[Met_help] [rt.rap.ucar.edu #63389] History for I am sorry to disturb you again.

[John Halley Gotway via RT]

----------------------------------------------------------------
  Initial Request
----------------------------------------------------------------

Hello!

First of all, thanks to your useful suggestions very much from my soul.

I want to plot the whiskers and outliers of boxplot, and you said I can the individual matched pair data and then plot the boxplot. but  the Matched pair data just one integer number in my Point_stat MPR, I do not know how to get the
extreme minimum and the extreme maximum form the Matched pair data. Maybe I think it is statiscal knowledge.

another question is about normal and bootstrap upper and lower confidence limits(Table 5-3 column 46-50), what is the different between the normal and bootstrap?

Best wishes!


Qiang Li


----------------------------------------------------------------
  Complete Ticket History
----------------------------------------------------------------

Subject: Re: [rt.rap.ucar.edu #63389] I am sorry to disturb you again.
From: John Halley Gotway
Time: Mon Oct 14 13:55:18 2013

Qiang Li,

Let me answer your second question first.  The statistics in the MET
output files are often followed by values for normal confidence limits
(_NCL and _NCU) and bootstrap confidence limits (_BCL and
_BCU).  The confidence intervals provide an estimate of the sampling
uncertainty.

For example, let's look at the Mean Error (ME) in column 53 of the
continuous statistics (CNT) output line.  Suppose you have 75 matched
pairs of forecast and observation values.  And you've computed
the average error to be 0.5.  That value of 0.5 is the mean error
computed over the 75 samples.  The confidence interval provides a 95%
confidence range around the computed mean error value.  So
you're 95% confident that the true statistic for the population falls
within that range.

Why do we list both parametric and bootstrap confidence limits?  There
are multiple ways of estimating the uncertainty for statistics.  For
some statistics, there is a parametric estimate where we
make some assumptions about the underlying distribution of the data
and use an equation based on the sample's mean and standard deviation
to estimate the uncertainty.  The bootstrap method makes fewer
assumptions about the underlying distribution but is more
computationally intensive to compute.  Both are reasonable estimates
for the confidence interval.  Bootstrap confidence intervals can
always
be computed but parametric ones are only computed on those statistics
for which they're well defined.

Regarding your first question, are you familiar with any statistical
scripting languages like matlab, idl, or R?  We use R extensively
here.  Using any of those tools, it's pretty easy the read in the
matched pair data and create boxplots of the forecast values,
observation values, or their differences.

Hope that helps.

Thanks,
John

On 10/12/2013 09:56 PM, Li, Qiang via RT wrote:
>
> Sat Oct 12 21:56:22 2013: Request 63389 was acted upon.
> Transaction: Ticket created by liqiang at ou.edu
>         Queue: met_help
>       Subject: I am sorry to disturb you again.
>         Owner: Nobody
>    Requestors: liqiang at ou.edu
>        Status: new
>   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=63389 >
>
>
> Hello!
>
> First of all, thanks to your useful suggestions very much from my
soul.
>
> I want to plot the whiskers and outliers of boxplot, and you said I
can the individual matched pair data and then plot the boxplot. but
the Matched pair data just one integer number in my Point_stat MPR, I
do not know how to get the
> extreme minimum and the extreme maximum form the Matched pair data.
Maybe I think it is statiscal knowledge.
>
> another question is about normal and bootstrap upper and lower
confidence limits(Table 5-3 column 46-50), what is the different
between the normal and bootstrap?
>
> Best wishes!
>
>
> Qiang Li
>

------------------------------------------------