Sedimentology (1993) 40,157-114

The statistics of counting clasts in rudites: a review, with examples from the upper
Palaeogene of southern California, USA

J E F F R E Y L. H O W A R D
Department of Geology, Wayne State University, Detroit, MI 48202, USA

ABSTRACT

The relative proportions of gravel sized particles of different lithology in rudaceous sedimentary rocks are
generally determined in the field by clast counting. Clast counts are usually carried out qualitatively in
order to assess sedimentary provenance. However, a review of the statistical aspects of clast counting
suggests that this technique also can be applied quantitatively, and to investigate a variety of other
objectives during basin analysis. Geographical and stratigraphical changes in the relative proportions of
clasts can be quantified statistically and used to characterize sediment dispersal patterns in space and time,
respectively. Statistical comparisons between clast assemblages can be used as a tool to match up rock
units. This approach may help to constrain tectonic or suspect-terrane models, or to document sediment
recycling. Both counting and sampling errors contribute to the total probable error of a clast count.
Sampling error results from the uneven distribution of clasts in outcrop, perhaps caused by selective
sorting. Counting error arises from a count of some number less than the total number of clasts in the
population. Sampling and counting errors can be minimized by counting in closely spaced subsets, and by
counting a total of at least 400 clasts, respectively. Thus, a useful procedure is to count four closely spaced
subsets of 100 each, and combine the results for a total of 400. Point counting should not be used because
differences in particle size produce biased results. A better method is to count all clasts above some
minimum size within a specified area of outcrop. Analysis of upper Palaeogene non-marine conglomerates
composing part of the Sespe Formation in California, using confidence intervals, hypothesis testing,
analysis of variance, ratio analysis and varietal studies, demonstrates that useful statistics can be derived
by counting clasts.

INTRODUCTION

Conglomerates, breccias and other rudaceous sedi- (Plumley, 1948; Sneed & Folk, 1958; Bradley 1970).
mentary rocks often contain gravel-sized particles of Tumbling-mill experiments show this to be particu-
different lithology which can be identified readily in larly significant when such clasts travel as a mixture
outcrop. Clast counting has been used for many years with very durable clasts (Abbott & Peterson, 1978).
to determine the relative proportions of different clast Thus, clast counts of rudites in different geographical
types in such deposits (Krumbein & Pettijohn, 1938), locations may be used to characterize changes in the
and it is still the most widely used technique today. proportions of various clast types, and thereby
The results are often combined with palaeocurrent establish a palaeotransport direction. Clast counts
data to make palaeogeographical or palaeotectonic also can be made at multiple stratigraphic positions
reconstructions (Potter & Pettijohn, 1977; Haughton in rudaceous strata in order to determine whether
et al. 1991). Clast counts have been applied most changes in the proportions of clast types have occurred
extensively to provenance studies, but there are many over time (e.g. Miall, 1970; Graham et al., 1986),
other useful applications. Studies of modern gravels perhaps in response to tectonism or eustasy. Another
have shown that less durable clasts generally decrease potential application of clast counts is to compare
in abundance with increasing transport distance clast assemblages of rudites in different locations.

157
158 J . L. Howard

This approach can be used to match up rock units for many sources because of the dirth of literature, and a
the assessment of fault offsets (e.g. Minch et al., 1976; field study was made to gather data on the probable
Bartling & Abbott, 1983), for suspect terrane analysis error of clast counting. The purpose of this paper is to
(e.g. Seiders & Cox, 1992), or for evaluating if present some general guidelines for clast counting
sediment recycling has taken place. If the aim is which can be used even by a field geologist with a
simply to establish the presence of a certain source- limited background in statistics. Simplified data sets
diagnostic lithology, as in a provenance study, clast and worked problems are presented using examples
counts can be made without particular concern for from upper Palaeogene non-marine conglomeratic
methodology. However, if the objective is to measure strata of the Sespe Formation, which crops out around
changes in the relative proportions present, or to Los Angeles, California (Fig. 1). The examples are
compare clast assemblages with one another, statistics used simply to demonstrate how clast counting can be
come into play. applied quantitatively to solve some common sedi-
Surprisingly few papers have been published in mentologic problems; the geological implications of
which clast count data are analysed statistically, and the results are not discussed. The examples also
the only ones known which deal specifically with the illustrate some of the statistical implications of certain
statistics of clast counting are those of Krumbein field procedures, as well as possible pitfalls when
(1942) and Krumbein & Miller (1953). I am not aware statistics are lacking. The effects of closure, which are
of any geostatistics text which covers this particular important in linear regression analysis (Kelley, 1971;
topic, although the statistics of point counting rock Aitchison, 1984), and applications of more sophisti-
thin sections and grain mounts have been discussed cated statistical procedures (e.g. discriminant analy-
(Chayes, 1956; Griffiths, 1967; Galehouse, 1971 ; sis) are beyond the scope of this report.
Kelley, 1971). The statistical problems associated with
clast counting in the field are different from those
involved in thin-section work, and have yet to be THEORY OF CLAST COUNTING
adequately addressed. The objective of this study was
to review the statistical aspects of clast counting, both The ultimate goal in clast counting is a quantitative
with regard to field procedure and data analysis. It determination of the proportions (percentages) of one
was necessary to borrow statistical methods from or more clast types within some target population of

119O

r" b
. :. C A IFORNIA

*LA
344
12
.:.-J
0
0
- - km 40
60
3

I - - -
,I 190
Fig. 1. Geographical distribution of upper Palaeogene non-marine conglomerates (fine stipple) in south-westem California.
Sampling locations: solid circles, Sespe Formation; open circle, Jalama Formation. Cities: LA, Los Angeles; SB, Santa
Barbara. Physiographical regions: A, Pine Mountain area; B, Santa Ynez-Topatopa Mountains; C, Simi Hills area; D, Santa
Monica Mountains; E, Santa Ana Mountains.
 The statistics of clast counting 159

clasts. Definition of what the target population is where CE =counting error or confidence interval
varies with the purpose of the investigation and could about the mean; t is the tabulated value of Student’s
be perhaps the total volume of rock comprising the t-distribution for n - 1 degrees of freedom and confi-
formation of interest, or the total volume of rock dence level 4 2 ; p=proportion of an individual
representing a given stratigraphic interval, or may be component; and n=total number of clasts counted
only that volume of rock found at a particular locality. (Natrella, 1966).
The most direct way to measure clast percentages As an example, suppose that out of a total of 100
would be to collect all of the gravel sized particles in clasts counted, 43 are granite. At the 0.05 confidence
the target population and count the number of each level with n = 100, t is 1.98, and the counting error
type. This is obviously impossible, hence a statistical associated with the ro ortion of granite clasts (0.43)
approach is required in which the true proportions in is:CE= 1 . 9 8 J ~ O = 0 . 0 9 8 . T h i s m e a n s
the population are inferred from counts of a limited that it can be stated with 95% confidence that the
number of clasts. percentage of granite clasts is 43 *9.8%, or that the
The uncertainty introduced by using a portion of true value lies somewhere between 33.2 and 52.8%.
the population (a sample) to draw a conclusion about Equation (l), however, generally holds only for
the whole population may be measured by probability p=0.10-0.90.
theory, however, there are two basic requirements : In practice, clast counting of an outcrop in the field,
(1) the percentages or proportions observed in the and indeed most geological sampling, is complicated
sample must be representative of those in the by the fact that parts of the target population are
population as a whole, and (2) the encounter with a physically concealed or otherwise inaccessible. Thus,
particular individual in that sample must be based a truly random sample is never obtainable because
solely on abundance and not biased by any other not all locations on the rock body have an equal
factor during the counting procedure. A sample is chance of being chosen for sampling. A visual
considered to be representative only if the population inspection of all available exposures, particularly
is randomly heterogeneous and the sample has been those exposing strata in three dimensions, could be
drawn at random, that is, each sample has an equal carried out to assess whether the available population
and independent chance of being chosen. In order for is the same as the hidden parts of the target population.
a clast count to have statistical validity, therefore, a It should be remembered, however, that because this
formal procedure must be used to assure that both the can never be ascertained with complete certainty,
exact point on the outcrop where the count is to be what we are obtaining with ‘95% confidence’ is simply
carried out, and the particular clasts to be counted at our best estimate of the true confidence interval about
that point, are selected at random. the mean.
From the standpoint of counting, a population of
conglomerate clasts may be considered to be dichoto-
mous and composed of one proportionp of individual
clasts having a certain characteristic, and another PROBABLE E R R O R OF CLAST
proportion 1-p which does not. A counting error COUNTS
arises from a count of some number n less than N , the
total number of clasts in the target population. The above discussion deals only with the statistics of
Assuming a random sample of n clasts, the best counting. The total probable error associated with the
estimate o f p is the observed proportion X / n , where X clast counting procedure is the result of both analytical
is the number of clasts having the characteristic. If n and sampling errors according to the relation :
is large ( 2 30), or if np and n(l -p) are larger than 5,
the sampling distributionofXis approximatelynormal % T E = J G F G (2)
with a mean equal to p and a standard deviation of where TE is the total probable error and el and e2 are,
J p m (Dixon & Massey, 1957). Using the respectively, the sampling and analytical errors in
proportion X/n=p with n > 30, the probable error percent (Krumbein & Rasmussen, 1941). The sam-
associated with counting may be read from a nomo- pling error is a function of the homogeneity of the
graph (Krumbein & Graybill, 1965; Van der Plas & deposit and of the precise point of sampling, whereas
Tobi, 1965) or calculated as follows: the analytical error is a function of clast identification
and counting. In contrast to particle size or heavy
CE = k t J P m , (1) mineral analysis, clast counts are carried out on the
160 J . L. Howard

outcrop, hence errors associated with mechanical the error by half. Judging from nomographs that show
sample collection, splitting, sieving, and so forth, are counting error as a function of n and clast percentage
avoided. (Dixon & Massey, 1957; Krumbein & Graybill, 1965),
Assuming that the error associated with clast the optimum number of clasts to be counted in any
identification is negligible, the analytical error is given sample is 400. Below this value precision
equivalent to the counting error, which may be declines markedly, but increasing the count to 1000
calculated according to Eq. (1). The counting error is probably does not increase the precision enough to
equivalent to the total probable error only if the clasts justify the additional labour required, except perhaps
are randomly distributed within a population, i.e. in very detailed studies.
clasts of the same type are not clustered together. Any Table 1 shows that in some cases sampling errors
non-random distributions of clasts in the deposit, are of comparable or greater magnitude than counting
perhaps due to the effects of selective sorting by clast error. In such instances, the observed total probable
shape, size, density and mass, will result in a sampling error was at least 41% larger, and observed to be up to
error even if the error associated with the selection of 85% greater than counting error. Thus, replicate clast
sampling location is negligible. counts probably should be used routinely in order to
Table 1 compares counting, sampling and total assess the magnitude of sampling error. Sampling
errors based on clast counts at five locations of upper error varies inversely with 4, hence four closely
Palaeogene non-marine conglomerate belonging to spaced replicate counts are required to decrease the
the Sespe Formation, and one locality of the Jalama error by half. Combining four replicate counts has the
Formation, a submarine fan conglomerate of Late same effect as composite sampling and will reduce the
Cretaceous age (Fig. 1). The magnitude of sampling sampling error by half regardless of its magnitude
error was assessed for a given site by means of a series (Krumbein & Pettijohn, 1938; Griffiths, 1967). If the
of replicate clast counts spaced several metres apart. sampling error can be reduced to 50% of the counting
The number of replicate counts ( k ) ranged from 3 to 6 error, the total probable error will only be 12% larger
and the number of clasts counted in each replicate (n) than the counting error.
varied from 100 to 200. The sets of replicate counts at These results imply that the best approach for clast
each location were normalized, and the means (maand counting at any given spot is to obtain a sample
sample standard deviations ( S ) were calculated for composed of four or more closely spaced replicate
each clast type. The sampling error was estimated counts. The replicate counts may be used to measure
using the relation for the standard error of the mean: the sampling error, or the total error derived from Eq.
SE= 1.96 (S/,,h), where SE is the sampling error in (2), and serve to minimize sampling error. The
percent; 1.96 is the tabulated t-value for n- 1 degrees replicates should consist of equal numbers of clasts, in
of freedom at the 0.05 level; and S is the measured order to give the results equal statistical weight.
standard deviation associated with some number of Counting error can be minimized by counting a total
replicate counts k . Counting errors and total probable of at least 400 clasts. Such a sample can be obtained
errors were calculated according to Eqs (1) and (2), by making four closely spaced counts of 100 clasts
respectively. The results at sites 15, 9 and 8b agree each, and then combining the results.
well with those of Krumbein (1942). He reported a Another important relationship to be noted in Table
total probable error of +_ 10.2%, based on a set of 24 1 is the fact that for small values of z, TE grows very
closely spaced counts of 100 clasts each, for a lithology large in comparison. Thus, when x=40-50% the
comprising 35% of the whole assemblage. coefficient of variation (CV = T E / a is,l6-30%, when
Tlie results in Table 1 show that both counting and x= 30% the CV is 30-50%, when x= %
1 the CV is
sampling errors contribute significantly to total prob- 50-100% and when X < 5%, the CV is 100-500%. This
able error. Counting errors are on average about 60% is due in part to the fact that although counting error
greater than sampling errors (range = 10-200% reaches a maximum value at 8=50% (Griffiths, 1967),
greater), hence total errors average only about 30% the CV ( C E / B decreases progressively as the clast
larger than counting errors. Counting error, therefore, percentage increases. For example, the counting errors
appears to be a good first approximation of total error. associated with clast percentageof 5 and 10%(n= 100)
Because counting error varies inversely with &, the are 4.3 and 4.7%, respectively, but the corresponding
total number of clasts counted is a critical factor CVs ( C E / B are 86% and 47%. These data indicate
controlling precision, however, an increase of sample that not only does precision increase with increasing
size by a factor of four is required in order to decrease clast percentage, but little interpretative value can be
 Table 1. Errors associated with clast counting. Sample locations given in Fig. 1. All values are percentages.
CIast tvDe
,a
Location 15 Location 9 Location 10 Location 14 Location 8a Location 8b Location 2
8 CE SE TE 8 CE SE TE 8 CE SE TE 8 CE SE TE 8 CE SE TE 8 CE SE TE
Metavolcanic* 32.0 9.2 5.8 10.9 - _ _ _ _ _ - - - - - - - - - - - - - -

Metarhyolitet - - - - 29.1 8.9 11.3 14.4 44.3 9.7 3.2 10.2 19.6 6.4 4.8 8.0 10.7 4.9 2.7 5.6 10.1 4.2 8.1 4.2
Meta-andesite _ - _ _ 19.9 7.8 12.2 14.5 5.6 4.5 2.7 5.2 2.2 2.3 1.8 2.9 2.3 2.3 1.7 2.9 - - - -
Metabasalt 4.8 4.2 1.9 4.6 _ - - _ _ _ _ - - - - - - - - - 0.7 1.2 1.4 3.4
Metatuff 2.4 3.0 3.4 4.5 1.0 1.9 1.4 2.4 0.9 1.8 1.8 2.5 1.3 1.9 1.9 2.7 2.5 2.5 1.6 3.0 9.4 4.0 4.9 6.3
Gneiss 3.3 3.5 2.6 4.4 _ _ _ _ _ _ - _ 2.2 2.3 1.9 3.0 3.8 3.1 2.8 4.2 - - - -
Quartzite 21.4 8.0 10.1 13.5 21.1 8.0 3.8 8.9
11.1 13.7 30.0 9.0 5.5 3.6 4.2 5.5 3.0 2.7 1.4 3.0 54.0 6.9 5.3 8.7
Marble _ - - _ _ _ _ _ 0.2 0.7 0.4 0.8 0.3 0.7 0.6 0.9 - - - -
Slate _ _ _ _ 1.1 2.0 1.1 2.3 - - - - - - - - - _ - -
Vein quartz 4.0 3.8 2.3 4.4 _ _ _ _ - - - - - - - - - - - -
Granite 2.5 3.1 1.9 3.6 _ - _ _ _ - _ - 10.0 4.7 5.0 6.9 31.8 7.4 6.1 9.6 - - - -
Graniteandgneiss - - - - 14.5 6.9 5.0 8.5 26.2 8.6 3.1 9.1 - - - - - - - - 11.7 4.4 2.4 5.0
Diorite 28.5 8.8 7.5 11.6 2.2 2.9 4.4 5.3 0.8 1.7 1.6 2.3 - - _ - 0.2 0.7 0.4 0.8 - - - -
Diabase _ - _ _ 1.9 2.7 1.9 3.3 0.9 1.8 0.9 1.0 - - - - - - - - - - - -
Basalt - _ - _ 0.2 0.9 0.4 1.0 1.5 1.9 1.9 2.7 0.2 0.7 0.4 0.8 - - - -
Chert 1.1 2.0 1.6 2.6 _ _ _ _ - - - - - - - - 3.4 4.4 2.6 5.1
Limestone _ _ _ _ - _ - _ 3.3 2.9 1.9 3.5 2.0 2.2 3.3 4.0 1.0 1.4 1.3 1.9
Sandstone _ - _ _ _ _ _ _ _ _ _ - _ _ _ _ 53.9 8.0 9.7 12.6 43.6 7.9 10.6 13.2 9.8 4.1 9.0 9.9
n 100 100 100 100 150 150 200
k 6 3 3 3 3 3 4
*Undifferentiated.
?Includes metadacite-metalatite.
Symbols: 8,mean; CE, counting error; SE, sampling error; TE, total error.
162 J . L. Howard

given to comparisons between clast types that compose select random distances from an origin on a coordinate
less than 5 1 0 % of the sampled population. system, and then use the corresponding point as the
site where the sample will be taken (Fig. 2). Once the
outcrop locations are established, it will be necessary
FIELD PROCEDURE to select at random the exact spots where the clasts
will be counted. As an example of one possible
Preliminary studies approach, suppose the objective is to randomly choose
four 1 x 1 m areas in which tocount clasts from a 10 x
A detailed study, in which clast counts are to be used 10 m areaof outcrop (Fig. 2). Since the 10 x 10 m area
rigorously, should begin with a field reconnaissance represents a total of 100 possible 1 x 1 m areas, each
of the areas to be sampled. The objectives are to could be assigned a number and the four selected
determine the extent of outcrop exposure and acces- using a table of random numbers.
sibility, and the identity of the clast types to be
counted. A field reconnaissance is critical for the
formulation of a sampling plan that will be most
effective for the particular situation. Representative
clasts must also be collected for thin-sectioning, or
perhaps slabbing and staining, and petrographic study
made prior to counting if clast counts are to be carried
out accurately. The confidence interval about the
mean defined above describes only the precision of
the clast count. Clearly, this has little statistical value I //
if the accuracy of clast identification is poor. Perfect
accuracy will obviously be unobtainable; even if a
thin section was made of every clast counted, errors P om
in identification would probably still result. The
overall accuracy of any clast count study, therefore,
will depend on both how many clast types were thin
sectioned, and how difficult it was to identify the
particular clast types present while counting was 200 600 m
taking place. These circumstances should be reported
along with the counting results. Fig. 2. A method for randomly selecting spot locations for
clast counting from an outcrop belt. Modified from Petersen
& Calvin (1965).
Selectionof sample location
The importance of obtaining representative samples If the objective is stratigraphic analysis, the map
from the available population of clasts has been noted location of a suitable section may also be selected at
above. Under ideal conditions, this will involve random using the above method. Either systematic
selecting random outcrop locations on a geological (Fig. 3A) or simple (Fig. 3B) sampling strategies may
map, as well as random areas on the outcrop surface then be used to choose the enact sampling points. In
where the clasts will be counted. Some examples of the former, the total thickness to be measured may be
possible approaches are given below, but inevitably divided by the number of sample sites desired to
there will be problems related to physically reaching obtain some number of vertical intervals of equal
the location of some pre-determined point. Because thickness. Each of these could be divided in turn into
each situation is likely to have a unique set of a number of vertical subintervals whose thickness is
circumstances, the only hard and fast rule is that an equal to that of the final area to be sampled. Clast
element of random sampling should be incorporated counts are then made at a fixed spacing within each
(whenever possible) at each level of the sampling main interval. The latter approach is the same except
process. that each subinterval is assigned a number, and one
Petersen & Calvin (1965) have described a method chosen at random from the set of all possible areas
for selecting sampling sites at random which can be within each main interval to obtain the desired
adapted to geological maps. The basic approach is to number of samples. Systematic sampling has the
 The statistics of clast counting 163

A B located, but in many cases the locations for study will

be predetermined simply by uncontrollable circum-
stances in the field area. If this is the case, and it is
impossible to evaluate how representative the sam-
4 pling sites are, an unknown amount of uncertainty is
introduced regarding the estimate of the true confi-
dence interval about the mean. The best guideline,
therefore, is to implement some sort of random
3 sampling method whenever possible. It is useful if the
methodology used is reported fully so that the readers
may judge for themselves whether or not random
sampling was accomplished.
2
Counting procedure

1 The goal in clast counting is to determine the number

of clasts of a given type present (a proportion or
percentage), independent of clast size. However, most
investigators have apparently adopted as ‘standard
Fig. 3. Sampling strategies for clast counting at multiple
stratigraphic positions. Four subintervals have been selected procedure’ a method for counting conglomerate clasts
by: A, systematic or fixed-interval sampling; B, simple analogous to the point counting technique commonly
random sampling. used for modal analysis of rock thin-sections. Typically
a grid is used at the outcrop to select at random the
particular clasts to be counted (Fig. 4A) until a total
advantage of producing more even coverage over the of 100 have been tabulated (e.g. Blatt, 1992). The
entire stratigraphic interval of interest. results are expressed as percentages, but point count-
Using the above methods, perhaps in conjunction ing was originally devised to obtain volume percent-
with sophisticated surveying techniques, it is possible ages of minerals in igneous rocks (Chayes, 1956). The
theoretically to reach the exact location on the ground volume percentages of clasts in a Conglomerate,
of some predetermined point. From. a practical however, are clearly dependent on particle size, which
standpoint, however, there is little justification for may vary non-randomly because of such factors as
doing this as long as a random sampling system is hardness, cleavage, etc. Point counting of a coarse
used at each scale in the process. Even the implemen- grained sedimentary rock results in a biased estimate
tation of a sampling system using a crude pace-and- of clast percentages which cannot be treated statisti-
compass method is better than no system at all. It is cally. This is because the probability of any given
very likely that occasionally the final points arrived at particle being touched by a grid point is proportional
using the above methods will be unsatisfactory because to the area of the visible surface, and since some clasts
of an obstruction, lack of suitable lithology, etc. The may be more deeply buried in the outcrop, they have
procedure could be repeated until suitable sites are less of a chance of being counted by this procedure

Fig. 4. Contrasting methods for counting clasts in rudites: A, grid method; B, line method; C, ribbon or area method.
164 J . L. Howard

than those with greater surface exposure. Even if all counts of unequal size. It also can be used graphically
clasts are exposed evenly, the method may still result to make plots in which the sizes of error bars or error
in biased data because a larger clast is more likely to boxes show the relative degree of precision of each
be counted than a smaller one. In either case, the measurement. Because each confidence interval con-
result is the same as point counting sand sized particles tains all values that are not significantly different from
in a grain mount, a ‘clast frequency’, which is larger the mean, statistical comparisons can be made rapidly
than the clast percentage for larger particles and by visual inspection. Hypothesis testing is an alterna-
smaller than the clast percentage for smaller particles tive method for calculating directly whether or not
(Van der Plas, 1962; Galehouse, 1969). An alternative statistically significant differences are present.
approach might be the ‘line method’ in which all Confidence intervals for individual clast counts may
individual particles encountered along a series of be calculated either according to Eq. (1) or (2). Ideally,
linear traverses are counted (Fig. 4B). However, this Eq. (2) should be used, but Eq. (1) is employed below
method also generates a clast frequency rather than a either because the sampling error is unknown, or
clast percentage (Galehouse, 1969). Because differ- because it is assumed to have been minimized by
ences in particle size would again result in a built-in counting in subsets. As an example, consider the
bias, such data are not amenable to rigorous statistical stratigraphic section of Sespe Formation at location
treatment. 13 (Fig. 1). Counts were made in ascending strati-
In view of these considerations, the best approach graphic order at four stratigraphic positions (A, B, C
for clast counting is probably the ‘area or ribbon and D), the data tabulated and counting errors
method’ (Fig. 4C). This method yields clast percent- calculated using Eq. (l), and a plot made showing
ages, which can be handled statistically as long as the clast percentagesas afunction of stratigraphic position
width of the band or area sampled is considerably (Fig. 5). Statistically significant differences are readily
greater than the diameter of the largest particle apparent where mean values fall outside of the
(Galehouse, 1971). In the case of conglomerates and confidence intervals of other means. The observed
breccias, the technique would involve counting each variations in abundances with stratigraphic position
individual clast within a specified area until the reflect depositional changes over time, perhaps due to
desired sample size is obtained. The size of the area base level change in response to eustasy or tectonism.
sampled should probably be at least 2.5 times the long Consistent, well defined patterns conceivably could
axis diameter of the largest clast. A lower clast size be used for stratigraphic correlation.
limit (e.g. 3 cm) may be used, below which identifica- Alternatively, the null hypothesis ( H J that propor-
tion is questionable, otherwise each gravel-sized
particle is counted regardless of size. A marker pen
may be used in the field to mark each individual clast
in the outcrop as it is counted so as to prevent counting I 1
the same particle twice. A systematic counting pattern
probably should also be adopted, for example, count-
ing from left to right and downward as if reading a
book. Seiders & Cox (1992) found it useful to
60 I ’I I 11
implement an upper size limit on the clasts counted
because their intention was to study clasts derived
only from very distant sources.

DATA A N A L Y S I S
t+
I I I

A B C D
yo
1 1
I

A B C D
L

A B C D
4 J

Confidence intervals and hypothesis testing Stratigraphic position

Confidence intervals can be used to make statistical Fig. 5. Confidence intervals (95%) for individual clast counts
at four stratigraphic positions in the Sespe Formation section
comparisons either between individual counts, or at location 13 (Fig. 1). Stratigraphic positions are in
between groups of counts. The technique is versatile ascending order. Clast types: G , granitoid; M, metavolcanic;
because it permits comparisons to be made between Q, quartzite. Data from Howard (1989).
 The statistics of clast counting 165

tions p , and p 2 in two samples are the same may be

tested using Student’s t-test. Given two random
samples of size n , and n,, and corresponding observed
proportionsp, = Xl/nl andp, = X 2 / n 2 ,where X,and
X, are the number of clasts of a given type, the
difference betweenp, andp, may be estimated by the
following confidence interval :
0 0 0 0 Marker bed
P1 -P2 ktJPl(1 - p , ) / n , +P20 -Pz)/n,, (3)
N o r m a l fault
where t is the tabulated value of Student’s t-distribu-
tion for 4 2 level of significance and n , + n 2 - 2 degrees
of freedom. If this interval covers zero, then there is
I % Attitude
Fig. 6. Geological sketch map (not to scale) showing alleged
no reason to reject the hypothesis that p 1 = p 2 ; if it Sespe Formation conglomerate marker bed (Belyea & Minch,
does not, the hypothesis should be rejected (Dixon & 1989) and stratigraphic sections I and I1 at locations 14 and
Massey, 1957). 13, respectively (Fig. 1). Geological mapping by Schoell-
As an example, consider the proportionsp, =0.067 hamer etal. (1981).
andp, = 0.271 of granite clasts at positions A and B in
Fig. 5 with n , = 252 and n, = 225. The appropriate
95% confidence interval for p 1 -p2 is : ‘Poway-type’clasts (reddish purple metarhyolite with
quartz and feldspar phenocrysts). The match-up was
0.271 -0.067k used to map out a conglomerate ‘marker bed’, and the
1.96J0.067(0.933)/252 + 0.271(0.729)/225, local stratigraphy was redefined on this basis (Belyea
or & Minch, 1989).
If the data are replotted showing 95% confidence
0.204-0.0658 <PI -p2 <0.204+0.0658, intervals (Fig. 7), beds A and B in section I do appear
or to match bed H in section I1 in terms of both ‘Poway’
and ‘pseudo-Poway’clasts (reddish-purple metarhyol-
0.138<p, -p2 <0’270.
ite containing only feldspar phenocrysts), although
Because this interval does not cover zero, the null the error bars are very wide. This is attributable to the
hypothesis is rejected and it can be stated with ‘95% large counting error associated with counts of only
confidence’ that there is a statistically significant about 150 clasts each. The lack of a statistically
difference between p1 and p 2 . On the other hand, the significant difference in this case, therefore, cannot be
confidence interval for the difference between the used with confidence to conclude that the beds are the
means at positions C and D is - 0.028 <pl - p 2 < same. This is because statistical analysis is designed
0.100, and the hypothesis is accepted that nodifference to show the likelihood that differences exist, not to
exists. prove that certain values are the same, hence it is
An example from the Sespe Formation of how such essential that confidence intervals be reasonably
methods might be applied to attack a lithologic narrow before the null hypothesis is accepted. The
correlation problem is illustrated in Fig. 6. Strati- ‘Poway-type’ clasts used in the alleged match-up also
graphic sections I and I1 (Fig. 1, locations 14 and 13, comprise at most only about 9% of the samples
respectively) are located approximately 6.5 km apart obtained. As demonstrated above, little interpretative
and previous mapping (Schoelhamer et al., 1981) value can be given to clast components making up 5
shows they are separated by a normal fault. At the 10% because sampling and counting errors are very
base of section I are two thick, massive conglomerate large by comparison. Precision increases with higher
beds that physically resemble somewhat the ‘marker component percentage, hence a more effective com-
bed’ located far above the base of section 11. Clast parison can be made by combining the proportions of
counts, mainly of 100-150 clasts each, were made at a ‘Poway’ and ‘pseudo-Poway’ clasts, which together
series of stratigraphic positions in each section, the comprise 31,24 and 20%of the supposed ‘marker bed’
results of which are summarized in Table 2. The in stratigraphic positions IA, IB and IIH, respectively
authors concluded that basal beds A and B of section (Table 2). A comparison between sample pairs using
I match bed H in section I1 on the opposite side of the Eq. (3) above shows that there are significant
fault based on the similarity in the proportions of differences, both in terms of the combined proportions
166 J . L. Howard

Table 2. Proportions of clasts composing Sespe Formation conglomerates at locations 13 and 14 (Fig. I).
Modified from Belyea & Minch (1989). Values are percentages.
Ascending stratigraphic position
Clast type A B C D E F G H I J
Section I : location 14
Poway 9* 8* 3 4 3 2 2 2 4 4
Pseudo-Poway 22* 16* 2 1 2 7 7 7 6 7
Andesi te-dacite 20* 26* 22 16 14 16 19 23 21 27
Aphanitic 22* 222 21 21 22 22 18 22 19 31
Granite 3* 2.5* 11 16 28 30 21 20 12 12
Quartzite 18* 20* 28 28 24 17 22 21 29 16
Metatuff 5* 3* 4 5 6 5 8 4 6 3
Miscellaneous 1* 2.5* 3 3 1 1 3 1 3 1
n 125 126 115 146 148 156 157 116 141 122

Section 11: location 13

Poway 2 1 2 4 3 2 3 8 *
Pseudo-Poway 9 9 5 3 9 6 10 12*
Andesite-Dacite 11 11 9 14 16 22 22 17*
Aphanitic 15 22 24 19 21 26 21 24*
Granite 21 21 23 15 12 16 12 11*
Quartzite 25 25 26 28 29 23 27 23*
Metatuff 6 3 5 5 4 4 4 4 *
Volcanic sandstone I 2 6 0 5 0 0 0 *
Miscellaneous 4 0 1 2 1 1 1 1 *
n 122 125 204 130 128 137 143 114
*Alleged marker bed.

I 1
ETION I of metarhyolite clasts and granite clast content
30 (Table 3). These results argue against the presence of
a Sespe conglomerate marker bed. The same results
also would have been obtained had granite clasts been
included in the original comparison (Fig. 7). This
10 correlation problem is not trivial because the postula-
tion of the 'marker bed' led to a rejection of previous
tectonic interpretations.
A J A J A J Clast counts made at multiple positions in a
stratigraphic section, or at multiple points within a

1 SECTION II I
t
Table 3. Tests for significant differences between sample
pairs in sections I and I1 at locations 14 and 13, respectively
Pseudo
30 Powoy b (Fig. I).
Confidence intervals Hypothesis
Sample pairs 011 - P 2 ) PI =P2
Combined metarhyolite clasts
IA versus IB - 0.04 <pl -pa < 0.180 Accept
IA versus IIH 0.001 < p l -pi <0.219 Reject
A H A H A H IB versus IIH -0.065 < p l - p 2 <0.145 Accept
Stratigraphic position
Fig. 7. Confidence intervals (95%) for individual clast counts Granite clasts
at multiple stratigraphic positions in the Sespe Formation IA versus IB -0.035 < p l - p 2 <0.045 Accept
sections I and I1 at locations 14 and 13, respectively (Fig. 1). IA versus IIH 0.015<p, -p,<O.145 Reject
IB versus IIH 0.021 <pl -pz <0.149 Reject
Data from Belyea & Minch (1989).
 The statistics of clast counting 167

geographical area, can be grouped together and In this particular case, the data may not be strictly
compared against similar groups from other areas. comparable because an upper size limit on clasts
Confidence intervals for groups of ccunts can be counted was used in one study, but not the other two.
calculated using the relation : CI = tS/Jk, where CI is Nevertheless, this type of approach is clearly an
the confidence interval in percent; t is the tabulated effective means for graphically displaying statistical
value of Student’s t-distribution for 4 2 level of comparisons of clast count data.
significance and k- 1 degrees of freedom; S is the
measured standard deviation; and k is the number of
clast counts in each group being compared. Casual Analysis of variance
observation of a table of t values shows that each Analysis of variance (ANOVA) has the advantages of
group should consist of counts from six or more being able to test simultaneously for significant
different places if possible. With smaller values of k, t differences among more than two samples, and of
is relatively large, possibly resulting in unacceptably running a lower risk of making a Type I error (rejecting
wide confidence intervals. H, when it is true) than with Student’s t-test. In a one-
Figure 8 shows how comparisons between groups way ANOVA test, the within-groups variance estimate
of clast counts can be used to address the difficult is compared against the among-groups variance using
problem of identifying recycled sediment. Each group the F statistic. The hypothesis tested is that the means
consists of counts from multiple geographical loca- of all groups are equal (R, = X 2 = F3= X4,etc.)
tions. Data from the Sespe Formation (Howard, 1987), against the alternative that at least one mean is
Palaeocene conglomerates (Colburn & Novak, 1989), different. The calculations are more readily carried
and Cretaceous conglomerates (Seiders & Cox, 1992) out using numbers of clasts rather than proportions,
in the mountains around location 11 (Fig. 1) were and using groups composed of counts of equal n-size
compiled, and a ternary plot was constructed showing (Krumbein & Miller, 1953).
95% confidence intervals for each mean. Significant To calculate the F statistic, first find the sum of
differences are readily apparent because no mean falls squares within groups: SSW =xX2-[(CXl)Z/nl +
within the error box of another mean. The results +
( x X 2 ) 2 / n 2 . . . 1, where CX’ means that each value
suggest that Sespe Formation conglomerates were not in every group is squared and the sum of all is
formed by recycling of older nearby conglomerates. obtained; and (CX1)’/nl is the sum of all values in
group 1 squared and divided by the number of values
in group 1, etc. The sum of squares among groups is
then calculated as follows : SSA = [(CXl)2/nl+
(CX2)’/n2+ . . . I- (CX>’/n,, where (CX)’/n, is the
sum of all values in every group squared and divided
by the total number of values in all groups. The total

/
sum of squares is : SST = C X 2- (CX>’/n,;the within-
groups variance estimate is Sw2 = SSW/n, - k , where k
is the number of groups; and the among-groups
variance estimate is S,’=SSA/k- 1. The F ratio is
computed as follows: F=Sa2/Sw2.The calculated F
O
z
e
trQ
;/,
value is compared against a tabulated F valueobtained
Sespe Fm. Conglom. with degrees of freedom among, df,=k- 1, and
degrees of freedom within, df,=n,-k. The test is

G
/ - ”
c
Palaeocene
Conglom.
\

v
one-tailed so if the calculated value of F exceeds the
tabulated value, the null hypothesis is rejected and it
may be concluded that at least one mean is different
(Minium &Clarke, 1982; Davis, 1986).
As an example, consider four groups of two or three
Fig. 8. Confidence intervals (95%) for the means (solid replicate counts of granitoid clasts representing differ-
circles) of groups of clast counts comparing clast assemblages ent stratigraphic positions (A-D) at location 13
in Sespe Formation and pre-Sespe Formation conglomerates
in the vicinity of location 11 (Fig. 1). See text for sources of (Table4). The results of analysis of variance using
data. Clast types: G = granitoid, M = metavolcanic, k=4, dfw=6, dfa=3 and n,=10 show that the
Q = quartzite. calculated F value of 16.0 exceeds the tabulated value
168 J. L. Howard

Table 4. Replicate clast counts (n= 100) at four stratigraphic Table 6. Clast count data (n= 150) for stratigraphic sections
positions in the Sespe Formation at location 13 (Fig. 1). of the Sespe Formation at locations 12, 13 and 14 (Fig. 1).
Values are numbers of granite clasts counted. Values are numbers of clasts counted. Some lithological types
are not represented. Recalculated from Belyea & Minch
Ascending stratigraphic Dosition (1989).
Replicate A B C D
~
Location 12 Location 13 Location 14
1 10 24 16 15 G Q M G Q M
2 3 31 13 10 ~~ ~

3 7 14 20 17 108 32 37 55 5 26 108
37 35 75 31 37 73 4 30 108
Mean 6.7 27.5 14.3 12.5
17 43 80
28 31 85 35 40 59
12 28 101 22 43 75 25 41 72
15 38 88 18 43 74 42 36 62
of 9.78 (Table 5). Thus, it can be stated with ‘99% 14 33 92 24 35 84 44 26 71
confidence’ that there are one or more statistically 18 40 85 32 33 68
significant differences in the granitoid clast content of 17 34 91 30 31 81
the Sespe conglomerate in the stratigraphic interval 17 45 76
17 23 66
studied.
If the purpose of an investigation is to make regional G, granite; Q,quartzite; M, metavolcanic.
comparisons between clast populations in different
areas, one approach is to use ANOVA and groups of with ‘95% confidence’, therefore, that there is no
data composed of clast counts from randomly selected evidence to indicate that there are significant differ-
stratigraphic positions at many localities within each ences in the mean numbers of granite, quartzite and
area. Using this approach, and the data of Howard metavolcanic clasts in the three stratigraphic sections
(1987), showed that very large within-group variances studied. These data suggest that there is relatively
result, thereby decreasing the power of the test. A little variation in the proportions of those clast types
better approach is to use groups of data composed of in the Sespe Formation conglomerate on the scale of
clast counts made at multiple stratigraphic positions a single mountain range.
in one or more sections from each area. This approach The same procedure may be used to test the
has the advantage of providing an integrated view of hypothesis that there are systematic differences in
a formation because data are collected over an entire clast populations when several mountain ranges are
stratigraphic interval. This method is illustrated using compared. Multiple counts were obtained for strati-
stratigraphic sections of the Sespe Formation at graphic sections of the Sespe Formation at locations
locations 12, 13 and 14 (Fig. 1). The original data 7, 9, 11 and 13 (Fig. l), and the results tabulated as
(Belyea & Minch, 1989)are based on counts of unequal counts of equal size for granite, quartzite and
size, hence they were first normalized using n = 150, metavolcanic clasts (Table 8). The mean granite clast
and then tabulated as groups representing counts of
equal numbers of clasts at different stratigraphic Table 7. Analysis of variance data for stratigraphic sections
positions (Table 6). The hypothesis tested is that the of the Sespe Formation at locations 12, 13 and 14 (Fig. 1).
means of the sections are the same. The results ~

Degrees
(Table 7) show that none of the calculated F values Sumof of Variance
exceed the table value of 3.47, so it can be concluded Source squares freedom estimate F*
Granite clasts
Table 5. Analysis of variance data for granite clast contents Among groups 45.3 2 22.7 0.2 (accept)
at different stratigraphic positions in the Sespe Formation at Withingroups 2556.0 21 121.7
location 13 (Fig. I). Total 2601.3 23
Degrees Quartzite clasts
Sumof of Variance Among groups 252.0 2 126.0 3.0 (accept)
Source squares freedom estimate F* Within groups 879.6 21 41.9
Total 1131.6 23
Among groups 529.8 3 176.6 16.0t (reject) Metavolcanic clasts
Withingroups 66.3 6 11.1 Among groups 1029.9 2 515.0 2.6 (accept)
Total 596.1 9 Within groups 4173.3 21 198.7
Total 5203.0 23
*Tabulated F values=4.76 (a=0.05); 9.78 (a=0.01). ~~~

?Highly significant difference. *Tabulated F values = 3.47 (a= 0.05); 5.78 (a = 0.01).
 The statistics of c h t counting 169

Table 8. Clast count data (n=400) for stratigraphic sections the within-groups variance estimate, and n is the
of the Sespe Formation at locations 7, 9, 11 and 13 (Fig. 1). number of values in each group (Minium & Clarke,
Values are numbers of clasts counted. Some lithological types 1982). For the above example, with a=0.05,
are not represented.
HSD=4.2,/-=91.7 in terms of granite clast
Location 7 Location 9 Location 11 Location 13 content. After the differences between all possible
G Q M G Q M G Q M G Q M pairs of sample means (xl,Tz,etc.) are computed and
tabulated, any values exceeding the critical HSD are
174 53 54 204 119 26 24 24 199 8 101 242
182 84 79 97 83 31 35 117 160 39 123 232 identified as significant. The results (Table 10)
129 27 140 53 129 108 55 58 269 29 129 202 strengthen the interpretation that granite clasts in-
101 86 180 129 150 146 73 98 188 crease in abundance northward, and metavolcanic
27 75 287 clasts southward, in the Sespe conglomerate. As
G, granite; Q, quartzite; M, metavolcanic. discussed previously, such geographical trends may
be useful in deriving a palaeotransport direction, or
for identifying differences in provenance.
content increases steadily northward, while metavol-
canic clasts increase in abundance southward. The
results of ANOVA, comparing means among the Ratio analysis
different sections, show that although there are no A count ratio is obtained by dividing the clast
differences in quartzite content, highly significant percentage or proportion of one clast type by another
differences exist for granite and metavolcanic clast in the same sample. Ratio comparisons are useful in
proportions (Table 9). basin analysis because changes in the relative abun-
The locations of the differences may be established dances of clast type often follow a predictable pattern.
using Tukey’s post-hoc HSD (honestly significant For example, the ratio of percent granite to percent
difference) test. The test is two-tailed and involves quartzite clasts in a conglomerate normally decreases
determining a critical value: H S D = q m , where with increasing transport distance because softer or
q is the value of the Studentized range statistic for the less durable clast types are abraded and reduced in
desired level of significance (a), within-groups degrees size and abundance more rapidly than harder clasts in
of freedom (df,) and number of groups ( k ) ; SW2 is the same assemblage. Thus, changes in count ratios
with geographical position in a basin may document
the pattern of sediment dispersion. Similarly, the
Table 9. Analysis of variance data for stratigraphic sections
of the Sespe Formation at locations 7,9, 11 and 13 (Fig. 1).
study of stratigraphic changes in count ratios may
~~ ~~~~ reveal changes in base level with time. For example,
Degrees if a source area was composed of both granite and
Sumof of Variance quartzite, and the former was uplifted by faulting, the
Source squares freedom estimate F*
ratio of percent granite to percent quartzite clasts in
Granite clasts syntectonic conglomerate would likely increase upsec-
Among 35 691.0 3 11 897.0 6.27 (reject) tion.
groups
Within 22905.0 12 1908.8 If x l , x2, x,, etc. and y , , y 2 , y,, etc. are the
groups proportions of two clast types with means .f and J and
Total 58597.0 15 standard deviations S, and S,,, respectively, and if
Quartzite clasts xl/yl, xz/yz, x 3 / y 3 ,etc. are the count ratios of the
Among 5727.0 3 1909.0 1.6 (accept) two clasts in some number of clast counts k , then . f / j
groups
Within 14787.0 12 1232.3 is an estimate of the mean of the count ratios
groups +
Fl,vwith variance SzXiy= ~ 2 / k @ ) 2 [ S x 2 / .Sy’/J2
fz -
Total 20514.0 15 2(r)(Sx)(Sy)/.fj],where r is the correlation coefficient
Metavolcanic clasts between xl, xZrx3, etc. and y , , y,, y 3 , etc. (Koch &
Among 52 487.9 3 17 496.0 6.7t (reject)
groups Link, 1971). A confidence interval for TxiV may then
Within 31 362.6 12 2613.5 becalculated using the standard relation: CI = t ( S , d
groups J k , where CI is t e o n f i d e n c e interval (dimension-
Total 83850.4 15 less); Sxiuis the JSzXly,t is the appropriate tabulated
*Tabulated F values = 3.49 (a= 0.05); 5.95 (a = 0.01). value for a12 level of significance and n - 1 degrees of
?Highly significant difference. freedom, and k is the same as above.
170 J . L. Howard

Table 10. Tukey's HSD test data for stratigraphic sections of the Sespe Formation at
locations 7 (x,),
9 (8'),11 (y3)and 13 (z4) (Fig. 1).
Granite clasts'
8,= 161.7 R2=113.8 y3= 60.8 y4= 35.2
8,= 161.7 0 47.9 100.9* 126.5t
8*= 113.8 0 53.0 78.6
x3= 60.8 0 25.6
F4= 35.2 0
Metavolcanic clasts'
8 , = 91.0 8,= 11 1.2 X 3 = 193.5 = 230.2
2 4
X I = 91.0 0 20.2 - 102.5 - 139.2*
X' = 111.2 0 - 82.3 - 119.0.
y3= 193.5 0 - 36.7
y4= 230.2 0
'HSD,,=91.7; HSD,, = 120.1. *HSD,,= 107.4; HSD,, = 140.6.
*Significant difference.
?Highly significant difference.

Using the data from Table6, mean count ratios dilution. If a stream is carrying granite and quartzite
(,Tx,,) were calculated for the three stratigraphic clasts, for example, and a tributary introduces abun-
sections of the Sespe Formation at locations 12, 13 dant greywacke clasts, the component percentages of
and 14 (Fig. I). The values of Sxiv and confidence quartzite and granite cla.sts will change markedly at
intervals are k0.2 or less, hence there is good this point, but the ratio between the two will be
resolution of differences in xxxiY between sites unaltered. As an example, McCracken (1972) made
(Table 11). The count ratios indicate random varia- clast counts of Sespe Formation conglomerates along
tions in granite and metavolcanic clasts from north to an E-W transect (Fig. 1, locations 1-6). The data
south across the range. Thus, the observed results show a complex mixture of two clast assemblages, one
agree with those reached above using analysis of comprised of quartzite, granite, gneiss and metavol-
variance (Table 7). Statistically significant differences canic rocks, and the other of chert, metabasalt,
were observed for metavolcanic/quartzite count ratios, greywacke and limestone. No systematic relationship
whereas analysis of variance found no differences.
This suggests that the use of count ratios may be
somewhat more discriminating.
Table 11. Ratio analysisof clast types in stratigraphic sections
The data from Table 8 were also used to compute
of the Sespe Formation at locations 12, 13 and 14 (Fig. 1).
count ratios for stratigraphic sections of the Sespe Recalculated from Belyea & Minch (1989).
Formation in different mountain ranges (Table 12).
The count ratios suggest a systematic decrease in Location 12' Location 13' Location 143
granite clasts from north to south comparable to Granite/quartzite
results obtained using analysis of variance (Table 9), TI,,> 0.69 0.64 0.70
however, statistically significant differences are ob- Con. Int. 0.15 0.06 0.10
sx/r 0.14 0.07 0.14
served only between comparisons of granite/quartzite
ratios. Overall, there is poor resolution of differences Granite/metavolcanic
in zx,v between sites because of wide confidence YA,Y 0.23 0.33 0.29
intervals. The data in Table 1 1 show that if k > 6 or 7, Con. Int. 0.06 0.04 0.03
then t / J k is about equal to 1.0 and confidence sx,, 0.05 0.05 0.04
intervals are then approximately equal to S+. Using Metavolcanic/quartzite
Sriras a measure of precision in Table 12, many more Fx,s 3.00 1.90 2.40
differences between sites are evident, and it is Con. Int. 0.21 0.09 0.14
recommended that no fewer than seven samples be S,,, 0.20 0.1 1 0.20
obtained if ratio analysis is to be carried out.
Count ratios can be used to overcome the effects of
 The statistics of clast counting 171

Table 12. Ratioanalysisofclast types instratigraphicsections the north, but the second suite including quartzite
of the Sespe Formation at locations 7,9,11and 13 (Fig. 1). may not be multicyclic. It could have been derived
~~~~i~~ Location Location Location from distant eastern sources, but was diluted by the
7' 92 112 133 other suite of clasts.
Granite/quartzite
'GY 2.96 1.09 0.70 0.33
Con. Int. 1.53 0.51 0.31 0.13 V A R I E T A L STUDIES
SXiY 0.96 0.32 0.20 0.11
When counting clasts, it is often possible to recognize
Granite/metavolcanic
l / Y 1.78 1.02 0.31 0.15 distinctive varieties of a particular lithology. For
Con. Int. 1.66 0.86 0.23 0.08 example, a subpopulation of granitoid clasts might be
SX/Y 0.67 0.54 0.15 0.06 subdivided further into monzonitic, syenitic and
graphic varieties, and each counted separately. The
Metavolcanic/quartzite
1.07 2.22 2.19
results will often be small (p < 0 4 - 0 . lo), low precision
X / Y 1.66
Con. Int. 2.18 0.64 1.14 0.41 clast percentages in terms of the entire clast popula-
SXIY 0.88 0.40 0.72 0.33 tion. This problem can be circumvented by counting
'k=3, t=4.30. *k=4, t=3.18. 3k=5, t=2,78. Conn. Int., only clasts of the varieties of interest, and then
confidence interval. expressing varietal abundances as proportions of the
subpopulation rather than the entire population (e.g.
is evident between component percentages and geo- in terms only of granitoid clasts). This approach
graphical location. Thus, his interpretation was a generates larger, more precise clast percentages which
nearby northern provenance for both assemblages, can be handled better statistically. An alternative
the latter assemblage being composed of first cycle approach, which has been used in heavy mineral
detritus derived from basement complex, and the analysis (Doeglas, 1940) but which is perhaps on less
former suite being reworked from older conglomerate. certain statistical ground, is to use ratios to increase
Ratio analysis, however, shows that the proportions the precision of small, low precision component
of granite, gneiss and metavolcanic clasts decrease percentages. In the case of rudites, this would be
westward along the transect relative to durable accomplished by first counting clasts of all types,
quartzite clasts (Fig. 9). An alternative intepretation, establishing the ratio between the subpopulation of
therefore, is that the two assemblages were derived interest and the entire population, and thereafter
from two different sources. The suite including counting only clasts of the subpopulation. For exam-
greywacke may have come from a nearby source to ple, if the proportion of granitoid clasts in the total
suite is determined to be 0.20 based on a count of
perhaps 200 clasts of all types, and then only granitoid
I I I I I I 1 clast varieties are counted until a total of 200 is
reached, this is equivalent to counting 1000 clasts of
.-
0
+ +I
the total assemblage.
0 Table 13 compares volcanic clast varieties in two
oc conglomerate facies of the Sespe Formation at location
11, which correspond to lower and upper depositional
sequences. The ratio of volcanic/non-volcanic clasts
(V/NV) was first established by counting roughly 400
clasts of all types, and then samples of about 100
volcanic clasts were collected from each facies. Each

A A -2
1 2 34 5 6
volcanic clast was subsequently sawed open and
classified as aphanitic (felsite), porphyritic or pyro-
clastic. Porphyritic varieties were further classified on
Site Location the basis of Munsell colour and phenocryst composi-
tion. Clast percentages of individual varieties are two
Fig. 9. Ratio analysis of clast counts in conglomerates along
an E-W transect in the Sespe basin. Site locations are shown or three times as large when expressed only in terms
in Fig. 1 . Clast types: G, granite; N, gneiss; M, metavol- of the subpopulation of volcanic clasts, hence statisti-
canic; Q, quartzite. Data from McCracken (1972). cal comparisons may be made using clast percentages
172 J . L. Howard

Table 13. Varietal analysis of rhyolitic volcanic clast suites Table 14. Tests for significant differences in rhyolitic volcanic
composing the lower and upper depositional sequences of clast suites composing the lower and upper depositional
the Sespe Formation at location 11 (Fig. I). Values are in sequences of the Sespe Formation (Fig. 1).
percent except for ratio V/NV.
Confidence intervals Hypothesis
Depositional sequence Location ( P I -P2) PI = P 2
Lower Upper Pyroclastic rocks
2 0.039<p1-pz <0.241 Reject?
Volcanic Total Volcanic Total
9 0.084 <pl -pz <0.236 Reject?
Clast variety suite suite suite suite
11 0.031 <p1 -p2 <0.229 Reject?
Pyroclastic 13 0.017<p1 -pz <0.163 Reject?
Welded tuff 3.0 1.o 15.0 6.8
Lapilli-ash tuff 3.0 1.o 5.0 2-4 Haematitic felsite and porphyry
Volcanic breccia 3.0 1.o 2.0 1.0 2 -O.OIO<p, -p,<O,O90 Accept
9 - 0 . 0 8 3 < ~ 1-pz<O.123 Accept
Porphyry 11 0.124<p1 -p2 <0.336 Reject*
Hqf 19.0 6.4 3.0 1.4 13 -0~115<p1-~~<0~135 Accept
Hf 10.0 3.3 7.0 2.9
Eqf 13.0 4.3 15.0 6.8 Epidotic felsite and porphyry
Ef 0.0 0.0 11.0 4.8 2 -0.023<pl -p2<0.163 Accept
Gqf 16.0 5.4 8.0 3.4 9 -0.044<p1 -p,<O.124 Accept
Gf 6.0 2.0 18.0 7.0 11 0.039<p1 -p2 <0.261 Reject?
13 0.153<p1 -p2<0.367 Reject?
Felsite
H 6.0 2.0 2.0 1.o Goethitic felsite and porphyry
E 1.o 0.3 3.0 1.4 2 - 0.045 <pl -pz < 0.205 Accept
G 17.0 5.9 8.0 3.4 9 0.043 <pI -pz <0.29 Reject*
11 -0.078<p1 -p,<O.178 Accept
Other 2.0 0.8 2.0 1.o 13 0.274<p1-pz <0.526 Reject*
n 133 - 91 -
- ?Upsection increase.
V/NV 0.34 0.44 -
Effective n - - *Upsection decrease
39 1 207
H, haematitic; E, epidotic; G , goethitic; q, quartz pheno-
crysts; f, feldspar phenocrysts; V/NV, volcanic/non-volcanic
ratio. SUMMARY AND CONCLUSIONS

This paper brings together statistical approaches from

a variety of sources and applies them to the method of
of greater precision. Using the other approach counting clasts using examples from non-marine
discussed above, it can be seen that although only conglomerates in southern California. The same
about 100 volcanic clasts were sawed open, the techniques may be applied to any type of rudaceous
effective sample size is doubled or quadrupled by sedimentary rock, as well as modern gravels. The
using the ratio of volcanic/non-volcanic clasts. main conclusion of this review is that clast counts can
Further analysis of the Sespe Formation at locations be applied much more quantitatively than has been
2, 9 and 13 was carried out using the same methods. done in the past, and in a greater variety of
The combined proportions of pyroclastic varieties, sedimentological contexts. Examples include charac-
and of felsite with porphyry, were used to test the terizing provenance, measuring palaeotransport direc-
hypothesis that the compositions of lower and upper tions, and documenting sediment recycling. Statistical
depositional sequence conglomerates are different. analysis of clast count data also can be used to attack
Using Eq. (3), 95% confidence intervals for the stratigraphic and tectonic problems more effectively.
differences in proportions were obtained (Table 14). In order for clast counts to be applied with statistical
A comparison of these results with the original data rigour, however, certain modifications are needed in
(not shown) suggests that clasts of pyroclastic rocks the field procedures that have been commonly used in
increase upsection at all sites, as does the proportion the past. Point counting, for example should not be
of epidotic porphyry at locations 11 and 13. Thus, used, replicate counts should be made to minimize
varietal studies can also generate quantitative data sampling error, and a total of at least 400 clasts should
useful in documenting changes in clast populations. be counted at any given spot sampled. If groups of
 The statistics of clast counting I73

counts from multiple stratigraphical or geographical In: Conglomerates in Basin Analysis: A Symposium Dedi-
locations are t o be compared, the groups should cated to A.O. Woodford (Ed. by I.P. Colburn, P.L. Abbott
& J.A. Minch), Pacific section, Soc. econ. Paleont. Miner.,
contain counts from a t least six or seven different
62,227-253.
places so that error bars are sufficiently narrow. DAVIS,J.C. (1986) Statistics and Data Analysis in Geology.
Confidence intervals a r e very useful in graphic Wiley, New York, 646 pp.
displays (e.g. orthogonal o r ternary plots), but it is also DIXON,W.J. & MASSEY, F.J. (1957) Introduction to Statistical
very useful t o the reader if the s a m e clast count d a t a Analysis. McGraw-Hill, New York, 488 pp.
DOEGLAS,D.J. (1940) The importance of heavy mineral
are reported i n table form as well. Unfortunately, analysis for regional sedimentary petrology. National Res.
most editors ab h o r publishing t h e s a m e d a t a set in Council Comm. Sed. Rept. 1939-1940, Exhibit G , pp. 102-
two forms i n the s a m e article. I t is hoped th a t this 121. US Government (Nat. Research Council).
paper will encourage not only the routine use of GALEHOUSE, J.S. (1969) Counting grain mounts: Number
elementary statistics when interpreting clast count percentage vs. number frequency. J. sedim. Petrol., 39,
8 12-8 15.
data, but also the application of more sophisticated GALEHOUSE, J.S. (1971) Point counting. In: Procedures in
methods (e.g. factor analysis) i n future basin analysis Sedimentary Petrology (Ed. by R. E. Carver), pp. 385-407.
studies. Wiley Interscience, New York.
GRAHAM, S.A., TOLSON,R.B., DECELLES, P.G., INGERSOLL,
R.V., BARGAR,E., CALDWELL, M., CAVAZZA, W., ED-
WARDS, D.P., FOLLO,M.F., HANDSCHY, J.F., LEMKE, L.,
ACKNOWLEDGMENTS MOXTON,I., RICE, R., SMITH,G.A. & WHITE,J. (1986)
Provenance modelling as a technique for analysing source
terrane evolution and controls on foreland sedimentation.
Thanks t o the following reviewers for providing In: Foreland Basins (Ed. by P. A. Allen & P. Homewood),
helpful an d constructive criticism of the manuscript: Spec. Publ. Int. Ass. Sedimentol., 8,425436.
I. P. Colburn, K.A. Grove, J. A. Harrell, R. D. GRIFFITHS, J.C. (1967) Scientific Method in the Analysis of
LeFever, L. B. Ronca and V. M. Seiders. Thanks also Sediments. McGraw-Hill, New York, 508 pp.
HAUGHTON, P.D.W., TODD,S.P. & MORTON,A.C. (1991)
t o Elaine Shelton for helping type the manuscript. Sedimentary provenance studies. In : Developments in
Sedimentary Provenance Studies (Ed. by A.C. Morton, S.P.
Todd & P.D.W. Haughton), Spec. Publ. geol. SOC.Lond.,
57,l-11.
REFERENCES HOWARD,J.L. (1987) Paleoenvironments, provenance and
tectonic implications of the Sespe Formation, southern
ABBOTT,P.L. & PETERSON, G.L. (1978) Effects of abrasion California. PhD thesis, University of California, Santa
durability on conglomerate clast populations-examples Barbara, Santa Barbara, CA, 306 pp.
from Cretaceous and Eocene conglomerates of the San HOWARD, J.L. (1989) Conglomerate clast populations of the
Diego area, California: J. sedim. Petrol., 48,3142. Upper Paleogene Sespe Formation, southern California.
AITCHISON, J. (1984) The statistical analysis of geochemical In : Conglomerates in Basin Analysis: A Symposium Dedi-
compositions. J l n t . Ass. Math. Geof.,16, 531-564. cated to A.O. Woodford (Ed. by I.P. Colburn, P.L. Abbott
BARTLING, W.A. & ABBOTT,P.L. (1983) Upper Cretaceous & J.A. Minch), Pacific section, Soc. econ. Paleont. Miner.,
sedimentation and tectonics with special reference to the 62,269-280.
Eocene, San Miguel Island, and San Diego area, Califor- KELLEY,J.C. (1971) Mathematical analysis of point count
nia. In : Cenozoic Marine Sedimentation Pacific Margin data. In : Procedures in Sedimentary Petrology (Ed. by R.E.
U.S.A.(Ed. by D. K. Larue & R. J. Steel), Pacific section, Carver), pp. 409-425, Wiley, New York.
SOC.econ. Paleont. Miner., 133-150. KOCH,G.S., JR. & LINK,R.F. (1971) Statistical Analysis of
BLA-IT, H. (1992) Sedimentary Petrology. Freeman, New Geological Data, vol. 11. Wiley, New York, 438 pp,
York, 514 pp. KRUMBEIN, W.C. (1942) Flood deposits of Arroyo Seco, Los
BELYEA,'R.R. & MINCH, J.A. (1989) Stratigraphy and Angeles County, California. Bull. geol. Soc. Am., 53,1355-
depositional environments of the Sespe Formation, north- 1402.
ern Santa Ana Mountains, California. In : Conglomerates KRUMBEIN, W.C. & GRAYBILL, F.A. (1965) An Introduction
in Basin Analysis :a Symposium Dedicated to A.O. Woodford to Statistical Models in Geology. McGraw-Hill, New York,
(Ed. by I.P. Colburn, P.L. Abbott & J.A. Minch), Pacific 475 pp.
section, Soc. econ. Paleont. Miner., 62,28 1-300. KRUMBEIN, W.C. & MILLER,R.L. (1953) Design of experi-
BRADLEY, W.C. (1970) Effect of weathering on abrasion of ments for statistical analysis of geological data. J . Geol.,
granitic gravel, Colorado River (Texas). Bull. geol. SOC. 61,510-532.
Am., 81,61-80. KRUMBEIN, W.C. & PETTIJOHN,F.J. (1938) Manual of
CHAYES,F. (1956) Petrographic Modal Analysis. Wiley, New Sedimentary Petrography. Appleton-Century-Crofts, New
York, 113 pp. York, 549 pp.
COLBURN, I.P. & NOVAK,G.A. (1989) Paleocene conglom- KRUMBEIN, W.C. & RASMUSSEN, W.C. (1941) The probable
erates of the Santa Monica Mountains, California: error of sampling beach sand for heavy mineral analysis.
Petrology, stratigraphy and environments of deposition. J. sedim. Petrol., 11, 10-20.
174 J . L. Howard

MCCRACKEN, W.A. (1972) Paleocurrents and petrology of PLUMLEY, W.J. (1948) Black Hills terrace gravels: a study in
Sespe sandstones and conglomerates, Ventura basin, Califor- sediment transport. J. Geol., 56, 526-577.
nia. PhD thesis, Stanford University, Stanford, CA, POTTER,P.E. & PETTIJOHN, F.J. (1977) Paleocurrents and
192 pp. Basin Analysis. Academic Press, New York, 296 pp.
MIALL,A.D. (1970) Devonian alluvial fans, Prince of Wales SCHOELLHAMER, J.E., VEDDER,J.G., YERKES,R.F. &
Island, Arctic Canada. J. sedim. Petrol., 40,556-571. KINNEY, D.M. (1981) Geology of the northern Santa Ana
MINCH,J.A., GIBSON,K.N. & PETERSON, G.L. (1976) Clast Mountains, California.Prof. Pap. US geol. Surv., 420-D,
populations in Sespe and Poway conglomerates and their 107 pp.
possible bearing on the tectonics of the southern California SEIDERS, V.M. & Cox, B.F. (1992) Place of origin of the
borderland. In : Aspects of the Geologic History of the Salinian block, California, as based on clast compositions
California Continental Borderland (Ed. by D.G. Howell), of Upper Cretaceous and lower Tertiary conglomerates.
Am. Ass. petrol. Geol. Misc. Publ., 26, 256-265. Prof. Pap. USgeol. Surv., 1526,80 pp.
MINIUM,E.W. & CLARKE, R.B. (1982) Elements of Statistical SNEED,E.M. & FOLK, R.L. (1958) Pebbles in the lower
Reasoning. Wiley, New York, 400 pp. Colorado River, Texas-a study in particle morphogene-
NATRELLA, M.G. (1966) Experimental Statistics. National sis. J. Geol., 66, 114-150.
Bureau of Standards Handbook No. 91, US Government VANDER PLAS, L. (1962) Preliminary note on the granulo-
Printing Office, Washington, DC, 355 pp. metric analysis of sedimentary rocks. Sedimentology, 1,
PETERSEN,R.G. & CALVIN,L.D. (1965) Sampling. In: 145-157.
Methods of Soil Analysis I (Ed. by B.A. Black), pp. 54-71. VANDER PLAS,L. & TOBI,A.C. (1965) A chart for judging
Agronomy 9, American Society of Agronomy, Madison, the reliability of point counting results. Am. J. Sci., 263,
WI. 87-90.

(Manuscript received 27 May 1992; revision received I2 September 1992)