Comparison of percentiles in SPSS

[Kai E]

Hello everyone,

I need some assistance with a statistical problem:

I have calculated the 99th percentiles with their 95% confidence intervals (using bootstrapping) for results of a blood test in a large population of healthy subjects. Analyses had been run in SPSS 21 and show that the 99th percentiles are quite different depending on certain background factors such as sex, age etc.

Does anyone know if SPSS 21 offers the possibility to compare the 99th percentiles in specific subcohorts, e.g. in men compared to women, age <65 compared to >65 years, and how to get this done?

Thanks in advance for your assistance.

Kai

[Jon K Peck]

If you used a procedure such as FREQUENCIES to compute the percentiles, you can use Data > Select Cases to focus on any subset of the data you want, and bootstrapping will work on that subset.

If you use a procedure such as EXAMINE (Analyze > Descriptives ? Explore), it takes factors in the specification, and, again, bootstrapping will use the same subsets.

HTH,


Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: [phone]




From:        Kai E <[hidden email]>
To:        [hidden email],
Date:        02/02/2013 07:46 AM
Subject:        [SPSSX-L] Comparison of percentiles in SPSS
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hello everyone,

I need some assistance with a statistical problem:

I have calculated the 99th percentiles with their 95% confidence intervals
(using bootstrapping) for results of a blood test in a large population of
healthy subjects. Analyses had been run in SPSS 21 and show that the 99th
percentiles are quite different depending on certain background factors such
as sex, age etc.

Does anyone know if SPSS 21 offers the possibility to compare the 99th
percentiles in specific subcohorts, e.g. in men compared to women, age <65
compared to >65 years, and how to get this done?

Thanks in advance for your assistance.

Kai



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Comparison-of-percentiles-in-SPSS-tp5717882.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

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

[Kai E]

Thanks Jon, for quick reply.

In fact, I had calculated the percentiles using the Analyze > Descriptives > Explore procedure. However, I realize that I should have been more specific in the description of my problem: what I need to do is to estimate at which levels of significance the 99th percentiles in specified subgroups (e.g. men and women) differ to each other.

Is this feasible in SPSS and how ?

Kai

[Kai E]

Thanks Jon, for quick reply.

In fact, I had calculated the percentiles using the Analyze > Descriptives >
Explore procedure. However, I realize that I should have been more specific
in the description of my problem: what I need to do is to estimate at which
levels of significance the 99th percentiles in specified subgroups (e.g. men
and women) differ to each other.

Is this feasible in SPSS and how ?

Kai



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Comparison-of-percentiles-in-SPSS-tp5717882p5717889.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

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

[Rich Ulrich]

There was a thread discussing CIs, etc., for quantiles, in the
sci.stat.math  Usenet group, from just this week.  See
https://groups.google.com/forum/?fromgroups=#!topic/sci.stat.math/QhE3qxl8E6Y

For the simple test, you simply look at the 2x2 contingency
table that is created by dichotomizing each group at their joint
99th percentile (which you might put in a variable by using options
in RANK). 

By the way ... this did not get asked in that sci.stat.math dialog ...
WHY do you want to compare an extreme quantile? 


--
Rich Ulrich

> Date: Sat, 2 Feb 2013 09:09:41 -0800

> From: [hidden email]
> Subject: Re: Comparison of percentiles in SPSS
> To: [hidden email]
>
> Thanks Jon, for quick reply.
>
> In fact, I had calculated the percentiles using the Analyze > Descriptives >
> Explore procedure. However, I realize that I should have been more specific
> in the description of my problem: what I need to do is to estimate at which
> levels of significance the 99th percentiles in specified subgroups (e.g. men
> and women) differ to each other.
>
> Is this feasible in SPSS and how ?
>

...

[Kai E]

Hello Rich,

the reason to evaluate 99th percentiles is that they by convention are the threshold to discriminate positive from negative test results for the blood test I am studying.

Best regards, Kai

[Andy W]

FYI extreme quantiles are some of the canonical examples of where the bootstrap fails. See http://stats.stackexchange.com/q/9664/1036 for discussion and further references to literature.