> Sorry, I did leave a little info out. The intervention was intended to
> increase the use of preventive medical procedures. The increased use was
> encouraged by rewarding their use with monetary incentives. The
> intervention also provided money to increase the population size
> overall, which contains a diabetic population. Lipid assays are a
> preventive measure for diabetics. So, we'd expect the number of lipid
> assays to go up if doctors are taking advantage of the monetary
> incentives. Hence the two hypothetical counts I gave below. The number
> of lipid assays went up, but is increase greater than would be expected
> if the intervention had no effect ( = the null hypothesis).
>

See chapter 6 of Statistics at Square One (section "Standard error of a
total")
http://www.bmj.com/collections/statsbk/6.dtl

They describe a simple method to compare two counts. Using your data:

* Standard error of a total (& difference between two totals) *.
DATA LIST LIST/n1 n2 (2 F8.0).
BEGIN DATA
280 365
END DATA.
COMPUTE zvalue = (n1-n2)/SQRT(n1+n2).
COMPUTE pvalue = 2*(1-CDF.NORMAL(ABS(zvalue),0,1)).
LIST.

HTH,
Marta García-Granero

>
> Thanks
> matt
>
> Matthew Pirritano, Ph.D.
> Research Analyst IV
> Medical Services Initiative (MSI)
> Orange County Health Care Agency
> (714) 834-3566
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Gene Maguin
> Sent: Thursday, November 06, 2008 12:19 PM
> To: [hidden email]
> Subject: Re: evaluating differences in count data
>
> Matthew,
>
>
>>> I'm trying to find out if there is a way to do inferential tests on
>>>
> count
> data. I am comparing the number of individuals who have had one lipid
> assay
> in the year prior to an intervention with the number of individuals who
> have
> had one lipid assay and in the year after the intervention. So I have
> two
> numbers which are counts. Let's say 280 individuals with one lipid assay
> prior to the intervention and 365 individuals with an assay after the
> intervention.
>
>
> I'm confused by your description. Are you saying that you had a pool of
> people (say, N=500) who were 'assigned' to a pre-intervention lipid
> assay,
> an intervention and a post-intervention lipid assay? 280 of 500 did the
> pre-intervention assay and 365 of the 500 did the post-intervention
> assay?
> And you want to compare the proportion who completed the pretest to the
> proportion who completed the posttest?
>
> Or, do I misunderstand and you are interested in something else?
>
> Gene Maguin
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>

David Greenberg

Re: evaluating differences in count data

I think this test is not appropriate. It is the test that would be appropriate if the data were being generated by a Poisson distribution. That was my own initial assumption when I responded to the initial inquiry. However, for each individual the response is binary. The assumption of independence of events that is characteristic of a Poisson distribution does not hold here. David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Marta García-Granero <[hidden email]>
Date: Saturday, November 8, 2008 5:25 am
Subject: Re: evaluating differences in count data
To: [hidden email]

> Pirritano, Matthew wrote:
> > Sorry, I did leave a little info out. The intervention was intended
> to
> > increase the use of preventive medical procedures. The increased use
> was
> > encouraged by rewarding their use with monetary incentives. The
> > intervention also provided money to increase the population size
> > overall, which contains a diabetic population. Lipid assays are a
> > preventive measure for diabetics. So, we'd expect the number of lipid
> > assays to go up if doctors are taking advantage of the monetary
> > incentives. Hence the two hypothetical counts I gave below. The number
> > of lipid assays went up, but is increase greater than would be expected
> > if the intervention had no effect ( = the null hypothesis).
> >
>
> See chapter 6 of Statistics at Square One (section "Standard error of
> a
> total")
> http://www.bmj.com/collections/statsbk/6.dtl
>
> They describe a simple method to compare two counts. Using your data:
>
> * Standard error of a total (& difference between two totals) *.
> DATA LIST LIST/n1 n2 (2 F8.0).
> BEGIN DATA
> 280 365
> END DATA.
> COMPUTE zvalue = (n1-n2)/SQRT(n1+n2).
> COMPUTE pvalue = 2*(1-CDF.NORMAL(ABS(zvalue),0,1)).
> LIST.
>
> HTH,
> Marta García-Granero
>
> >
> > Thanks
> > matt
> >
> > Matthew Pirritano, Ph.D.
> > Research Analyst IV
> > Medical Services Initiative (MSI)
> > Orange County Health Care Agency
> > (714) 834-3566
> > -----Original Message-----
> > From: SPSSX(r) Discussion [mailto:[hidden email]] On
> Behalf Of
> > Gene Maguin
> > Sent: Thursday, November 06, 2008 12:19 PM
> > To: [hidden email]
> > Subject: Re: evaluating differences in count data
> >
> > Matthew,
> >
> >
> >>> I'm trying to find out if there is a way to do inferential tests on
> >>>
> > count
> > data. I am comparing the number of individuals who have had one lipid
> > assay
> > in the year prior to an intervention with the number of individuals
> who
> > have
> > had one lipid assay and in the year after the intervention. So I have
> > two
> > numbers which are counts. Let's say 280 individuals with one lipid assay
> > prior to the intervention and 365 individuals with an assay after the
> > intervention.
> >
> >
> > I'm confused by your description. Are you saying that you had a pool
> of
> > people (say, N=500) who were 'assigned' to a pre-intervention lipid
> > assay,
> > an intervention and a post-intervention lipid assay? 280 of 500 did
> the
> > pre-intervention assay and 365 of the 500 did the post-intervention
> > assay?
> > And you want to compare the proportion who completed the pretest to
> the
> > proportion who completed the posttest?
> >
> > Or, do I misunderstand and you are interested in something else?
> >
> > Gene Maguin
> >
> > =====================
> > To manage your subscription to SPSSX-L, send a message to
> > [hidden email] (not to SPSSX-L), with no body text except
> the
> > command. To leave the list, send the command
> > SIGNOFF SPSSX-L
> > For a list of commands to manage subscriptions, send the command
> > INFO REFCARD
> >
> > =====================
> > To manage your subscription to SPSSX-L, send a message to
> > [hidden email] (not to SPSSX-L), with no body text except
> the
> > command. To leave the list, send the command
> > SIGNOFF SPSSX-L
> > For a list of commands to manage subscriptions, send the command
> > INFO REFCARD
> >
> >
>
>
> --
> For miscellaneous statistical stuff, visit:
> http://gjyp.nl/marta/
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD