infinite error bars on line plot

[thrillhouse86]

Hi, I have a relatively simple dataset where I have a collection of distances (the distances are discrete, and they form a finite set) and a measured signal parameter (measured at the distances). The number of measured signals is not the same at each distance.

I've tried to plot the distance vs the mean of the signal parameter with the standard error as the error bars. It works, but it tells me in the output:

"Warnings
One or more error bar calculations yielded infinite results. These error bars have been removed from the chart."

My question is (and I guess this is more of a statistics question than a SPSS question) how can you have infinite standard error ? I tried changing the error bar to the standard deviation - and I get the same error.
I don't see how these quantities can be infinitely large!

Thanks,
Thrillhouse

[Art Kendall]

  What is the n in the category represented by that bar?

Art Kendall
Social Research Consultants

On 9/30/2010 8:15 AM, thrillhouse86 wrote:

> Hi, I have a relatively simple dataset where I have a collection of distances
> (the distances are discrete, and they form a finite set) and a measured
> signal parameter (measured at the distances). The number of measured signals
> is not the same at each distance.
>
> I've tried to plot the distance vs the mean of the signal parameter with the
> standard error as the error bars. It works, but it tells me in the output:
>
> "Warnings
> One or more error bar calculations yielded infinite results. These error
> bars have been removed from the chart."
>
> My question is (and I guess this is more of a statistics question than a
> SPSS question) how can you have infinite standard error ? I tried changing
> the error bar to the standard deviation - and I get the same error.
> I don't see how these quantities can be infinitely large!
>
> Thanks,
> Thrillhouse
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/infinite-error-bars-on-line-plot-tp3046909p3046909.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
>

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

[thrillhouse86]

Hi Art Kendall,

Do you mean the 'n' in the standard error equation std/sqrt{n} ? if so the n in the bar is the number of measurements of the signal parameter at the particular distance.

Thanks,
Thrillhouse

[Art Kendall]

  I was asking what actual number went into the denominator of the SE?
I.e., how many cases are the in the category for the bar that is giving
you interest?
Any chance it is zero or one?

Art Kendall
Social Research Consultants

On 9/30/2010 9:54 AM, thrillhouse86 wrote:

> Hi Art Kendall,
>
> Do you mean the 'n' in the standard error equation std/sqrt{n} ? if so the n
> in the bar is the number of measurements of the signal parameter at the
> particular distance.
>
> Thanks,
> Thrillhouse
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/infinite-error-bars-on-line-plot-tp3046909p3047042.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
>

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

[thrillhouse86]

yeah its one - can you please explain to me why that causes an infinity ?

[Art Kendall]

  in many computer languages a floating point underflow or overflow is
called "infinity" i.e., too small or too big to be represented in a
register.
see
http://en.wikipedia.org/wiki/Division_by_zero

and

http://mathcentral.uregina.ca/QQ/database/QQ.09.04/katrina1.html

copy, paste, and run the following syntax into a new instance of SPSS.
It breaks the calculation down into steps. Results are system missing
when the computer cannot do what you instruct it to do.
* infinite or undefined standard error.
data list list /x(f2).
begin data
10
end data.
compute xmean = mean(x).
compute n=nvalid(x).
compute deviation  = x-xmean.
compute sumsquares = deviation**2.
compute popmeansquares = sumsquares/n.
compute poprootmeansquare = sqrt(popmeansquares).
compute nminusone = n -1.
compute samplemeansquares = sumsquares /nminusone.
compute samplestd =sqrt(samplemeansquares).
compute samplese = samplestd/sqrt(n).
list.


Art Kendall
Social Research Consultants




On 9/30/2010 1:00 PM, thrillhouse86 wrote:

> yeah its one - can you please explain to me why that causes an infinity ?
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/infinite-error-bars-on-line-plot-tp3046909p3047312.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
>

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

[Richard Ristow]

At 12:56 PM 9/30/2010, Art Kendall wrote:

>What actual number went into the denominator of the SE?
>I.e., how many cases are the in the category for the bar that is
>giving infinity? Any chance it is zero or one?

At 01:00 PM 9/30/2010, thrillhouse86 wrote:
>yeah its one - can you please explain to me why that causes an infinity ?

The SE is calculated from the dispersion of the values that are
observed. If only one value's observed within a category, there's no
observed dispersion of values within that category, and the SE can't
be estimated. 'Infinity' is a reasonable rendering of 'no estimate
can be made'.

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