Counting and Sampling Errors in Modal Analysis

by Point Counter
by M. SOLOMON

Department of Geology, University of Tasmania, Hobart

WITH TWO PLATES

ABSTRACT
There are three principal sources of errors involved in modal analyses by point counting:
(a) operator variation, (b) the determination of area by grid counting, (c) the determination
of volumes from areal analyses.
Operator variation is likely to be negligible but (/>) and (c) may combine to give a variance
equal to, or less than Q-44/MT, , c o / ^
~RA~i1+S'S\al
where A = measurement area, a — grid spacing, R = grain radius, and p = the fraction of
the particular mineral in the rock (Hasofer, 1963).

INTRODUCTION
T H E expansion of quantitative petrographic studies in recent years (e.g.
Whitten, 1961 a, b) has accentuated the need for precision in.all forms of com-
positional analysis. Because it is rapid and inexpensive, modal analysis by point
counter has become a widely used technique and yet the errors inherent in the
method are imperfectly understood. The establishment of this form of analysis
as a routine procedure is due almost entirely to Chayes (e.g. 1949, 1956) who
has undertaken experimental and theoretical studies of the basic principles and
the hazards of practical application. Some of Chayes's experimental results have
been re-examined by Bayly (1960 a, b).
Some basic weaknesses in point-counting theory were encountered by the
writer while attempting to apply point-counting analysis to coarse-grained,
ore-mineral assemblages. The effects of grid, rather than random, sampling had
not been examined in detail, and much of the previous work on sampling had
been largely empirical. As a result, difficulties arose when arranging analytical
procedures. Theoretical examination of these problems by A. M. Hasofer
(Department of Mathematics, University of Tasmania) provided some interest-
ing results (Hasofer, 1963), and a series of experiments were designed to test
his conclusions.
The principal sources of error in a typical modal analysis by point counting
are:
(a) The operator error, introduced by mis-identification of minerals, &c.
(b) The counting error, introduced during estimation of areal fractions (e.g. in
a thin section or mine face).
Journal of Petrology, Vol. 4, Part 3, pp. 367-82, 1963|
368 M. SOLOMON—COUNTING AND SAMPLING ERRORS IN
(c) The sampling error, introduced in estimating the composition of a volume
from one or more areal analyses.
Bayly (1960&) assumed that the total variance is the sum of the counting and
sampling errors, determined independently, but Chayes (1956, p. 91) was loath
to isolate them in his experimental work. Hasofer's results indicate that the two
principal components, counting and sampling errors, are not independent of
each other and both must be taken into account in designing sampling pro-
cedures.
Operator variation is essentially independent of the other two but it is likely
to be included in both counting and sampling errors; attempts to evaluate its
importance have been made by Chayes & Fairbairn (1951) and Griffiths &
Rosenfeld (1954), but their conclusions are not wholly acceptable. The experi-
ments conducted by Chayes & Fairbairn consisted of point-counter analyses
repeated by five operators on several slides; in this way the counting error was
introduced and it probably obscured operator error. In similar experiments by
Griffiths & Rosenfeld, the traverses run by individual operators were identical,
but apparently the position of the individual grid points was not necessarily
constant; thus in this case also, the counting error was probably incorporated.
During the experiments described later, and also during point counts made on
wolframite-quartz veins by the writer and assistants, operator error was roughly
assessed by repetitions involving two or three operators. The grid position was
fixed for any one series of estimations and both intra-operator and inter-
operator behaviour was studied. The variances calculated from these experi-
ments were insignificant compared with the other errors, and for the purposes
of this discussion operator variation will be ignored. However, it is likely that
it is important in complex mineral assemblages where there may be more oppor-
tunity for mis-identification.

ESTIMATING THE COUNTING ERROR

Theoretical study
In determining by point counting the fraction/; of an area occupied by a mineral,
the estimator of that fraction p1 is given by the ratio of the number of points
falling on that mineral, to the total number of points. Hasofer (1963) has shown
that the mean value of pl is p in all cases but that the variance of px depends on
the distribution of the mineral in relation to the grid; it may vary from zero to
p(\—p), these limits being independent of the number of grid points used.
However, by taking into account the factors of grain size, measurement area,
and grid spacing, it is possible to determine the upper limit of <72(/?,), assuming
a model in which (a) the grains are circular or oval in shape, (h) the centres of the
grains are distributed at random in the section, (c) the grains are entirely within
the section, and (d) p is low enough to make negligible the effect of possible
overlap between grains.
 MODAL ANALYSIS BY POINT COUNTER 369
An upper limit of the variances can be given in terms of the radius of the
grains R, the measurement area A, and the grid-side length a, by
^0-37/7o 3
RA 0)
If p is expressed as a percentage thenCT2(PJ)< 31pa3/RA. If R < \a, the variance
is given exactly by pa2/A.
Finally, if the grain diameters vary, then the upper limit is given by
0-31pa/[{R+[a%R)IR]}A] where CT2(/?) is the variance of the grain radius.
Although Chayes (1956, p. 91) apparently appreciated the likely effects of decreas-
ing the grid spacing, no attempt was made by Bayly (1960a) to improve his estima-
tion of <72(pi) which did not involve grain size, measurement area, or grid size.

—
- LjLJj
4—
~ H -*.
/ IT hrxtr
L__

• — - —

\ 1 \

[M
33
ii
~ —
-
I
( I / nzrtr
_—-—

FIG. 1. Method of preparation of 5-cm grains.

Experimental study
To test Hasofer's theoretical conclusions, a series of experiments were devised.
'Grains' were made from a sheet of thin cardboard, 60 cm square; this was cut
into four and then subdivided into 5-cm 'grains' by wavy lines (Fig. 1). These
grains approximated squares but possessed shallowly convex and concave sur-
faces in order to simulate natural grains or crystals more closely. This method
of cutting allowed selection of various percentages or fractions (particularly
\, \, | ) of the original square. Squares of sides 60, 40, 30, 20, and 15 cm were
marked out on white cardboard and selected numbers of grains thrown at
random on to the 60-cm square. The prepared 'sample' was then 'counted' by
placing over it a stable, plastic sheet (Astralon), on which a 1 -cm (and part |-cm)
370 M. SOLOMON—COUNTING AND SAMPLING ERRORS I N

TABLE 1
2
(Pi)for R = I'25 cm, and p = 8\, 25, and 50 per cent, with calculated values
(formula (/)) in parentheses
a (in cm)
A/a2

P — 8i per cent
44* 8-99 10-83
(18-46) (24-61)
100 106 206 4-71
(5-48) (8 21) (10-95)
225 0-93 0-62 1-40
(2-43) (3-65) (4-87)
400 0-32 0-46 0-28
(0-68) (1-37) (205)
900 013 016
(0-31) (0-61)
1600 007 007
(008) (017)
3600 001 002
(004) (0-76)
p — 25 per cent
36-30 31-00
(49-89) (66-52)
100 4-42 7-52 415
(14-80) (22-20) (29-60)
225 808 7-55 2-98
(6-58) (9-87) (1315)
400 0-85 2-37 1-67
(1-85) (3-70) (5-55)
900 0-57 1-22
(0-82) (1-64)
1600 0-26 0-25
(0-23) (0-46)
3600 0-35 0-20
(010) (0-21)

50 per cent
44J
100

225 4-71 8-54 21-41

(1315) (19-73) (26-31)
400 1-49 3-52 3-90
(3-92) (7-40) (11 10)
900 101 0-43
(160) (3-29)
1600 0-32 0-59
(0-46) (0-92)
3600 0-26 0-41
(0-21) (0-41)
 MODAL ANALYSIS BY POINT C O U N T E R 371
grid was drawn (Plate 1). Thus a could be varied from \ cm upwards and A could
be varied from 60 cm downward. Ten or fifteen counts were made for each
combination of a and A, the position of the grid in each case being arranged so
50
d
40
O = \
• =I
30
• = 2
X = 3
20
A = 4

10
0-9
0-8
0-7
0-6
OS
04

0-3

0-2

01
20 30 40 60 80 100 200 400 600 1000 2000 S000

FIG. 2. Results of counting at p = 25 per cent, R — 1-25 cm, and a — i, 1, 2, 3, and 4 cm.

that the grid lines were parallel to the margins of the sample areas. The variances
were calculated by the formula:

ff'lp,) =

where x = an individual reading of the number of grid points falling on the

grains, / = the total number of lattice points, and n = the number of repetitions
per combination of a and A. By using coloured cardboard 'grains' on a white
background, mis-identification (operator error) was reduced to a minimum; on
372 M. SOLOMON—COUNTING AND SAMPLING ERRORS IN
several occasions counts were repeated by two or three operators before chang-
ing grid positions, and it was found that the variations were negligible. Experi-
ments were made at an early stage of the theoretical studies to confirm the major
conclusion that the variance decreases with increasing A and decreasing a. The
results for R = 2-5 cm, p = 50 per cent, and a = 1, 2, 3, 4, and 6 cm, supported
this conclusion, but for practical purposes most of the variances obtained were
too high ( > 10). A further series of experiments were made with 1-25 cm grains

60
o=
-+- = 40 = 1

•= 30 = 1

•= 20 = 2
X = 20 = 3

10 20 30 40 50 60

FIG. 3. Relation between variance and percentage for a number of

counting arrangements, with R = 1-25 cm.

(2-5 cm grains cut in quarters as nearly as possible), with/? at 8|, 25, and 50 per
cent, and a = £, 1,2, 3; and 4 cm. The results from these runs, still on a pre-
liminary basis, were similar to the early series but with lower variances (Table 1;
Fig. 2); the variance is more or less inversely proportional to A/a2 and the
observed values show reasonable agreement with the values calculated from
formula (1).
The three values of p used in the experiments give an indication of the control
exerted by p on the variance (Fig. 3) except for values of A/a2 < 20.
 MODAL ANALYSIS BY POINT COUNTER 373
2 cm
Finally, with A/a constant, a was varied from \ to 4 cm, taking fifteen
counts for each set-up. This procedure was repeated six times, giving the results
shown in Table 2 and Fig. 4. To retain the 60 x 60 cm counting area for the 3- and
4-cm values of a, 1 -25 cm grains were used with proportionate reduction in
grid size.
TABLE 2
<j\pdfor R = 1-25 cm, p = 25 per cent, and A la2 = 900 (see Fig. 4)
a (in cm)
Experiment i 1 2 3 4
1 017 0-57 1-22 1-44 1-68
2 0-20 0-33 0-74 2-76 1-55
3 0-95 0-49 0-98 0-28 5-78
4 0-40 0-30 105 101 3-32
5 0-33 0-98 0-59 0-73 0-91
6 0-75 011 0-41 0-88 1-42

Hasofer (1963) has shown, using the F-test, that the observed average values
of variance are consistent with the theoretical model at the 5 per cent level.
In conclusion, it is clear that the controlling factor in reducing variance is a
and it seems advisable, judging from the experimental results, to reduce a to at
least Rj2 in order to maintain a useful variance for pv

The problem of grain size and shape

A vital factor in studying variance is knowledge of grain radius. In the experi-
ments just described it was possible to determine perimeters with a map-
measurer and, by using the average perimeter L, to find the average radius
(R = Ljl-n). Measuring perimeters in thin sections is practicable but tedious
(e.g. Bain, 1941) and it could hardly be seriously proposed without designing
special equipment. Where grains are roughly equidimensional, as in many
sandstones, the radius may be determined by estimating the circle of equivalent
area. Hornsten (1960) has described a microscope accessory for finding the
equivalent circle and determining 'nominal sectional diameter'; however, in
rocks of irregular grain this method is scarcely suitable. Chayes (1956, p. 72)
measures an average observed diameter for all mineral grains by counting the
number of grains crossed in a traverse of 25 mm (pers. comm.), the total being
the IC number; this method involves grains of major importance, of any shape,
and gives a measure of coarseness for the rock. This method is basically the most
useful but can be improved for point-counting purposes. Taking the 'traverse
intersection' number y for a particular mineral as the number of grain boundaries
intersected in a known length of traverse, divided by that length, then an un-
biased estimator of 1 //? (in the model described on p. 368) is given by 2(y)/3p
(Hasofer, 1963). Substituting for R in formula (1), p. 369, gives
374 M. SOLOMON—COUNTING AND SAMPLING ERRORS rN
thus eliminating the necessity of visually estimating p in the rock in order to
plan suitable arrangements.
All the theoretical conclusions apply to grains that are circular or oval in

0 1 2 3 4
a (cm)
FIG. 4. Behaviour of variance with A/a' constant, for varying values of a.
section and no prolonged attempts have yet been made to evaluate the effects
of irregular grain shapes. A preliminary experimental investigation was made
of the effects of highly irregular shapes (e.g. embayed quartz crystals) and of
 MODAL ANALYSIS BY POINT COUNTER 375
lath shapes (e.g. feldspars) by using jig-saw pieces and rectangular grains
respectively. All the jig-saw pieces were interlocking and possessed highly con-
vex and concave boundaries, though the pieces had similar areas (average radius
1-1 cm). The results (Table 3) indicate that low variances are still obtainable even
with irregular shapes, under the conditions applying to circular and oval grains.
In studying rectangular shapes, a series of values were arranged in the same
way as for Table 2, half of the 2|-cm grains being converted to rectangles with
sides in the ratio 1:2. The preliminary results (Table 4) are comparable with
those of Table 2.
TABLE 3

or jig-saw experiments with R = 1-1 cm and p = 11 per cent; calculated

values {formula (/)) in parentheses
a (in cm]
A/a2 i 1 2 3' 4
44£ 3-88 1400
(20-49) (33-26)
100 4-55 7-97 13-44
(7-40) (11 10) (14-80)
225 1-29 205 3-56
(3-29) (4-93) (6-58)
400 0-60 1 29 115
(0-92) (1-84) (2-77)
900 0-95 0-24 0-34 0-25 0-22
(0-21) (0-41) (0-82) (1-23) (116)
1600 011 0-27
(012) (023)
3600 003 011
(005) (010)

TABLE 4
a or
\Pi)f mixed grains, R = 1-25 cm and p = 25 per cent; calculated values
{formula {!)) in parentheses
a (in cm) i 1 2 3 4
0-59 0-47 0-67 0-53 1-28
(0-41) (0-82) (1-64) (2-46) (3-28)

Irregularity in form must affect the determination of R. Embayed grains tend

to give a higher y, which results in a reduction in R and thereby reduces the
calculated variance. For instance, on a counting model made up of strongly
'embayed' jig-saw pieces, occupying 25 per cent of the 60x60 cm base, the
average 'grain' radius (determined by measuring the area of the jig-saw and
dividing by the number of pieces) was 1-36 cm. The average figure for R, deter-
mined by the traverse intersection method and using a total traverse length of
1,200 cm, was 0-86 cm. Obviously, in planning counting arrangements some
376 M. SOLOMON—COUNTING AND SAMPLING ERRORS FN
account must be taken of grain shape by reducing y, but no formal recommenda-
tions can be made. Attempts to reorganize R so as tofitexperimental results are
mentioned on p. 378. Also involved in the problem of irregular shapes is the
result of compressing the grid so that it is rectangular in shape. No theoretical
studies have been made of this problem, but it seems unlikely that the effects of
such distortion would be very large. In an asymmetrical grid, a is taken as the
geometric mean of the two side-lengths.
Experimental work of potential interest was carried out on dolerite by Saha
(1959). During a comparison between the relative precision of point counting and
linear integration, Saha counted one dolerite slide eight times, changing the grid
position after each count, and thus carried out experiments on natural rock
similar to those just described on artificial models. His results (Table 5) are

TABLE 5
a
%Pi)for dolerite from Bombay State (Saha, 1959, table 2), compared with
calculated variances; average IC number = 58-9, a = 0-57, A — 425 mm2
Observed Calculated
Material P variance variance

Plagioclase 43-9 0-78 108

Pyroxene 38-4 i-27 0-94
Iron ore 6-9 008 017
Secondary minerals 5-6 0-20 014
Micropegmatite and glass 5-2 0-26 013

difficult to test because the IC values given indicate only an average grain
diameter for the several minerals present and no indication of grain size varia-
tion, except that the rock is non-porphyritic, is given. Taking the average IC
value and using the percentages given, the theoretical variances may be calcu-
lated (see Table 5). Considering the generalized nature of the data, the observed
and calculated variances are in reasonable agreement.

ESTIMATING THE SAMPLING ERROR

Theoretical study
The model set up by Hasofer, to study counting variance, is now extended to
that of a parallelepiped in which the centres of spherical grains are distributed
randomly in such a way that the spheres are contained within the parallelepiped.
If sections are taken from this block at random then the variance/^ of the frac-
tions for a particular mineral revealed on the sections ^ 4nR2pj5A. If the grains
are of unequal radius, then
(2)
5/4
where R is the radius of the largest grains.
 MODAL ANALYSIS BY POINT COUNTER 377
In petrography, the counting error becomes involved in the sampling error,
because the relative areas on individual sections are determined by counting.
In this case:
gL3[ ^ ) 3 ] (Hasofer, 1963). (3)
For2R<a, a\p3) < ^ . (4)
Formula (3) may be written as
0-44pa3 , 2-25pR2
~RA~ + ~4~'
the two expressions representing the counting and sampling errors, respectively.
If p is expressed as a percentage, then

Experimental study
Testing the applicability of Hasofer's conclusions with regard to volume
estimation involved two phases of study:
(a) The preparation and sectioning of an artificially prepared 'rock' and com-
parison with formulae (2) and (3).
(b) Comparison of Chayes' results for the Mt. Desert, Carnmenellis, and
Westerly granites with formula (3).
(a) A ' rock' model was prepared by cementing with plaster-of-Paris a frame-
work of'grains', in a mould with dimensions 20x20x50 cm. The grains con-
sisted of moulded plaster-of-Paris spheres and bricks, the spheres being 1 cm and
1-5 cm radius and the bricks approximately 3 x 2 x 2 cm. Five hundred of the
smaller spheres (about 11 per cent of the ' rock' volume) wer.e stained black.
The mould was filled systematically, using random number tables, and the
cementing plaster added at intervals. After setting, the block was cut by band-
saw on fifteen parallel surfaces chosen from a possible hundred cuts by random
number tables, and the areas occupied by the black spheres were measured; a
typical cut is shown in Plate 2. The variance from the fifteen results was 3-55
(o- = 1 -88) as compared with the theoretical value of 7-25 (a = 2-69). A second
slab, 20x20x25 cm, was made and the experiment repeated, using 100 black
balls of radius 1-5 cm. This yielded a variance of 4-52 as compared with the
theoretical value of about 1700. Both of the experimental results, and particu-
larly the second, were significantly lower than the theoretical upper limit, perhaps
due to the necessary condition of non-overlap of spheres and the difficulty in
arranging a truly random distribution within the block. At all events, the experi-
mental results did not exceed the upper limit dictated by theory.
In testing formula (3) each slice of the larger block was counted by the normal
378 M. SOLOMON—COUNTING AND SAMPLING ERRORS IN
grid technique and the variance of the results calculated. In three experiments,
the grid side was held at I- cm, thereby involving 1,600 points on the
20x20 cm sections. The three variances obtained were 3-54, 3-30, and 3-61, and
p values were 12-56 per cent, 12-55 per cent, and 12-45 per cent respectively. The
calculated variance in each case was 8-14, revealing a similar disparity between
theory and practice to that shown in the first block experiment for testing
formula (2).
(b) The exhaustive counts made by Chayes (1956) on three granites of vary-
ing coarseness provide useful data with which to check formula (3). From each
granite, Chayes made twenty thin sections, ten of which were used for counting
with A = 960 mm2 and ten for counting with A = 480, 320, and 160 mm2. Ten
counts were made for each measurement area for each granite. Counts were
made on quartz, K-feldspar, and plagioclase, and yielded the standard devia-
tions shown in Table 6 (Chayes, 1956, p. 66). The number of points counted
remained constant for each experiment at 660 (so that A/a2 = 660) and the grid
side length can be averaged by ^I(A/660). It should be possible, therefore, by using
formulae (3) and (4), to calculate the theoretical deviations. The theoretical
values for R were derived as follows: because the IC number increases once for
each mineral grain crossed, and there are three minerals involved in the experi-
ments whereas in y there are two counts per mineral grain, the IC number is
multiplied by two-thirds. Thus, y = (IC/25)x f. In adopting Chayes's IC num-
bers (1956, p. 75), converting them to R values, and comparing calculated and
observed variances, it was found (Table 6) that six of the thirty-six observed
values were outside, and above, the 99 per cent confidence limits of the calcu-
lated values but that, overall, eighteen were below, seventeen above, and one
equal to, the calculated values. By decreasing the IC number for Mt. Desert by
10 per cent and Carnmenellis by 30 per cent, only one anomaly remains (Carn-
menellis, K-feldspar, 160 mm2). It is possible that the irregularity in grain size
may have contributed to the six anomalies. Irregular boundaries increase the
IC number, decrease R and lower the expected variance. Five of these anomalies
were in the Carnmenellis granite, in which the abundance of accessory minerals
may contribute to irregularity in grain of the major constituents.

PRACTICAL APPLICATION
It is now a little easier to arrange an analysis so as to accommodate a required
precision, giving certain limiting conditions: the grains should be randomly
distributed, of more or less regular shape, and p should be small. Chayes's
results indicate that significant departures from regularity in shape and relatively
high percentages of particular minerals are permissible.
The recommendations made by Chayes, based only on his experimental work,
can be improved upon by taking into account the percentage of mineral present
and by suitable arrangement of a and A.
The design of analytical procedure is likely to be controlled largely by
 TABLE 6
Comparison of calculated standard deviation, Chayes's experimental values for standard deviation, and 99 per cent
confidence limit for the calculated deviation o
D
760 mn 320 mm* 480 mm2 960 mm'-
Minerals 99% 99% 99% 99%
from Calcu- Chayes's confidence Calcu- Chayes's confidence Calcu- Chayes's confidence Calcu- Chayes's confidence
granites lated value limit lated value ' limit lated value limit lated value limit
r
Ml. Westerly •<

Quartz 200 1-78 3-01-0-96 204 205 3-07-0-98 202 1-52 3-04-0-97 203 1-57 304-0-97
K-feldspar 2-39 205 3-60-115 2-37 202 3-57-114 2-34 1-40 3-52-112 2-38 1-64 3-58-114
Plagioclase 215 1 92 3-23-103 2 15 1 90 3-23-103 218 1 39 3-28-105 215 1-27 3-23-103

Ml. Desert
Quartz 3-62 3-63 5-45-1-74 310 2-94 4-66-1-41 2-76 2-76 415-1-32 2-25 1-68 3-39-1 08 2
K-feldspar 3-62 3-60 5 45-1-74 2-98 2-49 4-48-1-43 2-95 418 4-44-1-42 2-35 2-70 3-54-1-13 H
Plagioclase 2-71 3-65 4-07-1-30 202 2-42 304-0-97 201 •3 05 3-02-0-96 201 2-67 3-02-0-96 n
o
c
Carnmenellis z
Quartz 615 •10-66 9-26-2-95 4-68 6-20 7-04-2-24 419 •6-46 6-31-2-01 306 •5-21 4-60-1-47 H
K-feldspar 5-34 •12-29 804-2-56 403 518 606-1-93 2-85 •5-52 4-29-1-37 2-86 419 4-30-1-37
Plagioclase 4-59 6-35 6-91-2-20 318 2-80 4-78-1-53 303 2-67 4-56-1-45 2-52 2-65 3-79-1-21
m
* Results outside 99 per cent confidence limit.
380 M. SOLOMON—COUNTING AND SAMPLING ERRORS TN
(a) available grid spacing (the Swift point counter, for instance, allows spacing at
only £ mm x 2V, TV, &, and -3- mm) and (b) the economics of producing the desired
measurement areas. For instance, a great number of the writer's thin sections
cover about 20x20 mm; to produce a variance of 2 (a = 1-41), assuming
R = 01 mm and p = 25 per cent, and knowing that A must be kept below
400 mm2, a might be conveniently taken as 0-5 mm, and 1,300 points counted.
If a smaller area is taken, say 75 mm2, then to keep a constant error a must be
taken at about 0-2 mm, and about 1,800 points counted. For a coarser rock like
the Carnmenellis granite, which has a grain radius near 0-5 mm, a variance of 2,
assuming p = 25 per cent, might call for A = 1,900 mm2, with a = 1 mm. This
area would require two of the largest thin sections (24 x40 mm) used by Chayes,
and the counting of 1,900 points. This would be preferable to Chayes's recom-
mendation (1956, fig. 11) that for the same circumstances, about six slides would
be required and 660 points counted. However, in rocks like the Carnmenellis
granite, it appears likely that higher variances than expected might occur and it
would be wise to increase A to counteract this (by adding, say, at least a section
25 x 20 mm). With coarse granites, many of the practical difficulties are lessened
by working on polished slabs (e.g. Jackson & Ross, 1956).
Some typical combinations of a and A for given circumstances are shown in
Table 7. Where the grid opening is asymmetrical, a is taken as the geometric
mean of the unequal sides.
TABLE 7

Effect of varying a and A for (a) R — 01 mm and (b) R = 0-5 mm; p — 25 per
cent and o2(p3) = 2
a (mm) A (mm2) Points

Jt = 01 mm
005 32 13,000
0 1 37 3,700
0-2 75 1,800
0-4 200 1,250
0-5 310 1,250
0-6 450 1,350
R = 0-5 mm
0-25 800 13,000
0-5 950 3,650
10 1,900 1,900
20 5,000 1,250
30 11,000 1,250

The precision for each mineral is dependent on p and R, so that ideally each
mineral is counted separately; however, for ease of working it is advisable to
measure several at a time provided each can be measured accurately enough
under one grid arrangement. It may be necessary to split the analysis and count
under two different grids.
 MODAL ANALYSIS BY POINT COUNTER 381
In arranging analyses it is essential that sufficient lengths of traverse are used
to determine y—from tests it seems necessary to have crossed several dozen
grains before calculating R.
The technique may be applied regardless of scale, as recognized by several
workers who have examined polished and etched slabs of coarse-grained rocks
(e.g. Jackson & Ross, 1956).

Previous analyses
Examination of point-counting techniques as used by several petrographers
reveals considerable variation in precision. As far as can be determined from
available information, the analyses made by Barringer (1953) on a synthetic,
screened concentrate of sulphide ores with a symmetrical grid (a = 0-5 mm)
should have given variances of 13 {a = 3-6) for chalcopyrite (p = 40 per cent),
8 for pyrite (25 per cent), and 5 for sphalerite (15 per cent). Increases in the
sample area would have improved considerably the precision. Jackson & Ross
(1956) fared worse in their counting of slabs of coarse-grained rocks; using
Zip-a-tone paper as a grid they normally counted 1,000 points and used a
sample area about one hundred times the area of the largest grain. For instance,
using Zip-a-tone no. 3 (a = 1-5 mm) and taking p = 25 per cent and R —
2-5 mm, the variance is likely to be about 20—obviously higher than desired.
Whitten (1961o) quoted measurements by Saha on the Wollaston granite, which
has an IC number between 25 and 42. Points at 0-3-0-4 mm spacing were counted
on traverses 1 mm apart (a = 0-6 mm) and 1,500-2,000 points were counted.
Taking p = 30 per cent, the maximum variances probably ranged from 12
(IC = 25) to A\ (IC = 42) if 2,000 points were taken. On the Donegal granite
(Whitten, 1961ft), under similar conditions, only 1,411-1,425 points were
counted, giving an even poorer precision. Irregularity in grain shape might lead
to higher variances in practice. However, McDougall (1962), when point-
counting Tasmanian dolerites, used areas greater than 300 mm2 and counted
3,000 points; for plagioclase, taking R = 0-25 mm and a = i mm, the expected
variances should be less than 1, a precision probably greater than required.
Another satisfactory arrangement was recommended by Griffiths (I960), who
has described a general technique for sedimentary rocks. This involves counting
several thousand points on areas varying from 625 mm2 to 400 mm2. Taking
p = 25 per cent, a = 0-3 mm (from his description), and assuming an average
area of 500 mm2, gives maximum variances of 11-5 if R = 1 mm, 3 if R =
0-5 mm, and 1 if R = 0-25 mm.

ACKNOWLEDGEMENTS

Thanks are due to Mr. A. M. Hasofer (Mathematics Department, University

of Tasmania), who has guided this study and provided the statistical inspiration,
and to Dr. A. Spry (Geology Department, University of Tasmania), who has
0233.3 C C
382 M. SOLOMON—COUNTING AND SAMPLING ERRORS
commented usefully from a.petrographer's viewpoint. The cost of materials and
student assistance was met by a research grant from the University of Tasmania.

REFERENCES
BAIN, G. W., 1941. Measuring grain boundaries in crystalline rocks. J. Ceol. 49, 199-206.
BARRINGER, A. R., 1953. The preparation of polished sections of ores and mill products using diamond
abrasives, and their quantitative study by point counting methods. Trans. Instn. Min. Metall.
Lond. 63, 21-41.
BAYLY, M. B., 1960a. Modal analysis by point-counter—the choice of sample area. J. geol. Soc. Aust.
6, 119-30.
19606. Errors in point-counter analysis. Amer. Min. 45, 447-9.
CHAYES, F., 1949. A simple point counter for thin-section analysis. Ibid. 34, 1-11.
1956. Petrographic modal analysis. New York: John Wiley & Sons, Inc.
& FAIRBAIRN, H. W., 1951. A test of the precision of thin-section analysis by point counter.
Amer. Min. 36, 704-12.
GRIFFITHS, J. G., I960. Modal analysis of sediments. Rev. Ge'ogr. phys. 3, 29-48.
& ROSENFELD, M. A., 1954. Operator variation in experimental research. /. Geol. 62, 74-91.
HASOFER, A. M., 1963. On the reliability of the point-counter method in petrography. Aust. J. Appl.
Sci. 14, 168-79.
HORNSTEN, A., 1960. A method and a set of apparatus for mineralogic-granulometric analysis with
a microscope. Bull. geol. Instn. Univ. Upsala, 38, 105-37.
JACKSON, E. D., & Ross, D. C , 1956. A technique for modal analyses of medium- and coarse-grained
(3-10 mm) rocks. Amer. Min. 41, 648-51.
MCDOUGALL, I., 1962. Differentiation of the Tasmanian dolerites: Red Hill dolerite-granophyre
association. Bull. geol. Soc. Amer. 73, 279-316.
SAHA, A. K., 1959. On the precision of modal analysis of rocks. Quart. J. geol. Soc. India, 31, 1-6.
WHITTEN, H. H. T., i96ia. Systematic quantitative areal variation in five granite massifs from India,
Canada and Great Britain. J. Geol. 69, 619-46.
19616. Quantitative area modal analysis of granite complexes. Bull. geol. Soc. Amer. 72, 1331-60.

EXPLANATION OF PLATES

PLATE 1
Typical experimental arrangement for determination of the counting error.

PLATE 2
A typical cut from the artificial rock model.
Journal of Petrology Vol. 4, Part 3

4/ » • *,

* • ^ • ~ 4 4

4
9 * 4
4
M. SOLOMON
Plate 1
Journal of Petrology
Vol. 4, Part 3

M. SOLOMON
Plate 2