 |
Download a pdf copy of this help file here |
|
Download a pdf copy of this help file here |
Format: VoseUniform(min, max, U)
A Uniform distribution assigns equal probability to all values between its minimum and maximum. Examples of the Uniform distribution are given below:
The Uniform distribution is used as a very approximate model where there are very few or no available data. It is rarely a good reflection of the perceived uncertainty of a parameter since all values within the allowed range have the same constant probability density, but that density abruptly changes to zero at the minimum and maximum. However, it is sometimes useful for bringing attention to the fact that a parameter is very poorly known.
Sometimes we want to get a rough feel for whether it is important to assign uncertainty to a parameter. You could give the parameter a Uniform distribution with reasonably wide bounds, run a crude sensitivity analysis, and see whether the parameter registered as having influence on the output uncertainty: if not, it may as well be left crudely estimated. The Uniform distribution assigns the most (reasonable) uncertainty to the parameter, so if the output is insensitive to the parameter with a Uniform, it will be even more insensitive for another distribution.
There are some special circumstances where a Uniform distribution may be appropriate, for example a VoseUniform(0, 360) distribution for the angular resting position of a camshaft after spinning; or a VoseUniform(0, L/2) for the distance from a random leak in a pipeline of segments of length L to its nearest segment end (where you'd break the pipeline to get access inside).
Sometimes you might have a complicated function you wish to plot for different values of an input parameter, or parameters. For a one parameter function (like y=GAMMALN(ABS(SIN(x)/((x-1)^0.2+COS(LN(3*x))))) for example), you can make two arrays: the first with the x-values (say between 1 and 1000), the second the correspondingly calculated y-values. Alternatively, you could write one cell for x: =VoseUniform(1,1000) and another for y using the generated x-value, name both as outputs, run a simulation, and export the generated values into a spreadsheet. Perhaps not worth the effort for one parameter, but when you have two or three it is. Graphic software like S-PLUS will draw surface contours for {x,y,z} data arrays.
A Uniform distribution is often used as an uninformed prior in Bayesian inference.
The unit Uniform distribution, i.e. Uniform(0,1), is used in the generation
of nearly all other distribution types.
Fitting
a Uniform distribution to data is problematic and not recommended. ModelRisk
does not provide the ability to fit a Uniform distribution to data, because
it uses maximum likelihood methods for all distribution fitting. In the
case of the Uniform distribution, this would make the estimated parameters
for the Uniform equal to the minimum and maximum observed values, which
is a counterintuitive result (why would it not be possible to have values
outside of the range of observations?). If you wish to fit a uniform distribution
to your data, the most pragmatic approach, based on order statistics,
is to use the following:
Let
be
the number of observed values, 
and
be
the lowest and highest values observed, then the fitted distribution is
equal to:
VoseUniform generates values from this distribution or calculates a percentile.
VoseUniformObject constructs a distribution object for this distribution. Professional and Industrial editions only.
VoseUniformProb returns the probability density or cumulative distribution function for this distribution. Professional and Industrial editions only.
VoseUniformProb10 returns the log10 of the probability density or cumulative distribution function. Professional and Industrial editions only.
