normalized?

[mpirritano]

SPSS gurus,

 

I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.

 

In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?

 

I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.

 

Does this question make sense?  Need more info?

 

What am I missing?

 

Thanks

Matt

 

Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

 

[Ling Ting]

I will be out of office till Nov. 1 with limited access to email.   If you need immediate assistance please contact 479-575-2905.  Thank you.

[Tesiny, Ed]

I think what's missing is what you're IT person means by normalized.  He maybe thinking database and you're doing research, no?

Edward P. Tesiny
Director of Evaluation and Outcome Management
New York State OASAS
1450 Western Ave.
Albany, NY 12203
518-485-7189
[hidden email]

________________________________

From: SPSSX(r) Discussion on behalf of Pirritano, Matthew
Sent: Thu 10/28/2010 7:47 PM
To: [hidden email]
Subject: normalized?



SPSS gurus,



I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.



In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?



I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.



Does this question make sense?  Need more info?



What am I missing?



Thanks

Matt



Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

=====================
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

[Harris, Betty A.]

That's an interesting question.  In my experience IT people speak a different language than data analysts, so asking for clarification would be a good idea.

That is the same metric expressed as the rate per 100 admissions.

One of the techniques I used to use back when I was working with death and inpatient data was to calculate rates per 100,000 population. When I calculated death rates for various geographic areas with different age distributions, I calculated age-adjusted rates which control for differences in age of the underlying population.  However, I don't remember doing that as much with the inpatient data.  Here's an old summary I put together using HCUP and Oklahoma data where I used percentages to compare the state to the US data on various characteristics (e.g. discharges, LOS, PtDays, charges, deaths, etc) within Clinical Classification System diagnosis groupings. The goal was to provide a reference (the US percentages) for looking at the Oklahoma percentages.
http://www.state.ok.us/~chs/draft/1998/stacsum/

Note I translated most of the dx grouping labels into terminology that laypeople should find more understandable. Note also, that I didn't control for anything because HCUP is an estimate based on the total US population and so was the Oklahoma data (tho back then, that was iffy given it was a brand new data collection effort in Oklahoma with data missing for a subset of hospitals, typically for the smaller, more rural hospitals and women's specialty hospitals).

Here in Oklahoma, there are considerable differences between hospitals based on their size, specialty and where they're located geographically.
http://www.state.ok.us/~chs/draft/1998/peer/

Hope this helps,

Betty Harris
Senior Research Associate
OU E-TEAM
http://eteam.ou.edu/
http://webELQA.eteam.ou.edu/

Matthew wrote:
>I have been asked by a healthcare IT person if I normalized my data. To
>me normalized means one thing, you put everything on the same scale,
>usually z-scores.

>In this case I am reporting the percentage of inpatient hospital stays
>with a particular diagnosis. And I am comparing it to previous research
>that presents percentages for the same diagnosis. My confusion comes
>from the fact that as far as I know, percentages are normalized. Right?

>I was asked if I made my data comparable to the other data. The N of my
>sample is much smaller than the other sample. I was told that I needed
>to extrapolate my data to the larger sample size. But I can't see why
>this would matter because we are talking about percentages.

>Does this question make sense?  Need more info?

>What am I missing?

=====================
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

[Art Kendall]

sometimes "standardized" is confused with "normalized".
A standard score (e.g., a z-score) is (score-meanscore/sdscore).   (can be added to file with /save on descriptives.)

A normalized score is found by converting the score to a normal distribution by finding the cumulative proportion of scores and finding z-score on a normal distribution that corresponds to it. (can be added to a file with rank  and /normal).

to see a simulation that demonstrates the distinction
Open a new instance of SPSS.  Copy the syntax below to a syntax window.  Run it.

*Simulation to demonstrate the distinction between standardized and normalized scores'.
* using canned procedures and by specifying steps.
*Art Kendall    Social Research Consultants.
*generate random integers from 1 to 25.
INPUT PROGRAM.
LOOP id=1 TO 300.
COMPUTE score = rnd( rv.uniform(.5,25.5)).
END CASE.
END LOOP.
END FILE.
END INPUT PROGRAM.
FORMATS id (F3.0)  score(F2).
descriptives variables= score
    /save.
formats zscore (f6.2).
compute constant = 1.
aggregate outfile= * mode=addvariables/ break = constant
 /meanscore = mean(score)
 /sdscore = sd(score).
compute demoz = score-meanscore/sdscore.
formats demoz (f6.2).
rank variables = score /proportion into cumprop  /normal into normscore.
formats cumprop (f6.4) normscore (f6.2).
compute demonorm = idf.normal(cumprop,0,1).
formats demonorm (f6.2).
display variables /variables=score Zscore  demoz cumprop normscore demonorm .
var labels
     score 'simulated score'/
     Zscore 'standardized score from DESCRIPTIVES'/
     demoz 'standardized score from COMPUTE'/
     normscore 'normalized score from RANK'/
     demonorm 'normalized score from COMPUTE'.
list variables = id score Zscore  demoz cumprop normscore demonorm  /cases from 1 to 20.

* notice the straight line.
* Chart Builder.
GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=score Zscore MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: score=col(source(s), name("score"))
  DATA: Zscore=col(source(s), name("Zscore"))
  GUIDE: axis(dim(1), label("simulated score"))
  GUIDE: axis(dim(2), label("standardized score from DESCRIPTIVES"))
  ELEMENT: point(position(score*Zscore))
END GPL.
*notice the curved line.
* Chart Builder.
GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=score normscore MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: score=col(source(s), name("score"))
  DATA: normscore=col(source(s), name("normscore"))
  GUIDE: axis(dim(1), label("simulated score"))
  GUIDE: axis(dim(2), label("normalized score from RANK"))
  ELEMENT: point(position(score*normscore))
END GPL.

*compare the line in this graph to the previous one.
* Chart Builder.
GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=Zscore normscore MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
  SOURCE: s=userSource(id("graphdataset"))
  DATA: Zscore=col(source(s), name("Zscore"))
  DATA: normscore=col(source(s), name("normscore"))
  GUIDE: axis(dim(1), label("standardized score from DESCRIPTIVES"))
  GUIDE: axis(dim(2), label("normalized score from RANK"))
  ELEMENT: point(position(Zscore*normscore))
END GPL.



Art Kendall
Social Research Consultants


On 10/28/2010 7:47 PM, Pirritano, Matthew wrote:

SPSS gurus,

 

I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.

 

In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?

 

I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.

 

Does this question make sense?  Need more info?

 

What am I missing?

 

Thanks

Matt

 

Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

 

===================== 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

[Art Kendall]

PS in database technology it is a completely different concept having to do with assuring quality control.

Art Kendall
Social Research Consultants

On 10/28/2010 7:47 PM, Pirritano, Matthew wrote:

SPSS gurus,

 

I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.

 

In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?

 

I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.

 

Does this question make sense?  Need more info?

 

What am I missing?

 

Thanks

Matt

 

Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

 

===================== 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

[Matthew Pirritano]

What the IT person (she) said was that the other study I was comparing my data to was based on a larger number of claims. So while my study may have been based on 1,000 claims the other study was based on 10,000 claims. She was saying something about how I would need to adjust my numbers for that. But of course percentages are comparable across studies.

I think she must just be misunderstanding something. And yes, I usually represent my data in per 1,000 numbers. But the data I'm comparing my numbers to are percentages. And so my data are percentages.

Thanks for confirming my understanding. I'm confident my numbers make sense as percentages.

Thanks
Matt

 

Matthew Pirritano, Ph.D.
Email: [hidden email]

---

From: "Tesiny, Ed" <[hidden email]>
To: [hidden email]
Sent: Fri, October 29, 2010 4:19:09 AM
Subject: Re: normalized?

I think what's missing is what you're IT person means by normalized.  He maybe thinking database and you're doing research, no?

Edward P. Tesiny
Director of Evaluation and Outcome Management
New York State OASAS
1450 Western Ave.
Albany, NY 12203
518-485-7189
[hidden email]

________________________________

From: SPSSX(r) Discussion on behalf of Pirritano, Matthew
Sent: Thu 10/28/2010 7:47 PM
To: [hidden email]
Subject: normalized?



SPSS gurus,



I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.



In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?



I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.



Does this question make sense?  Need more info?



What am I missing?



Thanks

Matt



Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

=====================
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

[Art Kendall]

As point estimates it makes sense to present both estimates the same way.

Of course interval (error) estimates are needed also.� If I recall correctly, there are PYTHON extensions for comparing proportions (percentages/100) under special circumstances. But with those numbers of cases in the two studies, the confidence intervals should me of moderate size.

Of course, the overall estimate is not the simple average of the two percentages.� It is the sum of the numerators over the sum of the denominators.

Art Kendall
Social Research Consultants

On 10/29/2010 11:58 AM, Matthew Pirritano wrote:

What the IT person (she) said was that the other study I was comparing my data to was based on a larger number of claims. So while my study may have been based on 1,000 claims the other study was based on 10,000 claims. She was saying something about how I would need to adjust my numbers for that. But of course percentages are comparable across studies.

I think she must just be misunderstanding something. And yes, I usually represent my data in per 1,000 numbers. But the data I'm comparing my numbers to are percentages. And so my data are percentages.

Thanks for confirming my understanding. I'm confident my numbers make sense as percentages.

Thanks
Matt

�

Matthew Pirritano, Ph.D.
Email: [hidden email]

---

From: "Tesiny, Ed" [hidden email]
To: [hidden email]
Sent: Fri, October 29, 2010 4:19:09 AM
Subject: Re: normalized?

I think what's missing is what you're IT person means by normalized.� He maybe thinking database and you're doing research, no?

Edward P. Tesiny
Director of Evaluation and Outcome Management
New York State OASAS
1450 Western Ave.
Albany, NY 12203
518-485-7189
[hidden email]

________________________________

From: SPSSX(r) Discussion on behalf of Pirritano, Matthew
Sent: Thu 10/28/2010 7:47 PM
To: [hidden email]
Subject: normalized?



SPSS gurus,



I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.



In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?



I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.



Does this question make sense?� Need more info?



What am I missing?



Thanks

Matt



Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

=====================
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

[Jon K Peck]

Yes, there is an extension command called PROPOR available from Developer Central (www.spss.com/devcentral) that provides two finite-sample proportion confidence intervals and ci's for differences.  It does not have a dialog box interface, but executing PROPOR /HELP displays the syntax chart.   The command requires the Python plugin and works with SPSS version 16 or later.


Jon Peck
Senior Software Engineer, IBM
[hidden email]
312-651-3435



From:        Art Kendall <[hidden email]>
To:        [hidden email]
Date:        10/30/2010 05:43 AM
Subject:        Re: [SPSSX-L] normalized?
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

As point estimates it makes sense to present both estimates the same way.

Of course interval (error) estimates are needed also. If I recall correctly, there are PYTHON extensions for comparing proportions (percentages/100) under special circumstances. But with those numbers of cases in the two studies, the confidence intervals should me of moderate size.

Of course, the overall estimate is not the simple average of the two percentages. It is the sum of the numerators over the sum of the denominators.

Art Kendall
Social Research Consultants

On 10/29/2010 11:58 AM, Matthew Pirritano wrote:
What the IT person (she) said was that the other study I was comparing my data to was based on a larger number of claims. So while my study may have been based on 1,000 claims the other study was based on 10,000 claims. She was saying something about how I would need to adjust my numbers for that. But of course percentages are comparable across studies.

I think she must just be misunderstanding something. And yes, I usually represent my data in per 1,000 numbers. But the data I'm comparing my numbers to are percentages. And so my data are percentages.

Thanks for confirming my understanding. I'm confident my numbers make sense as percentages.

Thanks
Matt
Matthew Pirritano, Ph.D.
Email: matthewpirritano@...

---

From: "Tesiny, Ed" <EdTesiny@...>
To: [hidden email]
Sent: Fri, October 29, 2010 4:19:09 AM
Subject: Re: normalized?

I think what's missing is what you're IT person means by normalized. He maybe thinking database and you're doing research, no?

Edward P. Tesiny
Director of Evaluation and Outcome Management
New York State OASAS
1450 Western Ave.
Albany, NY 12203
518-485-7189
EdTesiny@...

________________________________

From: SPSSX(r) Discussion on behalf of Pirritano, Matthew
Sent: Thu 10/28/2010 7:47 PM
To: [hidden email]
Subject: normalized?



SPSS gurus,



I have been asked by a healthcare IT person if I normalized my data. To me normalized means one thing, you put everything on the same scale, usually z-scores.



In this case I am reporting the percentage of inpatient hospital stays with a particular diagnosis. And I am comparing it to previous research that presents percentages for the same diagnosis. My confusion comes from the fact that as far as I know, percentages are normalized. Right?



I was asked if I made my data comparable to the other data. The N of my sample is much smaller than the other sample. I was told that I needed to extrapolate my data to the larger sample size. But I can’t see why this would matter because we are talking about percentages.



Does this question make sense? Need more info?



What am I missing?



Thanks

Matt



Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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