You are on page 1of 7

ANALYSIS OF INCOMPLETE BLOCK.

DESIGNS WITH
REFERENCE SAMPLES IN EVERY BLOCK

M. C. GACULA, JR.

ABSTRACT Table l-Construction of design with reference samples


Sensory scoresare relative and their magnitudesare highly dependent
on the test samplesbeing compared. In order to anchor the scores,the Basic design for BIBD:
test samplesare evaluated with reference to a control or standard. In Blocks (Panelists) Design parameters:
this paper, balancedincomplete block and complete-incompleteblock Treatments 123456 t=4 k=2
designsmodified by the incorporation of a reference samplein every A x x x r=3 b=6
block are reviewed in view of their potential usein sensoryevaluation. B X x x A= 1 N=bk
Examplesof the designand the statistical analysisare provided. C x x X = 12
D X x x

Basic design augmented with reference sample R:


INTRODUCTION Blocks (Panelists) Design- parameters:
Treatments 123456
THE EXPERIMENTAL DATA derived from sensory evalua- R xxxxxx t+l=5 k+l=3
tion techniques are correlated with quality, that is, the score a A x x x r =3 b=6
sample receives is highly dependent on the quality of the B X x x h=l N=b(k+l)
samples to which it is compared in a given set or block. In C x x X = 18
order to anchor the scores, a reference or standard sample is D X x x
incorporated in each block. In incomplete block designs, not
all samples are contained in each block, hence a reference
standard is highly recommended.
When one of the objectives of a study is to compare the statistical analysis can proceed. When t and k are given, Yates
reference standard or control to every other treatment, then it (1936) showed that r, b and h are obtained by taking all
is necessary that the reference sample appear in every in- possible selections of k from t as follows:
complete block.
The purpose of this paper is to describe the balanced in- t!
complete block (BIBD) and complete-incomplete block
(CIBD) designs and their statistical analysis when a reference k! (t -k) !
standard is contained in each block. From the author’s (t-l) !
knowledge, such designs are not available in textbooks today. = (k - 1) ! (t - k) !
DISCUSSION (t-2) !
= r(k - l)/(t - 1) (3)
Balanced incomplete block design (BIBD) = (k - 2) ! (t - k) !
Since the introduction of incomplete block designs by
Yates (1936, 1939, 1940), numerous publications dealing with Consider a BIBD with t = 4 and k = 2. Substituting the
these designs have appeared and textbooks by. Kempthorne value of t and k into formulae (1) through (3), we obtain r = 3,
(1952), Federer (1955) and Cochran and Cox (1957) have b = 6 and h = 1. The layout of the design with these param-
become the standard references in this area. The potential use- eters is given in the upper half of Table 1, where the letter X
fulness of BIBD in sensory evaluation has been described denotes the particular treatments that, are compared within a
(Galinat and Everett, 1949; Hanson et al., 1951; Marquardt et block. This layout is often referred to as the basic design,
al., 1963; Gacula and Kubala, 1972). In order to comprehend (Federer, 1955; Li, 1964). If the basic design is repeated p
the sections to follow, we shall briefly review some properties times to achieve adequate replication, the design parameters b,
of the BIBD. r and h are multiplied by p to become pb, pr and ph. A
The BIBD, as is known today, is defined by five integer catalogue of BIBD plans is found in Cochran and Cox (1957)
values called design parameters: and in Fisher and Yates (1963).
The calculation of the design parameters for BIBD and
t = Number of treatments. CIBD augmented with a reference standard is complex; there-
k =Number of treatments appearing in a block, k < t. fore, it is appropriate to introduce a simpler general procedure
b =Number of blocks or panelists. applicable to both incomplete block designs with and without
r = Number of replications or judgments per treatment. a reference standard (Pearce, 1960; Searle, 1971; Trail and
X =Number of times a pair of treatments appear in the same Weeks, 1973). The general procedure starts by constructing
block. the incidence matrix of the design layout. Using the BIBD
The value of these parameters should be known before the layout in Table 1, the incidence matrix N is

r 1010107
Author Gacuia is with the Armour Research Center, Scottsdale, AZ (4)
85260.

0022-1147/78/0005-1461$02.25/O
0 1978 Institute of Food Technologists where a 1 denotes the presence of the treatment and 0, its
absence in the layout. -Text continued on page 1462

Vol. 43, No. 5 (1978) - JOURNAL OF FOOD SCIENCE - 1461


The transpose of 8. symbolized by _N’,is obtained by inter- and NY’ is shown by Pearce (1960) to be,
changing the rows and columns (Searle, 1966) to read,

1100

I 1
0011
I$ =. 1010 (5)
0101
100 1
0110
In (9), the value of the design parameters are b = sR = 6, r = s
and the product between 3 and E’, if all elements are either 0 = 3, X = 1 and Au = 3. The parameter Xu is used later in the
or 1 (Pearce, 1960), yields a matrix with the main diagonal analysis of CIBD with reference sample. The number of repli-
elements (underlined) equal to r and the off-diagonal elements cates for the reference standard is always equal to b.
equal to X: Table 2 contains the intrablock analysis of variance for a
BIBD with a reference sample. In this table, CF (correction
factor), SST, SS,, SSbZp’SS, adl. and SS, are computed by
the standard procedure described by Yates (1936), Rao (1947)
and Cochran and Cox (1957). Because of incomplete blocking,
that is, not all treatments are contained in a given block (k <
Only the upper half in (6) is shown because the matrix is t), it is well known that the treatment totals Tl should be
symmetrical. adjusted for block effects accomplished by calculating,
BIBD with reference sample
The idea of having a reference sample in every block of
Qi&$ , i = 1,2, . . ., t + 1
BIBD’s traces back to the independent work of Pearce (1960)
and Basson (1959). The construction of BIBD with a reference where B(t) refers to block totals in which treatment i occurs.
sample is achieved by simply adding the reference sample to all The calculations of effects due to treatments and reference
blocks of BIBD in a manner shown in the lower half of Table sample, as well as their variances, are obtained by the formulae
1. The order of tasting of the three samples within each block derived by Basson (1959) and are given as formulae (11)
is determined at random and not as shown in the layout. The through (15) below. The estimate of effect due to the refer-
addition of the reference sample modifies the parameters t and ence sample (tR) is
.k to t + 1 and k + 1, respectively. The BIBD’s are widely
catalogued, therefore, the construction of the augmented (11)
design, that is, the BIBD with the added reference sample is
greatly facilitated. Note that formulae (1) through (3) do not with variance,
apply to the augmented design.
The incidence matrix of the augmented design (lower half, (12)
Table 1) is
111111 and that for the ith treatment,

and its transpose,


N=

[ 1 101010
100101
011001
010110
(7)

with variance,
A

ti = (Qi + +,, (Ok + M/(k + 1))

(t - 1)’ (k + 1) uz. (14)


tr (k&l) e

(8) If contrasts between the ith treatment and the reference


sample are desired, the variance of the contrast is:
(k + 1) (k + t - 2) 2
r(kt-1) “e ’ (15)

The standard error of an effect is the square root of its


Table 2-lntrablock analysis of variance for a 61613 with reference
variance. The value of o, ’ in the above formulae is estimated
samples
by the error mean square in the analysis of variance. It should
Source of variance DF Sum of squares be recognized that when the basic design is repeated p times,
the value of b, r and X where these appear in formulae (11)
Total N-l S.ST=EX&-CF, i=1,2, . . . . t+l through (15) should be multiplied by p.
j=1,2, b The adjusted treatment mean is obtained by Xi = ; + ti, i =
II= 1.2, ::::p
CXfijP

-CF
1, 2, . . .) t, where s is the grand meanexcluding the reference
Repetitions P-l ss, =
b (k+ 1)/p sample, and the reference sample by Xu = $ + t^u. Standard
ZB;
procedures, such as the Duncan’s Multiple Range Test, the
Panelists p(b - 1) SSb,, = -~ Least Significant Difference Test (LSD), are used to perform
within repetition b (k + 1 j/p multiple comparisons of effects or the adjusted means.
Example of BIBD with reference sample
Treatments t st adi. = B 4Qi
(adj. for panelists) The data in Table 3, taken from the author’s file pertaining
Error By difference SS, = By difference to a nitrate/nitrite study on canned chopped ham, are a part of
a larger body of sensory data. For each panelists, the order of

1462 - JOURNAL OF FOOD SCIENCE - Vol. 43, No. 5 (1978)


INCOMPLETE BLOCK ANAL YSIS WITH REFERENCE SAMPLES.. .

Table 3-Nitrate-nitrite study designed as a BIBD with reference SamPIeS. Design Parameters: t+ 1 =$k+l=3,p=2,pr=6,pb=12,ph--2

Repetition Panelists R 1 2 3 4 Bj RQ

I 1 4 5 4 13
2 5 3 6 14
3 5 7 6 18
4 4 4 4 12
5 5 6 5 16
6 4 5 3 12 85
II 7 5 6 3 14
8 3 4 4 11
9 5 6 7 18
10 4 6 5 15
11 4 7 5 16
12 6 5 2 13 87

Ti TR=54 37 27 ’ 29 25 G= 172
BO) 172 95 79 88 82
Bcij/k + 1 57.3333 31.6667 26.3333 29.3333 27.3333

B(i) t+1
Qi = Ti -- -3.3333 5.3333 0.6667 -0.3333 -2.3333 Z Qi=O
k+l

ii tR = -6.4167 0.9843 -0.0357 -0.2500 -0.6786 ~;i = 0

xi = j + ;i ZR = 4.50 5.88 4.88 4.67 4.24

CF = (172)=/ 36 = 821 .JJJ8 13* + l4a +. . + 16* + 13’


%:p = - SS, - CF
3
SST = (4’ + 5’ + . . + 2’ ) - CF = 874.000 - 821.7778
= 52.2222 = 841.3333 - 0.1111 - 821.7778
= 19.4444
ss = (85’ +87*) sst adj. = (-3.3333) (-0.4167) + (5.33331 (0.9643) + . . .
P
- CF = 821.8889 - 821.7778
18 + (-2.3333) t-0.6786) = 8.1748
= 0.1111~ SS, = 52.2222 - 0.1111 - 19.4444 - 8.1748 = 24.2919

sample presentation was randomized.


The initial step in the analysis is to calculate the marginal Table 4-Analysis of variance of chopped ham flavor
totals and the quantities B(t) and Qt, i = 1, 2, . . ., t + 1. For Source of variance DF ss MS F-ratio
instance, for treatment 1:
Total 35 52.2222
B(r)=13+18+16+14+18+16=95 Repetitions 1 0.1111 0.1111
Q, = 37 - (95/3) = 5.3333 Panelists IO 19.444 1.944
within repetition
Treatments (adjusted) 4 8.1748 2.0437 1.67
From formula ( 13), Error 20 24.4919 1.2246

5.333 - 3.333 Calculation of standard errors:


Standard error of treatment means (square root of formula 14):
4
;, = = 0.9643
4.667 3’ (3)
SE, = 4(6) (8 _ , ) 1.2246 = 0.4436
1
and from formula (1 l),
Standard error of treatment versus reference standard
2, = 3/(12) (2)) (-3.333) = -0.4167. (square root of formula 15):
i
sEt-R=+=] 1.2246=0.5915
The adjusted mean for treatment 1 is x1 = 4.9167 + 0.9643 =
5.88, and for the standard is xu = 4.9167 - 0.4167 = 4.50.
The analysis of variance for flavor in the chopped ham is
displayed in Table 4. At 4 and 20 degrees of freedom, the 5%
tabular F value is 2.87 which is not exceeded by 2.04/1.22 = augmented with controls. However, attention should be given
1.67. Thus, there is no evidence that the five means are to the increase in the size of the block which may defeat the
statistically different. main purpose of incomplete blocking, i.e., to reduce block
size. The block size increases from k + 1 to k + 1 + d for the
CIBD with reference sample augmented CIBD, where d is the number of duplicate observa-
Cornell and associates (1972, 1975) developed the com- tions within blocks.
posite complete-incomplete block designs for use in sensory The development of the CIBD is an attempt to segregate
evaluation. These designs belong to a family of designs known the nonadditive effects from the experimental error and thus
in the statistical literature as the extended complete block obtain an estimate of pure error which is not possible in a
designs first reported by John (1963). The CIBD also may be randomized complete block design (RCBD) with one observa-

Vol. 43, No. 5 (1978) -JOURNAL OF FOOD SCIENCE - 1463


tion per cell. The result of this development is a more precise BIBD designs as reported by Trail and Weeks (1973) and then
test of statistical significance of contrasts. A CIBD is formed show the resulting design obtained when a reference standard
by an appropriate combination of RCBD and incomplete is added in every block.
block designs. We shall consider the CIBD generated from According to Trail and Weeks (1973), a CIBD has the fol-
lowing properties: (a) each treatment is applied either no or ni
times in a block; a value of 1 or 2 for no and ni is practical
from the standpoint of smaller block size; (b) replacement in
Table 5-Design layout for a CIBD
the RCBD incidence matrix, of no by zero and nl by 1 results
Panelists (Blocks) in the incidence matrix of a BIBD. It follows from (a) and (b)
that the incidence matrix for CIBD is
Treatments 1 2 3 4 5 6
bJ=“*~*+n, (g-N*) (16)
A X X X X X X
X X X where N * = incidence matrix of the generating BIBD, and M =
B X X X X X X inciden”ce matrix of the generating RCBD with matrix ilie-
X X X ments all 1’s.
C X X X X X X Using matrix (4) as the generating BIBD, and substituting
X X X into formula (16), we have
D X X X X X X
X X X

Design parameters: t = 4 b=6 no = 2


k = 6 h = 13 “, = 1
r=9 N = bk = 36

Taking ni = 1 and no = 2, the CIBD incidence matrix is


r121212i
Table 6-Analysis of variance of CIBD

i I
122121
Source of E= 211221
variance DF Sums of squares 212112
Total N-l SST = XX; - CF i = 1,2, ., t and the layout of this matrix is given in Table 5. T?ie design
j=l,2 I ., b parameter h is the off-diagonal elements of matrix w--
NN’’ which
Panelists b-l Ss,=ZBf/k-CF is equal to 13:
(blocks)

i 1~
Treatments t - 1 E 13 13 13
SS, adj. = (k/U1 ZC? 1
(adjusted)
Interaction fb - I) ft - 1 I SSht = ZXi/nfj - CF - SSh - SSt aoj. -15 - 13
Pure error N - bt SSe = Zdifi2 15
NV’ = 15 13 -13

CF = (CXij)s/N, correction factor; N is the total number of observa- The main diagonal, to be defined later, is a design parameter
tions. for CIBD with a reference standard. The parameter X is also
a,= Ti3!L obtained by (Trail and Weeks, 1973)
k
h = h* (n, - n,)’ + n,(2r - bn,,) (18)
nij = Number of times the ith treatment occurs in the jth block and
is equal to the elements of the incidence matrix N,. where A* is the parameter of the generating BIBD (upper half,
dij Xii, - Xij,, difference between duplicate samples where fi,j) Table 1).
cell is duplicated. The analysis of variance for the CIBD layout in Table 5 is
shown in Table 6 and follows closely the standard calculations
in incomplete block designs. A numerical example of CIBD is
found in Cornell and Schreckengost (1975).
Table 7-Layout of CIBD with reference R When a CIBD is augmented with a reference sample, the
result is a supplemented balance design (Type S) described by
Panelists (Blocks) Pearce (1960) as shown in Table 7. The augmented CIBD has
Treatments 1 2 3 4 5 6
12 samples less, per basic design, than the RCBD with two
observations per cell. As in the BIBD’s, the sample presenta-
R X X X X X X tion within panelist is randomized. The incidence matrix of
X X X -X X X the augmented CIBD is
A X X X X X X
222222

[ 1
X X X
B X X X X X X 121212
X X X
(19)
C X X X. X X X
X X
211221
X
D X X X X X X N= 212112
122121
X X X and,
Design parameters: 2418181818
t+1=5 r=9 h= 13 s= 15
k=8 ‘R= 12 AR= 18 SR = 24
EN = (20)
b=6
N=bk=48
-

1464 - JOURNAL OF FOOD SCIENCE - Vol. 43, No. 5 (1978)


INCOMPLETE BLOCK ANAL YSIS WITH REFERENCE SAMPLES. ..

Table 8-Analysis of variance of CIBD with reference samples Table g-Data and statistical calculations. Design ParameWS:
t+~=5,k=8,b=6,r=9,rR=12,h=13,h~=,18,S=15.s~ =
Source of 24
variance DF Sums of squares
Treatments
Total N-l SST = XX; - CF i=1,2,...,t+l
j=1,2,...,b Panelists R A 0 ‘2, D Bj
Panelists b-l SSb= ZB;/k-CF
(blocks) 1 2 3 4 2 1 16
A
Treatments t SS, adj. = CtiQi 1 2 1
(adjusted) 2 3 3 2 3 2 19
Interaction t(b - 1) SSbt = zX&/nij - CF - SSb - SSt adj. 2 2 2
Pure error N-bb(t+l) SS, = Edi/ 3 2 2 2 4 2 19
1 3 3
4 1 3 3 3 2 18
1 2 3
5 2 1 1 3 2 14
where each matrix element is given by (9) and equals the 1 2 2
following design parameters (Pearce, 1960): sR = 24, defined 6 1 2 1 2 1 13
as the number of self-concurrences for the reference standard; 2 2 2
AR = 18, the number of times that the reference sample and
treatment appear together in a block; and s = 15, the number Ti 19 20 20 24 16 G=S9
of self-concurrences for the treatments. When the design lay- ; = 2.0625
out in Table 7 is repeated p times, b, r, rR, h, hR, s and sR are nijBj 198 149 151 147 147 t+1
each multiplied by p. Note that rR is the number of replicates kQi = kTi - nijBj -46 11 9 45 -19 ZkQi=O
for the reference standard. t+1
The analysis of variance of the layout in Table 7 is shown a, -5.750 1.375 1.125 5.625 -2.375 zq = 0
in Table 8. With repetitions of the basic design layout by
different panelists, the source of variance will include repeti-
tions (SS,), and the panelists sum of squares (SSb) become ti -0.4792 0.1526 0.1240 0.6383 -0.2760
nested with repetitions (SSb:p). Both SS, and SSb:p are ob-
tained by the formulae given in Table 2; however, a minor xi=i+ii 1.58 2.22 2.19 2.70 1.79
change in the denominator of these formulae should be made
by using k instead of k + 1. The estimates of standard errors CF = 9S2/48 = 204.1875
SST=(2*+12 ...+12+2z)-CF=235-CF
and effects due to treatments including the reference standard
= 30.8125
are obtained by the formulae given in Pearce (1960); these
16’+19’+.. +13*
formulae are given by (21) through (24). Ss,= L- CF
The estimate of effect due to the reference standard is 8
determined by = 4.1875
bkIR = 6(8)18 = 864 rRA - rhR = 12113) -S(18) = -6
iR = rQR/bhR fh+*R=4(13)+18=70

and that due to the ith treatment by iR = S(-5.75)/6(18) = -51.75/108 = -0.4792


.
tA = [864(11) + 6(-46)1/(70) (864) = 0.1526
^
;B = [8&l(9) + 6(-46)1/60,480 = 0.1240
t, = WAR) (kQi) - { @Rh - ‘*R) (kQd}, i =, 1
I
2
>
.
.9
t (22) tc = [864(45) t 6(-46)1/60,480 = 0.6383
1
(th + AR) WAR) tD = [864-19) + 6(-46)1/60,480 = -0.2760

SSt adj. = t-0.4792) (-5.750) + . . + I-0.2760) (-2.375)


The adjusted mean <or the ith treatment or reference standard = 7.3507
is calculated by y + t i, i = 1, 2, . . ., t + 1, where pis the overall sSbt=[(32/2)+(51/2)+... + (2’11) + (3'/2)I - CF - SSb- SSt adj.
mean. The standard error of the difference between two = 13.2743
adjusted treatment means is ss, = [(2 - 112 + (3 - 212 + + (1 - 2F1/2
= 6.000

between the five groups of pork chops are statistically signifi-


and that between treatment - reference contrast is cant at the 1% level. The reader may perform a multiple
comparison test on the effects or adjusted means by using the
th+hR standard errors given in Table 10.
SE,.R=J (1+iR) MS,/( ). (24) The reference-treatment comparisons can be determined by
k using the least significant difference test (LSD). In this
example, the critical value of the LSD test is
Example of CIBD with reference sample
Table 9 contains the data pertaining to visual color evalua-
tion of pork chops seven days after storage in display cases.A t .01,,8 (SEteR) = 2.878 (0.2561)= 0.7371
five-point color scale is used where 1 = no discoloration and 5
= extremely gray or brown; score of 1 is most desirable.
The statistical calculations are given in the bottom half of where t is the value of the Student’s t distribution at the 1%
Table 9. The analysis of variance shows that the differences significance level with 18 degrees of freedom for error. The

vol. 43, No. 5 (19781 -JOURNAL OF FOOD SCIENCE - 1465


Table lo-Analysis of variance of pork chop experiment Cochran. W.G. and Cox, G.M. 1957. “Experimental Designs.” John
Wiley and Sons, New York
Cornell. J.A. and Knapp, F.W. 1972. Sensory evaluation using com-
Source of variance DF ss MS F-ratio posite complete-incomplete block designs. J. Food Sd. 37: 976.
Cornell, J.A. and Schreckengost, . J.F.. 1975. Measuring
. . . panelists’
. . con--
Total 47 30.8125 sistency usmg composite complete-mcomplete MOCK aeslgns. J.
Panelists 5 4.1875 0.8375 Food Sci. 40: 1130.
Treatments (adjusted) 4 7.3507 1.8377 5.514, P < 0.01 Federer, W.T. 1955. “Experimental Design.” The MacMiIIan Co., New
York.
Interaction 20 13.2743 0.6637 Fisher. R.A. and Yates, F. 1963. “Statistical Tables for Biological. Agri-
Pure error 18 6.0000 0.3333 cultural and Medical Research.” Hafner Publishing Co.. Inc.. New
York.
Standard error of two adjusted treatment means: Gacula, M.C. Jr. and KubaIa. J.J. 1972. Data analysis. Interblock and
intrablock estimates of variance on taste panel data. J. Food Sci. 37:
832.
Galinat, W.C. and Everett, H.L. 1949. A technique for testing flavor in
SE = d2 (0.3333)/(70/8) = 0.2760, i # i’ sweet corn. Agronomy J. 41: 443.
xi- xi*
Hanson, H.L., Kline. L. and Lineweaver, H. 1951. Application of
balanced incomplete block design to scoring of ten dried egg sam-
ples. Food Technol. 5: 9.
Standard error of treatment versus reference standard:
John, P.W.M. 1963. Extended complete block designs. Au&al. J. Stat.
5: 147.
SE t _ R = fi2%0.3333)/(70/8) = 0.2561 Kempthome. 0. 1952. “The Design and Analysis of Experiments.”
John Wiley and Sons, New York.
Li. CC. 1964. “Introduction to ExDerimentaI Statistics.” McGraw-Hill
. Book Co. New York
Marquardt, R.A.. Pearson, A.M.. Larzelere. H.E. and Grieg, W.S. 1963.
differences that exceed 0.7371 are declared significant. The Use of the balanced lattice design in determinhm consumer prefer-
comparisons of interest are: ences for ham containing 16 different combin&ions of s& and
sugar. J. Food Sci. 28: 421.
R - A = 1.58 - 2.22 = -0.64 Pearce. S.C. 1960. Sumdemented balance. Biometrika 47: 263.
R -B = 1.58 - 2.19 = -0.61 Rae. C.R. 1947. General methods of analysis for incomplete block
designs. J. Amer. Stat. Assoc. 42: 541.
R - C= 1.58 - 2.70 = -1.12** Sesrle, S.R. 1966. “Matrix Algebra for the Biological Sciences:” John
R - D= 1.58 - 1.79 = -0.21 Wiley and Sons. Inc., New York.
Searle, S.R. 1971. “Linear Models.” John Wiley and Sons, Inc. New
Thus, the comparison between R and C is statistically signifi- York.
Trail. S.M. and Weeks, D.L. 1973. Extended complete block design
cant. Baaed on the scale used, the negative difference indicates generated by BIBD. Biometrics 29: 565.
that treatment C is significantly inferior to the reference Yates, F. 1936. Incomplete randomized blocks. Ann. Eugenics 9: 121.
standard. Yates, F. 1939. The recovery of interblock information in variety trials
arranged in three-dimensional lattices. Ann. Eugenics 9: 136.
REFERENCES Yates, F. 1940. The recovery of interblock information in balanced
incomplete block designs. Ann. Eugenics 10: 317.
Basson, R.P. 1959. Incomplete block designs augmented with a re- MS received 11/16/77; revised 3120178; accepted 415178.
peated control. M.S. thesis, Iowa State University, Ames, IA.

PHENOLIC ANTIOXIDANTS IN FREEZE-DRIED FOODS. . ., From page 1460

Eastman Kodak Company. 1972. Tenox TBHQ anitoxidant for oils, Labuza, T.P., McNally, L., Gallagher. D., Hawkes, J. and Hurtado. F.
fats, and fat containing foods. Publication No. ZG-201. Eastman 1972. Stability of intermediate moisture foods. 1. Lipid oxidation J.
Chemical Products, Inc., Kingsport. TN. Food Sci. 37: 154.
Eastman Kodak Company. 1963. Tenox Eastman food grade antioxi- Labusa. T.P., Tsuuuki, H. and Karel. M. 1969. Kinetics of Iinoleate
dants. Publication No. G-109. Eastman Chemical Products, Inc., oxidation in model systems. JAOCS 46: 409.
Kingsport, TN. MeNsir. H.M. and BoneBi. E.J. 1969. “Basic Gas Chromatography.” 5th
FIink. J. and Karel. M. 1970a. Effect of processing variables on reten- ed. Varian Aerograph. Palo Alto CA.
tion of volatiles in freeze-drying. J. Food Sci. 35: 444. MassaIdi. H.A. and King, C.J. 1974. Volatiles retention during drying of
Fhnk. J. and Karel, M. 1970b. Retention of organic volatiles in freeze- &nthetic emulsions. J. Food Sci. 39: 438.
dried solutions of carbohydrates. J. Agr. Food Chem. 18: 295. Menting, L.C. and Hoogstad? B. 1967. Selectivity of carbohydrate films
Furia, T.E. & BeIIanca, N. 1977. The properties and performance of as influenced by their moisture content. Experientia 23: 738.
Poly A0 -79; A nonabsorable, polymeric antioxidant intended for Schock. T.J. 1964. Fatty substances in starch. In “Methods in Carbohy-
use in foods. JAOCS 54: 239. drate Chemistry.” Vol. 4, p. 56. Academic Press,NY.
Gejl-Hansen. F. and Flink. J.M. 1977. Freeze-dried carbohydrate con- Sherwin, E.R. 1972. Antioxidants for food fats and oils. JAOCS 49:
taining oil-in-water emulsions: Microstructure and fat distribution. J. 468.
Food Sci. 42: 1049. Sherwin. E.R. 1974. Private communication Eastman Chemical Prod-
Karel. M.. Tannenbaum. S.R.. Wallace, D.H. and Maloney. H. 1966. ucts, Inc;, Kingsport. TN.
Antioxidation and methyl Iinoleate in freeze-dried systems. 3. Ef- Thijssen, H.A.C. and RuIkens, W.H. 1968. Retention of aromas in dry-
fects of added amino acids. J. Food Sci. 31: 892. ing food liquids. De Ingenieur 80: 45.
Kinlg, 3c;; 1970. Freeze-drying of foods, CRC Critical Rev. Food Tech. MS received l/14/78; revised 4129178; accepted 616178.
: .
King. C.J. and Chandrasekaran. S.K. 1973. Analysis of volatile loss
from liquids during freeze-drying and evaporative drying as a ternary
diftusion process. In “Proceedings of the 12th International Con-
erss of Refrigeration,” Vol. 3, p. 649. Avi Publishing Co., Westport.
Michigan Agricultural Experiment Station Journal Article No. 8397.
Lak% T P Heidalbaugh N D Silver M and Karel M 1971. Oxida- Supported in part by Dairy Research Inc.
tion’& m&mediate mo&ure’Aonten&. JAOCS 48: ‘86:

1466 - JOURNAL OF FOOD SCIENCE - Vol. 43, No. 5 (1978)

You might also like