next up previous contents
Next: The Variogram Up: The Geostatistical Language Previous: Regionalized Variable and Random   Contents


Mean and Variance

A random function $Z=Z(/omega,/boma{x})$ depends on two arguments, one is for the elementary random events $/omega$ and the other one for the geographical place $/boma{x}$. In this subsection we only consider a particular place $/boma{x}$, such that $Z(/boma{x})$ may be seen as a usual random variable. An important tool for the description of the distribution is the notion of the (mathematical) expectation $EZ$ of the random variable $Z$. It is defined by the weighted mean of all possible values, weighted by the probability density $f$, namely

/begin{displaymath}EZ=/int^{/infty}_{- /infty}zf(z)dz/ / / ./end{displaymath}

The integration is over the whole domain of values, e.g. in the case of grades in % it would be from 0 to 100%. In case of only finite many possible values $c_{j}$ with corresponding probabilities $f_{j}, j=1,/ldots,k,$ we simply use the sum:

/begin{displaymath}EZ=/sum_{j=1}^{k} c_{j}f_{j}/ / / ./end{displaymath}

This theoretical expectation of a random variable is often simply called mean, $m=EZ$.

The variance $/sigma^{2}$ of a random variable is defined by the expected value (i.e. the mean) of the quadratic deviation from the mean $m$. In the general case we have

/begin{displaymath}/sigma^{2}=Var Z=E[(Z-m)^{2}]=/int^{/infty}_{- /infty}(z-m)^{2}f(z)dz/end{displaymath}

and in the case of finite many possible values

/begin{displaymath}/sigma^2=/sum_{j=1}^{k} (c_{j}-m)^{2}f_{j}/ / / ./end{displaymath}

The square root of the variance is also called standard deviation or (sometimes) dispersion. It represents a measure for the width of the distribution, i.e. for the variability of $Z$. The unit of the dispersion is of course the same as that of the variable $Z$, and we may define e.g. a tolerance interval $m /pm 2
/sigma,$ which says that a realization will fall in this interval with a certain (high) probability.

In practice, if $n$ empirical measurements (realizations) $z_{1},/ldots,z_{n}$ of a random variable are at hand, the expectation may be estimated by the simple calculation of the mean. For this estimated mean $/hat{m}$ we have

/begin{displaymath}/hat{m}=/hat{E}Z=/frac{1}{n} /sum_{i=1}^{n} z_i/end{displaymath}

and for the estimated variance

/begin{displaymath}/hat{/sigma}^{2}=/hat{E}(Z-/hat{m})^{2}=/frac{1}{n-1} /sum_{i=1}^{n}
(z_{i}-/hat{m})^{2}/ / / ,/end{displaymath}

where the denominator in general is $n-1$ to compensate the loss of information because of estimation of $m$.


next up previous contents
Next: The Variogram Up: The Geostatistical Language Previous: Regionalized Variable and Random   Contents
Rudolf Dutter 2003-03-13