Published September, 1959
A Correlation and Path-Coefficient Analysis of Components
xof Crested Wheatgrass Seed Production
2Douglas R. Dewey and K. H. Lu
SYNOPSIS.Fertility and plant size were the most
important componentsof seed yield. Inasmuch as fer-
tility and plant size were negatively correlated, r ~
--.665, a compromisemust be reached in selection for
these two characters if maximum seed yields are to be
obtained. The methodof "path coefficients" proved use-
ful in analyzing correlation coefficients in this systemof
interrelatedvariables.
CONSIDERA’BLE emphasis is currently being placed in
a hum’bet of grass-breeding programs upon the im-
provement of seed characters, especially seed size and seed
yield. These traits are particularly important in the estab-
lishment of range grasses, which .are usually seeded under
adverse conditions or which may be expected to reseed
themselves. Rogler (4) and Kneebone (1) have demon-
strated a positive relationship between seed size and seed-
ling vigor and have inferred that selection for large seed
will result in strains with superior ability to establish. A
wide range of variation in fertility exists in manygrasses
and provides an important criterion of selection for seed
yield. A knowledge of the inter-relationships among seed
size, fertility, and other seed and forage characters is neces-
sary if selection for the simultaneous improvementof these
traits is to be most effective.
This investigation was designed to furnish information
on the nature of these associations in crested wheatgrass
and thereby suggest appropriate selection procedures. A
second purpose of this paper was to demonstrate the appli-
cation of the method of "path coefficients" in the analysis
of correlation in a system of related variables. This tech-
nique has been used quite widely by animal breeders and
geneticists, but only sparingly by plant breeders. Descrip-
tion of the technique was first published by Wright (5)
1921. Recent summaries of the basic features of the method
and its applications are given by Li (2, 3), to whomthe
reader is referred for more comprehensive descriptions and
derivations of formulas.
MATERIALS AND METHODS
Open-pollination progenies of 79 standard crested wheatgrass
plants selected for large seed, goodseedliog vigor, and high fer-
tility were planted in a spaced nursery in 1954 on the Evansfarm
of the Utah Agricultural Experiment Station near Logan, Utah.
Twocomposite progenies were also included and brought the total
number of progenies to 81. The planting was madein a 9 by 9
balanced lattice design, and each progeny was represented in 10
replications by an ll-plant plot. During the summerof 1955, each
plant’ was sampledby selecting three representative spikes upon
which the following observations were made: 1. Numberof spike-
lets; 2. Fertility (seeds per spikelet); 3. Seedsize (weight of
seeds); 4. Seedweight per spike. In the interest of simplicity and
X Cooperative investigations of the Crops Research Division,
ARS, USDA,and the Utah Agricultural Experiment Station,
Logan, Utah. Approvedas Journal paper No. 78, Utah Agr. Exp.
Sta. ReceivedDec. 19, 1958.
a Research Agronomist, Crops Research Division, ARS,USDA,
and Associate Professor of Statistics, Utah State University, re-
spectively. The authors acknowledgethe efforts of Dr. Wesley
Keller whoinitiated this study.
515
ease of expression, seeds per spikelet and fertility ~are being con-
sidered synonymous, as are also weight of 100 seeds and seed size.
Data were not collected in 1956 because of a change in per-
sonnel. In 1957 the same data as those taken in 1955 were taken
as well as seed yield and plant size as measuredby mature-plant
weight.
A variance analysis wascomputedfor each variable and a covari-
ance analysis for each pair of variables. All analyses were computed
from meansof ll-plant plots. Phenotypic and genotypic correla-
tion coefficients for all possible comparisonswerecalculated from
the variance and covariance components.Correlations were Based
on one year’s data for seed yield and plant size and on two-year
averages for the remainingcharacters.
¯ Path-coefficient Analysis
A path coefficient is simply a standardized partial-regression
coefficient and as such measuresthe direct influence of one variable
uponanother and permits the separation of the correlation coeffi-
cient into components of direct and indirect effects. Theuse of the
methodrequires a cause and effect situation a~nongthe variables,
and the experimenter must assign direction in the causal system
based upon a priori grounds or experimental evidence.
Five variables were included in the path-coeffcient analysis. The
nature of the causal system is represented diagrammatically as
follows:
/
(1) Seed size; (2) Spikelets per spike; (3) Fertility;
(4) Plant size; (5) Seedyield; (X) Residualfactors.
In the path diagram the double-arrowed lines indicate mutual
association as measuredby correlation coefficients, ru, and the
single-arrowedlines represent direct influence as measuredby path
coefficients,
The path coefficients in this particular instance were obtained
by the simultaneous s~lution of the following equations, which
express the basic relationships between correlation and path
coefficients.
1. r~ ---~ P~ + r~P~+ r~aPa+ r=~P~
2. r~ = P=~+ r=P~+ r=aPa~+ r.o~P,~
3. ra ~-- Pa .+ r~P~+ r=~P=+ r,,P~
4. r~= ~ P~ .+ r~P~ + raPm+ tamPa
5. 1 ~--t~xs.+ P~ +’P=~+ P~m+Fa~+ 2P~ra~P~+ 2P~r~aP,~
.+ 2P~r~P~+ 2’P~r~Pm+ 2P~r~P,.~ + 2Pz~r,~P~.~
Seed weight per spike wasomitted from the diagram, since it is
merely an intermediary step and is completelydetermined by vari-
ables 1, 2, and 3. The only componentof seed yield that was not
included ’was the numberof spikes per plant. The fact that data
were not collected on this particular character was unfortunate,
since its inclusion wouIdhave madethe causal system morerigid.
Plant size (variable 4) was included as a factor contributing to
seed yield inasmuchas it is obviously related to numberof spikes
per plant. TheX variable consists of all residual factors that influ-
516
encedseedyield andin this caseincludessamplingerrors plus the
failure of plant size to be perfectly correlated with numberof
spikesper plant.
RESULTS AND DISCUSSION
Whenstudying correlations it is of prime importance to
recognize the nature of the population under consideration,
inasmuchas the magnitudeof the correlation coefficient can
often be influenced by the choice of individuals uponwhich
the observations are made.Theeffect of selection of parent
clones upon the variation in the population under study is
not known.It wouldseem reasonable that variation would
be diminished in those characters for which the parent
clones had been selected, namelyseed size and fertility. If
this is true, correlations involving these traits wouldlike-
wise be reduced somewhat in comparison with those
obtained from a similar unselected population.
Phenotypicand genotypic correlation coefficients for all
possible comparisonsare presented in table 1. The pheno-
typic and genotypic coefficients agreed very closely in each
instance. This similarity was probably due to the relatively
large numberof replications and plants within plots which
tended to reduce the error (environmental) variance
minor proportions. Had the environmental variance and co-
variance been reduced to zero, the phenotypicand genotypic
correlation coefficients wouldof necessity be identical.
Throughoutthe remainder of this paper reference will be
made only to genotypic correlations in order to avoid
unnecessaryrepetition.
A number of interesting relationships can be observed
from table 1. One of the most important to the grass
breeder is that betweenseed size and fertility, whichhad a
correlation of --.706. The less fertile progenies tended to
produce larger seeds than did the more fertile progenies.
As an example, the 10 smallest seeded progenies ranked in
the upperone-fourth on the basis of fertility. This associa-
tion was not entirely unexpectedsince it appears reasonable
that as more seeds are produced per spikelet, the average
seed size should decrease because of competition among
seeds £or food reserves. Selection for seed size alone will
likely be done at the expenseof fertility, and vice-versa.
Althougha substantial negative association existed between
these two characters, it was not so strong as to makelarge
seed and high fertility entirely incompatible, although
simultaneous improvement of both traits is made more
difficult by this unfavorablecorrelation.
Anothercorrelation worthy of someattention is that be-
tweenfertility and plant size, r = --.665. Progenies char-
acterized by large, vigorous plants were less fertile than
progenies comprised of smaller plants. This lends some
Table 1--Phenotypicand genotypic correlations betweenseed,
spike, and forage characters in crested wheatgrass.
Seed Spikelets/ Fertility Seed wt./
size spike spike
Seed size .260" -.640 -.280
.274 -.706 -.285
Spikelets/spike -.279 .268
-.300 .197
Fertility .823
,825
Seed wt./spik~
Plant size
Seed yield
* The upper correlation in each cell is phenotypic, The correlation coefficient must
exceed . 218 and . 285 to be significant at the , 05 and . 01 levels, respectively.
AGRONOMY
JOURNAL
support to the frequently stated claim that h!gh seed pro-
duction and high forage production are incompatible.
Althougha negative relationship existed betweenfertility
and plant size, seed yield was positively correlated with
plant size, r = .440. Obviously, the seed-producing advan-
tage of large plants must come from their having more
spikes per plant. Althoughthese relationships were J!ound
to exist underspaced-plant conditions, there is no asse.rance
that the sameassociations hold for solid seedings.
Plant size and seed size had a positive correlation, r --
.525. Large plants on the average producedlarger seed than
small plants. The reason for this association maybe that
large plants with a greater leaf surface are capable of
greater photosynthetic activity and more photosynthetic
products are directed into seed formation. This, corr~bined
with the fact that larger plants produced fewer seeds per
spikelet, should give large plants a decided advantagewith
respect to averageseed size.
Although fertility is undoubtedly an important factor
contributing to seed production, the relationship between
fertility and seed yield was negligible, r = .256. The cause
of this small correlation is that fertility was negativelycor-
related with plant size, r = --.665, which in turn was
positively correlated with seed yield, r = .440, resulting in
the net effect of decreasing the expected size of the corre-
lation betweenseed yield and fertility.
It is apparent that manyof the characters are correlated
because of a mutualassociation, positive or negative, with
other characters. As more variables are considered in the
correlation table, these indirect associations becomemore
complex, less obvious, and somewhatperplexing../it this
point, the path-coefficient analysis provides an effective
meansof untangling direct and indirect causes of associa-
tion and permits a critical examinationof the specific forces
acting to produce a given correlation and measures the
relative importanceof each causal factor.
The diagram shownin figure 1 greatly facilitates the
understanding of the nature of the cause and effect system.
The diagramshows, in essence, that seed yield is the result
SEEDSIZE
SPIKELETS
/ ~,’
(2) o -
SEED.~
= 1.12o
P35 FERTILITY
plant Seed 151
size yield
,495 .007
.525 .005
°443 .432
.474 .477
-.568 .257 PLANTSIZE
.256
-.665 (4)
-.273 ,579
-. 311 .617
~,504
,440 RESIDUAL
FACTORS
Figure1--Apath diagramand coefficients of factors influencing
seed yield in standardcrested wheatgrass.
 DEWEYAND LU: CORRELATIONAND PATH-COEFHCIENTANALYSIs OF COMPONENTS
OF PRODUCTION 517
of seed size, num’berof spikelets per spike, fertility, plant
size, and a compositevariable that includes all other factors
affecting seed yields in this study. The first four variables
are themselvesinterrelated; consequently,each factor influ-
ences seed yield by a direct contribution and by acting in
combinationwith the three other variables with whichit is
correlated. The residual variable, X, is assumedto be inde-
pendent of the remainingvariables.
Specific exampleswill illustrate the utility of the method
as an aid in analyzingcorrelation coefficients. Thecorrela-
tion coefficient of seed size with seed yield, r = .005, con-
sists of four components,the relative contribution of which
can be exposed with the use of equation 1 as follows:
Seed size vs. seed yield .................. - r = .005
Direct effect, PI~ ....................... 228
Indirect effect via spikelets/spike, q2P25 .... 087
Indirect effect via fertility, r~sP~, ........ --.790
Indirect effect via plant size, r14P~ ....... 480
Total ............................ 0O5
With the correlation coefficient partitioned into its com-
ponents, one can clearly see what is contributing to the
observed correlation. The direct effect points out the
already obvious fact that, with other variables held con-
stant, increasing averageseed size will increase seed yield.
However,the more subtle indirect effects play a more im-
portant part and maskthe direct influence. A strong nega-
tive influence, --.790, wasregistered indirectly by seed size
uponseed yield through seeds per spikelet due to the fact
that the seed size was negatively correlated, r = --.706,
with seeds per spikelet, which in turn had a large direct
effect, 1.120, uponseed yield. It will be noticed that path
coefficients, like other regressioncoefficients, maybe greater
than 1. Theindirect effect throughspikelets per spike, .087,
was negligible; and the influence via plant size, .480, was
quite sizeable and positive. The net effect in this systemof
opposing influences was that one negative effect counter-
balancedthree positive ones, makingthe over-all correlation
betweenseed size and seed yield essentially zero.
Withthe application of formulas 2, 3, and 4, the corre-
lations of spikelets per spike, fertility, and plant size with
seed yield can ’be partitioned in a similar manner.
Spikelets per spike vs. seed yield ........... r -- .477
Direct effect, P_o~ ....................... 317
Indirect, via seed size, rmP~._ ........... 062
Indirect, via fertility, r_oa’P~,~ .............. .336
Indirect, via plant size, r24P45 ............ 434
Total .............................
477
Fertility vs. seed yield .................... r -- .256
Direct effect, Pa5 ...................... 1.120
Indirect, via seed size, rlaP~ ............ --.161
Indirect, via spikelets/spike, r2~P~, ....... --.095
Indirect, via plant size, ra4P~ ........... --.608
Total ............................ 256
Plant size vs. seed yield .................. r ---- .440
Direct effect, P~ ....................... 914
Indirect, via seed size, r14P~ ............. 120
Indirect, via spikelets/spike, r~4P_o~ ........ 150
Indirect, via fertility, ra4Pa~............. --.744
Total ............................ 440
An examination of the correlation componentsreveals
that fertility and plant size werethe twofactors that exerted
the greatest influence both directly and indirectly uponseed "
yield. Thesetwo traits were important componentsin every
correlation that involved seed yield, whereasseed size and
spikelets per spike were relatively unimportant. This analy-
sis gives a somewhatdifferent picture than does the simple
correlation analysis. For instance, the correlation between
fertility and seed yield, r -- .256, gives the misleading
impression the fertility had little to do with seed yield,
whereasthe path analysis exposesfertility as a majorinflu-
ence. The apparent conflict betweenthe ’two analyses arises
largely from the fact that the two methodsare measuring
different things. Whereas correlation simply measures
mutual association without regard to causation, the path-
coefficient analysis specifies the causes and measurestheir
relative importance. This latter technique is obviously the
most useful whenconditions permit its application.
Of particular concern to the breeder is the fact that in
each correlation fertility and plant size exerted opposite
influences. Hadit not beenfor the strong indirect effect of
plant size, the correlation betweenfertility .and seed yield
wouldhave been very high. ’Conversely, the correlation be-
tween plant size and seed yield was greatly reduced as a
result of the indirect negativeinfluenceof fertility. If maxi-
mumseed yields are to be obtained, a compromisemust be
reached in the selection programfor the two characters.
Inasmuchas a favorable relationship existed betweenplant
size and seed size and since large plants have forage-pro-
ducing advantages, more emphasis should probably be
placed on plant size than on fertility. These observations
mayonly be valid when applied to similar spaced-plant
populations. Further investigations are in progress to deter-
mine these relationships in other populations grownunder
different conditions.
SUMMARY
Eighty-one standard crested wheatgrass progenies in l 1-
plant plots replicated 10 times in a spaced planting were
studied to determine the inter-relationships amongseveral
seed, spike and forage characters and to establish the rela-
tive importance of each as they affected seed yield. Seed
size, spikelets per spike, fertility, and seed weightper spike
were measuredon each plant in 1955and 1957, while plant
size and seed yield were measuredonly in 1957.
Phenotypic and genotypic correlation coefficients were
calculated from variance and covariance analyses for all
possible combinations of characters. The phenotypic and
genotypiccorrelations .agreed very closely in each compari-
son, Important negative genotypic correlations existed be-
tweenfertility and seed size, r -- --.706; and betweenfer-
tility and plant size, r -- --.665; while seed size and plant
size werepositively correlated, r = .525.
The use of the methodof "path coefficients" was illus-
trated as a meansof analyzingcorrelation coefficients. This
technique was employedto establish the relative importance
of seed size, fertility and plant size as determinersof seed
yield. Fertility and plant size had strong influences, direct
and indirect, uponseed yield; whereasseed size and spike-
lets per spike were relatively unimportant. Becauseof the
negative correlation betweenfertility and plant size, selec-
tion must be based on a compromisebetweenthe two traits
if maximum seed yields are to be obtained.
518 AGRONOMY JOURNAL