You are on page 1of 21

- ~-

.
Iber 1960 November 1960 Journai of the South African Institute of Mining and Metailurgy 231

ON THE DEPARTURE OF ORE VALUE DISTRIBUTIONS


FR01\1 THE LOGNORJ.\¡IALMODEL IN SOUTH AFRICAN
GOLD MINES

By D. G. Kri~e. :\LSc.(Eng.) (l\Iember)*

SYNOPSIS

A larg-e num\'('r of gold. llranillm I1lldpyrite \"l1lnedistribntions on mines in the Kinross area,
o the \Vitwatersrand and exten!<ions, and tho Klel'ksdorp and Orange Free Stuto fiolds are analy!<eJ
and it is shoWTlthat tltere ¡",a ro~ulal' pattorn of d('partn:-e from the pI'edously aecepted h.gnormal
model which, ir ignor"d. eall leud to iJiasscd ami inefficit>nt."tatistiC'al estima te;; of the poplIlation
mean. lt is Ru~g..sted that this dcpnrture can be deolt with by adrling a cOllstunt a to 011values.
¡.I>.by normalizing the distr¡blltiúns on the \':Hi...of thr, ':ariable log. (;:;+ a). AA analysis is r:-,cn
of the eITed of the valne of a on thfJ biassos in ami the efficiencies of estimates based 011 this
model, and snggcstions are made for the a priori aaswul'tion of valuos of a rOl'borehole valllations
in specjf\c flelrls.
'Ce8S8.ry
. Somo prcliminary Rnggc«tions aro also mado regardin~ the estimation or confidence limitA rOl'
estimat~ of the popullltion mean calculated on tltis a¡'prolJ.ch.

IN'rRODUCTION
ili in the
Statistical methm!s of ore ,-aluation have, ta a limited extent, been employed 011
the South African gold mines for about ten years and, except for the statisticaJ control
of mine sampling.7 these have all heel1 hased 011the lognormal frequency di"trihution
f
pa.ttern first applied to goJd values by I-;ichel.~ Various difficulti~R wcre, howover,
eneountered from time to time In applying the Jognor'11al mouel in certain mines to J

ortl distributions in respcct of ore blocks and ~tol,3 faces under-ground and dcpz.rturcs
J
from this model w¡;re also obsen'ed for value distt-i;"utions from whúle sectiúl1s of !
those min~s. It was, however, a>;gumed that th~ model wou\d. still be suitable tor f
dealing with di"tributions in respect of large : .ef a-n:as, e.g., borehole valll~s trom
extensive areas such as whole mining properties or a combination of snch propcrties.
A critical review of distributio11 pattems throughout t~e South African gold
fielili has, however, so far been lacking and this lcd the author to collect. with the I
valued assistance of the other Gronp Head Offices coneerned, data from a large
number of gold mines stre~ching from Kinross in the east to Klerksdorp in the west
and souchwards to the Orallge Free State field. On analYf\is this survey suggests that r
with suitable adjustmems, the lognOl'mal mOQeJ can now be applied with more
eonfidence and on a wider basis.

PATTER OE ORE DISTRIBUTIONS

Oí the valua distributions collected, 35 are sho"'n in detall in Annexnres


1 and II and cover the foHowing:-
. Gold :
Kinross area Three mines (borehole valnes)
East Rand ... Three mines

· Financial Engineer, Anglo Transvaal Consol. lnvest. Co. Ltd.

~ '".._...-... .J
p.l
j
!

,
.!

--- ---- --
.-.------. - - ----- ---......-

Novemb
232 Journal of the South African Institute of Mining and Metallurgy November 1960 Qnu..
o,. 11,,- ,Ieparlure of ore m/lle '¡j8tributiotl' from thP. lognormal madel in SOIlI1, ""frican Go/d Mine8-D. G. Kriqe

Central and West Rand. Three mines


Far West Rancl One mine
Klerksdorp field Four mines
O.F.S. field Ten mines and boreholes
Uranium :
Klerksoorp &nd OF.S. fidds Six mines
Pyrite :
O.F.S. field o" ... ... ... One mine. .,..
ti>
These distributions have been "..na]ysed by the p]otting of the customary histo-
grams (on arithmetic and logarithmic scales for catE'gory limits) and of the cumu- :II
:z: ...
I lative frequencies (on logar~thmic. probabiJity paper). Without ex ception , but to u..J..
various extents, aU the histograms and log-probability
deviation from the theoreticallC'gnormal
plots show a consistent
pattern. This is well iIlustrated in character Ci
"
\f
=
5:i
0:::.::.
by the gold va]ue distribution for the Merriespruit mine based on 1,000 values u.. >-
::;liT
f<elected from raises a'1d drives on a grid covering the who]e of the mine workings <.>
<C
-1 up to October, 1956 (No. 28 in Annexures). The relevant histograms and log- .'.
e
probability curve for this mine as shown on Figs. 1, 2, 3 and 4, together with the
theoret;callognormal curves (markf'.~ A on each figure) reveal that :- ""
(i) The basic value distribution (see Fig. 1) has a high concentration of very .
10
low values.
I
Ip (ii) On a log.scale (Fig. 2) tr.e .~istribution is not symmetrical but has its peak
to the right of that of the corresponding normal curve, and has its longer Fla. 1-8
based 00
tail towards the ]owcr vahes (Le., it is negatively skew); in addition it
t has a sccond mode (not shown on the graph) in the lo\\" "alue categories
t due in this p::.rticular case to a 6,4 per cent frequency of trace values.
(iii) On log-probability papel' (Fig. 4) the distribution is not a straight line
(Le., the thcoretical ]ognorma] modcl) but curves downwards with increasing
curvature towards the lower values.
Tbe data in Annexure l furthp.r.more confirm tha,t this departure from log-
normality is general and applies equalIy to go]d, uranium amI pyrite values; further-
more, it cannot readil,v be attribuwJ either to assay errors of a higber order lor the
lower values or to an a bnormal mixt;'re of high and low va]ue sections in the relevant
mines. The only valid conclusiol1 to be drawfl is that this departure is inherent in a.ll
the dist.ributions and that this pattcrn has w be accepted as basic to the ore diotri-
butions concerned.

EFFECTS OF DEP,\,RTURE FROM LOGNORMALITY

As will be seen from Annexure 1, t,he main effect of a.pplying lognormal theory
to the rlistributions of individual values is that the statistical mCHIl (i.e., the "t"
estimator of Sichel6) is posith-ely L~,.ssed in all cases to an extcnt ranging from a.
fmction of a per cent (Blyvooruitzicht) to ovel' 100 per cent (Cjty Dgep). Furthermore,
the relevant samp]ing distributions r..~ l\IClTiespruit (Nos. 3; and 38 in Annexures
l and 1I) show that, apart from the bias, and cven if this eould be corrected for by
-
tbe applicatioll 01 an overall correction factor, tbe statistica] mean can under extreme
conditions prove to be less efficient than the arithmetic mean, the relative efficiencies
being measured by th., logarithmic varia.nces shown. A similar effect wa.s observed o
for the President Brand mine referred to again laOOr.
It is necesaary to stress at this point that in many cases the dcparture from
FI.. 2- H
lognormality and its effect on statistical estimatcs, is not significant, and this wss based on
particu]arl)' the posit,iun for the tw ~ mines (Le., East Gcdu]d snd Blyvooruitzicht)

..
i
¡;

I
I
--- - - --- -- - ----- - --- - - - -
.- - '--'~"
-.

)~r 1960 November 1960 Journal of the South African Institute cf Mining and Metallurgy 233
Onthedepartur.%,e t'allUdi.t,ibuU",..f,,,,,. th, log'I<.rma/
'I/Iod./in South Af,ican Go/d.Uine,-]). G. Krige
'{rige

....
.... !il
histo- ;r. ...,
curou-
out to ::z:
U-I
= e
sistent "' 1- ='",.....
lracter j V L.U I
o:: ;:.
...
values ::o
.rkings
....
I
d log-
th tho -I
e;
,
.1 very '0 4070100 200
I .v. I
300 400
61 .'39
500 600 700
-
20

800
9
a
1000
ARITHMETiC SCALE - INCH OWT
8 peak
Fla. l-lIlslo~ram 01 1000 ~old values from Ihe I\ferrtesprult mine wlth Ihe theoretleal normal eurves
Ii
longer based nn tbe loa.trhhms oC the values and on Ihe lo~artthms of the (values plus 55) marked A and B ¡
tion it respeetlvely.
~gories
.
lt tine
-easing

)1 log- i
Irlher-
:01'the
levo.nt
: in all
distri-
{) -o >- I
C-')
:z:
:
theory
le "t"
a I :.
;'oro a B:.
::. J
[fiOre, -<
exures _II- c..>
for by -.
;t1"eme
[encies . L
oerved . o
, froro
is "'as FI¡&.:3- HlstoCram 01 1000 ~old values from tbe Merrlesprult mine wltb tbe theorellcal normal eurves
based on tbe loaarltbms 01 Ibe values and on tb.. lo~arltbms 01 Ibe (value plus 55) marked A and B
respe-.:tJvely.
tzic~t)

- ---- -- - ,. ~ _._~--- - _. .'-.- ... -- -- -- -~ , -

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


--- - .- ~ -'.- _.. -- -'-'-' '. -, ~

234 Journal of the South African Institute of Mining and Metallurgy November 1960 Novemtl
On tI.edtpart!lreóf oreMllle rli,'ribv'ion. frol/! lilelog,.o",UJ1
'H{~ltl;/1S01.tI.Afriefln Gol"]Jfi",-D. G. K~ o.. lA

8cale (~
on which most of the author's original investigations3 were based. It is also for this
negligil
reason that the need for a more widespread amI critical inve8tigation did not become cases tj
apparent until more recently. cent on
Nevertheless, no mine valuator would like to employ a basis of valuation which dealinc
would tcnd to lend a consistent positive bias to his grade estimates; in fact, the solved~
natural tendenc,r is to prefer, if anything, a negatiye bias, particularly in the case unattM
of borehole e¡;timates on which the investment of large capital snms has to be risked.
The need is, therefore, clear for a new approach to the problem. What is required
for cstimating the mcan from a small number of values is a relatively simple,
unbiased and efficient estimator, which wiII silo", a significant improvement on the
straightforward arithmetic mean. I
I 11 SUGOESTED NEW APPROACH c. o
In the search for a new estirüJtor the 100 sets of 10 values each from the
~
Merrií'spruit mine already referred LJ (distributions Kos. 3i to 'u in tbe Annexures)
¡ were used to test firstly the possibility of using either tlle geometric mean 01' the
median of a set of yalue¡; as the basic factor in a new estimator. Reference to the
,I
~
logarithmic variances of the sampling distrillUtions for the geometric mean and the
median (Nos. 39 and 40) shows that for this mine both of the;se are less cfficient than
the arithmetic mean (Xo. 3i) and alternative approaches \Vere, therefore, conddered.
I
The possibility of fitting a lognormal curye on a truncated basis,I,2 on the
upper yalne portion of the distributions only, was tried amI gave sati8factory resulta
in several ca:;es. Tlle replacemcnt in effect. of the long lefthand tail 011thc logal'ithmic

Ii
I
'":- (} ,
>- ..,
c:..;>
:z: "" / 133
/
LL.I
::>
c::f
-.
-c
...
;:::
/ -1-.
'1 L.L.I
ex: :::!
t IJ,.. ou.
::r:: {
i 1-

i.
r'
"
..'"
, .t
,1

112 78 708
43 "-
13
1I~2 1718
4
2818
l
4461
-
,
7080

SCALE OF LOG (Z+55) - INCHDWT Flg. f-C


FII1.3- Hlstogram 011000 Merrlesprult values wlth the theoretical Dormal curve (B) plotted OD a value
probablli
s\.ale 01 lul1arlthm or the (\'alue plus 55).

- - - - -- -- ---
- -- --- ... .- - - -- '- - ... :, __o

Der 1960 November 1960 Journal of the South African Institute of Mining and Metallurgy 235
K~ Qn t}¡3d,-porture of ore l'IllU~di"tributiolls fro", the lognOT1IIa!
",oo,¡ in Sout}¡ .dfrÚ'an GtJLlJfjnu-D. G. Krige

scale (see Fig. 2) by one similar to the righthand one was found to introduce a
I
for this
become
negligible pm!itive bias in the estima te of the mea.n in sUéh cases. However, in other
cases the fit of the truncated lognormal f)urve even if confined to the upper 50 per
cent onIy of the observed values, was not satisf.-.ctory. In any event tohe prohlems in
~lwhich dealing with smaH truncated logno!'mal samplel' ha ve apparemly not yet been
~~t, the solved and furthermore the los8 in efficiency \ViII prohably render such a basis
he case unattractoivc. I
risked.
2000
equiroo I
simple,
on the
-/
/ ,,, I
/ .....;,/, ""'1000
i
oro the
lexures)
or the
- .. d ¡"",.
:.,..--
.. '/

I
,,,,"/ I
I 40e j
to the ¡I
3.nd the i ..J "'7 : 500
ütthan
- I
.. .
B. THEORETlCAL
lOGMORMAL
CN LOG.
(z . ,,
:-idered. \ ,,'." ;/ I 200

V
on the
. results ......
,..
:,,) ui
...
I
rithmic ,..... .. ...
.. .... / / tOO
I \
......:..:...:. t
,..." Q
/' A. THEORETltAL lOGHO¡Ud,,:
I
%
u
./'" // I I I I 50

y' V 40

i/ 10

/ 70

/ .mll
\ DI""'''''' " 'OO' "LO".
I

I I fO
-f
I I I

¡ I
I 5

.t
5
5 10 20 30 40 ~O 60 lO 80 '0 98
..-J..-
~67 1080
.. CUt.fULAT\VE FIlEQUfNCY PER CENT.
!Jgarithmic ProbabililY~p'er:
Pla. "-Cumulatlve F,'equency Dlstrlbutlon of 1000 values from the l\ferrlesprult mine on lo~arlthmlc
probablllty paper wlth tbe correspondlnC theoretlcal loanormal equivalen.s based on the values and on
111a value the (value plus 55). rr.arked A and 8 respectlvely.

, ~ _. ,~..., ~ ...-..-.-- ~.-

-------
---"----- ~ r__________ --- -.- ._---
I
, I

236 Journal of the South African Institute of Mining and Metallurgy November 1960 November
o" IlUlJ'lH,rttlr,010" ..du. '¡j8lributioll8lramtlUllo.Jr.ormalIilOlI,1
in SouthAlrican GoldMira-D. G. Krif/t Onlhe I

The relatively simple device of normalizing the distributions by dealing with tba the varÍJ
variable: logarithm of the (value plus a constant), was subsequcntly analysed and should b
has proved satisfactory from a practical point of view. This device is evidently not less thar
new,),2 but the author was unable to trace any treatment of small ¡;ampling theory infiniteh
for this case. Although a natural ju~tification for such a procedure is difficult, it can
be argued that 1\ basic 10gTJormal di¡<t.riLution of ore values if subjected to leaching
where a
mino (6..
which remo ves the same <]ua:1tity of mineral per unit arca throughout, will be retical d
changed into a distri l ~tion úf the type reljuiring normalization on the basis of tbe tion me!!
transformed variable log (value plus a constant), Le., log(z+a). and othe
As suggested by Siche}? in dea~ing with the ratios of check sampling valucs, a.
depat'ture of this type may also be explained by assuming a mixture of two or more
10gnormal distributions witn différent means and variances. The evidence in UsÍI
Annexure 1 suggests, however, that ifthis is the case, it will be completely impractical mately ;
to at.tempt any separation of such distributions except possibly in the special cases
of distinct and well-defined oresboot:;, such as on the East Rand.
o. were no]
Merriesp
B in cae
Various methods ~~ave been suggested for estimating the best value of a, oucb tbis basi
as maximum likelihood, moments, quantiles, Kemsley's method and the grapbical be evide
meihod. Aitchison ano Brown,l after discussing these in some detail, recommended positiva
tha quftntile method. Only this ml:thod and the graphical method are, therefore, elimina:
explained briefly :- and the
(i) Graphical trial and eITor : Using logarithmic-probahility paper (see Fig. 4), reasonat
the basic distribution of :.. )s first plotted, and then the distributions oí cases, pE
( z -+-a ) with the constant a being increased in stages until t.he resultant The X2 t
I plot yields the most satisfactory straight line tit. For all practical purposes fl\ctory
tbis procedure will be found adequate, as the fit is evidently not very sensi- (No. 12
tive to small variations in the value of a eithcr side of the optimum value. cent wi1
cumulat
(ii) The t.heoretical solution to the above graphical determination can be tho 20 (
arrived at. as follows. If the median valne (i.e., the 50 per cent cumulative mines t}
I fl'equency yalue) for the actual distrihut.ion is interpolated (say, m) and the ] 3 and 2
values (say, l'l and F2) p ~ any t.wo cumulative frequencies 81'Rced equi-
11 distantly from and on eithp,r sirle of Hm" ftre al so int.erpolated (or read off probabil
distribut
the log-probability plot) then varions .
t _m2_~~ 1 ) the obse
a - Fl +}'2-21n ,t of proba
~
Where F) and F2 are the values at the 10 }Jer cent and 90 pe.. cent cumu- c. do not
populat.j
".

lative frequencies, or at 20 per eent snd 80 per eent, 01' at 5 per cent and mine as
95 per cent, ete.
routillO I
~ 'l'his follows sinee the distribution of log (z+i:t) has to be normal with Ac
"1 mean=log (median+a), and thus requiring that borehole
r,
10g (median+a)-log (Fl+a) = log (F2+a)-log (m+a) Here ag;
cent con
Le., 2 lo? (:t+a) = log (F2+a)+log(Fl+a) The
(m+a)2 = (FI+a) (Ji'l+a)
merelv a
m2-FIF2 of a ; tI
-)+ F 2--m')
~
~
"
and hence a = valut'.'It
From a purely thcoretieal point. ofview the snggested normalization with log, (z+
a) as the transformed variable suffcrs from the disadvantage of having a lower finite The
limit for the distribut.ion when z = o, Le., at !og ea for the normal or at a for the and ana
eOITesponding lognormal distributi(,'-!. When log, a is small relative to the mean of 16 to 28
I

-----
,..-- - ---

..

=r 1960
November 1960 Journal of the South Arrican InstitUte of Mining and Metallurgy 237
.rit/d
07&the dt?parture 01 ore tVJlue d.i8tributimt~ /rom the loglJormal -uwd,-I ti' .'"nl/; ,.1/,,;,"1,' ,:.,Itl Jfinrlf-D. G. Krige

th the
:d s.nd the variable log, (z+a) the effect is negligible, but when this is not the case, it
should be noted that the values in the tail of the normal distribution with values
ly not less than log ,a will 0.11occur at log ,a, thus telescoping the tail values into an
theory
it cs.n infinitely small category at the value log,a. Thi<:;oceurs, e.g., with a distribution
where a sigréficant proportion of the values are traces such as for the Merriespruit
aching
lill be mine (6.4 per cent-see Annexures 1 and II). Even in such cases, however, the theo-
of the retical disahility is apparently of no practical consequence in estimating thc popula-
tion mebn as will be evjdent from subsequent anll.Jyses of the data from this mine
and others where conditions are similar (e.g., CI...yDeep).
lues, .s. APPLICATIONS OF THE SUGGES'n:n NEW APPROACH
r more
:lce in Using t.hjs new approach and determining the value of the required a approxj.
actical mately as suggested above; the 35 distributions (Nos. 1 to 35) already referred to
1 cases o were normalized with the results tabulated in Annexure I. The position for the
Merriespruit distribution (No. 28) is i!lustrated on Figs. 1 to 4 by the curves marked
B in each ea~~e.With no significant exception the curves fitted to the distributions on :1
, such
this basis gave better fits than the corresponding straight lognorrnal curves as will If.
¡phical be evident from the relevant X2 probabil;¡.ies sho'.vn (Annexure 1). Furthermore, t!..e
lended
positive bias in the statistical estima te ("t")6 of the population mean is virtually I.
!refore,
eliminated in all cases and the agreement between the medians of the actual values
and the theoretical medians [=geometric means of the variable ( z + a) less a] is 1
~g. 4), reasonably close. The X2 test!'! djd not show satisf. I~tory levels for acceptability in all I¡
...
ions of
"
cases, particularly where the distrihutions consis'~ of very large numbers of values.
sultant 1Íle X2 test is, however, too stringent in such cases although the fit is quite satis- l.
II
Irposes factory for 0.11practical purposes; e.g., in the case of the Hartebeestfontein mine
; sensi- (No. 12 in Annexure n 22,500 values gave a PíX2>observel1 value) of only 0.1 per
value. 1
cent with fairJy coarse grouping of values, altnough the actual and theoretieal o{
:mn be cumulative frequencies ditTered by a maximum of only 0.i2 per cent at any one of
illative the 20 eategory limits used. AIso for the Hartt.~eestfontein and President Brand
md the mines the fits for the distributioIl.'! of grids of 1.000 and 740 selectcd values (Nos.
i equi- 13 and 23) gave probabilitie!'< of 14 per cent and ;:'.' per ceut respectively compared to
'ead off probabilities of 0.1 per cent and < ,05 per cent for th: complete development. value
distributions of 22':500 and 14,5\)0 values respectively (Nos. 12 and 22). There Me
various practical reasons why the theorctical C\1r',es ca:mot be expected to represent
(1) the observed distributions for such large numbers of value,¡ \\ithin the strict limits

. cumu-
mt and
o of probability; e.g., the fact that all the developrnent valueR to date for a whole mine
do not strictly rcpresent either a random or a systematic sele"tion from the parent
population due to variations in the development, grid from section to section in the
mine as well as in the procedUl'es and control of unrlerground sampling and of the
routine a"'saying of these samples over the long periods of time covered.
al with A case not listed in Anllexure 1 is that uf the tlistriLution of the \)1 O.F.S.
borehole values (first intersections) analysed by the author in a previous paper.3
Here again the X2 test on the basis of log (z+9) ¡raye a probability level of 74 per
cent compared with only :3\)per cent on the basis of log "-.
The X2 tests referred to ahove and recorded in Annexure 1 shoulrl be regarded
merely as relative tests of goodness of fit for the curve above the lower limiting value
of a ; the hypothesis is naturally rejected in all c~'SeRwh~re there are in faet any z
.j values below tha value accepted for a.
og, (z+ 11 EFFECT OF VARIATIONS IN THE CONSTANT a
er finite
for the The practical effects of varying the value of tha constant a have been observed
nean of and analysed for the gold distl'Íbutions in the O.F.S. aml Klerk;;dorp fields (Xos.
16 to 28 in Annexures 1 and lI) and the overalI results are summarized in Table 1.

.. --- -- - -.- --- -.-.

.
1
-...--...----

238 Jour"al of the South African Institute of Mining and Metallurgy November 1960 Nave'T
of "re mIllo diBtributio...
On 1MrlOparlllTf fruti/I,," lognormalmod.¡in 8uuth African GoúJMin..-D. G.Krifl. Q,¡

TABLE 1 1
say,6
on Sit
pr CE
Value or a(Ba.9u;
(inch-dwt) O I 20 40 60 80 100 I(]erk
O.}'.S.Ficld Reef) . As m(
Biasinstatistical ,1% -2% -3% -21% -4% -1)%
estimate or tu to to to to to of a
mean +30% +9% +2% +1% +!% +1% KJerk
0.6 .5 .45 .4 ,35 .3
sho\\i.
Variance or log. (z+a) to to to to to to
towar
2-f¡ 1.7 1.5 1.3 1.2 1.1
s:
Klerksdorp Field I

J ( J'aal Reef):
Bias in statiRtical
cRtimate or
0%
tu
.- 1%
to
- 2%
to - - -
G (O .A
the IK
mean ' 6% + 11% 0% (i) Fo
I Variance or log. (z+a) 1
o.; to 1.2 ,55 to ,9 ,5 to ,8 - - - t
I
I \1
In eompiling Tab]e J a]]o"".Illce was made for the fact that t11ereef areas covel'ed
I are not aJI of the same size ami t~'\t all the inch-uw!' values do not correspond to 1
single umlel'ground s:¡,mp]es and are therefore not a]] comparable to individua] q,W
borehole intoJ'8t'ctions. The figurei' shown can, however, be accepted as practica]
limits a pplica lile to borehole estimates for reef areas ,::'overing who]e mining pr~perties 1
01' substantial parts thereof.
F~rthermore, the position on hili.s as analysed a bove, although based on distri.
(ta +
bntion<1 of large numbers 01' va.lues for cach case dcalt with, can also be accepted as
covering the position for smal1 "sarrp]es" (Le., small sets of values) as it will t~ (2) Fo
evident frcom the [ollowing two test cases that thc bias is likc]y to be smaller in dea.ling
wlth small "samples":-
.TABLE n

("/ Pre~idcnt Brand mine (500 single values ~~Iected on a grid from distribution No. l:-Jj
\VherE
O 40 80 200
.1'\
Value or a (inch-dwt) . ~.
I and iO
Bias in ..tat.ist.i('ul mean basedon :--
(a) Total 500 values ... ... .. ... +20,1% +.0.6% -3.9% -8.0%
(b) 50 sets of 10 values elWh .. . ... +0'5% -2,4% -5'7% The ~
+14.7% I
I
I

T
\BLE II A straigl

(i) Merriespruit mine (Í,OOO singlE' vlilues-diRtribution No. 28)

Valne or a (inch-ctwt) u 55
Bias in statistical mean based on:-
la) 'fotal 1,/100 values +29.7% -1-5% .
(b) 1UOsets or I O values each +24,1 % - .8% oh~',
(z-r uj

,...---.

--- ---

- ..

,be,. 1960 November 1960 Journal of the South African Institute of Mining and Metallurgy 239
K'.!lIte an/M departure01ore val,,,,di./ributi",,, Irn... /heloounrmalm",leli.. .sou/hAfriL"" voltl .Hiuts-./J. 11.Kriue

The above allalysis, therefore, suggests that if the value of a is accepted at,
say, 60 inch-dwt for tbe Barial Hecf in the O.F.S. field then a borehole estimate bascd
on Sichers "t" estimator shollld not l::e bia¡'sed b~' moro than an insignificant -~!
100 ..'
--- p~r \Jcnt 01' +4 per cent; similarly for a =40 inch-dwt fOl' the Vaal R':Jcf in th'3
Klerksdorp fit>ld. the bias should also be ini!ignificant (-:2 per cent to O per cent).
--l)% As more data becomcs avaiJa.ble it may be possible to analyse the regional variation
to of a in more detail, e.g., a pr'eJiminary review of the position on thc Vaal Reef at
1% Klp:J-..;dorp sllggcsts that both for gold and uranimn valnes there is un overall trend
.3 ShOWl11gan ¡nerease in the valne of a whcn proeeeding fl'Om Stilfontt:in in the north
to towards the south and south-west.
1-1
11 STATISTICAJ. MEA~8 AND THEIR CONFIDENCE LIMITS BASED ON log (z+a)

&suming now lognorrnality for (z+a), the relevant statistical estimates [01'
the p:>pulation mean wiU be: 3,4, 5, 6
(1) For population "arianee unlmown :
la = antilogro [logro (z + aH4>\ V)] -a (2)*
Where log (z+a )=mean of common logs of observed (values+a) in a sample
covered of size n,
pond to V =5.3019 X unadjm~ted varianee of ('ommon logf<of ob8erved (vall1es+a), II.nd
:li'7idual
,racrica 1 if>(V)=the sum of a mathematical series defined and tabulated by Siche16.
operties Tha estima.ted varianee o[ (ta+a) wiH=

1 distri-
:pted as
(lo+a)2 {ednor[ 1+:2(:~l) + 2!:!2(n!;)(n-t-l) + ] -1 } (3)
will be (2) :F'ür population varianee known a priori : 3, .. .1
dealing " . .4342945 (11 --.1 )O'a2
t a=antJlog [
log (z+a)+ 2n - a J

)i
. ., .2171472 (n-l )O'-:t~ . ,
:0. 13)
=antrlog [log (z-t-a), n J -a (4) i
\Vhere O'a2=known variancc of natural Iog3 of thd valu~s in the population of
200 v (z+U) values,
and iog (z+a) has the same meallÍng as for (2) above.
t-
The varia.nee of (l"a+a) = Wa+a)2 [exp( 0'~2) -lJ (5) .j
The varianee of the above estimators may now be compared with that of the I
straight aritl.metie mean (Z)3,~ whieh
-
= ~ [eXP(O'a2)
- 1] (6)
.;. ~
n- 1 (unadjusted sample varianee) .(7)

· It is intorcsting to note
observed values 1'01'(z + a ) as wen n.~<!>(
th,Ü as tha \"alue 01' a approl\ehe.. infinity,
V) \Viii approach zcro, and antilog log (::+a) wiU approach
tho variKnec 01' the

(z + a) and therofore ta \Viii approadl ::. i.o.. the straight arithmetic mean.

__o __ _.-- - -

----
j

---~...~... - .--._-_.. ------

240 Journal of the South Afríean Institute of Mining and Metallurgy November 1960 Nov~n
Qn l!tedeparlure01"re .'fJl,,,,di.'ribution.lro1llti/elognormal"wdelin South ¿jfrieanGo/d.\finu-D. G. Krige VII&

Where (J = population mean for the variable (z+a)


Ua2 = variance of naturallogs of the population of (z+a) values.
z=mean of observed values for z in a sample of size n.
Basic d
and population variance ofz and of (z+a) values = (J2[exp (ua2)-I] (8) 8ampli.'
No.of
No. 0(;
As pninted out a9cve in referring to the actual samp1ing distributions obst>rved Theol'"
in several cases for "samples" or sett. with from 7 tú 12 values each (l\os. 36 tú 43 Actual
in Annexure II), tbe distribution of tbe aritbmetic mean and of tbe ta estimates mE
can be regardcd as 10gnOl'mal for prac:tical purposes.
1f, therefore, t1.e above estimatúrs are now all assumed tú be lognormally T
distribl1ted witb "ariances which can be expressed in tbe same form as (8) above,
then the estimated varia.nces of these estimators, Le., Nos. (3), (5) and (7) above l~an
be expressed as follows .-
o. e statist
T
confid.
Variance of ta=i=NC'.(3) above interv:
limitt
=i=(to.+a)2[ exp(ut~) -1] ..(9) t.heml
Yariauce of t"a=i=N~ (5) above
=i=(t"a+a}2[exp(ut~a)-I] (10)
I A
1
paragr
and variance of Z =i=n-=-1(unadjusted sample variance) meao
I of )(1
f =i= (zHexp(~)-l] .(11) No. 2~
¡t
.1 In all the above equations (9) to (11), the onl)' unlmowns in dealing with a set
!r (or !':ample) of values are UI~' Ut'~' lJ~ and estimates of these can, therefore, be
¡¡ solved for and \ViIIbe cquiw,lent to the estimated variances (i.e., 82)of log. (t.:.+a),
log. (t"a+a) and log. z respectively.. With the necessary adjustment between the
1I
.1 mean" of the corresponding normal and lognormal distrillUtions. Le., exp (u2/2) =i= Arithm
exp :'-~/2) a first. apIJroximation to the exact confidence limits can be obtained using t p-s:inu
,1
the Studt'nt distribution for sma]) sample" and the normal distribution for large 'a estiJ
!! samples; in a future papp,r,8 tablcs wi!! be included for deterrnining these confidence .'"'
limits.

11 E.g. For P(90 per cent), (Jbased on, say, 10 values wHI not be lesa than T
JI metic
approximately exp [log<~f.<x+a)-1'383 sla+1S2ta] oí o.~,
= (ta+o) [exp(ts2ta-l.383sta)] .(12) I1gaj11
wjth a
.Where S2ta = t.he estimat~ of Ut20:obtained from (9) above and 1.383 is the close t
10 pel'ce~t fractile of the Student's distribution for (10-1) degrees of freedom.
The above estimators amI their val'iances are all based on randoll1 sampling
theory whh,h, as pointed out by the author in a previous 113per.4 is not. strietIy B
applicable to Rystematic grid sampling as in boreholes or tú systematic interval the O.:
sampling as in ul1derground development. This is due to the stratified nature of the ore shown
deposit,s in the South Afi'ican gold mines, but t.he overall effect is sueh that for
practical purposes in borehole estimates random sampling theory can be applied. tI
This is supported by the following practical examples (Table 111) :-- arithim

."..,..~ ,.,.'_ .~_~,.~~___"',..,. -,.0'.'-

-- - - - ------- ----
p

~r 1960 November 1960 Journal o( the South A(rican Institutp :)( Mining and Metallurgy 241
ri~ On tht tÜparlure of ore ralue di.lrib"ti(/II. fr011i Ih;. lo<}norwl11ll0</et ," S.mth African (}(/úlltlille8-D. G. Kr;ue

TABLE !II

Arithmetic means la estimate


I
ti
t
Basic distribution No. ... .. . ... 13 28 '1 2 21)
(8) Sampl'ng distribution No.
No. of values per set (01' sample)
oo. oo.
oo.
33
10
I 37
10
42
12
,:1 41
10
¡ rved No. of sets taken ... oo. ... 100 100 300 100
Theoretical u" for mealla of seta t .10 I .1.4 I .16 60
.18 I .12
to 43- Actual log variance (u') found for
mates means of seta oo. oo. oo. I ,09 I .16 I .14 I .18 I .15

!Cany 'l.'he differences between the theoretical 'tnd observed values for 0-2 are not
.bove,
'e ~an statisticall) significant at a 9;")per cent confidence level.
11
The 100 sets of 10 drawn fl'om d:itributinn No. 28 wcrc :lIso used to tef~ the li
11
..
confidence limits as suggestcd in Xos. (9) and (12) above. For ¿Ion80 per cent confidence 11
interval (i.e.. with 10 per cent upper amllower limits) 7;")of the 100 sets of confidence '1
limit!> straddlcd the actual mean value, wherca" for a 90 pelocent, confidence interval h
(9)
tbe mean was covered by 85 sets of confidence limits.

PRACTICALTESTS FOR ta
..(10)
Apart from the case of the )ferriespruit fi!ine (No. 28) dealt with in the pl'evious i
pal'agraph and showing an improvement of the ta estimai:ol' over the al'ithmetic j
mean (see also ~o. 37 amI 41 in Annexul'es) a ~;milar test was ctlnied out on j}Osets
of 10 values each dmwn on a grid from the P"esident. Brand wol'kings (di;;;tl'ibution
..(ll) No. 23) with the fol!owing reslllts (Table IV) :-
TABLE IV
a set
?, be
..+a), Corrected logarjthmic
n the Variance:¡:,u:
2) ~ Arithmetic means of seta 0'20!
using t est;.nate" of sets 0.213
large ta "stimates : (1 = 40 0.l93
:lence (1 = 80 0.l9S
(1 = 200 0.183

These observed variances compare with the theoretical variances for the arith-
metie means and ta estimator8(fol' a = 60) Cltlculated from the parent distribution,
of 0.243 amI ü.2lO respectively, the differcnces bet~"~en actual" ami theol'etical"
..(12) again being insignificant. Thi8 example al so iHustrates tbat apart from bias dealt
with above, the efficiency of ta does not appear to be very sensitive to values of a
'< the close to 01' wel! in exeess of the optimum figure.
domo

~ling FURTHER IMPROVEL\I":NTS


TO ta
rictly Bearing in mind the praetical observed limits fol' the 0-2of value distributions in
erval
16ore the O.F.S., it i8 possible to furthf'r improve on these estimates by applying the limits
t for sbown fol' a=60 in Table I (Le., .4 to 1'3) to the ta estimates for ~lerriespl'uit
Jlied. . t Ba..¡ed on populatioll varjanl!e -:- I1l1mber of valnes pcr sct e'{uated to (U) ahovo (fol'
arithmetic means) and 011 the 801utioll to No. (9) above for the la estimates.

.-....--- - _.- ....-

-- --- --- -- - -- -
- - -_........ --- .-. .....

242 Journal of the South African Institute of Mining and Metallurgy November 1960 NovembeJ
On the depaTture 01 ore value distributi<J1I.lrolll tJ.e W¡¡normol model in Soulh ¿jlri~an Gol<l Mine.-D. G. Krige Onw,

(a=55) and the limits shown for a=40 and a=80 to the relevant President Brand . i Means :
ta estimates, referred to above, wherever the "sample" variance (colTected) of a &et
of 10 values falls outside these Ijmits. In such cases the relevant upper or lower
limiting value for (72\Vas used in the t"a formula (Ko. (4)) anrl yielded in the vast
majority of cases further improvemel:!ts on the straight ta estimates. Significant
improvements were effected mainly to sets of values which gave exceptionaJly high
or low arithmetic means a!ld ta v:>lues. The following is the overall position
(Table V) :-
TABLE V

Corrected logarithmic
variance -:-cr
-
MZRRIESPRUIT:
100 ta eRtimates ...
100 ta estimates with limita for cr
... ...
...
...
...
...
...
...
...
...
...
0.163
0.148
a o
a=40 a=80 The
PRESID};XT BRAND :
50 ta estimates ... ... -.. ... ... ... ... 0.193 0.H18 for (72,LE
50 ta estimates ,,-ith limits for cr ... ... ... ... ... 0.138 0.144 fO])OW8,.
0.2
This clearly shows the advantage of applying some limiting values for (72even
where these. as is the case in the O.F.2., cover a relativelywiderange. The confidence log (
limits for the:se impmved tc;:;(-stimate<.¡\Viii, therefoJ'e, on the whole be closer than
those for the snaight la estimates but as these cannot be caiculated, the more
conservative limits based on the straight ta estimator are, it is suggested, to be
Anti
appli~d. subject to the limiting values for (72used. 'l'his aspect is also covered in the
tabl~s ta be published8.
':XAMPLES
-
t"a
Presidu¡f
The following are typical exampics of the improvements in estimates obt..:.ined OW)
by applying the suggested methods a;nd indicate in detail the procedure followed ;- Set No. ::
Vah
lIf er¡ ;espruit .-
Overall mean of 1,000 values=261 inch-dwt. '10 2
.,
Se~ No. 39 :-
:1
Values (z) Values+55 Logs of (z+55) Logs2 of (z+55) 3
46 101 2-0043 4.01722 8
105 ]60 2.2041 4,85806 b
129 184 2.2648 fH 2932 1(1
168 223 2.3483 5.51451 l
206 261 2.4166 5.83995 ]64
225 280 2-4472 5,98879
244 :Means 21
299 2.4757 6.12909
292 347 2.5403 6.453l2
313 368 2'5658 6,58333
320 375 2.5i40 (i.6:?;i48

l\J('an~ 204'8 2'31:141= log (z+a) 5'713887

---- --- ----


- - ~._-.. .

r 1960
Novemb'!r 1960 Journal of the South Afri-::anInstitute of Mining and Metallurgy 243
Un tM dep..rtureo/ ore miue di.tribution./rom the lOf/normal
moo.l in South A/rimn GoldJfinn-D. a. Krige
rige

:;rand )lean" 204-8 21)9.8 2.3841 = log (z+a) 5.7]3887


.1 "('f L,',~ (:!.:¡S~1)2 =.,;"j'I:~::!1:::;
"Jwer
~ vast =0.029954
ficant x5'3ú19 =0.1588= V
- higil From Sichel's tables6 for V=0.1588 and n= 10
.sition
cf>(V)=0.0342
log (z+a) = 2.3841
=2-4183
:Die Antilog10=262=(ta+55)
" J
...,
y
-55
la = 207
-
!sO The con-ected variance~a2= lO~-1 X V=0.176 which is less than the lower limit
-
98 for a2, i.e., 0,4. Therefore the t"a estimator becomes applicable and is calculated as
44 follows, from formula (4) above:-
- n-l
0.2171472 X - Xaa2 =0.2171472
9
x-x''!
n 10
: even =0.0782
.dence log (z+a) =2.3841
than
more
2.4623
to be Ant.ilog =290=t. a+55
lU the -55
t.a =235

Pre..~ident Bmnd :
:ained Overall mean of 740 values= 1026 inch.dwt.
ed:-- Set No. 35 :--
Values (z) Values+80 Logs of (values+80) Logs2 of (z+80)
67 J47 2.16i3 4.69719
o 229
~98
309
378
2'4900
2.5775
6.20010
6.61351
311 391 2.5922 6.71950
341 421 2.6243 6,88695
825 905 2.9566 8.74148
829 909 2.9586 8.57331
1071 lli5] 3.0611 9'3703~ "
1217 1297 3.1129 !H9015
16-167 16547 4.2187 17.79743
Means 2~66 2246 2'8759=log (;¿+80) 8.5500
le8s (2'8759)2 =8,2709
=0.2791
(X5'3019) =1'4796~V
From Sichel's tablcs6 for V=I.4796 and
n=lO

_. - ----
j

-- -- ----
0.__'_"'-'__ _ _ a..--.__

244 Journal of the South African Institute of Mining and Metallurgy November 1960
On ti.. departureof ore ,'al". di81ributilJlI.fr"m11",WglIlJTmal
modd ill Soutl..AfricanGoldMina-D. G. Erige

,p( V)=0'3024
log (z+80) =2.8759
3.1783
Dlstrl-
Antilog =1508= (ta+80)
-80 No.

la = 1428 --
butlan I MI:
KI:oI
3 tnm
-
EA
The corrected vl1.:iance ~.r= 1~1 X V = 1.644 which is greater than the upper 2 :&:1
3 Wel",
limit for u2 whan a=80 (see Table 1) i.e., 1.2. 4 I E.R.I
-
Thcrcfore the tNa estimator becomes applicable and is calculated al) in the ('E:o;y
6 CroW1
abo\'e example
- -" n-l
:
9
c ('
.....-
6
.-
Cily

.2] , 14¡z X-Xua~


n
=0.2345 7
Wr.3
Rand
8
=z.8759 - -
+log (z+80) Fa
9 DI)"\'.
=:;.1l04 10
--. -
Antilog =1289 =tNa+80 11
KLEE
!'>tUfo
-80 1Z
13
HartA
di
14 DufI'e
tNa = lz09 --- 15
- ":ui

ORA!
16 Lorai
SimiJ..r1y for this set of values, for a=40, ta= 143G and tNa= 1303. 17 tlo.
18 Frffi.
19 F.ee
CO:S-CLUSION 20 \\,,,,,\
.,., 21 "di!
2Z Pr
The autJwr again wishes to expreRs his appl'eciation of the assistance :'.mdered 23
24 Prt..i
by the val'ious Group Head Offices c()ueerned D.l1dtrusts that this papel' ma:; prove 25 Well.
usefull10t only in futurc borehole valuatious fol' new propositions, but also in over- 26 Han!
27 Yiflo
coming certain of tbe difficulties encountel'ed to date in the straight application oí -- 2S
- Y'!tt
the lognormal curve in valuation problems vn the mines. t:RA:

REFERENCES
29
30
31
I Dulf,
I
Stllf,
1;Jnrt
32 1.01'''
1. AITCHISON,J., and BROWN, J. A. C. The lognormal distribution. Univeraity Press, C..mbridge,
1957.
2. lLu.D. A. Statistical theory with enginee!,ing applications. .lo1m Wiley and SOll." 1952.
a. e 33
34

35
I
I -Vlr
FTed
MCJT

Pnr.:
3. KRIGE. D. G. A statistical appronch to some basic mine y¡¡luation problems on t.hc 'Vitwaters-
rand. Jnl. Ohem. .Uet. ulId loJin. Soco S.Afr. (December, 1!J51). Disf'lIssions Ilnd replies, March, SAII)
.
1!J52 May, 1952 ; July. 1952; nnd August, 1952.
4. KRIGE. D. G. A statistic-al analy,.¡is of some uf the borehole vahu.'s in the Oral1ge Fre.' State
36
37
38
I .Yerr
Han

39
goldfieldR. Jul. Chem. Met. and Jlfin. Soco S.Afr. (September, 1952). Dil'Cu8bions ond repl"" 40
Xovembor. 11152,and :FebnloIJ', 1953. H
5. SICHEL, H. S. An experimental and theoretieal investigotion of bias error in mino sampling
with spocial reference to narrow gold reefs. Trans. 1nst. lo/in. (¡lid .Uet., London (Fobruary,
42
43 I F.Jtst
Wel¡

1947).
6. SIC'HEL.H. S. New methods in the ,¡tatistical evaluntion of mine sampling data. Tram. 1m!.
JJin. c/IId .Mel.. London ¡:\Iarch, 1952).
7. HOWLA:-w, H. STo .1.. ono RIOIEL. H. S. SI"ti~ji"'li CJlIality cont.rol of routine \Ulderground
sampling. JlIl. S.A. In..'. .11'/.. I/J"I .\Id. (.JlIlllltlr~', 1!lIHI):di"wIIR..ion(l\tarch. 1111;0).
8. KRIGE, D. G. Dovelo¡mwlIts in the valuation 01 gold mining properties from oOI'<'holeresults.
Awaiting pnblication for the Seventh Commonwealth Mining and l\1etaJlurgical Congress, 1961.

- '-- ---- -- ---


- -'"

334 Journal of the South African Institute of Mining and Metallurgy January 1961 January 1961 le
.
,
0.'1 Ihe IÚparlure o/ ore .'Olm rlislribmio>.. /rolll Ihe l"l!1!(>nnnl fllooel ill SOIJlh .4friC'llt Gold JIim,-D. G. Krige On lhe IÚparlure o/
f

I
1
~
,
t
.>-
"
l
t ~
\)'
IJ ..
~
l&.

,
~
..
I Ir para meter a>
normal <Estributio,
-55 O
~z (1..."," D , J-- will be'J-shaped, a¿
FIIt.I - Merrlesprult The straight log-!
negativc skewness 1.
The disagreement between ma themati cal model and observations is elearly
brol1ght out. by both thc tran8Íormed and original observations. The histogram of the
original data (:Fig. 1) suggest!< that, due t.o the large number of value.'.! below 4.5
inch-rlwt, the basie modcl should bt> J-shapcd rather than unimodal. When the In the limit, (2)
o!"iginal observations are transforr.led according 10 MI' Krige's proposal the eorrect 1l00'mality without St.
modd should be a combined J- and be!l-shaped distribution as the obscrvcd Mcnie-
Fil1i\lIy, transfor.l11
spmit histogram indieatc~ in Fig. 'l. Furthermore, if the old log-transformation is
used, the model should be unimodal and should disp)¡ty moderate negative skewness
as \Vas shown in MI' Krige's papel'.
To iIlustratc that the l\1erriespruit data are no mere freak and that they behavc This distributioll 1:
/1> 1 whieh is the Me
as they should,. 1 ha ve derived the following model:
The original ineh-dwt (z) valnes are deseribed by a 3-parameter law 1'he (,h1'OOrelevant
a speeifie set of const
P(z)=.iJfza-1 exp [-bzc] . , oo...,. ,.oo (1)

where .iJf= t~:) and O < z < + 00

.~~ - ".,.- ....


-

---
,
.1

!urgy January 196! January 1961 Journal of the South African Institute o( Mining and Metallurgy 335
nld.\1;,..,-D. G. Erige Qn the/Upar/u,. o/ oremllle di./ribu/ion:;freno/he1000llOfmal
fII~lelin S".ith A/rican Gold"\1iJ..,-D. G. Erige

I
I

,."
l
I ,..
tt
..
V.
II

t
10c;... 55 - ':1c. l.OG,. ('Z +-SS)__
!IC. :l - Merrl""l'rult

If parameter a> 1 the distriLution ha.s a mode and looks very much like a log-
. normal distribut;,:m characterized by its extreme skewnes~. Ir a::;1 the distribution
will be'J-shaped, as for example suggested by the :\'Ierriespruit observations.
'Ihe straight 10g-transfQrmation x=[og. z yields the distribution with the rcql1ired
negative skewness Lo.:
r,¡;ervationsis clearly p (x)=Uej;p [ax-be..'] (2)
_ The histogram of the f where -co~ x < + co
r of values below 4-5
unimodal. Whcn the In the iimit, ~2) bocomes normal and would covar those mines which display
'" proposal tho correct normality without addition of a constant a
:;;the observed Menie- i
; 10g-tl'ansformntioll is
, Finally, transformation y=log. (z+a) leads to
~atc negative skewness i f (y)=.M (eL-a)a-l exp [y-b {e!/-a)C] (3)
where logea :::;y < + CO
~and that they behavo t This distribution is unimorlal if a< 1 but takcs on a cOlllbined J- and bell-shape ir
a>l which is the l\[erriespruit case shown in Fig. 2.
!arameter law The three rclevant distributions of equations (1). (2) and (3) are shown in Fig. 3 for
a specific set of constants .
........................ (1) a -1
-2
b=iG
c=!

'-

--- --- ----


_11III"'1' --

II
I
¡

336 Journal of the South African Inst¡tute of Mining and Metaliurgy January 1961 January 1961 Jo
¡
On the departllre 01 ore mllle a'slribulion8 Iro", Ihe lognor",,,l ",OOpl¡Il 50ull. Alrieall Golti Mine_D. G. Krige On lhe departure 01 "
í
I 1 would like
t p. 534, Vol. 55, ir
. pc;z) _ M. ~a.-I. e-b"'c
~ of a being posit.ive
VH,.R" M. c:bt-
T( i) MI' Krige ha:'
O.Ez ~oo ]})odel, the statÍsi
I ordinary aritl).met
I (>rocedure used. T
that, due to a con
actions I have fail
f rstimates even if
I -z (I:NCH-DwT.)--
normality. Fig. 4 ¡
p("JC)" M:eGx-bec"OC. ShOWIlas a solid a:
i -OO$::c.:S OO

~
r
f t
,.
,
V
l
w
J
,- I
a --:x. & Loa.. Z ~
~
1: I
" I
I J
,

i
,I
- y_ Loc.. (z..~)-
FI~.3
,
'!
-_/
" ;'

Although this particular model appears to be more logical than 1he 10gi1ormal, we
ha ve achieved ollly moderate success with it so far and the sampling properties ofthl'
estimates for its three parameters based on a small number of ohservations, as
required 101'borehole valuations, are unknown. I ha Yt1to expre:
Returning now to MI' Krige's technique i1. ;;honld be pointed out that Merr¡"- \\henever it.s estim3
f'pruit is a rather extreme case and that the 10gnol'mal model is approximating the Is "Ol1tral'Y to an ac
observations of the ot.her minei:. rather better. I¡mi!;.; and associate.
I suggest the foHowing criterion for the use of the transformation l()ge ~¿-ra): An anaJogy to
a<exp [~a--2'326 Ua] (4) t...;!ing for si({niticam
,," a signiticance levE
This 0oudit.ion will ensure that the area under the curve for the non-existillg ,Iuthodtics and the
negative ore values will .be les3 than 1 per cent. The estirr3tion of l)(lrameter a is. .j.. tlw test.
as pointed out by MI' Krige, full of difficulty. MI' Krige's equatiol1 (1) is only valitl
if z starts at -a. In practice this is illlpossible. Howevcr, as long as my conditiotl JIJ addi tiOll, 1 J
(4) is fn1filled, the a estimat.e will Ilot be gl'o:ssly out. In mallY cases the graphil-al ! . l-stimator may be
solution might be prcferred to MI' Krige':s equation (1). :,,,:,\'1t¡'j1,l1ce i~ u2+ L

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


<::allurgy January 1961 January 1961 Journal af the South African Instiwte cf Mining ~nd Metallurgy 337
" GoldJfi,,,,,--lJ, a. Krig, ()n the depurture o/ ore ,,,,lu. dislribulions /rollO lhe lQfJllorllOulmodel in Soutl. .l/riwn Gold J1iues--D. G. Krige

1 would like to draw attontion to a ¡'ecunt papel' by Fuinleib in the .T.A.S.A..


-t _bs.' p. 534, Vol. 55, in which a maximum likelihood solution is offel'ed for the problem
.e:
.:bi- .
of a being positive, as is the present case. .
Mr Krige has sho\\'n experimentally that, in spite of a \Hong mathematical.
-rÜ-) tllodel, the statistical estimates for average grade are remarkably close to the
.0
IIrdinary arithmetic means with virtually no bias due to the maximum likelihood
procedu!'e used. The reportcd results \Vere good even for Merriespruit, It appears
J that, dt:e to a complicsted chain of compensatillg errors whose mathematical intcr-
:lctions 1 have failed to pin clown exa,ctly, tJ\C 't' technique gives reasonable gráde
estimates even if applied to a population whose logarithms ¡>,refar removed from
1 normality. Fig. 4 illustrates some of these points, In this di[~gram the pnpula,tion is
shown as a so lid and the fitted model as a broken eurve. .
,
CO~~EC7 t'10DEt.

NORMI/t. MOLJEL

I
I
I
I
.- I
I
I

($2 I I

~
1 I
I
// }.... ,
r

I 1

/
/
I
i
/
I l I
J
j
I
!
_.-""'/

.al tban th~ lognormal, we


I - y . l.OGe (z +IX)--
'.amplir,g propel'ties of the Flg.4

,i
nber of ohservations, as
1 havE>to express disagreement with l\f¡-Krige's suggestion to assume (72a priori
-;)ointed out that l\ferrie- whenever its estimate V excoeds 01' falls short of a predetermined qusntity. This
i~ contrary tn all accepted statistical philosnpy and wil1 result in actual confidence
.del i8 appt'oximating the hmits and associated probabilities not in line with those predicted from theory.

;ormation log. (z+a): An analogy to :Mr Krige's suggestion may be found in the falla{'ious way of
(4)
............................. tcsting for significance of difference, looking at the result of the test and then deciding
(m a significance level. This procedurc i8 strongly conúemned by allleading statistical
:lTve for the non-existing authorities and tho correct procedure is to fix the signiticance level Erst and then
matiol1 of para.metor a is. do tho test.
equation (1) is only val id
In addition, 1 havo recentl,' shown that associated significanee lovels for the
as long as my condition
1" estireator may bo grossly out if (72is assumed a priori, when in faet the true
:nany cases the graphica! lug-variance is (72+ 6,,(72. .

"'_"'1"~""

----
,.- '''"'-

Fcbruary 1961 Journal


338 Journal of the South p.frican IlIstituteof Mining and Metallur'gy January 1161
On 1Mdpparlur"o/ o", mi"" di.tribU/io".from Ihe¡ognar",,,l¡¡¡",lelin o:>ou.lh
A/rican GoldMines-D. G. Erige

In spite of my criticimn 1 wish to reiterate that MI' Krige's transformation, PROCEEDE'


couplcd with my 't' estimation method. iR at present the best available technique
for statistical grade evaluation. 1 wish to congratulate the author on a most significant
advancement of our knowledge.

Prof A. Linder (Visitor): 1 have folll.wed with great int.crcst the work done by
1\11'Kl'ige and DI' Sichel, for improving the methods oí mine valulltion thlOugh the The .Tanuary Ge
use of }>l'oper statistici1.l toals. 'l'his intcrest derives par tIY from the fa,ct that 1 have Johannesburg, ori We:.
to tcach statistics to students of geology at the University of Geneva, statistics being
compulsory for geologists at this university. Mr. W. S. Findla
1'h6 new step taken by MI' Krige in his prescnt papel', namely the transformation There \Verealso p
of gold values using log, (z+a) iustead of lag ,Z, is a further refinement. 'l'he rt sulting Messrs.H. E. Cross, j
, improvement may w~il be quite substantial and MI' Krige is to be commended for
Dr. J. de V. Lambrech
i t
Messrs. H. F. Allan,
I his investigation into the intluence of a upon the efficiE;ncy of estimates.
W'hile this refinement is a most useful one, as far as the estimation of the gold
I W. F. Keyes, E. K. !<,
I
Robcrts, W. T. Stohar
eontent of a certain bl.ock of ore is con cerned, the main problem t.úbe sol ved, it seems
to me, is to devise proceJures for cliscriminating between ore Llocks which are "payable" Two Associate M
,I
I
I and those which are noto I Four Affiliate 'ln
.f
Th~ proLlc/lJ is cleady very c('~plex; it involvcs fixing the size of blocks, deter. A. C. Nesbitt and E. j
mining the number of samples to be t.aken from one block and to give rules of "accep- 1 TweJve visitors, II
,,
I
tance" (J;""rcjection" of blocks ac(;crding to some given pa:¡-limit. It st:cms to mc,
that the methods developcd by DI' O. J~. DII.vies(l) and his co-workers at l'_1perial
t, Lentt, Dr. A. A. LiE~
Adams (Sccretary) an¡:
I Chcmical Industries, for screening che~llical compounds in vie\\ of their therapcutic
Total present: Fo
! valuo might be u:oefulJy studied 1y mb'1Íng engineers. The esscntiaJ poinb> of DI'
ioI t ])a\-ies's procedures are to procerd in two stages. In terms of the problem of sp,lectioll
¡ of ore blocks it would consist of a first prelíminary inspection of ore block!:!, USillg
the President sail
relativcly few samples. the aim Lcing to rejed blocks with very low gold contento
From those block:,; which arc acccptcd atthc first st3.ge more samples are takon in a
, JOhllston, a Past Pres
i 3.;cond ;;t.agc', so as t.ú arri,-e at a final class¡fi~atioll of ore hlocks. This method 6pahles
f 5th .Tanuary, 1961, an
I (lne t.o work aceording to given risks; one risk being to rcject a block with paya bIt:
J Institute on I hh AUgl
i ore content, thc other heing to accept a block w¡th unpa.yablo ore eont..ent. 1960. As a mark of i'¡
The ult,imatc goal :Ü which all efl'orts should be dircct,ed, is to arrive bt a pro- i meet.ingrose and obsc'
I
'ceduro which allO\vs one to select ore bloeks in the most economica.l way. -.:think
that a suita lile modification of DI' Davies's rncthods ¡¡hould prove of great ralne to
I
I
the golf] mining industry.
f
J ,f The Minutes of th
in the January, i96i, is
I

,
I DftVies, O. L. ; Some aspects of thc economies of anal~'tical tcsting. Technometrics, J, 1959, p. 49.61
i
¡
I
t" The PresidcnL an
¡
I ¡, having been published
, thcm to mcmbership it
I

, i
! Members: Francois G.
l'
i
. A.\'.fociateMemberJ': K
Alfi!iares: Leofw¡n L:
Hons. Ric!;ard Benjamín L

t
t\..tallurgy March 1961
March 1961 Journa! of th.. South African Institute of Mining and Met1ll/urgy 401
<:<.-1'4.¡v. .RO"'~anJ D. C. SIt!)'/' On ,,' ueparture o[ Ore!'a/ue Di.5Iribllliolu{rOl" the Lognormal Modl!l in South African Gold l\1ines-D. G. Krige

\' to some of the mining


)' here is for each gron!)
ON THE DEP ARTURE OF ORE VALUE DISTRIBUTIONS
om the mines are from FROM THE LOGNORMAL MODEL IN SOUTH AFRICAN
oubling the capac!ty of
) accommodating those GOLD MINES
an overnight stay for
jn per week in technic:ll By D. G. Krige (Mf'mber)

ea] College otTt:rsmany Published in the Journal, November, ]960


ove their status in tlieir
Comribution 10 discussioll
Richard F. Link* and Georgc S. Koch, Jr. * We have read the several valuable
article~ by Mr. Kige on mine valuation that ha ve appeared in the 5íowh African
Minin;; and. Engineering Journal in recent years and welcome the opportunity o(
comme'.lting on thc prescnt interesting paper.
Th~ basic problcm attacked in Mr. Krige's paper is simply this: how can a
reliabk estímate of the gold content of these South Atrican mines be procured from
a lin:ít~d amount of information. We believe that the exact nature of the distributiw
which these assay values follow is sOr.1ewhatacademic. In particu!:Jr, it ~eems more
important to stress the value of interval estimatíon for thís prob!em than to obtain
B.SC., Ph.D., F.G.S., best point cstimates of óistributions whose very natUre is questionable. Furthermorc,
Published by Oldhams since the robustncss of Student's "t" statistic is weí! known, it appears that contidencc
intervals based on this statistic calcuJated dirccth fram the untransformcd data would
lead tv practicaliy useful answers, even thouglÍ. this procedure may be statistically
.1 of thc various cutti:w sOr.1ev':ultinefficícnt. Referríng to Mr. Krígc's data, we compute the 95 ~{, confidence
:11gunits for insta¡¡atio~ interval for th¡;:mean bascd on Merriespruit se~ ~39 to be (U9, 271), that is to say
. extent where bord and we would believe, on the basis of this samplc, that thc truc mean lies bctween. 139
and 271 inch-dwt, and 1'0; the Prl"sident Brand set ~ 35 it is (0,5772). Note that
in b01;l cases the ¡fIJe vaJue of the mean is within the confidencc interval. For the
lachines are clearly and President Brand set:í; 35 the interval is very braad, indicating that the average
ldcnts and for enginccrs value is not well k!1('wn. Of course, whenever there are extren.e vaJues in the sample,
the confidence interval will be wide.
We ar~ examining a large number of assays fram two large Mexican mixed metal
:;J of mechanized equip- mine:; (lead, silver, copper, zinc. gold): the Frisco Mine of San f'rancisco Mines of
:' Toof control and dust MexícC',Ltd., and the Fresnillo Mine of the Fresnillo Co. Our work show$ that these
i
A.E.F. ! data a.-c ncither normal nor lognormai whether taken for cntirc veins or groups of
! veins 0f for various portions of veins. Whik there is a definite posjtive skewncss for
J these ddt<t,it i:. not ~0 large as it would be if the data were lognormal. A square-root
transfcrmation more nearly normalizes these data than does a logarithmic trans-
formation; but it 1s not entirdy satisfactory since the transformed data still do not
i foIlow a normal distrib.ution. . .
FinaIly, we wish to congratulate Mr. Krige on his interesting analysis of these
J data and also for presenting in dctail the original frequency tables of unaltered data.
The form of geologic distributions is a subjecI that has recently attracted wide attention
and it only by studying such distributions from a variety of geú!ogic settings that
t progress can be made.
f

.
'1
I 'Oregon State College, CorvaIlis, Oregon, U.S.A.

- -- ,..."

"'--
~
',' .,-.--

~
,
t.,lIur¡:y Januuy 1'161
(:. ".rad,.,t. /). fO.C'. A\lrlAr.'""
..
J
hnuary 1961 Journal of the South African Instltute of Mining and Mecallur¡:y 333

\ ¡
\
...
'~II studied in full. I.:aI
kimbcrlit.c miJl('ral ..r
Q ~nl11c Jlle~h size.
ON THE DEPARTURE OF ORE VALUE DISTRII3UTIONS
luch \"it.~Hnll, challgil\;!
;dle thc large tI'lIn;I~I"
FROM THE LOGNORl\1AL MODEL IN SOUTH AFRICAN'
ro tlle water no\\', '1'1.., GOLD MINr\.S
00 crit,il:a1 IIn(l thl' 1(','1
uoke; it \Vas 1'01'1111"'1'
;;t.itoution of IIn alluvial By D, G, Kri~c (!\'femLcl')
1I
"I\ounccu impl'O\'on\l:al '1
I'lIl.Ii:;lwd ill t.ho .!ullrlUlI, r\o\'('lIIh('(', II/IiO l.

"(1 illcreaso 01' diuJlIO:ul 11

.', L)\\t ilo lllay he addc'"


Contribulions to Discussion :1

¡¡
f}jcl1cy sho\\'ed a pru. "
H. S. Sichcl pr"I1II.cr): The Iqgllf)l'Inalla\\' \\'as inlrollll('c(l hy tIw \\'ril,1'1' t:I tiro
f ndualiol1
rcprrsl.nted
in JU.17 uceausc the Ingarilhms
by lhc lIormal fl'cqucnéy
of illch.(h\t.
distrihutioll.
ndul's apJ!('¡u'(,(l t.o he \\"ell
Su¡'sequclI(,ly ;\I'('S:;I':; Krig(' 111111
11
.1
"
1t Itll"~ iI\(I('pcnd(,III Iy ,"el'i liNI mj' 'filldi IIgS, Latcr, o\'er,.;cas in \'cstign t"l'''; alxo I'el'0l'(,('d !I

.
t
I
l'onsiderablc ,.;ueccs:; \\'ilh lIds modd. Oncc ,intcre,.;t \nlS !1roused ill ~;talistical ndlla.
lioll IIICt,III/lls, mol'c :tnd 1I10l'C dat.a eall1e t.n hand alld dl'\'ialillll'; 1'1'0111tlll' ol'igillal
!\
1:
i lIIodl'l IIIIHlo lIll'ir II,pl)('arnllI'CS on sl'\'Cr'l~ (lecllxion,.;. In t.lle ahs(Onco 01' an." olhl'l' o'

t
t
III('I,hod \\'e en I'l'ied on \\'ith lognol'lnal
IIl'eded ill (,he solutillll of pracl;eal
thcol'Y I\S the he,.;(, possi bit: npproxi (1111
pl'Obll'lI1s,
1iOIl

! 1\fr Krige, in his prl'~ent. papel'. has madc u \'ulllahle cOlllrilJlltioll lo lhe xllhjct'!,
l ".Y su:~g('si.ii1g the additioll of a (~..nslant a tll' o¡'scr\'ed ineh.dll't, ndlll':; 1)('1',.1'1'
¡ ~\\'itehi,jg inlo logarilhms, By ~o doing lhc lIIodcl'ate I}('gati\'l) ske\\'lIcs,,; nl'ising {'mili
f .
thc eon\'clltionallog-trall,.;fol'mation
01' lognormal
is lal'gcly cIilllillaü.d alld ti", ('xi~tillg kllo\\'h'dge
theol'Y n)¡1Y sl.ill be uH'd. as in tlle pa,.;t, hce¡¡,use t.he l..gari:,hl.1:; of illt:h.
r lI\\'t, pj¡I~: rLare 1I01'lnall)' distrihuted for 1I10-;t of the oh";cr\'ati,,ns exa'lIÍlH'd to.date.
ii 1.('0;', I be IIliSlllldl'l'st.ood ill \\'hat r shnll ha\'..' lo sav jater 011. \\'11(\11re\'i(~\\'ill~'
nili('ally :\11' l~rige's tIl.\\, J)['oJIIlsal. ] should like to statl; lhat ] am \\'holehcartedly
iu fu\'()ur of this IIIcthod of trall,.;f"rJlwlion.
1 han\ tried it OU(, in S('\'('r:d practical 1;f()l>lems \\'ith \\'hich J \\'as rl'cently
('()lIfl'..nted ami J alll satisfi('d t.ha(, the additioll of e(,II,.;tan(, a Icads lo hdler g/';\(I('
est¡mat.es alld a,.;sociated eOllfidencc lill1its than hi(,hcrto ohlailled \\'ith the eonn~lI.
tit>l\al IIIcthod.
I'rofcs,~or 1(erri('h hn,.; 011 nlall)' oceasions slres,.;ed t.llal. lile ,,.;t.at.ist i('iall is tilo
f loonsejl'n.:e 01' the I'c;;carch \\'urker, ] 11 carrying 0111. t.llis pl'Ofc:;;;ional fllndioll
(OOllseilOnce 1 may pcrha ps he [orgÍ\'cll if 1 on'cl' a fe\\' scarehillg r('lIIark".
or a

fí .rIIch-d \\'1-\'all1l's 111


a y rUII froJII ".ero t.o \'cry large posiLi\'e ndllcs 11\1(,I)wy ol".jollsly
('allllol he lH'galiH'. 1r use is IIllldl' of 1\fr l:l'igc':; tl'llw~forlllat iOIl y lo!!, (::, u),
.
¡
\
~
\\'Ilcrl' :: (knolex
IS 'l.ero.
irieh.dwl, the y \':1l'iable terminatcs ai. [(J[j,a if the
o

o LscJ'\'cd. \'aluc

Tllc Ilol'lllal disll'ihl1tioll fitll'd t.n t,he !I"~, hO\\'c\'cr, \\'ill ('xlellll lo lIe~nli\'(\
III~al'illll!ls alllt 11'1\1'1\ Ihl':-;I' aro I,r:lIIsl'ol'lIlI.d IJal'k (,0 (,111' ol'igillal ,.;cale Uw lIíallw.
III:\Iieal Ii1mlcl \\'ill 11I'(,dlleo negali\'(, illl'h.d\\'t, '.'allles which do lIot exis!. SemI ¡-rolll
tl.is allgle, t.lle SlI¡!¡!('sll'll 1It'\\' IllOdei is dcfillildy illcorn:ct \\'hcl'I'as tl1l: 0111 nuul(,1
<lid 1H1l slIl1'el' rl'OII1 llli" illeoll,;i,.;II'IIl'Y. :\1... Kl'ige's hislo¡.;rallls 01' 1.11(: :UI'ITi('Slu'lIil,
\'allll's are I'l'dl'all'lI ill Fig, 1 ami ~ /l'ilJwllt 1)I'(:,,~killg lile alJ.scissa scalc as Iw has dOJI('.

-;; ,.,-.-- ~'"..-, ~~~ ~, ..., ~ ~,,--~..- ~~.,.-;-r.."..~"-_____

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

You might also like