next up previous contents
Next: The Lognormal Distribution Up: Some Theoretical Distributions Previous: Some Theoretical Distributions   Contents


The Normal Distribution N( $/mu ,/sigma ^{2}$)

The values of many quantitative variables are concentrated around a certain value and larger deviation are rather rare. This concentration (better density) could be represented by a function as shown in Figure 2.7. Traditionally the normal distribution is used as approximation of such a behavior. Under certain assumptions it is also possible to show mathematically that the distribution of a sum of many independent small errors tends to the normal distribution.

The density depends on 2 parameters $/mu ,/sigma ^{2}$ and is defined by

/begin{displaymath}f(z)=/frac{1}{/sqrt{2 /pi} /sigma} e ^{- /frac{(z- /mu)^2}{2 /sigma^{2}}},
/; /sigma > 0/ / / ,/end{displaymath}

where

/begin{displaymath}EZ=/int ^{/infty}_{- /infty}zf(z)dz=/mu, /; / / / Var(Z)=/sigma^{2}/ / / ./end{displaymath}

The location parameter $/mu $ corresponds to the expectation (mean) and the scale parameter $/sigma $ is called deviation or standard deviation (see Figure 2.7).

Figure 2.7: Density of the Normal Distribution with Mean $/mu $ and Standard Deviation $/sigma $.
/begin{figure}/begin{center}
/mbox
{/beginpicture
/setcoordinatesystem units <1....
...3174 -2.5 .01753
-2.75 .00909 -3. .00443 /
/endpicture}
/end{center}/end{figure}

The distribution function

/begin{displaymath}F(z)=/int ^{z}_{- /infty} f(t)dt/end{displaymath}

and also its inverse (quantile) are difficult to compute. Let us consider the standardized variable

/begin{displaymath}Y=/frac{Z- /mu}{/sigma}/end{displaymath}

with the density $g$ and distribution function $G$, then

/begin{displaymath}E Y=0 /; /mbox{and} /; Var(Y)=1, /mbox{and}/end{displaymath}


/begin{displaymath}g(y)=/frac{1}{/sqrt{2 /pi}} e^{-y^{2}/2}/ / / ./end{displaymath}

Figure 2.8: Density of the Standard Normal Distribution.
/begin{figure}/begin{center}
/mbox
{/beginpicture
/setcoordinatesystem units <1....
...3174 -2.5 .01753
-2.75 .00909 -3. .00443 /
/endpicture}
/end{center}/end{figure}

This distribution is also called standard normal distribution $N(0,1)$, and $G$ resp. the inverse is tabled in many text books. For the transformation we simply need

/begin{displaymath}F(z)=G(/frac{z- /mu}{/sigma})/ / / ./end{displaymath}

The model of the normal distribution is used in practice, if e.g. the investigated mineralization, as in iron-ore, has high grades and the variability is relatively small, as also in the thickness of stratified ore deposits or the density of ore.

Example 2.4: Let us consider the data in the example at page [*], and the model of normal distribution. How large are presumably the shortest intervals (tolerance intervals), which consist of approximately 70%, 90%, 95% or 99% of the values? From the table of probability distributions (see Appendix) we find

$/begin{array}{lll}
Q_{.85} & = & 1.04//
Q_{.95} & = & 1.64//
Q_{.975}& = & 1.96//
Q_{.995}& = & 2.58.
/end{array}$

In case of a standard normal distributed random variable $Y$ we would get intervals of the sort [ $-1.04, 1.04], [-1.64, 1.64$], etc. From the data (grouped in the Stem and Leaf Display) the mean is estimated as $/bar{z}=23.5$ (as approximation of $/mu $) and the standard deviation $s=1.86$. Therefore the transformed intervals of

/begin{displaymath}Z=Y /sigma +/mu/end{displaymath}

are computed to be:

$/begin{array}{llcl}
70 /% : & [-1.04 /times s + /bar{z}, 1.04 /times s + /bar{z...
....58 /times 1.86+23.5, 2.58 /times 1.86+23.5] & = & [18.7, 28.3].//
/end{array}$


next up previous contents
Next: The Lognormal Distribution Up: Some Theoretical Distributions Previous: Some Theoretical Distributions   Contents
Rudolf Dutter 2003-03-13