You are on page 1of 10

bs_bs_banner

Journal of Sensory Studies ISSN 0887-8250

COLLECTING SAMPLES FROM THE SHELF: DOES THIS


CONTRIBUTE TO SHELF-LIFE KNOWLEDGE?
GUILLERMO HOUGH1,3, PERSIS SUBRAMANIAM2, CHANCHAL NARAIN2 and CINDY BEEREN2
1
Comisión de Investigaciones Científicas de la Provincia de Buenos Aires, 9 de Julio, Buenos Aires 6500, Argentina
2
Leatherhead Food Research, Leatherhead, Surrey, UK

3
Corresponding author. ABSTRACT
TEL: +54 2317 524140;
Fax: +54-2317-431309; Samples with different storage times were collected from supermarket shelves for
EMAIL: guillermo@desa.edu.ar three products: a strawberry fruit drink (SFD), digestive biscuits and shortbread.
Sensory descriptive analysis and chemical measurements were performed on the
Accepted for Publication November 20, 2012
samples. Consumers measured sensory acceptability and answered an accept/
doi:10.1111/joss.12022
reject question for each sample. These last data were used to estimate the sensory
shelf life of the SFD applying survival analysis methodology. For the digestive bis-
cuits and shortbread, this estimation was not possible because of the low rejection
probability and that acceptability differences between samples did not follow the
expected trend in relation to dates. Partial least squares regressions showed the
relationships between acceptability (Y-matrix) and trained panel descriptive
analysis + chemical measurements (X-matrix). Collecting samples from the shelf
could be a means of obtaining cut-off points for shelf-life estimations. However, in
two of the three examples presented, confidence intervals were wide. This was due
to the relatively low number of experimental points in the regression and/or the
batch variability inherent in the sample collection method. Overall, collecting
samples from the shelf contributed limited information to shelf-life knowledge.

PRACTICAL APPLICATIONS
If samples with different storage times could be collected from the company’s
deposits and/or from supermarket shelves, this would mean an important saving
in resources, especially as this would mean not having to wait for the different
storage times to elapse. The title of this article proposed answering the question of
whether collecting samples from the shelf contributes to shelf-life knowledge. The
answer is that it does, although with limitations. Depending on the product, it can
provide sensory shelf-life estimations or establish if current “best-before” dates are
adequate. Correlations between acceptability and objective sensory or instrumen-
tal data can help in understanding modes of deterioration.

companies base their products’ shelf life on previous experi-


INTRODUCTION
ence and knowledge rather than on shelf-life testing.
With the exception of a few food and drink products, such In shelf-life testing, to ensure that statistical-based knowl-
as distilled spirits and cheese, the quality of most foods and edge is obtained, as well as within budget, good experimen-
drinks decrease with storage time. Trying to determine tal planning is required. Gacula (1975) described a series of
when these quality changes in a product occur has been an statistical approaches that take into account controlling the
area of interest to the food industry. In order to study the number of necessary measurements. The designs he pro-
process of shelf life, four main areas have to be considered. posed were the partially staggered, staggered and completely
These are the product, statistical design, instrumental mea- staggered designs, which have been adopted by experiment-
surements and sensory evaluation. The culmination of these ers over the years. In all of these designs, it is assumed that
different factors results in shelf-life testing being relatively the samples are stored by the experimenter in such a way
costly and time consuming. For this reason, the majority of that all of them have been exposed to the same conditions

Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc. 47


COLLECTING SAMPLES FROM THE SHELF G. HOUGH ET AL.

and only stressed as the experimenter desires. In a recently shelf lives, which facilitated the search for samples with dif-
published review on methods for sensory shelf-life estima- ferent best-by dates. They were also going to be the object of
tion (Hough and Garitta 2012), in all reviewed publications a further experiment on the influence of formulation varia-
the starting point was known and the statistical units were tions on shelf life.
followed longitudinally. To follow this procedure, controlled Samples were collected from local supermarkets with
storage chambers are necessary and, more importantly, the appropriate best before dates. In total, five aged variants of
time necessary for product deterioration has to be available. each product were obtained. All samples were stored in a
In a dynamic marketplace, this time resource is often scarce. temperature-controlled facility at 20C until testing. Manu-
If samples with different storage times could be collected factures provided information on the period stated as being
from the company’s deposits and/or from supermarket suitable for their product. Samples were collected with the
shelves, this would mean an important saving in resources, following storage times:
especially as this would mean not having to wait for the dif- Strawberry fruit drink (SFD): 8, 12, 16, 20 and 28 weeks.
ferent storage times to elapse. Digestive biscuits (DB): 2, 10, 14, 19 and 39 weeks.
When collecting samples from supermarket shelves, Shortbread (SB): 4, 17, 21, 34 and 39 weeks.
neither the original quality of the production batches nor It was not possible to have samples with storage time = 0
the storage conditions are known. Thus, when differences as samples with this storage time were not to be found in
between samples are found, the effects relating to batch-to- the supermarkets. Recommended best before dates as stated
batch variation, storage conditions and storage times can by the manufacturers for the SFD, DB and SB were 26, 39
become statistically confused. It could be hypothesized that and 52 weeks, respectively.
if quality standards are high, and storage conditions are
generally uniform, then any differences between collected
Sensory
samples would be due to storage time. It could be assumed
that mass produced food products made by leading compa- Sensory evaluation was conducted using the Leatherhead
nies are generally uniform; also, that if the products are all Food Research’s sensory trained panel (15 females). This
collected within a 25-km radius in a country such as the experienced panel generated consensus vocabularies for the
U.K., the storage conditions would be generally uniform. three products (Table 1).
Thus, in principle, collecting samples with different “best- Sensory profiles were developed based on the principles
by” dates from supermarket shelves would be a valid alter- of quantitative descriptive analysis. Assessors were provided
native to storing samples under controlled conditions. It with 30 mL aliquots of the SFD served in a small clear
could even be argued that shelf-life estimations based on plastic glass. Approximately 5 g of the biscuits was served on
these collected samples would be truly representative of paper plates. All samples were presented with three digit
what the consumer will expect to find when consuming the codes in a randomized order. Measurements were by tripli-
product from the marketplace. cate assessed on different days. Data were recorded on
10-cm unstructured line scales with appropriate anchors
using Compusense 5, version 4.8 (Compusense Inc.,
MATERIALS AND METHODS Guelph, ON, Canada).
Sample Collection
Consumer Testing
Commercially available biscuits (shortbread and digestive)
and juice-based noncarbonated drinks (strawberry fruit The following numbers of adult consumers who were non-
drink [SFD]) were selected as examples of the types of rejecters of the products were recruited from a local
products to investigate. These products have relatively long database: 79 for the SFD, 76 for the DB and 87 for the SB.

TABLE 1. DESCRIPTORS USED BY THE TRAINED PANEL FOR EACH ONE OF THE PRODUCTS

Attribute Strawberry fruit drink Digestive biscuit Shortbread biscuit


Appearance Red color and clarity Color (light-dark) and color uniformity Color, quantity of sugar, quantity of holes
Aroma Overall intensity, artificial strawberry Overall intensity, butter/margarine, Overall intensity, butter, margarine, raw
and red fruit cereal and stale pastry, oily/rancid and caramel
Flavor Overall intensity, artificial strawberry, Overall intensity, butter/margarine, Overall intensity, butter, raw pastry,
red fruit, sweet, acidic and bitter cereal, stale, sweet and salty oily/rancid, fat, sweet and salty
Mouthfeel (drink)/ Body and drying in mouth Hardness on first bite, crispiness, Hardness of first bite, crumbly,
Texture (biscuits) crumbliness, bitty and breakdown rate breakdown rate, dry and greasy

48 Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc.


G. HOUGH ET AL. COLLECTING SAMPLES FROM THE SHELF

Consumers were asked to taste each of the samples and significant flavor chemicals of the product that showed an
measure their acceptability for overall liking, appearance, interesting trend were focused on. Finally, all the values cor-
flavor and texture (for biscuits) on a 9-point scale (1 = responding to the youngest (freshest) sample were rational-
dislike extremely, 9 = like extremely). In addition to rating ized and given a starting point of 1. In this way, the older
acceptability, consumers were asked if they would accept or samples then could be compared with the fresh product by
reject each sample by indicating “yes” or “no” on their ballot dividing their area ratios by the area ratio of the fresh
form. product. This allowed an easy comparison of results as all
Products were presented blind, with three digit codes, trends were given the starting value of 1.
and in a balanced order to remove bias from order effects.
Rancidity Indicators. Peroxide Value. Anisidine value
Crackers and water were provided with the samples as
and free fatty acids in biscuits were determined by methods
palate cleansers to be used between samples. Consumer data
BS684 Section 2.14.1987, ISO 6885:1998 and ISO 660:1996,
were recorded on paper ballots.
respectively.
As with the sensory profiling, individuals were provided
with 30 mL aliquots of each of the five SFDs, served in clear
plastic glasses or 5-g portions of the biscuits served on indi- Statistical Analysis
vidual paper plates. The acceptance/rejection data were processed using survival
analysis statistics.
Chemical Measurements Methods of survival analysis have been developed to
evaluate times until an event of interest, often called sur-
Sugar Profile of the SFD. Sugar profile was measured vival times, taking into account the presence of censored
using an in-house developed and validated procedure using data (Gómez 2002). These methods have been applied to
a Dionex ICS3000 Ion Chromatograph (Dionex, Sunnyvale, shelf life of foods (Hough et al. 2003; Curia et al. 2005;
CA). Salvador et al. 2007; Araneda et al. 2008) and have been
described in detail with corresponding software by Hough
Gas Chromatography and Mass Spectrometry (2010).
(GCMS) of Biscuits. Distilled water was placed in a Analysis of variance (ANOVA) was applied to the con-
round bottomed flask fitted with a Snyder column. The water sumer data and trained panel descriptive data considering
was heated until it was boiling and then left for 1 h. This consumer or assessor as a random effect and sample
volatile-free water was cooled and sealed until required. as a fixed effect. Partial least squares (PLS) regression
An aliquot of each sample was placed in 20 mL of (Martens and Martens 1986) was used to relate acceptability
volatile-free water in a test tube. A total of 180 ng of deuter- values (Y-variables) to sensory and physicochemical values
ated toluene and 180 ng of deuterated ethyl benzene were (X-variables). For the PLS regressions, only sensory descrip-
added to the sample and it was sealed and mixed. The tube tors with significance levels ⱕ10% were included.
was fitted with a purge head, heated to 40C and purged with The cut-off point methodology has been used to estimate
clean nitrogen at a flow rate of 20 mL/min. The nitrogen the shelf life of different food products like vegetable oil
was allowed to exit through a trap containing Tenax TA (Ramírez et al. 2001), dulce de leche (Garitta et al. 2004)
(Scientific Instrument Services Ltd., Ringoes, NJ). The and a human milk replacement formula (Curia and Hough
purge was continued for 20 min after which the Tenax trap 2009). The methodology has been described in detail by
was purged directly with dry nitrogen to remove any water Hough (2010). The first step toward estimating a product’s
and the trap was inserted into a thermal desorption and sensory shelf life by this method is the determination of the
cold trapping injector attached to a GCMS. The volatiles cut-off point, which is calculated as follows:
from the trap were transferred into the GC column for (1) Obtain samples with an increase in a sensory defect
analysis. and/or a decrease in a desirable defect.
The mass spectrometer was operated in full-scan mode to (2) Measure the acceptability of these samples with
record complete mass spectra from compounds eluting consumers.
from the GC column. The compounds were tentatively (3) Measure the intensity of the defect and/or desirable
identified by comparison of their mass spectra with those in attribute with a trained sensory panel. Alternatively, a
a literature database. Approximate relative concentrations physicochemical index can be measured.
were calculated by comparing the area of a GC peak to the (4) Obtain the mean square error from the ANOVA on the
area of the internal standard eluting closest to it. consumer data and apply the following formula:
The peak areas were expressed as a ratio to show the peak
area of the corresponding internal standard. The area ratios 2 MSE
S = F − Zα
that did not change with time were left aside and only the n

Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc. 49


COLLECTING SAMPLES FROM THE SHELF G. HOUGH ET AL.

where (2) When performing an experiment of acceptance/


S = value below which the sensory acceptability of the rejection with the objective of estimating sensory shelf life
most preferred sample is significantly reduced; (SSL), the fresh sample is usually included in the sample set.
F = acceptability of most preferred sample; It is to be expected that all consumers would accept this
Za = one-tailed coordinate of the normal curve for sample; however, there are always a proportion of consum-
a significance level; ers who reject it. Hough et al. (2003) justified excluding the
MSE = mean square of the error derived from the data of these consumers from the SSL calculations. For the
ANOVA of the consumer data using consumer and sample SFD, this would mean eliminating the data of 30 consum-
as variation factors; and ers, leaving us with a data set of 49 consumers.
n = number of consumers. Two sets of calculations were performed, adopting both
(5) Correlate the means of the consumer data versus the the above criteria. For each one, the Weibull, log-normal
means of the trained panel data. and normal distributions were tested. Choosing the distri-
(6) Perform an inverse prediction by introducing the value bution with the lowest log-likelihood value (Hough 2010,
of S in the above correlation. If the sensory descriptor is a Chapter 4) resulted in the Weibull distribution for criteria
defect, this will provide the cut-off point above, which (1) and the log-normal distribution for criteria (2). SSLs
acceptability is <S. If it is a desirable descriptor, it will were estimated considering rejection probabilities of 25 and
provide the cut-off point below, which acceptability is <S. 50% (Hough 2010, chapter 4). Results are shown in Table 2.
A means or covering point (1) above would be to collect If the consumers who rejected the freshest sample were
samples from the manufacture and supermarket shelves included (criteria [1] above), this resulted in a shorter esti-
with different storage times. This was carried out in the mated shelf life than if the consumers were not included
present project and, as explained previously, the samples (criteria [2] above). For a 50% rejection, 22 weeks versus 27
were evaluated by a consumer panel, and sensory and weeks. When the data of all consumers who rejected the
chemical properties were also measured. freshest sample were eliminated, we effectively eliminated
data from some fussy consumers who genuinely rejected the
8-week sample and would have accepted the 0-week sample.
RESULTS AND DISCUSSION
By eliminating the data from these fussy consumers, it is
logical that the estimated shelf life increased. Confronted
Shelf-Life Estimations
with this situation where there are no data available for the
truly fresh sample, it is recommended to consider data from
Strawberry Fruit Drink. When collecting samples from all consumers (criteria [1] above) and thus report a shorter,
the shelf with different storage times, it is quite possible that more conservative shelf life.
the sample corresponding to storage time = 0 is not avail- The SFD used in the present study had a label “best
able. In the case of the SFD, the freshest sample that was before” time of 26 weeks (6 months). This is close to the
collected had 8 weeks of storage. Of the 79 consumers who estimation of 22 weeks (see Table 2) derived from samples
evaluated this sample, 30 rejected it. The different ways of collected from supermarket shelves. It is supposed that the
dealing with these rejections are as follows: 26 weeks currently used by the manufacturer has either
(1) Consider that these 30 consumers were particularly sen- been measured or is derived from distribution experience.
sitive to storage changes in the SFD and thus had they had The present study collecting samples from the market
the opportunity to taste a sample with storage time = 0, they shelves allowed confirming their present “best before”
would have accepted it. If this were the case, these consum- recommendation.
ers would be considered left-censored (Hough et al. 2003). Overall acceptability scores measured on a 1–9 scale were
In practice, this means adding a time point = 0 to the data as follows for each storage time: T8 = 6.1, T12 = 6.1,
for which it is assumed that all consumers accept the T16 = 6.2, T20 = 6.0 and T28 = 5.7. The ANOVA on these
product. consumer data indicated differences at a significance

TABLE 2. MODEL PARAMETERS AND ESTIMATED SHELF LIFE (⫾95% CONFIDENCE INTERVALS) FOR THE STRAWBERRY FRUIT DRINK
CONSIDERING DIFFERENT CRITERIA REGARDING CONSUMERS WHO REJECTED THE FRESHEST SAMPLE

Adopted criteria for consumers who Model’s m Model’s s Shelf life for Shelf life for
rejected the freshest sample parameter parameter 25% rejection 50% rejection
Include them as left censored (Weibull model) 3.42 0.860 11 ⫾ 4.2 22 ⫾ 5.6
Not include them (log normal model) 3.302 0.534 19 ⫾ 3.3 27 ⫾ 5

50 Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc.


G. HOUGH ET AL. COLLECTING SAMPLES FROM THE SHELF

level = 12%; thus, it is barely significant. However, they do


Acceptability versus Sensory and
show a tendency to decrease from approximately 20 weeks
Instrumental Measurements
of storage time, coinciding with the estimated shelf life for a
50% rejection of 22 weeks (Table 2). The tendency was
similar for flavor and appearance acceptability scores. Strawberry Fruit Drink. Figure 1 shows the PLS corre-
lation of overall and flavor acceptability (Y-matrix) versus
Digestive Biscuits and Shortbread. Table 3 presents
sensory descriptors + concentration of sugars (X-matrix).
overall acceptability and percent rejection for DB and SB for
The first two PLS dimensions accounted for 89 and 10% of
a combination of samples from the manufacturer and
the Y-variation, respectively.
supermarket shelves with different “best before” dates.
During storage, the sucrose originally present in the SFD
ANOVA on acceptability scores resulted in significant levels
as an ingredient was split into fructose and glucose; this
of 2 and 0.1% for DB and SB, respectively. However, the dif-
reduced the sweetness of the drink which did not seem to
ferences between samples did not follow the expected trend
appeal to consumers. The descriptor defined by the trained
in relation to dates. For DBs, it was not clear which sample/s
panel as artificial strawberry a priori would have a negative
were outliers. Either T10 or T14 was from poor batches or
connotation; however, consumers liked it. Consumers also
suffered severe storage conditions, or T19 was from an out-
liked the red color, and when this was lost during storage,
standingly good batch or had suffered mild storage condi-
the clarity appearance increased. The red-fruit aroma had a
tions. For SB, T21 seems to be the outlier with lower
significance level of 7%; however, the sample means had a
acceptability and higher percent rejection than samples
range between 15 and 17 on a 0–100 sensory scale, hardly of
stored for longer; this sample could have been from a poor
practical importance.
batch or it could have suffered severe storage conditions.
It would not be reasonable to submit the data from these Digestive Biscuit. Figure 1 shows the PLS correlation of
samples to shelf-life estimations based on methods such as overall and flavor acceptability (Y-matrix) versus sensory
survival analysis or cut-off point (Hough 2010, Chapters 4 descriptors + concentration of chemical compounds (X-
and 6), as the acceptability does not follow the expected matrix). The first two PLS dimensions accounted for 69 and
trend over time. However, there is valuable shelf-life infor- 22% of the Y-variation, respectively.
mation in the data. The DB had a label “best before” time of Relationships between sensory descriptors and chemical
39 weeks. As samples collected over this period had percent compounds were generally as expected: stale aroma and
rejections ⱕ25% (Table 3), manufacturers can either leave stale flavor were related to hexanal, pentanal and furanal;
the “best before” date as it is with the confidence that it is a these last chemical compounds are associated with fat oxi-
reasonable shelf life, or they can initiate a shelf-life study dation. Butter flavor was associated with high levels of
under controlled storage conditions to analyze if the “best acceptability.
before” date can be extended. The SB had a label “best In Fig. 2, the ranking of the samples on the first PLS com-
before” time of 52 weeks. With the percent rejection values ponent did not follow the expected storage time trend:
shown in Table 3, it would be wise for manufacturers to sample T19 had less negative attributes than samples T10
conduct a study under controlled storage conditions to and T14; this followed the mean acceptability scores as
determine if their current “best before” date is adequate. A shown in Table 3.
detailed backlog of consumer complaints could be of addi- PLS correlations were also performed for:
tional help in the decision of conducting further shelf-life (1) Appearance acceptability (Y-variable) versus color, uni-
studies. formity, pyrazines and pyrroles (X-variables); uniformity

TABLE 3. MEAN OVERALL ACCEPTABILITY AND PERCENT REJECTION FOR DIGESTIVE BISCUITS AND SHORTBREAD COLLECTED WITH DIFFERENT
STORAGE DATES

Digestive biscuits Shortbread


Storage time Overall acceptability Percent Storage time Overall acceptability Percent
(weeks) (1–9 scale) rejection (weeks) (1–9 scale) rejection
2 7.1a 13 4 6.9a 17
10 6.6b 19 17 6.2b 39
14 6.6b 19 21 6.7a,c 28
19 7.0a 17 34 6.6a,c 25
39 6.6b 25 39 6.5b,c 32

Column means with different letters differed significantly based on Fisher’s least significant difference (P < 5%).

Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc. 51


COLLECTING SAMPLES FROM THE SHELF G. HOUGH ET AL.

1
red_aroma

overall_aroma
Fructose T16 OVERALL
overall-flavor FIG. 1. PARTIAL LEAST SQUARES REGRES-
Glucose T20 artificial_aroma SION OF OVERALL AND FLAVOR ACCEPT-
PLS2

Furfural artificial_flavor
red_color ABILITY (Y-MATRIX) VERSUS SENSORY
-1T28 T12 sweet1
clarity DESCRIPTORS + CONCENTRATION OF
T8 body SUGARS (X-MATRIX) FOR THE STRAWBERRY
Sucrose
FRUIT DRINK
FLAVOR
Sensory descriptors are in lower case letters:
artificial strawberry aroma and flavor, body,
clarity of appearance, overall aroma and
flavor, red color and red-fruit aroma. Sugar
concentrations are in italics. T8 to T32:
-1
samples collected at different storage times
PLS1 in weeks.

was positively correlated with acceptability, and pyrazines acceptability was due to storage time, batch variation or a
and pyrroles (Maillard reaction products) were negatively combination of both factors.
correlated.
(2) Texture acceptability (Y-variable) versus crispiness, Shortbread. Figure 3 shows the PLS correlation of
hardness and bitty (X-variables): crispiness and hardness overall and flavor acceptability (Y-matrix) versus sensory
were positively correlated with acceptability and bitty descriptors + concentration of chemical compounds (X-
negatively correlated. matrix). The first two PLS dimensions accounted for 53 and
As for overall and flavor acceptability, in the appearance 37% of the Y-variation, respectively.
and texture PLS correlations, the ranking of the samples on Relationships between sensory descriptors and chemical
the first PLS component did not follow the expected storage compounds were generally as expected: rancid flavor and
time trend. Thus, for the DBs, the PLS regressions are of stale aroma were related to anisidine, hexanal and pentanal,
value as a relation of acceptability to different objective and to a lesser degree, peroxide value. Diacetyl is usually
indices, but it cannot be ascertained if the variation in described as having a buttery flavor and thus it would be

FLAVOR
pentanal
T19 stale_aroma
OVERALL
hexanal
butter_flavor
T10
PLS2

FIG. 2. PARTIAL LEAST SQUARES REGRES-


-1 T14 1 SION OF OVERALL AND FLAVOR ACCEPT-
T02 T39
furanal ABILITY (Y-MATRIX) VERSUS SENSORY
stale_flavor DESCRIPTORS + CONCENTRATION OF
CHEMICAL COMPOUNDS (X-MATRIX) FOR
THE DIGESTIVE BISCUITS
Sensory descriptors are in lower case letters:
butter aroma, stale aroma and stale flavor.
Chemical compounds are in italics. T2 to
-1 T39: samples collected at different storage
PLS1 times in weeks.

52 Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc.


G. HOUGH ET AL. COLLECTING SAMPLES FROM THE SHELF

FLAVOR
OVERALL pentanal
rancid_flavor
hexanal
T4 T39
diacetil
stale_aroma
butter_aroma

PLS2
anisidine
FIG. 3. PARTIAL LEAST SQUARES REGRES-
-1 T21 1
SION OF OVERALL AND FLAVOR ACCEPT-
ABILITY (Y-MATRIX) VERSUS SENSORY butter_flavor T34
peroxide
DESCRIPTORS + CONCENTRATION OF T17
CHEMICAL COMPOUNDS (X-MATRIX) FOR
THE SHORTBREAD
Sensory descriptors are in lower case letters: free_fatty_acids
butter aroma and flavor, rancid flavor and
stale aroma. Chemical compounds are in
-1
italics. T4 to T39: samples collected at
different storage times in weeks. PLS1

expected to be closely correlated with butter descriptors; for samples and thus its correlation to hardness and
however, the correlation was not high (Fig. 3). crumbly was not tested. Overall acceptability did not corre-
In Fig. 3, T17 is opposed to overall and flavor acceptabil- late significantly with hardness nor crumbly.
ity; it was the sample with the lowest mean acceptability The aroma and flavor descriptors and their associated
(Table 3), lower than samples stored for longer period. As chemical compounds did not adequately explain the low
mentioned in the section Digestive Biscuits and Shortbread, acceptability of sample T17, as shown in Fig. 3. Simple
this sample could have been from a poor batch or it could linear regressions of acceptability versus color, hardness
have suffered severe storage conditions. However, it was not and crumbly could not explain that low acceptability
the sample with the highest values for negative attributes either.
such as rancid flavor or high hexanal concentration.
The other sensory descriptors that were significant for
Cut-Off Points
samples were color, hardness and crumbly. Neither overall
acceptability nor appearance acceptability correlated signifi- Three cut-off points will be presented for illustration
cantly with color. Texture acceptability was not significant purposes.

6
T8
5.9
T12
5.8
Flavor acceptability

5.7
T16
5.6 S

5.5
T20
FIG. 4. FLAVOR ACCEPTABILITY VERSUS
ARTIFICIAL STRAWBERRY FLAVOR FOR 5.4
Cut-off point
STRAWBERRY FRUIT DRINK SAMPLES
The curve corresponds to an exponential 5.3
T28
regression, S = value below which the
sensory acceptability of the most preferred 5.2
35 40 45 50 55
sample is significantly reduced, thus defining
the cut-off point. Artificial strawberry flavor (0-100 scale)

Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc. 53


COLLECTING SAMPLES FROM THE SHELF G. HOUGH ET AL.

7.2

7.1 T02

T19
7
Overall acceptability

6.9

S
6.8

6.7
FIG. 5. OVERALL ACCEPTABILITY VERSUS
Cut-off point
PENTYL FURAN CONCENTRATION RELATIVE
6.6 TO T02 FOR DIGESTIVE BISCUIT SAMPLES
T10 T14
T39 S = value below which the sensory accept-
ability of the most preferred sample is signifi-
6.5
cantly reduced, thus defining the cut-off
1.0 1.5 2.0 2.5 3.0 3.5
point; 95% confidence intervals above and
Pentyl furan below the cut-off point are indicated.

Figure 4 presents the cut-off point for artificial straw- Figure 5 presents the cut-off point for pentyl furan con-
berry flavor of the fruit drink. The following exponential centration in the DBs to show how a cut-off point can be
equation was fitted using nonlinear procedures from determined for an instrumental parameter such as a chemi-
Genstat (VSN International, Hemel Hempstead, U.K.): cal concentration. Pentyl furan is indicative of fat oxidation
(Evans et al. 1971) and thus its negative correlation with
Overall acceptability
acceptability. The linear correlation was significant at <1%
= 5.27 + 1.68 × 10−6 exp (0.256 × artificial _ strawberry )
and the regression explained 97% of the variance. An
The regression significance level was 7%, and the regres- inverse prediction was performed, entering the regression
sion explained 87% of the variance. An inverse prediction with an acceptability value of S = 6.75, which gave an esti-
was performed, entering the regression with an acceptability mated cut-off point of 2.49 for the pentyl furan concentra-
value of S = 5.53, which gave an estimated cut-off point of tion. This would mean that when the pentyl furan
46 on the 0–100 artificial strawberry sensory scale. This concentration was above 2.49, there would be a significant
would mean that when the artificial strawberry flavor was decrease in overall acceptability in relation to the most pre-
below 46, there would be a significant decrease in overall ferred sample. Confidence intervals for the inverse predic-
acceptability in relation to the most preferred sample. The tion can be estimated using the method proposed by Draper
regression procedure also calculated 95% confidence inter- and Smith (1981) whose formulas coincide with those
vals for the inverse prediction and these were ⫾38; being so developed by Graybill (2000); the 95% lower and upper
wide they were not drawn on the cut-off point plot (Fig. 4). intervals were 2.23 and 2.77, respectively. As shown in
These confidence intervals were wide due to the following: Table 3, average acceptability values did not follow the
(1) The small number of experimental points: in this case, expected trend in relation to storage time: T19 had higher
there were 5; and as three parameters were calculated by the acceptability than T10 and T14. Figure 5 shows that when
exponential regression, this left only 2 degrees of freedom. pentyl furan concentration increased, overall acceptability
Increasing the number of samples would mean finding decreased; however, it is not clear whether this trend was
them on the shelves and this was not an easy task. Manufac- strictly due to storage time or whether, e.g., T19 was made
turers do not always have samples available and most with fat with an extra dose of antioxidant. Thus, before
samples found in supermarkets are close to manufacturing relying on this cut-off point for sensory shelf-life determi-
date. nation, further tests are needed to confirm that pentyl furan
(2) Samples had different storage times but, as stated changes are due to storage time for this biscuit.
earlier, they were also from different production batches Figure 6 presents the cut-off point for butter aroma in the
and probably had undergone different storage conditions. SB biscuits. The linear correlation was significant at <9%
This probably led to an increase in unexplained variability and the regression explained 57% of the variance. An
between the samples. inverse prediction was performed, entering the regression

54 Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc.


G. HOUGH ET AL. COLLECTING SAMPLES FROM THE SHELF

7.0
T4

6.8

Overall acceptability
T21
6.6 T34

S
T39
6.4
FIG. 6. OVERALL ACCEPTABILITY VERSUS
BUTTER AROMA FOR SHORTBREAD
SAMPLES 6.2 T17
S = value below which the sensory accept- Cut-off point
ability of the most preferred sample is signifi-
cantly reduced, thus defining the cut-off 6.0
point; 95% confidence intervals above and 30 32 34 36 38 40 42 44 46
below the cut-off point are indicated. Butter aroma

with an acceptability value of S = 6.65, which gave an esti- SFD, this allowed shelf-life estimation, but it did not for the
mated cut-off point of 39.6 for the butter aroma. The DB and SB. Thus, it can be concluded that as a means of
method proposed by Draper and Smith (1981) to calculate modeling shelf-life estimation, sampling products with dif-
confidence intervals in an inverse prediction is based on the ferent best before dates form the market shelves is not to be
intersection of the horizontal line drawn from the S value recommended for most food products. In the case of prod-
with the upper and lower confidence curves of the linear ucts such as the SFD, which presents very small batch-to-
regression. As observed in Fig. 6, these confidence curves batch variations, it may be worth performing the exercise
were so wide that the intersections do not exist. The regres- if resources for a conventional shelf-life study are not
sion did not improve if T4, a possible outlier, was left available.
out. No other regression between acceptability values and PLS regressions showed the relationships between accept-
sensory or chemical concentrations was significant; thus, ability (Y-matrix) and trained panel descriptive analysis +
cut-off points could not be derived from the SB samples. chemical measurements (X-matrix). For the SFD, the rank
of the samples on the PLS map followed the storage time
rank; this was not the case for the DB and SB. Thus, for the
CONCLUSIONS
DB and SB, the PLS regressions were of value as a relation of
Collecting samples from the shelf allowed for a reasonable acceptability to different objective indices, but it cannot be
sensory shelf-life estimation of the SFD. Batch-to-batch ascertained if the variation in acceptability was due to
variations for this product are small, and thus when samples storage time, batch variation or a combination of both
with different storage times were collected, these variations factors.
were not confused with batch-to-batch variations. This was Collecting samples from the shelf could be a means of
not the case for the biscuits, products where batch-to-batch obtaining cut-off points for shelf-life estimations. However,
variations are higher and thus confusion of this variable in two of the three examples presented, confidence intervals
with storage time was important. Different storage condi- were wide. This was due to the relatively low number of
tions of each one of the batches could also have led to unex- experimental points in the regression and/or the batch vari-
pected trends in relation to dates. In spite of not being able ability inherent in the sample collection method.
to estimate shelf-life values for the biscuits, valuable infor- The title of this article proposed answering the question
mation was obtained in relation to the current “best before” of whether collecting samples from the shelf contributes to
dates used for these products. For the DB, the current “best shelf-life knowledge. In the sense that all experiments, if
before” date was adequate, and for the SB, it would be advis- reasonably designed and analyzed, provide knowledge, the
able for them to conduct a controlled study. answer is yes. However, is this information of value? For
In the Introduction section, it was hypothesized that if only one of the three studied products were shelf-life esti-
quality standards are high, and storage conditions are gener- mates obtained. For the other two products, only a picture
ally uniform, then any differences between collected of the state of samples with different “best before” dates was
samples would be due to storage time. However, what was obtained, and further conventional shelf-life studies were
obtained was more akin to a description of what the distri- recommended. Only for samples with very small batch-to-
bution of product is at that point in time; in the case of the batch variations would collecting samples from the shelves

Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc. 55


COLLECTING SAMPLES FROM THE SHELF G. HOUGH ET AL.

be recommended for shelf-life estimations. The correlations EVANS, C.D., MOSER, H.A. and LIST, G.R. 1971. Odor and
between acceptability and objective sensory or instrumental flavor responses to additives in edible oils. J. Am. Oil. Chem.
data are of value as they can help in understanding modes Soc. 48, 495–498.
of deterioration. GACULA, M.C. 1975. The design of experiments for shelf life
study. J. Food Sci. 40, 399–403.
GARITTA, L., HOUGH, G. and SÁNCHEZ, R. 2004. Sensory
shelf life of dulce de leche. J. Dairy Sci. 87, 1601–1607.
ACKNOWLEDGMENTS GÓMEZ, G. 2002. Análisis de Supervivencia. Apuntes del curso
de la Licenciatura en Ciencias y Técnicas Estadísticas de la
This project was funded by the Sensory and Consumer
Facultat de Matemátiques I Estadística. Barcelona: Universitat
Forum, Leatherhead. Special thanks also to the sensory
Politécnica de Catalunya (in Spanish).
trained panel and the consumers who participated in this GRAYBILL, F.A. 2000. Theory and Application of the Linear
project. Model, Duxbury Press, Park Grove, CA.
HOUGH, G. 2010. Sensory Shelf Life Estimation of Food Products,
Taylor & Francis Group, Boca Raton, FL.
HOUGH, G. and GARITTA, L. 2012. Methodology for sensory
REFERENCES
shelf-life estimation: A review. J. Sensory Studies 27, 137–147.
ARANEDA, M., HOUGH, G. and WITTIG DE PENNA, E. 2008. HOUGH, G., LANGOHR, K., GÓMEZ, G. and CURIA, A. 2003.
Current-status survival analysis methodology applied to Survival analysis applied to sensory shelf life of foods. J. Food
estimating sensory shelf life of ready-to-eat lettuce Sci. 68, 359–362.
(Lactuta sativa). J. Sensory Studies 23, 162–170. MARTENS, M. and MARTENS, H. 1986. Partial least squares
CURIA, A. and HOUGH, G. 2009. Selection of a sensory marker regression. In Statistical Procedures in Food Research (J.R.
to predict the sensory shelf life of a fluid human milk Piggott, ed.) Elsevier Applied Science, Oxford, U.K.
replacement formula. J. Food Qual. 32, 793–809. RAMÍREZ, G., HOUGH, G. and CONTARINI, A. 2001.
CURIA, A., AGUERRIDO, M., LANGOHR, K. and HOUGH, G. Influence of temperature and light exposure on sensory shelf
2005. Survival analysis applied to sensory shelf life life of a commercial sunflower oil. J. Food Qual. 24, 195–204.
of yogurts. I: Argentine formulations. J. Food Sci. 70, SALVADOR, A., VARELA, P. and FISZMAN, S. 2007. Consumer
S442–S445. acceptability and shelf life of “flor de invierno” pears (Pyrus
DRAPER, N.R. and SMITH, H. 1981. Applied Regression communis L.) under different storage conditions. J. Sensory
Analysis, Chapter 1, John Wiley & Sons, New York, NY. Studies 22, 243–255.

56 Journal of Sensory Studies 28 (2013) 47–56 © 2013 Wiley Periodicals, Inc.

You might also like