SERVICES FOR

1. Product Testing

Situations in product testing are diverse. Different test situation needs different statistical model to fit. Model selection plays a vital role in product testing. Valid model leads to a valid conclusion. Misuse of models may lead to a misleading conclusion.

SRS provides and uses a series of valid experimental designs and statistical models for product testing in various situations including:

--Consumer, panel and instrument tests
--One, two and multiple products tests
--Replicated and non-replicated tests
--Independent and dependent sample tests
--Yes/no, categorical, ranking and continuous scales
--Univariate, multivariate and time curve data
--With or without prior test information
--For different discrimination methods and designs, e.g., the forced-choice, double forced-choice, A-Not A, same-different and degree of difference methods.

Some profound and sophisticated models are used in product tests, for example:

--Beta-binomial (BB) model for replicated difference and preference tests
--Dirichlet-multinomial (DM) model for replicated ratings
--Some test statistics adjusted by BB and DM models, e.g., adjusted McNemar, Bennett, Stuart-Maxwell statistics for replicated observations, two or multiple matched samples, binary or categorical scales
--Generalized linear models (GLM), e.g., logistic regression model for binary or categorical response variables
--Mixed-effects models for analysis of group data of continuous response variables
--Bayesian approach for the sensory discrimination tests with small sample size and enough prior information. This is particularly suitable for the routine tests conducted regularly


2. Consumer Research

Consumer segmentation--

SRS provides and uses partition of chi-square technique for consumer ratings and replicated ratings data. Partition of chi-square tests can be used to detect differences in location and dispersion effects of consumer responses. The dispersion differences indicate consumer segmentation.
SRS provides and uses model-based clustering technique to segment consumers and provide graphical presentation of the segmentations.

Modeling consumer purchasing behavior--

SRS provides and uses stochastic models to study consumer purchasing behavior including repeat-buying and brand choice. The main data analyzed in the study are consumer panel data, which are purchasing records of the same people or households over extensive periods of time for some fairly frequently bought branded products, like the various lines of food and drink, of soap and toiletries, of cigarettes and so on. The main models used are the compound distributions, e.g. the Gamma-Poisson (or Negative binomial), beta-binomial and Dirichlet-multinomial distributions. With the models, the consumer panel data can be fit and some useful predictions, e.g., penetration, i.e., the proportion of buyers in a long time period, and the proportion of loyal consumers, can be made. Using the predictions, some insights for market of a brand can be obtained.

Linking consumer liking with sensory attributes--

SRS provides and uses various statistical techniques to link consumer liking with sensory attributes and to find out the sensory attributes which can drive the consumer liking.


3. Evaluation of Panel Performance

Trueness and precision are two main criteria for evaluation of any measurement including sensory measurement. Trueness refers to the closeness of agreement between the consensus value of measurement results and the true or accepted reference value. Precision refers to the closeness of agreement between measurement results.

SRS provides and uses different indices (statistics) to evaluate performances of panel/ panelists in different sensory analysis tasks (yes/no, ranking and rating).

For example, in order to evaluate performance of a quantitative descriptive panel, based on the principle of classical reliability theory and new development in psychlogical measurement, two indexes, i.e., reliability coefficient (intraclass correlation coefficient) and agreement coefficient (interrater reliability coefficient) are used to measure the quality of ratings. Both numerical results and graphical results are provided.


4. Time-Intensity and Shelf-Life Data Analyses

SRS provides and uses HANOVA technique to analyze time-intensity data and Survive Analysis technique to analyze shelf-life data.

HANOVA is an adaptive hi-dimensional analysis of variance technique. The advanced technique can be used to compare multiple groups of time-intensity cures. The advantage of the technique is that the approach can properly combine the evidence at different time points to obtain a powerful overall test. The technique can analyze the data that the conventional multivariate technique cannot deal with, for example, the data with curve dimensions larger than the sample sizes.

Survive Analysis is a well-developed statistical methodology for analysis of failure time data. The methodology can model shelf-life data and estimate mean failure time and predict the failure probability of your product at any time in storage. Statistical comparisons of shelf-life distributions of a product in different storage conditions or multiple products in same storage condition can be made.


5. Sensory Threshold

Sensory threshold is a stimulus intensity that will produce a response with a 0.5 probability. Sensory threshold is a measure of sensitivity.

SRS provides and uses a series of techniques including parameter and non-parameter methods to determine individual and population (or group) thresholds, detection and difference thresholds. The precision of the estimated threshold is also given. Statistical comparison of two or multiple thresholds can be made.


6. Thurstonian Model

Thurstonian delta or its estimate d’ is a measure of sensory difference or sensitivity, which is theoretically independent of methods and scales used. Thurstonian model is useful especially in comparison of test results using different methods or scales.

SRS provides and uses a series of techniques for analysis of d’s data. The techniques are based on variance components models of d’s.

--Estimate of individual d’s and their variances from different methods
--Estimate of population or group d’s based on random and fixed effect models
--Comparison of two or multiple d’s
--Comparison of several sets of d’s
--Multiple comparison for d’s data


7. Meta-Analysis in Sensory and Consumer Research

Meta-analysis is a collection of techniques whereby the results of two or more independent studies are statistically combined to yield an overall answer to a question of interest. Meta-analysis is important especially in sensory and consumer field because the results of any single sensory or consumer study are probabilistic and could have occurred by chance. Meta-analysis can provide a foundation for conclusions.

SRS provides and uses statistical techniques based on random or fixed effects models to analyze, summarize and integrate individual experiment results. The experiments may be conducted previously or conducted in different positions using different samples by different consumers or panels. The experiment results may be inconsistent.


8. Data Processing and Routine Statistical Analysis

--Quick and accurate data process including data summaries, making tables and charts
--Routine statistical analyses for sensory and consumer data using appropriate univariate and multivariate, analysis and visualization techniques