Está en la página 1de 14

Contents of DISTRIB 2.

0 Help System

DISTRIB is a computer program designed to aid in the fitting of independent data to a theoretical
probability distribution. The program as ritten in !isual Basic"T#$ for the %indo"T#$ operating
system and is based upon the te&t 're(uency and Ris) *nalysis by +.%. ,ite "%ater Resources
-ublications. /ittleton C0$. The ob1ecti2e of this program is to pro2ide an easy to use and intuiti2e
interface for the analysis of data. The program as ritten by Dr. Ron 3aglin at the 4ni2ersity of Central
'lorida.

%hat is a Distribution5

Types of Distributions

4sing DISTRIB


Types of Distributions
6ormal
/og 6ormal
7 -arameter /og 6ormal7
-earson
/og -earson
+umbel

#ethods of *nalysis
#a&imum /i)elihood
#ethod of #oments

4sing DISTRIB

4sing 'iles
DISTRIB 'iles
Importing Data

3diting Data and 0ptions
-asting Data from Spreadsheets
3diting *ctual Data
/e2el of Confidence
-lot Te&t

-lotting *nalysis
Comparison -lots
Histograms


%hat is a Distribution5

4nderstanding distributions can best be done by e&ample. So...
In the study of hydrology it is 2ery ad2antageous to predict a flood. It is. hoe2er.
almost impossible to predict if a flood ill occur. say. ne&t year. Instead e try to
predict the probability of a flood. If a flood has occurred 8 times in the last 900 years
then e can 2ery simply state that there is probably a 9 in 2: chance that one ill occur
ne&t year. Hoe2er. is the last 900 years representati2e of the probability5 %hat if one
of the floods occurred last year. If e ere predicting a flood based on 900 years the
year before the flood e ould ha2e thought a flood occurred e2ery 77 years and
therefore there as a 9 in 77 chance of the flood. To alle2iate these inaccuracies hy
don;t e analy<e all storms that occurred in the last 900 years. 0nly those storms hich
are the largest ill cause a flood. but no our analysis is based on a lot more data. I6
'*CT e can probably use only data from the last 90 years to predict a flood e2en if a
flood has not occurred in the last 90 years. %hy5 Because the data ill probably fit a
distribution. Rainfall data typically fits any number of distributions and fit them ell.

%e run DISTRIB and e find our data fits a 2 parameter log normal distribution 2ery
nicely. 0dds are any storms in the future ill also fit nicely. %e can predict storms
hich occur e2ery 900 storms. e2ery 9000 storms. etc.. %e ha2e fit our data to a
distribution and e can ma)e predictions based on that distribution. =ou may be
familiar ith the most common distribution > the normal distribution. =ou )no. the one
that is shaped li)e a typical bell cur2e. %ell. surprise? there are many distributions
hich fit many different types of data. This program ill help you fit your data to the
distributions a2ailable in the program.

6ormal Distribution

The probability density function for the normal distribution is@


here@
A mean of the population of &
A 2ariance of the population of &

The *ssumptions of the 6ormal Distribution are@
9. The 2ariable is continuous
2. Consecuti2e !alues are Independent
7. -robabilities are Stable

The Prediction for any 2alue of a normal distribution can be found from@


here , "fre(uency factor$ is the standard normal de2iate.

The Standard Error of a normal distribution can be found from@


!arious methods for the esitmation of the delta parameter e&ist.The results for the
-rediction and Standard 3rror using the ma&imum li)elihood and method of moments
for the 6ormal distribution are identical.

/og 6ormal Distribution "2 -arameter /og 6ormal$

The probability density function for the /og 6ormal Distribution is@


here@
y A ln"&$ > natural log of &
y A mean of the population y
y A 2ariance of the population of y

The *ssumptions of the 6ormal Distribution are@
9. The 2ariable is continuous
2. Consecuti2e !alues are Independent
7. -robabilities are Stable
8. *ll 2ariables are non><ero

* number of methods ha2e been used to handle 0.0 2alues. DISTRIB does not perform
any of these con2ersions for you and may crash if you attempt to fit 0 data to a log
distribution. To alle2iate this problem you may@
9. *dd 9.0 to all data
2. *dd a small positi2e 2alue to all data.
7. Substitute 9.0 in place of all 0 data.
8. Substitute a small positi2e number in the place of all <ero readings.
:. Ignore all <ero obser2ations.
B. Consider the probability distribution as the sum of the probability mass at 0.0 and a
probability distribution o2er the remainder of the range. This method is described in
Cennings and Benson.

7 -arameter /og 6ormal

The probability density function for the 7 parameter log normal distribution is@



here@
y A ln"&>a$ > natural log of "&>a$
y A mean of the population y
y A 2ariance of the population of y


-earson Distribution

The -earson distribution is represented by folloing probability density function. The
mode of this function is a & A 0. This e(uation is a selecti2e case of the three parameter
gamma distribution.



here@
A difference beteen mean and mode " A > Dm$
Dm A mode of popolation &
A Scale parameter of distribution
po A 2alue of p&"&$ at mode


/og -earson Distribution

Substituting yAln"&$ for & in the -earson distribution gi2es the /og -earson type III
distributionE



here@
y A difference beteen mean and mode " A y > =m$
=m A mode of population y
A Scale parameter of distribution
pyo A 2alue of p&"y$ at mode


+umbel Distribution

The +umbel distribution "also referred to as 'isher>Tippett Type I. Double 3&ponential.
+umbel Type I. and +umbel 3&tremal distribution $ is characteri<ed by the probability
density functionE



here@
A Scale parameter of the distribution
A /ocation parameter of the distribution

Standard 3rror of Distribution

The standard error can be calculated by the e(uation.



%here
SS3 A the sum s(uared error difference beteen the actual and predicted data.
n A 6umber of -oints.

Standard 3rror of -rediction

The Standard 3rror of the -rediction is calculated specific to a prediction at a certain
probability. It is important to the calculation of a confidence
inter2alH/-FC06'ID36C3. This statistic is output in the probability
analysisH/-F-R3DICTI!3 and the Comparison *nalysisH/-FC0#-*RIS06. This
statistic should not be confused ith the Standard 3rror of
DistributionH/-FST*6D*RD3RR0R hich is for the entire distribution.


Confidence Inter2al

* confidence inter2al for any prediction can be obtained by using the e(uation


here@
Dt A -rediction !alue of 32ent
St A Standard 3rror of 32entH/-FST*6D*RD3RR0R-R3DICTI06
t A Standard 6ormal De2iate Corresponding to Confidence /e2el

The t statistic ill be found from the le2el of confidence. The le2el of confidence can be
changed but it must be between 0.50 and 0.999. The le2el of confidence can be
changed using@ 3dit. /e2el of Confidence.

Determination of -lotting -osition

The most common method of determination of plotting position is the %eibul #ethod.
0ther methods e&ist and can be used in DISTRIB. To change the method clic) on the
name of the plotting position on the DISTRIB main screen. * plotting position ill be
calculated for e2ery data point in an array of data. The data must be sorted prior
"DISTRIB ill automatically sort the data$.
The a2ailable plotting position formulae are@

%eibull -robability



here
m A sorted number of data "9.2. ....n$
n A number of data points


California



'oster



3&ceedence


-rediction *rray

The prediction array in DISTRIB 2.0 uses only one type of distribution. and a forms
predictions based on a number of probabilities contained in the prediction spreadsheet.
The reuslts are placed in the same spreadsheet.

-redicti2e *nalysis

-redicti2e analysis based on a return period is automatically gi2en in the prediction
spreadsheet. If you ish to change the return periods used in prediction. enter the
corresponding probability in the first column and the clic) the type of distribution
desired. The calculations ill be automatically performed.

Distribution *nalysis

* Distribution analysis is automatically performed on the data in the main spreadsheet.
simply by clic)ing the button corresponding to the distribution type desired. The
probabilities for the indi2idual points are calculated by %eibull
e(uationH/-F%3IB4//.

#a&imum /i)elihood 3stimation

The ma&imum li)elihood estimates the distribution parameters such that product of the
li)elihoods of the indi2idual e2ents "/$ is ma&imi<ed. In terms of an e(uation this
becomes an estimation of ,... such that


is ma&imi<ed.

#ethod of #oments 3stimation

The method of moments uses the calculation of the rth moment about the origin of a
distribution.



The probability function p"&$is then directly substituted into the e(uation and the
distribution parameters are sol2ed for directly. 'or e&ample in the case of a 6ormal
DistributionH/-F60R#*/ it can be shon that distribution parameters sol2e into the
form.

0pening and Sa2ing DISTRIB "G.DST. G.DD2$ 'iles

DISTRIB 2.0 opens and sa2es all files ith a DD2 e&tension. These files are sa2ed in a
*SCII format. 'iles ill be sa2ed as a list of numbers for each of the data points in the
analysis. Data can also be importedH/-FI#-0RTI6+D*T*.



Importing Data

Data can be imported in both DISTRIB 9.0 and DISTRIB 2.0. The data must be in
*SCII format ith one number per line. There should be no e&tra spaces. 'or e&ample
if you ish to import a set of data ith H pieces of data then the file ould loo) li)eE

GGGGG Beginning of 'ile > Do 6ot -ut this /ine in =our 'ile GGGGGG
99.27
92.:8
98.B:
9B.IJ
H.:8
99.JB
98.72
9H.:8
92.IJ
GGGGG 3nd of 'ile Do 6ot -ut this /ine in =our 'ile GGGGGGGGGGG

=ou can create this file using %indos 6ote-ad or any *SCII te&t editor. be sure there
are no e&tra spaces at the end of your file.
3diting Data

Data can be edited directly by changing the data on the spreadsheet in the distribution
indo. %hen the distribution is selected the data ill be sorted. This data can also be
sa2ed to dis).

3ntering -lot Te&t

-lot Te&t Cannot be modified in DISTRIB 2.0

-lotting the Distribution *nalysis

The distribution *nalysis can be plotted. The plot ill contain the actual data and the
prediction The plot can be sent to the %indos clipboard using the copy button. To
print a plot use the -rint button.

-lotting Histogram

DISTRIB 2.0 cannot plot Histograms.

-asting Data from Spreadsheets

Data can be pasted from a spreadsheet directly into DISTRIB. To do this you must ha2e
the data in columnar format in the spreadsheet. In your spreadsheet highlight your data
in columnar format and copy it to the clipboard. Sitch to DISTRIB K*lt Tab or Ctrl>
3scL. Select 3dit. -aste from the DISTRIB menu. The data ill be sorted and displayed
in the spreadsheet control in the distibution indo. =ou may edit the data ithin this
display. Do not orry about blan)s in the data. as DISTRIB ill remo2e these
automatically. Clic) 0) from the -aste data dialog. =ou may no analy<e the data.

También podría gustarte