You are on page 1of 3

Published September, 1971

Estimation of OptimumField Plot Shape and Size for


1
Testing Yield in Crambeabyssinica Hochst.
2V. D. Meier and K. J. Lessman

ABSTRACT error in cratnbe performance tests, a uniformity trial


A uniformity lest using Prophet crambe(Crambeabys. was conducted to determine optimum plot size and
sinica Hochst.) in a "solid seeding" of 1,296 basic units shape for evaluating differences in yield potential
(one basic unit equaled 1.47 ~) was conducted to deter- among crambe lines and introductions.
minean appropriateplot shape and size for use in cramhe
yield testing.
In analyses using Smithsregression procedure,the soil MATERIALS AND METHODS
heterogeneity index, b, was calculated as 0.5361 and the
2.
optimumplot sDe was estimated to be 6.70 m A uuiformity trial was conducted on the Purdue University
Results of analyses using the modified maximum curva- AgronomyFarm near Lafayette, Ind. in 1968. Approximately
ture technique developed by Lessmanand Atkins (1963) 0.45 ha of Prophet crambe was broadcast-seeded. Before
-
indicated a plot size of 5.35 m as optimum. harvest 0.97-m-widealleys were cut, leaving 36 ranges extending
Plot shape had an effect on plot-to-plot variability. the width of the trial area. Thirty-six plots 0.97 m wide by
Long, narrowplots arranged with their greatest length 1.52 m long (one plot was considered a basic unit) were har-
in the direction of greatest variation resulted in a lower vested in each range, allowing a total of 1,296 basic units.
estimate of amongplot variance. Plants from each basic unit were bagged and threshed, and
the resulting seed was cleaned and weighed. Weight values
Additional index words: Uniformitytrial, Soil hetero- from adjacent basic units were combined for estimating yields
geneity, Coefficient of variability. from plots of different lengths and widths. Length was con-
sidered to be in the direction perpendicular to alleys while
width was considered to be in the direction parallel to alleys.
In the following discussion plots with their greatest dimen-
C RAMBE (Crambe abyssinica
potential
Hochst.) has showh
as a new oil seed crop. Crambe seeds
sion (in number of basic units) perpendicular to alleys are
called "long" plots, and plots with their greatest dimension
contain 35 to 50~ oil that is high in erncic acid (50 to parallel to alleys are called "wide" plots.
60%). The refined oil itself, the erucic acid consti- Two main methods were used to estimate optimum plot
tuent, and oil derivatives have wide potential for nse in size. The first was that developed by Smith (1938). In this
industry (Bruun and Matchett, 1963). Papanthana- method the optimum plot size is determined by the equatiou
sion, Lessman, and Nyquist (1966) evaluated 11 initial X0 = bK~/(l--b)K,, [1]
introductions of crambe for yield, plant height, num- where ~ (expressed in number of square meters) is the opti-
ber of primary branches, number of pods, and seed nlunl plot size, b is the soil heterogeneity index, K~(expressed
2.
weight. The harvested plot size used was 4.9 m in man-hours per plot) is that part of the cost dne to number
Plants were grown in row spacings of .36 and .97 m. of plots, and K2 (expressed in man-hours per square meter)
Significant differences were found among introduc- that part of the cost due to size of plot. The cost factors, K~
and K,,., were determined by itemizing costs in man-hours,since
tions for number of primary branches. Differences for labor is usually the major cost component when conducting
all other characters were nonsignificant. tield plot experiments. Smith (1938) derived equation [1]
The inability to detect differences of statistical sig- after proposing an empirical relationship between plot size
nificance among introductions for so many characters and plot variance and expressing it as:
may have been the result of one or more factors. In- ~
V~ : V~/X [2]
troductions may have been samples from the same where Vxis the variance on a per-nnit basis of plots consisting
population, ff this were correct, their average perform- of X basic units and V1 is the variance of plots of one basic
ance would be expected to be similar. Another pos- unit. Equation [2] may be expressed in logarithms as:
sibility may be that this species of crambe contains log Vx = log V, -- b log X
little or no genetic variability for the particular char-
acters chosen for study. However, this would seem in which b is estimated as a linear regression coefficient. In
unlikely, considering the experience with many other this study b was weighted according to number of degrees of
freedomavailable for each log V.~ as proposedby Federer (1955)
plant species in which natural variability is abundant and was computed using the Weighted Regression Analysis
when plants representing a broad sample of the species Program in the CDC~500 computer. Variance per basic unit
are studied. Finally, variation due to experimental was calculated as
~
V~ = V~)/X
error may have been too large to detect significant
differences among introductions for traits evaluated. where V(~ is the among-plot variance.
In field research technique, number and size of repli- The oth.er procedure used was the modified maximumcurva-
cations, care and handling of individual plot samples ture technique presented by Lessman and Atkins (1963).
as well as size and shape of plots are all factors that this methodthe optimumplot size is determined by first locat-
ing the region of maximmncurvature of the curve relating
influence variation attributable to experimental error. coefficients of variability with their corresponding plot sizes.
Since no information is available as to the effects The general equation y = aXw closely estimates this relation-
of plot size and shape in minimizing experimental ship wherey is the coefficient of variability and X is the corre-
spondingplot size in basic units. If the above equation is con-
verted to logarithms, the constants a and b can be found by
~ Contribution from the Department of Agronomy, Purdue a weighted regression analysis as above. Lessmanand Atkins
University, Agricultural Experiment Station. Journal Paper No. (1963) reasoned that the r.egion of maximum curvature of the
4345. Received Feb. 17, 1971. given curve could be located between the two successive tangent
~Graduate Research Assistant and Professor of Agronmny, lines with the largest angle of intersection. However,slopes
Department of Agronomy, Purdue University, Lafayette, Ind. of each successive tangent line were determined at equal incre-
47907. ments along the X-axis, which would bias the region of maxi-

648
MEIER & LESSMAN: OPTIMUM FIELD PLOT SHAPE AND SIZE 649

mumcurvature toward lower values of X with a curve defined RESULTS AND DISCUSSION
by y ~ a/Xu. A better method would be to take equal incre-
lnents along the length of the curve. Curvature at any point Amongplot variances, variances per basic unit, and
along a line can be determined using the equation: coefficients of variability for each plot size and shape
K = y"/(1 are presented in Table 1. The relationship between
where K is the curvature of the line, and y and y" are the first coefficients of variability and their correspondingplot
and second derivatives of the function defining the line (Gran- sizes was that of" a decrease in the coefficients of vari-
ville, Smith, and Longley 1962). Curvature is maxilnum where ability with an increase in plot size.
the derivative of K with respect to X is 0. For the function
y = a/X b the point of maximumcurvature, X~, of the line The relative rate of reduction was less with larger
depicting the function is plots. This relationship is similar to that reported
previously in other field plot technique problems
X : [a~be(2b q- ])/(b q- ~/(~b + " (Smith, 1938; Lessrnan and Atkins, 1963).
The above equation converted to logarithms is With regard to plot shape, a general trend indicated
log X~___ 1/(2b -t- 2) log [a2b~(2b -t- 1)/(b -5 2)]. more variation associated with "wide"plots than with
After X. is determined, it is modified by cost factors using K~ "long" plots. To study the effects of plot shape on
and K.. which were used in Smiths procedure as follows: plot yield variances F tests were used to determine
X0 =_ X~K~/XoK2A whetherdifferences existed betweenequal size plots of
where A is the area of the basic unit and Xo is the optimum different shapes. The F test consists of the ratio of the
plot size in numberof basic units. Lessmanand Atkins modified larger amongplot variance over the smaller among
X, using only K~ and K~ which were calculated on a percentage plot variance. "Long"plots consistently showed sig-
basis. Marani (1963) pointed out that calculating K~ and nificantly less variation amongplots than "wide"plots.
iu percentages biases the results. The above equation simplifies
tO To determine the effect of plot shape on the regres-
~/~.
Xo -~- (X.Ka/Ko~A) sion coefficient, b, separate b estimates were obtained

Table 1. Variances and coefficients of variability of various plot sizes and shapes, calculated from crambeuniformity trial yield data.
Plot size and shape Plot size and shape
Size Length Width Total Among Corr. Size Length Width Total Among CoeL
(No. o[ (No. of (No. of number plot Variance of (No. of (NO. of (No. of number plot Vartance o~
basic busic basic of variance per basic wtr. basic basic basic of variance per basic wit.
units) units) malts) plots V~) unit V ~ units units) units plots V(x) unit V ~
x x
1 l 1 1,296 2,805 2,805 20.9
25 5 5 49 324,618 519 9.0
2 t 2 648 8,260 2,065 17.9
2 2 l 648 7,157 1,789 16.7 27 3 9 48 447,922 614 9. S
27 9 3 48 253,825 348 7.4
3 1 3 432 16,159 i,795 16. 7
3 3 i 432 13,031 1,448 15.0 28 .1 7 45 517,631 660 10.1
28 7 4 45 277,476 354 7.4
4 t 4 324 25,323 1,583 15.7
4 2 2 325 21,504 1,344 14.5 30 5 6 42 471,359 524 9,0
4 4 1 324 20,099 1,256 14.0 30 6 5 42 386,159 429 8,2
5 1 5 252 36,528 1,461 15.1 32 4 8 36 669,805 654 tO.l
5 5 1 252 24,843 994 i2.5 32 8 4 36 391,970 383 7.7
6 t 6 216 52,642 1,462 35 5 7 35 599,829 49{) 8.8
15.1
6 2 3 216 43,192 13.7 35 7 5 35 353,934 289 6,7
1,200
6 3 2 216 40,274 1,119 13.2 36 3 12 36 864,787 667 10,2
6 6 1 216 28,856 802 11. 2 36 4 9 36 758,928 586 9,5
7 l 7 18(I 63,244 1,291 14.2 36 6 6 36 563,987 435 8.2
7 7 1 180 658 36 9 4 36 419,369 325 7.1
32,225 I0. I 36 I2 3 36 426,284 329 7.2
S l 8 144 81,500 1,273 14. 1
S 2 4 162 66,393 1,073 12.7 40 5 8 28 774,686 484 8.7
8 4 2 162 62,394 975 12.3 40 8 5 28 493,035 308 6.9
8 8 1 144 41,050 641 10.0 42 6 7 30 701,948 398 7.9
9 i 9 144 89,203 1,iOi 13.1 42 7 6 30 529,553 30(I 6.8
9 3 3 144 81,772 1,010 12.5 45 5 9 28 775~105 383 7.7
9 9 ~ 144 45,702 564 9.4 45 9 5 28 538,456 266 6.4
10 2 5 126 95,503 955 12.2 48 4 12 27 1,407,139 611 9.8
I0 5 2 126 74,983 750 10.8 48 6 8 24 911,672 396 7.9
L2 L 12 108 157,586 1,094 13.0 48 8 6 24 740,804 322 7.1
12 2 6 10S 140,783 978 12.3 48 12 4 27 697,717 303 6.9
12 3 4 108 i28,489 892 Ii. 8 49 7 7 25 620,458 258 6.4
12 4 3 108 127,878 888 it.8
12 6 2 I08 36,907 604 9.7 54 6 9 24 936,625 32[ 7.4
12 12 1 108 69,625 484 8.7 54 9 6 24 795,818 273 6. 5
I4 2 7 90 i68,762 861 lt.6 56 7 8 20 722,177 230 6.0
14 7 2 90 95,070 485 8.7 56 8 7 20 919,690 293 6.8
15 3 5 84 186,840 829 11.4 60 5 12 21 1,586,793 441 8.3
60 12 5 2i 903,844 251 6.2
16 2 8 72 213,240 833 11.4 63 7 9 20 680,633 171 5.2
16 4 4 81 203,367 794 11. 1 63 9 7 20 968,611 244 6.2
t6 S 2 72 121,019 473 8.6 64 8 8 16 1,082,963 264 6.4
1~ 2 9 72 226,392 699 10.4 72 6 12 18 1,823,823 352 7.4
t8 3 6 72 274,879 848 t1.5 72 8 9 16 1,071,991 207 5.7
18 6 3 72 174,125 537 9.1 72 9 S 16 1,240,959 239 6.
18 9 2 72 i30,350 4O3 7. 9 72 12 6 18 1,377,189 266 6.4
20 4 6 63 280,805 702 81 9 9 16 1,171,593 179 5.3
20 5 4 63 236,485 591 9.6
84 7 12 15 1,511,982 2[4 5.8
21 3 7 60 336,218 765 tO. 9 84 12 7 15 1,642,797 233 6.0
2I 7 3 60 179,150 406 8.0
96 ~ 12 12 2,272,326 247 6,2
24 2 12 54 424,949 738 10.7 96 12 8 12 2,029,495 220 5.9
24 3 8 48 427,460 742 10.8
24 4 6 54 425,401 739 10.7 108 9 12 12 2,800,305 240 6.1
24 6 4 54 285,817 496 8.8 108 12 9 12 2,021,090 173 5.2
24 8 3 48 24i, i04 419 8.1 144 12 12 9 4,465,671 215 5,8
24 12 2 54 205,512 357 7.4
650 CROP SCIENCE, VOL. 11, SEPTEMBER-OCTOBER 1971

using data from "long" and "wide" plots. Data from length in the direction of more variation were con-
plots with equal number of basic units in length and sidered as optimum for plot shape.
width were not used. These estimates were 0.5828 and With regard to plot size, Smith's procedure was the
0.4983 for "long" and "wide" plots, respectively. To first method used. The b estimate was calculated as
determine if the difference between the b estimates 0.5361, which indicated intermediate soil heterogeneity
was statistically significant, the following t-test was in the experimental site. Cost factors calculated were
applied (Steele and Torrie, 1960); 0.3822 man-hour per plot for K! and 0.0659 man-hour
per square meter for K2. Equation [1] indicated an
t = (hi - b2) / Vs2,, (1/Sx*,, + l/Sxz,) optimum plot size of 6.70 m2 using these values for
KI and K2.
where s2p is the pooled variance about regression and With the modified maximum curvature technique
2x2 is the corrected sum of squares of log X. A t value the constants a' and b' were computed to be 21.08 and
of 2.80 was calculated. Such a t value is highly un- 0.2678. Using these constants the point of maximum
likely on the basis of chance if the b estimates were curvature, Xc, was determined as 3.36 basic units.
not truly different. It was concluded that estimates Modifying Xc by the cost factors resulted in an opti-
of b may be affected by plot shape. mum plot size estimate of 3.64 basic units or 5.35 m2,
Smith stated that b should range from 0, indicating which is 20% smaller than the plot size estimated by
complete uniformity, to 1, indicating random soil Smith's procedure. Since Smith noted that little effi-
variability or a strong productiivty gradient. Thus, ciency was lost by using a plot size up to 50% larger
less variability among plots would be expected in a or smaller than the optimum plot size, very little loss
fairly uniform experimental area. Conversely, more in efficiency would result using a plot size calculated
variability among plots would be expected with more by the modified maximum curvature technique.
random soil variability. In this investigation data
from "long" plots showed less among plot variability
and a larger b estimate than that from "wide" plots.
These contradictory results may be partly explained
in that ranges were separated by 0.97-m alleys, and
"long" plots were oriented with their greatest dimen-
sion perpendicular to the alleys. Therefore, the actual
length of "long" plots was greater proportionally than
that of "wide" plots, even though their lengths were
the same in terms of basic units. Since basic units
were farther apart in "long" plots, their composition
could be considered more random. Thus, a larger b
estimate would be expected for "long" plots. Also,
basic units were farther apart in simulated "long"
plots; hence, more variation should occur within
"long" plots resulting in less among plot variation.
Since "long" plots had the least variation among plots
at any given size, long narrow plots with their greater

You might also like