Geochemical Journal, Vol. 21, pp.

75 to 84, 1987

NOTE

The need for standardization of normalized multi-element

diagrams in geochemistry: a comment

N.M.S.Rocx

Department of Geology, University of Western Australia, Nedlands 6009, WA, Australia

(Received January 6, 1987; Accepted May 28 , 1987)

A potpourri of mutually inconsistent normalized multi-element diagrams ('spider-diagrams') has been

used in geochemistry, on the basis of different choices of: (i) element order; (ii) normalizing materials
(chondrite, `primitive' mantle, MORB, composite shale, etc.); (iii) normalizing values (many incompatible
sets are used for conventional chondrite-normalized rare-earth element (REE) plots, which generate
`enrichments' varying by as much as 65 % from a single rock , and Eu anomalies of quite divergent strengths).
Standardization is urgently required, not least because (1) many published diagrams are unnecessarily
difficult to compare; (2) different element orders can drastically affect patterns, or lead to artificial
'pseudo -anomalies' lacking petrogenetic significance. One type of chondrite and MORB-normalized
spider-diagram (each based on one set of normalizing values) should suffice to represent most igneous rocks.
Arguments favouring the chondrite-normalizing REE values of Nakamura (1974), the chondrite-normalized
19-element plot of Thompson (1982), and the MORB-normalized 15-element plot of Pearce (1982, 1983) are
preferred, hopefully as an impetus to wider discussion. Consistently ordered sub-sets can be chosen from
either multi-element plot if fewer elements are in fact analysed. A FORTRAN77 program has been written to
realize these suggestions.

appreciate the current position: included therein

THE PROFUSION AND INCONSISTENCY OF PUBLISHED
are over 300 spidergrams arranged in no fewer
NORMALIZED DIAGRAMS
than 130 spider-diagrams of 19 different types!
Normalized log-linear multi-element dia A necessarily cursory examination of other
recent literature suggests this is merely the tip of
grams (Figs. 1-3) are becoming increasingly
widespread, abundant and useful in the petrolo the iceberg, and that a veritable galaxy of
spider-diagram varieties is now actively pub
gical and geochemical literature. The terms
spidergram, for a normalized plot of one rock or lished by different authors.
mineral sample, and spider-diagram, for a series The parameters causing these perplexing
of such plots on one figure (Thompson, 1982), variations are as follows:
will be employed here for brevity. Thompson et (1) Choice of normalizing material: In igneous
al. (in Hawkesworth and Norry, 1983) summa petrology and geochemistry, 3 normalizing
rised the pre-eminence of spider-diagramsin the materials are most popular: (i) chondrites; (ii)
`primitive' or `primordial' mantle; and (iii)
literature with the aphorism `arachnids rule
OK', but the present author wonders whether an MORB. Sedimentary and metamorphic
alternative anagram-`anarchids rule OK' geochemists also use N.American composite
might not be a closer approximation to the shale, seawater, average crust, and other norma
truth, in view of the plethora of mutually lizing materials.
inconsistent diagrams now being promulgated. (2) Choice of normalizing values: Even diagrams
One need look no further than invaluable normalized to purportedly the same material use
recent books by Hawkesworth and Norry (1983) a bewildering range of normalizingvalues. Table
and Menzies and Hawkesworth (1987), to 1 compiles some commonly used sets of values

75
 76 N.M.S.Rock

400
300 NORMALIZING VALUES
A& Schmitt et al.(1964)
200 Haskin et al.(1 968),
Nakamura(1974)

Masuda et al.(1973)
100
80 Evensen et al.(1978)
60

40
30

20 Eu anomaly virtually absent

w
w
ir
10
8
6

4
La Ce Pr Nd Sm Eu Gd Tb DU Ho Er Tm Yb Lu

Fig. 1. Illustrating the influence of normalizing values on enrichments and Eu anomalizes for REE
spidergrams. The same rock, a type monchiquite from Serra de Monchique (Portugal), is plotted versus five of
the seven different sets of chondrite-normalizing values from Table 1. At this scale, the patterns resulting from the
sets of Haskin et al. (1968) and Nakamura (1974) are indistinguishable. The patterns derived from the sets of
Schnetzler & Bottino (1971) and Taylor and Gorton (1977) are omitted for clarity, both lying within the range
shown; moreover, the former is necessarily exactly parallel to that derived from the Masuda et al. (1973) set. Raw
data-values (in ppm) for the monchiquite are as follows: La 103; Ce 253; Pr 27.0; Nd 110; Sm 14.2; Eu 2.0; Gd
7.7; Tb 1.1; Dy 5.9; Ho 0.91; Er 1.5; Tm 0.22; Yb 0.9; Lu 0.15. Produced using SPIDER (Rock, 1987c).

for chondrite-normalized REE plots. No less whence this peculiar 'golden ratio' derives is not
than six of these sets appear in different papers explained!
within Menzies and Hawkesworth (1987)! Some (3) Choice of element order: Only in the case of
individuallauthors use inconsistent normalizing REE plots is the element order pre-determined
valueseven within their own opera (cf. plots 4 and (by atomic number). For other spider-diagrams,
5, Table 2), while others do not state the sourceof element order is necessarily more arbitrary,
their normalizing values at all (e.g., plots 7-8, since elements of adjacent atomic number (e.g.,
Table 2). What might euphemisticallybe termed Rb, Sr, Y, Zr) are now chemicallytoo dissimilar
`numerical adjustments' introduce even more to plot contiguously. As with so many other
confusion-for example, Erlank et al. (in Men arbitrary orders in geochemistry (e.g., that for
zies and Hawkesworth, 1987, p.283) state they Si02, A1203 etc. in analyses), a plethora of
have used Taylor and Gorton's (1977) REE mutually inconsistent personal choices has
chondrite values divided by 1.2, but Taylor and arisen. Indeed, few authors cited in Table 2
Gorton (1977) already use the Leedey chondrite justify their chosen order at all, and some
values of Masuda et al. (1973) divided by 1.2 authors even state they have adopted the order
(see Table 1). It thus seems that Erlank et al. are of others, but in fact change it in arcane and
using Masuda's values divided by 1.44 but unexplained ways (cf. plots 1 and 2, Table 2).
 Standardization of normalized multi-element diagrams 77

Table 1. Examples of varying chondrite-normalizing values used for published REE plots

Source Schmittet Haskin et Schnetzler Masuda et Taylor & Nakamura Evensen et al. (1978)") Robust Dominant
al. (1964) al. (1968)") & Bottino al. (1973)") Gorton (1974)") estimatesd) cluster
mode e)
(1971)") (1977)")

La .30 .33 .378 .315 .329 .2446 .2446 .319-.330 .329

Cc .84 .88 .934 .976 .813 .865 .6379 .7272 .83-.89 .859
Pr .12 .112 .116 .0964 .1000
Nd .58 .60 .683 .716 .597 .63 .4738 .4716 .61-.64 .65
Sm .21 .181 .218 .23 .192 .203 .1540 .1640 .200-.211 .217
Eu .074 .069 .077 .087 .072 .077 .0580 .0597 .073-.079 .078
Gd .32 .249 .308 .311 .259 .276 .2043 .2453 .264-275 .271
Tb .049 .047 .049 .375b) .0390
Dy .31 .366 .39 .325 .343 .2541 .2478 .328-.344 .334
Ho .073 .07 .073 .0567 .0590
Er .21 .2 .235 .255 .213 .225 .1660 .1736 .218-228 .216
Tm .033 .03 .03 .0256 .0267
Yb .17 .2 .235 .249 .208 .22 .1651 .1487 .216-.226 .214
Lu .031 .034 .040 .0387 .032 .0339 .0254 .0262 .0349-0356 .0389
Method`) INAA INAA ID ID (ID) ID ID+INAA ID ID ID
Notes Accuracy Composite Composite Leedey Previous Average Statistical Stat. av. Range of Single
low of 10 of ?10 chondrite column/1.2 of 10 CI average of 56f) 18 averages average
alone of 56+22 of 56
(-10
av. of 17

a) These sets all used in different chapters of Hawkesworth and Norry (1983) or Menzies and Hawkesworth
(1987).
b) This value should read 0.03745.
c) INAA = instrumental neutron activation analysis; ID = isotope dilution.
d) Range for 18 different robust estimates of location (averages): 4 of Hampel's 3-part redescending series; 3 of
Huber's one-step series; Andrew's sine estimate; one adaptive estimate; the 10 %, 20 %, 25 % and 50 %
trimmed means; the Shorth mean; John's estimate; the Gastwirth median; the Trimean; and a
multiply-skipped estimate (Andrews et al. 1972). Uses the same database of 56 ID chondrite analyses as
Evensen et al. (1978).
e) Ellis et al. (1977); an iterative robust estimate which rejects data at successivelyfewer standard deviations
from the mean until the estimatestabilises. Closelyanalogous to the statisticalprocedure used by Evensen et
al. (1978).
f) NAA values in table 3 of Evensen et al. (1978) multiplied by 0.2446, i. e., same factor as used by these authors
themselves to derive "CI average" (previous column).

The extraordinarily inconsistent element elements (e.g., XRF data alone) could be
orders in Table 2 are in no way `excused' by the consistent with those for more comprehensive
3 different normalizing materials. If all these (e.g., XRF+INAA) data, no plot in Table 2 in
orders reflected some objectivephysico-chemical fact appears to be an exact sub-set of any other.
parameter measurable on at least an ordinal (i.e. On the contrary, many orders are totally irre
rankable) scale, they should be immutable concilable.
under varying normalization. In fact, several
element pairs appear reversed on different plots
THE NEED FOR DIAGRAM CONSISTENCY IN
(see for example Ba,Rb and K,Nb in Table 2 PETROGENETIC INTERPRETATION
where, in each case, two plots differ from all the
others); such choices are thus revealed as These variations are unimportant only if
entirely arbitrary. Although it would be reason spider-diagrams are used merely for comparing
able to plot subsets of some omnibus order, so those rocks or minerals plotted within the
that diagrams for smaller numbers of analysed diagram (or within a fully consistent set of
 78 N.M.S. Rock

diagrams). Spider-diagrams of whatever kind artificial, can therefore easily arise. Yet even if
can then be a highly effective means of conde readers are alerted to the possibility of
nsing and displaying multivariate chemical data apochryphal differences, recognising them is
in two dimensions, and this author is by no often impeded by authors' reluctance to state
means averse to spidergrams in principle (e.g., their normalizing values clearly. At best, these
Rock, 1986, 1987a,b). However, variations be vital values will be hidden in the `small print' of
come not only more confusing and frustrating, the respective figure captions; more likely, they
but actively hinder genetic interpretation, in will be given in neither of the actual captions
circumstances outlined below: being compared, but hidden in the captions of
Comparison of different plots. It becomes earlier REE plots in the same respective papers;
difficult or impossible to compare different plots at worst, normalizing values will not be given at
if different normalizing materials, values or all in one or other paper (e.g., plot 8 in Table 2),
element orders are used. Fig. 1 shows how altogether preventing meaningful comparisons.
different normalizing values from Table 1 affect A universal, consistent set of normalizing values
REE patterns. Although the overall pattern for on all plots is thus needed, which will be
this rock is not fundamentally altered, normal understood implicitly, and thus need no longer
ized enrichments vary by as much as 65 % be explicitly quoted.
between the different plots; this variation ex (ii) Effect of elementorder: Nb anomalies, as on
ceeds that for entire igneous rock suites in some Fig. 2a, are statistically quite different from the
published spider-diagrams! Put another way, a often quantitatively interpretable Eu anomalies
rock may appear `more enriched' on one REE on REE plots. Whereas the X-axis on REE plots
plot than another, even if its raw REE abund is a higher-ordered metric variable (Coxon,
ances are as much as 65 % less, if the two plots 1982)-i.e. Xl < X2 <...X15 and {X„-X„-1} is
use different sets from Table 1. constant, with the element symbols effectively
Significanceof element anomalies. Anomaly proxying for ranked integers (atomic numbers)
recognition is an increasinglywidelyused techni from 57 to 71-the X-axis on other types of
que in the nascent `science' of spidergram spider-diagrams is only a nominal variable-i.e.
interpretation. Probably the most widely recog one can no longer say X,< Xi but only that that
nised are Eu (Fig. 1) and Nb anomalies (e.g., X, ~ Xj, since atomic numbers are no longer in
Fig. 2a)-the latter supposedly a hallmark of any logical order (Fig. 2). It is thus possible to
subduction-derived magmas (Pearce, 1982, generate (or suppress) a `negative Nb anomaly'
1983). However, 'pseudo-anomalies' can also be merely by judicious choice of element order
generated by purely artificial (graphical) rather (Fig. 2). Once again, some orders (e.g., 2 and 4
than real magmatic effects: in Table 2) will be more prone to generate
(i) Effect of excessivelyhigh or low normalizing negative Nb anomalies than others (e.g., 3 and
values. This applies to all types of spidergram, 12), from an analysis of the type displayed in
including REE plots. For example, the chon Fig. 2, because Nb appears much earlier in the
drite values of Schmitt et al. (1964) are intrinsi sequence, among the elements of higher normal
cally less prone to generating negative Eu ized tenure.
anomalies, because Sm/Eu (2.84) and Gd/Eu Significance of overall spidergram shapes:
(4.32) are significantlyhigher than in the other Pearce (1983) devised an empirical scheme for
sets in Table 1: in Fig. 1, the Eu anomaly evident `dissecting' spider -diagrams into their compo
in three patterns has all but disappeared in the nent parts. He inferred for example that rocks
fourth, Schmitt et al. pattern. Clear, graphical yielding spidergrams similar to Fig. 3 contained
differences, of a type normally imbued with 3 source components: a mantle component (A),
petrogenetic significance but in actuality purely a sublithospheric component (B), and a subduc
 Standardization of normalized multi-element diagrams 79

(a) (b)
C
a)

E CD
CD

CD
N

E
L0
Z

Rb Ba Nb K Sr P Zr Ti Y Sr K Rb Ba Nb P Zr Ti Y
37 56 41 19 38 15 40 22 39 (Atomic numbers)

Fig. 2. Illustrating the major influenceof elementorder on the overallshape of a spidergram. (a) and (b) use the
same normalized value for every element, but contrasting orders followingplots 3 and 4 from Table2. (The two
plots use identical subsets exceptfor Ca, which is here omitted). Note particularly than an Nb anomaly appears
only in (a). Atomic numbers are added to show the lack of any logical sequence.

tion zone component (Q. This scheme has been try to establish standards which, though not
widely imitated (e.g., Macdonald et al., 1985). absolutely `perfect', can nevertheless be
However, the above remarks concerning accepted and employed consistently by all. The
anomaly interpretation apply with equal force remainder of this note is intended as both
here. For when some elements appear at such impetus and exhortation towards such progress.
disparate positions in different sequences (e.g., The present section outlines objective reasons
Sr, varying from 1st to 8th position in Table 2) for preferring particular plots or normalizing
consistency of spidergram shapes is inconceiv parameters in Tables 1-2 to others, while the
able: it is still possible to move elements final section proffers a more pragmatic
between the various inferred components by approach, via computerization.
juggling the sequence. It is proposed that four basic types of
spidergrams should be sufficient for most pur
poses: (i) chondrite-normalized REE (normaliz
CRITERIA FOR PREFERRING PARTICULAR ing values of Haskin et al., 1968and Nakamura,
NORMALIZED PLOTS
1974); (ii) composite shale-normalized REE
The irrationalities and inconsistencies out (normalizingvalues of Gromet et al., 1984);(iii)
lined above seem to ally spider-diagram inter MORB-normalized multi-element (plot 12 in
pretation more with alchemy than with modern Table 2); (iv) chondrite-normalized multi
geochemistry. Indeed, the situation strongly element (plot 1). These is no need to justify
resembles the imbroglio in geochronology, prior normalized REE plots (i) and (ii) per se, as they
to Steiger and Jaeger's (1977) highly effective have been the most widely-used form of spider
rationalization of decay constants. Similar re gram since the pioneering Japanese REE work
commendations are badly needed in geochemis of the 1960's. However, the rationale behind the
 N.M.S. Rock
80

Table 2. Examples of the plethora of multi-element 'spider-diagrams' in two recent textbooks

Source of diagram' Element order Normalizing values from

Chondrite-normalized spider-diagrams
1) Thompson et al., in H & N, Ba,Rb,Th, K, Nb, Ta,La,Ce,Sr,Nd,P,Sm,Zr,Hf,Ti,Tb,Y,Tm,Yb Thompson (1982)
p.168
2) Lloyd in M & H, p.115 Rb,Ba,Nb,La,Sr,Zr,Y,Cr,Ni Thompson (1982)
Mantle-normalizedspider-diagrams
3) Dawson in M & H, p.134 Ca,Sr,K, Rb,Ba,Nb, P,Zr,Ti,Y Jagoutz et al. (1979),
Wood et al. (1979)
4) Erlank et al., in M & H, p.281 Rb,Ba,Nb, K, Sr,P,Zr,Ti,Y Rogers (unpubl.thesis)
5) Harte et al., in M & H, p.183 Sr,Rb,Ba,Ta+REE Jagoutz et al. (1979),
Nakamura (1974)
6) Harte in H & N, p.60 Cs,Rb,Ba,K,Sr+ REE Wood et a!. (1979)
7) Menzies et al. in M & H, p.330 Rb,Ba,Th,Ta,K,La,Ce,Sr,Nd,P,Sm,Eu,Hf,Ti,Tb,Y,Yb,Lu Values quoted but source
not given
8) Hawkesworth et al. in M & H, Rb,Ba,Th, K, Nb, La,Ce, Sr,Nd,P,Zr,Ti,Y Not stated`)
p.375
9) Le Roex in M & H, p.398 Ba,Rb, K, Nb, La,Sr,Zr,P,Ti,Y,Yb,Sc, V ,Fe,Cr,Co,Ni Sun and Nesbitt (1977)
10) Weaver and Tarney in H & N, Rb,Ba,Th, K, Nb, La,Ce, Sr,Nd,P,Hf,Zr, Sm,Ti,Tb,Yb Wood et al. (1981)
p.211
11) Thirlwall and Jones in H & N, Rb,Ba,Th,Nb, K, La,Ce,Sr,Nd/P,Zr/Sm,Ti/Eu,Gd,Dy,Y,Er,Yb Sun (1980)
p.188
MORB-normalized spider-diagrams
12) Pearce in H & N, p.232 Sr,K, Rb,Ba,Th,Ta,Nb, Ce,P,Zr,Hf,Sm,Ti,Y,Yb Pearce (1982)0

a) M & H = Menzies and Hawkesworth (1987); H & N = Hawkesworth and Norry (1983).
b) A subset of this diagram is also used by Cox in Hawkesworth and Norry (1983, p.144).
c) Also used by Norry and Fitton in Hawkesworth and Norry (1983, p.7); source of normalizing values again
not stated.
Italics and bold type merely emphasise inconsistencies between element orders.

other recommendations now follows. Schmitt et al. (1964)constitute pioneering analy

tical work, but are now largely of historical
Complementary use of MORB-, chondrite and interest (Evensen et al., 1978) as their accuracy
shale-normalized spidergrams (=10 %) can now be bettered; Fig. 1 has already
Use of all 3 types is justifiable because they shown how these values tend to generate diffe
respectively serve to compare geological mate rent patterns from the other sets in Table 1.
rials with `the most abundant volcanic rock', Those of Schnetzler and Bottino (1971) can also
`mantle source' and `representative crustal mate be excluded as the least complete, the lack of a
rial', and are therefore complementary. Fur value for La being particularly deleterious;
thermore, as only one set of shale-normalizing furthermore, they are partly based on the more
values has been widely used, it requires no complete set of Haskin et al. (1968). Those of
further justification. Masuda et al. (1973, p.240) were based on
"every consideration", claimed as suggestingthe
Choice of chondrite-normalizing values Leedey chondrite to be the "safest" normalizing
The `best' set of chondrite-normalizing REE material, but these `considerations' were not
values (from the medley in Table 1) does explained. In view of the notorious variability
however require careful consideration. These and inhomogeneity of chondrites, it is difficult to
sets are based on varied and largely irreconcil choose a single example as `typical', and both
able criteria: some represent analyses of single this set of values and that of Taylor and Gorton
chondrites, some averages of several chondrites, (1977-derived simply from them) are surely
and others mixtures of chondrites. Those of therefore less reproducible than the remaining 3
 Standardization of normalized multi-element diagrams 81

sets, which represent averages of several or location with a wide range of real geological
many chondrites. data-sets (Lister, 1985, 1986; Rock, 1987d).
At first sight, the set of Evensen et al. (1978) Like hierarchical cluster analysis, they are prone
would seem the optimum choice, since these to unstable solutions, which may just as easily
authors compile a large database of 22 INAA home in on atypical as typical results, early on in
and 56 ID chondrite analyses; clearly justify the iteration. The present, highly skewed and
their choice of compiled analyses; amply allow outlier-ridden, REE data appear to have pro
for the different analytical methods in Table 1; duced just such a result, for computer recalcula
and make the only serious attempt so far at a tion (Rock, 1987e), using the original database
rigorous statistical estimation, by censoring of Evensen et al. (1978), confirms that the DCM
erroneous and atypical analyses in order to is discordant with 18 other robust estimates of
generate robust averages. Unfortunately, three location for Nd, Sm, Er, Yb and Lu (Table 1).
other factors undermine the value of their With Lu, for example, the DCM lies not only
recommendations: well outside the narrow range for the 18 robust
(i) The Tb value is given as 0.3745 instead of estimates, but even outside 95 % confidence
0.03745; even if this is merely a typographic limits for both the arithmetic mean and median.
error, it is most unfortunate in a set of standard The consistency of the other 18 estimates for all
values, and begs the question of whether other REE (Table 1), which arises despite their use of
errors have been overlooked. a wide diversity of censoring procedures, is
(ii) Their calculations involve several `numerical evidence in itself of their greater reliability.
adjustments', such as 1.0412 to `reconcile' the A more detailed description of these robust
ID and INAA data, and 0.2446 to produce a procedures is presented by Rock (1987d) and
least-squares best fit to CI chondrites. As the Rock et al. (1987), being out of place here, but
final figures are significantly lower than all Table 1 suggests that Nakamura's (1974) norma
others in Table 1, these factors may well need lizing values are the most reasonable, from a
reassessing. Furthermore, least-squares is an purely statistical viewpoint. Since Nakamura's
inappropriate statistical technique here because values are also favoured by the geological and
it assumes data normality, whereas these data analytical elimination process above, there are
distributions are not only very strongly skewed good overall grounds for preferring them.
and leptokurtic (Vbl skew and b2 kurtosis up to Although Nakamura's set does lack values for 4
+4.9 and +29.6 respectively), but also include REE, these can be interpolated from the set of
gross statistical outliers (Rock, 1987d). Haskin et al. (1968), these two sets again being
(iii)They chose an iterative robust statistical almost identical for the remaining 11 (Fig. 1;
technique, similar to the Dominant Cluster Table 1). Nakamura's set also has the compen
Mode (DCM) of Ellis et al. (1977): after sating major advantage that it is used in Thomp
calculation of an initial arithmetic mean, results son's (1982) multi-element plot; vital consisten
k standard deviations from the mean were cy between these two recommended plots is
eliminated and the process repeated with de therefore achieved.
creasing k at each iteration, until the estimate
stabilised. This is an equally unfortunate choice, Choice of chondrite rather than mantle
for DCM-type estimates are known from de normalization for multi-element spidergrams
tailed Monte Carlo studies to be much less Chondrite-normalization and mantle
efficient than, for example, Hampel's robust normalization serve effectively the same pur
location estimates (Andrews et al., 1972; Ham pose (i.e. to compare with `mantle source
pel et al., 1986). Furthermore, DCMs in ,practice material'); they are competitive, not com
produce demonstrably unreliable estimates of plementary. Chondrite-normalization is prefer
 82 N.M.S. Rock

red for 4 reasons: (a) chondrite-normalizedplots reason, Rb/Ba have become inverted and Cr and
have precedence and are probably more numer Ni added). Thompson's plot can thus be recom
ous in existing literature; (b) REE plots are mended without further discussion.
almost universally chondrite-normalized, and Earlier arguments also support use of a
consistency is thereby maintained between these MORB (rather than mantle-normalized) plot to
two types of plot-vital when nearly all multi complement Thompson's. It is therefore fortun
element plots also include REE (Table 2); (c) ate that the only so far widely-used MORB
the plethora of mantle-normalized plots is a normalized form (no.12 in Table 2) is also the
symptom of their imprecise, subjective most objectively based (Pearce, 1982, 1983):it is
rationale; (d) agreement on the one widely-used ordered on the basis of elemental ionic poten
type of chondrite-normalized plot (Thompson, tials and lherzolite/melt distribution coefficients,
1982) is more feasible than on one of the such that the incompatibilities of both mobile
mutually inconsistent mantle-normalized plots. and immobile elements increase towards the
centre of the pattern (Fig. 3). This is surely
further justification (if any were needed) for
preferring Pearce's plot.
k Of course, the compatibilities and mobilities
c .
of chemical elements are not as immutable as
a) their atomic numbers: for example, Zr and Nb
E
a) .
(the archetypal incompatible, immobile ele
a)
ments) can become compatible, mobile ele
N ments in alkaline, halide-rich media (e.g., agpai
O tic syenite magmas) which favour transition
E metal complexing and crystallisationof complex
LO
Z Zr or Nb minerals (eudialyte, pyrochlore, etc.).
A The characteristic Nb `anomalies' on the spider
Sr K Rb Ba Th Ta Nb Ce P Zr Hf Sm Y Ti Yb grams of destructive plate-margin igneous rocks
MOBILE IMMOBILE appear precisely because Nb has a different
.4-4 Increasing incompatibility <-< compatibility and mobility in their precursor
magmas relative to the tholeiitic normalizing
material (MORB); that is, Nb is in the `wrong'
Fig. 3. Example of Pearce's (1982, 1983) MORB
normalized spidergram, illustrating empirical pet position for such rocks. However, to change
rogenetic interpretation from spidergram shape (after element order with different rock-types would
Macdonald et al., 1985). Such a shape is considered to be to annul the entire logic of normalized plots:
comprise 3 components A, B and C derived from they only have comparative value when the same
different sources. The elemental constitution of A, B order is used for all types.
and C is clearly dependent on displayed element order.
Pearce's order was chosen on a quite diffe
rent basis to Thompson's in Table 2: the latter
Choice of element order on multi-element spider gives "the smoothest overall fit to data for
grams Icelandic lavas and North Atlantic ocean-floor
All the above arguments lead to preference basalts" (Thompson, 1982, p.105). Although
for a chondrite-normalized multi-element plot. these two recommended orders consequently
The two such plots in Table 2 both use Thomp bear little relation to one another (Table 2, nos.
son's (1982) normalizing values, and Lloyd's 1 and 12), they are based on two quite different
element order is stated to be a sub-set of normalizing materials, and should not therefore
Thompson's (although for some unstated be regarded as incompatible but as com
 Standardization of normalized multi-element diagrams
83

plementary, illuminating quite different aspects HUBER, P. J., ROGERS, W. H. and TUKEY, J. W.
of the rocks they are used to depict. (1972) Robust Estimates of Location. Princeton
University Press: New Jersey.
COXON, A. P. (1982) The User's Guide to Multi
COMPUTER IMPLEMENTATION dimensional Scaling. Heinemann: London.
ELLIS, P. J., COPELOWITZ,I. and STEELE, T.W.
It is one thing to suggest recommendations (1977) Estimation of the mean by the dominant
such as the above, and quite another to achieve cluster method. Geostand. Newsl. 1, 123-130.
standardization in practice. A computer prog EVENSEN, N. M., HAMILTON, P. J. and ONIONS, R. K.

ram, which can plot consistent spider-diagrams (1978) Rare-earth abundance in chondritic
meteorites. Geochim. Cosmochim. Acta 42, 1199
quickly and easily, is a hitherto little-explored 1212.
but powerful means to this end. Such a program GROMET, L. P., DYMEK, R. F., HASKIN, L. A. and
can effect standardization by itself, since many KOROTEV, R. L. (1984) The "North American
geologists may be prepared to trade a minor shale composite": its compilation, major and trace
element of personal choice for its far greater element characteristics. Geochim. Cosmochim. Acta
work-saving powers. This course has already 48, 2469-2482.
HAMPEL,F. R., ROUSSEEUW,P. J., RONCHETTI,E. M.
been taken with amphibole nomenclature: a and STAHEL,W. A. (1986) Robust Statistics: the
program to output International Mineralogical Approach Based on Influence Functions Wiley: New
Association-recommended names (Rock and York, 502p.
Leake, 1984) has already been implemented HASKIN, L. A., HASKIN, M. A., FREY, F. A. and
worldwide, and now appears to be catalyzing WILDERMAN,T. R. (1968) Relative and absolute
terrestrial abundance of the rare-earths. Origin and
rationalization in the literature (e.g., Horak and Distribution of the Elements, 889-912. Edited by L.
Gibbons, 1986). H. Ahrens, Pergamon: New York.
Another program (SPIDER) has therefore HAWKESWORTH,C. J. and NORRY, M. J. (1983)
been written in standard FORTRAN77, to Continental Basalts and Mantle Xenoliths. Shiva:
Orpington.
produce the four basic types of spider-diagram HORAK, J. M. and GIBBONS, W. (1986)
advocated above. SPIDER is documented else
Reclassification of blueschist amphiboles from
where (Rock, 1987c), and is already in use at Anglesey, North Wales. Mineral. Mag. 50, 533-535.
several UK institutions. It is sufficientlyflexible JAGOUTZ, E., PALME, E., BADDENHEUSEN, H.,
to generate subsets of each spider-diagram type BLUM,K., CENDALES,M., DREIBUS,G., SPETTEL,
where only a subset of elements have been B., LORENZ, V. and WANKE, H. (1979) The
analysed. It can even generate user-defined abundances of major, minor and trace elements in the
earth's mantle as derived from primitive ultramafic
spider-diagrams for any whom the present re nodules. Proc. 10th Lunar Planet. Sci. Conf. 2,
commendations do not convince. It has been 2031-2050.
made as machine and system-independent as LISTER, B. (1985) Looking at analytical data. Geo
possible, although, of course, portability for stand. Newsl. 9, 263-274.
LISTER, B. (1986) Best estimates from interlabora
graphics programs is at present strictly con tory data. Anal. Chim. Acta 186, 325-329.
strained.
MACDONALD,R., THORPE, R. S., GASKARTH,J. W.
Should geochemists ever achieve consensus and GRINDROD,A. R. (1985) Multi-component
that another set of normalized plots is preferable origin of Caledonian lamprophyres of northern Eng
to those advocated here, adjustments to SPID land. Mineral. Mag. 49, 485-494.
ER would be easily made. The author would MASUDA, M., NAKAMURA, N. and TANAKA, K.
therefore welcome comments on the subject. (1973) Fine structure of mutually normalized rare
earth patterns of chondrites. Geochim. Cosmochim.
Acta. 37, 239-248.
REFERENCES MENZIES, M. A. and HAWKESWORTH, C. J.
(1987) Mantle Metasomatism. Academic Press:
ANDREWS, R. W., BICKEL, P. J., HAMPEL, F. R., London.
 84 N.M.S.Rock

NAKAMURA,N. (1974) Determination of REE, Ba, posal. Isot. Geosci. in press.

Mg, Na and K in carbonaceous and ordinary chon SCHMITT, R. A., SMITH, R. H. and OHEHY, D. A.
drites. Geochim. Cosmochim. Acta. 38, 757-775. (1964) Rare-earth, Y and Sc abundances in meter
PEARCE,J. A. (1982) Trace element characteristics oritic and terrestrial matter. Geochim. Cosmochim.
of lava from destructive plate boundaries. Andesites, Acta 28, 67-86.
525-548. Edited by R. S. Thorpe, Wiley: New York. SCHNETZLER, C. C. and BOTTINO, M. L.
PEARCE, J. A. (1983) Role of the sub-continental (1971) Some alkali, alkaline earth and rare earth
lithosphere in magma genesis at active continental element concentrations and the Rb-Sr age of the Lost
margins. Continental Basalts and Mantle Xenoliths City meteorite and separated phases. J. Geophys. Res.
230-249. Edited by C. J. Hawkesworth and M. J. 76, 4061-4066.
Norry. Shiva. STEIGER, R. H. and JAEGER, E. (1977)
ROCK, N. M. S. (1986) The geochemistry of Lewi Subcommission on geochronology: convention on the
sian marbles. The Lewisian and Comparable High use of decay constants in geo and cosmochronology.
Grade Terrances. Edited by R. G. Park and J. Tarney, Earth Planet. Sci. Lett. 36, 359-362.
Spec. Publ. Geol. Soc., London. 27, 109-126. SUN, S. S. (1980) Lead isotope study of young
ROCK, N. M. S. (1987a) Kimberlites as varieties of volcanic rocks from mid-ocean ridges, ocean islands
lamprophyres. Proc. 4th Intl. Kimb. Conf., Perth. in and island arcs. Phil. Trans. Roy. Soc. Lond., Ser A
press. 297, 409-450.
ROCK, N. M. S. (1987b) The nature and origin of SUN, S. S. and NESBITT, R. W. (1977) Chemical
lamprophyres: An overview. Alkaline Igneous Rocks. heterogeneity of the Archaean mantle, composition of
Edited by J. G. Fitton and B. G. J. Upton, Spec. the earth and mantle evolution. Earth Planet. Sci.
Publ. Geol. Soc. London. 30 in press. Lett. 35, 429-448.
ROCK,N. M. S. (1987c) SPIDER: a general purpose TAYLOR, S. R. and GORTON, M. K. (1977)
FORTRAN77 program to generate normalized multi Geochemical application of spark-source mass spec
element abundance diagrams in petrology and geoche trometry. III: element sensitivity, precision and
mistry. J. Geol. Educ. in press. accuracy. Geochim. Cosmochim. Acta 41, 1375-1380.
ROCK, N. M. S. (1987d) Estimating averages in THOMPSON,R. N. (1982) Magmatism of the British
geology: an empirical study of robust estimates of Tertiary volcanic province. Scott. J. Geol. 18, 49-107.
location and scale. in press. WOOD, D. A., JORON,J. L., TREUIL, M., NORRY,M.
ROCK, N. M. S. (1987e) ROBUST: a FORTRAN77 and TARNEY,J. (1979) Elemental and Sr isotope
package for exploratory data analysis using robust variations in basic lava from Iceland and the surround
estimates of location and scale, normality tests and ing ocean floor. Contrib. Mineral. Petrol. 70, 319
outlier assessment. Comput. Geosci. 13 in press. 339.
ROCK, N. M. S. and LEAKE, B. E. (1984) The WOOD, D. A., TARNEY, J. and WEAVER, B. L.
International Mineralogical Association amphibole (1981) Trace element variations in Atlantic Ocean
nomenclature scheme: computerisation and its con basalts and Proterozoic dykes from northwest Scot
sequences. Mineral. Mag. 48, 211-227. land: their bearing upon the nature and geochemical
ROCK, N. M. S., WEBB, J., MCNAUGHTON,N. J. and evolution of the upper mantle. Tectonophysics 75,
BELL, G. D. (1987) Estimating averages and 91-112.
errors in isotope geo-science: a nonparametric pro