Professional Documents
Culture Documents
Survival Analysis Applied To Sensory Shelf Life of Yogurts-I: Argentine Formulations
Survival Analysis Applied To Sensory Shelf Life of Yogurts-I: Argentine Formulations
Survival Analysis Applied To Sensory Shelf Life of Yogurts-I: Argentine Formulations
Introduction
he consumption of yogurt in Argentina has increased substantially over the past decades. As in many other countries, yogurt in Argentina is sold with live bacteria, which makes this product particularly sensitive to changes during storage. If manufactured under good hygienic conditions, as is the case with most industrialized yogurt, bacterial changes during storage do not impose a health hazard; rather the sensory properties will change until they reach a limit beyond which the consumer will reject the product. Sensory shelf life of yogurt has received little attention in the literature. Yadab and others (1994) studied the effect of soymilk added to buffalo milk on biochemical and sensory characteristics during storage. They defined the shelf life as the number of days that soyogurt could be stored before the mean flavor score decreased below 6, rated on a 9-point hedonic scale. Vargas and others (1989) analyzed shelf life of a soy-whey yogurt. Changes in sensory, chemical, and microbiological characteristics were determined on samples stored at 5 C and 15 C. They used a 7-member trained panel and evaluated the product for color, consistency, and flavor changes. They estimated shelf life on the basis of major flavor and microbiological changes. Karagl-Yceer and others (1999) compared carbonated yogurt samples stored at 4 C and measured sensory attributes as a function of storage time. In none of these studies was the shelf life estimated on the basis of the consumers acceptance or rejection of the product. As discussed by Hough and others (2003), food products do not have sensory shelf lives of their own; rather they depend on the interaction of the food with the consumer. A yogurt with 30-d storage at 10 C will be accepted by 1 consumer who is not concerned with a high-acid flavor but rejected by another consumer who does not like high-acid yogurts. Hough and others (2003) applied survival analysis statistics to determine sensory shelf life based on consumer acceptance or rejection of products with different storage
MS 20050022 Submitted 1/11/05, Revised 5/1/05, Accepted 5/22/05. Authors Curia, Aguerrido, and Hough are with the Comisin de Estigaciones Cientficas de la Pcia. Buenos Aires, Argentina, H. Yrigoyen. Author Langohr is with Dept. d Estadstica i Investigaci Operativa. Univ. Politcnica de Catalunya, Pau Gargallo 5, 08028 - Barcelona, Spain. Direct inquiries to author Hough (E-mail: guille@ghough.cyt.edu.ar).
times. Their key concept was to focus the shelf life hazard on the consumer rejecting the product, rather than on the product deteriorating. To apply the method, they used a data set obtained from an accelerated storage study of yogurt stored at 42 C. Manufacturers give their yogurts the same shelf lives regardless of their composition or variety. In Argentina, the standard shelf life is 35 d. Studies of the shelf life of commercial yogurts of different compositions at real storage conditions have not been reported. The objective of the present work was to establish the shelf lives of commercial Argentine stirred yogurts of different composition by applying survival analysis statistics to consumer data.
Consumer study
People who consumed either whole-fat or fat-free stirred yogurt at least once a week were recruited from the city of 9 de Julio, Buenos Aires, Argentina. Eighty consumers tested each of the yogurt types; thus, 320 consumers were recruited in total. Each consumer received the 7 yogurt samples (corresponding to each storage time at 10 C) monadically in random order. Fifty grams of each sample were served
2005 Institute of Food Technologists
Further reproduction without permission is prohibited
S442 JOURNAL OF FOOD SCIENCEVol. 70, Nr. 7, 2005 Published on Web 8/25/2005
in 70-mL plastic cups. Water was available for rinsing. For each sample, subjects had to answer the question, Would you normally consume this product? Yes or No? It was explained that this meant that if they bought the product to eat it or if it was served to them at their homes, would they consume it or not? They were also asked to evaluate acceptability of appearance, flavor, texture, and overall acceptability, scoring the product using a scale with 9 boxes anchored on the left with dislike very much, in the middle with indifferent, and on the right with like very much.
Calculations
Survival analysis methodology was used to estimate the shelf life of samples stored at 10 C using the results obtained from consumers when asked if they would normally consume the samples. In food shelf life studies, samples with different storage times are presented to consumers. Assume that we define a random variable T as the storage time at which the consumer rejects the sample. The rejection function F(t) can be defined as the probability of a consumer rejecting a product before time t, that is F(t) = P(T t). Usually, survival times are not normally distributed; instead their distribution is often right-skewed. For the evaluation of survival times, often a log-linear model is used: Y = ln(T) = + W (1)
a but rejects the sample stored at time b, then a < T b, and the data are interval-censored. Finally, if a consumer accepts all samples, then T > c, and the data are right-censored. To estimate shelf life, the probability of a consumer rejecting a product [that is, F(t)] must be chosen. Gacula and Singh (1984) mentioned a nominal shelf life value considering 50% rejection, and Cardelli and Labuza (2001) used this criterion in calculating the shelf life of coffee. In the present study, shelf life was calculated for F(t) = 25% and 50%. Calculations of survival analysis were performed according to Garitta and others (2003), where concepts and instructions of S-Plus software for calculations of survival analysis are presented. To establish whether the fat content and/or flavor influenced rejection times, the following log linear regression model with inclusion of covariates was applied (Klein and Moescherger 1997). Its form is analogous to Eq. 1: (3) where T is the storage time at which a consumer rejects a sample; 0, 1, 2 are the regression coefficients; Z1 is the covariate indicating the fat content: fat-free (Z1 = 0) and whole-fat (Z1 = 1); Z2 is the covariate indicating the flavor: strawberry (Z2 = 0) and vanilla (Z2 = 1); does not depend on the covariates; and W is the error distribution. As mentioned before, if T follows a Weibull distribution, W is the extreme value distribution with probability density function:
where W is the error term distribution. That is, instead of the survival time T, its logarithmic transformation is modeled. In Klein and Moeschberger (1997) or Lindsay (1998), different possible distributions for T are presented, for example, the log-normal or the Weibull distribution. Choosing the Weibull distribution, the rejection function is equal to
(2) is the survival function of the smallest extreme value where distribution , and and are the models parameters. To date, there are no statistical tests to compare the goodness-offit of different parametric models used for interval-censored data. Therefore, visual assessment of how parametric models adjust to the nonparametric estimation is the common practice in choosing the most adequate model. For the present data, the following standard distributions were compared: smallest extreme value, normal, logistic, Weibull, log-normal, and log-logistic. Details about each of these distributions can be found in the literature (Klein and Moeschberger 1997; Meeker and Escobar 1998). Because of the discrete nature of the storage times, T will never be observed exactly and thus the censored nature of the data (Hough and others 2003). Suppose that consumers are presented with samples stored at times a, b, and c. If a consumer rejects the sample at the 1st storage time observed, then T a, and the data are left-censored. If a consumer accepts the sample stored at time
URLs and E-mail addresses are active links at www.ift.org
Apart from comparing the general effect of the covariates on the rejection time distributions as modeled by Eq. 3, it could be of interest to compare the shelf lives directly. For example, it might be of interest to compare whether the shelf life of the whole-fat vanilla yogurt is different to the shelf life of the fat-free vanilla yogurt. The shelf life of a given yogurt was defined as the time corresponding to a predefined percentage of rejection by consumers, which was fixed at 25% and 50% in the present study. Thus, to compare shelf lives of different yogurts, it is necessary to compare the 25% and 50% quantiles. For example, for comparing whole-fat vanilla versus fatfree vanilla, the following equation for the confidence interval of the shelf life difference (Meeker and Escobar 1998) was used: (4)
where Z(1 /2) is the (1 /2)-quantile of the standard normal distribution and sedifference is the standard error of the shelf life difference, calculated by the following equation:
Table 3Shelf life values estimated for a 25% and a 50% of rejection by consumers Shelf life (d) 95% confidence intervals Yogurt For 25% rejection 41 39 36 28 10 10 6 7 For 50% rejection 60 59 48 43 9 9 5 7
Whole-fat stirred strawberry Whole-fat stirred vanilla Fat-free stirred strawberry Fat-free stirred vanilla
berry yogurt and established a shelf life of 41 d for all the companys products, the manufacturer would receive more complaints than expected from fat-free-stirred-vanilla yogurt consumers.
Influence of covariates
The fat content covariate (Z1 in Eq. 3) was significant, and the flavor covariate (Z2 in Eq. 3) was not. Figure 1 shows that fat-free yogurts received higher rejection rates than whole-fat yogurts. A possible explanation for fat-free yogurts having higher rejection rates than whole-fat yogurts could be that the non-nutritive sweeteners used in fat-free formulations deteriorate over storage time, thus increasing the perceived acidity of the products, which led to consumer rejection. Direct comparison of shelf lives using Eq. 4 also can be of interest. Table 4 shows that overall there were no differences between strawberry and vanilla yogurts in Argentina. For fat-free yogurts, there was a significant difference in the 25% quantile (that is, shelf life assuming a 25% consumer rejection) tested by applying Eq. 4. The quantile comparison provided complementary information to the overall covariate model.
Figure 1Percentage of consumers rejecting the product versus storage time S444
Figure 2Acceptability versus storage time for whole-fat stirred vanilla yogurt
URLs and E-mail addresses are active links at www.ift.org
Yogurt Fat-free strawberry versus whole-fat strawberry Fat-free vanilla versus whole-fat vanilla Whole-fat strawberry versus whole-fat vanilla Fat-free strawberry versus fat-free vanilla
was a significant difference of shelf life between formulations. A log-linear model (Eq. 3) was introduced that allowed comparing rejection time distributions of the different yogurts. Overall, fat-free yogurts had lower rejection times than whole-fat yogurts. A formula that allowed direct comparison of shelf lives was also introduced (Eq. 4), giving complementary information to the rejection time distribution comparisons.
Acknowledgments
Financial assistance came from the Programa de Ciencia y Tecnologa para el Desarrollo (Project XI.16) and from the Agencia Nacional de Promocin Cientfica (PICT 99).
References
Cardelli C, Labuza TP . 2001. Application of Weibull hazard analysis to the determination of the shelf life of roasted and ground coffee. Lebensm Wiss Technol 34(5):2738. Elliott RP, Clark DS, Lewis KH, Lundbeck H, Olson JC, Simonsen B. 1982. Microorganismos de los alimentos. Tcnicas de anlisis microbiolgico. Vol. I. Zaragoza, Spain: Editorial Acribia. Gacula MC, Singh J. 1984. Statistical methods in food and consumer research. New York: Academic Press. Garitta L, Gmez G, Hough G, Langohr K, Serrat C. 2003. Estadstica de supervivencia aplicada a la vida til sensorial de alimentos. Tutorial introductorio y clculos a realizar utilizando S-PLUS. Madrid: Programa CYTED. Hough G, Langohr K, Gmez G, Curia A. 2003. Survival analysis applied to sensory shelf life of foods. J Food Sci 68:35962. Karagl-Yceer Y, Coggins PC, Wilson JC, White CH. 1999. Carbonated yogurt sensory properties and consumer acceptance. J Dairy Sci 52:13948. Klein JP, Moeschberger ML. 1997. Survival analysis, techniques for censored and truncated data. New York: Springer-Verlag. p 32474. Lindsay JK. 1998. A study of interval censoring in parametric regression models. Lifetime Data Anal 4:32954. Meeker WQ, Escobar LA. 1998. Statistical methods for reliability data. New York: John Wiley & Sons. Vargas LHM, Reddy KV, Da Silva RSF. 1989. Shelf-life studies on soy-whey yogurt: a combined sensory, chemical and microbiological approach. Lebensm Wiss Technol 22:1337. Yadab VB, Jha YK, Garg SK, Mital BK. 1994. Effect of soymilk supplementation and additives on sensory characteristic and biochemical changes of yogurt during storage. Austr J Dairy Technol 49:348.
Conclusions
urvival analysis methodology was used to estimate the shelf life of Argentine yogurts varying in fat level and flavor. A unique shelf life for this product does not seem reasonable because there
S445