FS 131 Module 10: Relating Sensory Data With Other Types of Data

Introduction to Sensory Evaluation of Foods

Overview of Module 10

• This module focuses on relating sensory data with other types of data in food research.

• The discussion is structured around four main topics: introduction, association tests, correlation tests, and regression analysis.

Defining Key Concepts: Association, Correlation, and Regression

Understanding Association

• Association measures the relationship and strength between nominal or categorical (non-numeric) data.

• It primarily focuses on potential relationships among non-numeric variables.

Understanding Correlation

• Correlation assesses the relationship between ordinal, interval, or ratio (numeric) data.

• It is a subset of association but specifically deals with numeric variables.

Understanding Regression

• Regression combines correlation with ANOVA to establish predictive power between variables.

• It provides a coefficient that quantifies the strength of the relationship after conducting correlation analysis.

Importance of Analyzing Relationships in Sensory Data

Reasons for Analysis

• Analyzing relationships helps identify significant differences among treatments in product development projects. For example, understanding what contributes to differences in formulation acceptability.

• It allows researchers to correlate sensory data (like acceptability scores) with physicochemical characteristics to explain observed differences among formulations A, B, and C.

Examples of Relationship Analysis

Practical Applications

• Measuring how cocoa concentration affects perceived sweetness in fruit juice samples illustrates analyzing relationships effectively.

• Another example includes assessing how gender influences food preferences through consumer surveys and statistical testing methods like chi-square tests for independence.

Conducting Association Analysis

Steps in Association Testing

• Begin by determining if a significant relationship exists using chi-square tests for independence on categorical variables like gender and food preference tendencies.

• If significant results are found, calculate Cramér's V to measure the strength of association between these variables.

Transitioning from Association to Correlation

Understanding Correlation

• Correlation measures linear relationships between two quantitative variables that can be continuous and normally distributed.

• Scatter plots are used to visualize these relationships; an R value close to +1 indicates a strong positive correlation while values near -1 indicate a negative correlation.

Exploring Spearman's Rank Coefficient

Application Example

• Spearman's rank coefficient is useful when dealing with non-normally distributed data; it ranks independent and dependent variable associations.

Introduction to Regression Analysis

Overview of Regression

• Regression extends correlation analysis by predicting one variable based on another through fitting lines to observed data points.

• Distinguishing independent (predictor variable X) from dependent (response variable Y), regression estimates how changes in X affect Y outcomes.

Understanding Regression Analysis in Experiments

Introduction to Regression

• Regression analysis is used to describe the relationship between variables by fitting a line to observed data.

• It incorporates random error into statistical analysis, allowing for estimation with some wiggle room for error.

Independent and Dependent Variables

• The independent variable (X) is set by researchers before experiments, such as ingredient amounts in formulations.

• The dependent variable (Y), like overall acceptability, measures the output of the final product based on changes in X.

• Understanding the difference between these variables is crucial for effective regression analysis.

Types of Regression

• Simple linear regression involves one predictor and one response variable; multiple linear regression includes more than one predictor variable.

• For example, using both cocoa powder concentration and brown sugar percentage as predictors can analyze their combined effect on a response variable like color intensity.

Assumptions of Regression

• Four main assumptions must be met:

• Homogeneity of variance (homoscedasticity): residual variance should remain constant across all values of X.

• Independence of observations: measurements from different samples should not influence each other.

• Normality: Y should be normally distributed for any fixed value of X.

• Linearity: there must be a linear relationship between X and Y for effective regression implementation.

Performing Simple Linear Regression

Calculating Simple Linear Regression

• Simple linear regression estimates relationships between two quantitative variables, predicting Y based on given values of X using a specific formula: y_i = beta_0 + beta_1 x_i + texterror .

• The slope (beta_1) indicates how much Y changes with an increase in X, while the intercept (beta_0) represents the predicted value when X equals zero.

Components of the Formula

• Each component has specific meanings:

• y_i: predicted value for dependent variable Y at given independent variable X.

• beta_1: regression coefficient indicating expected change in Y as X increases.

• Error term accounts for variation not explained by the model.

Best Fit Line Calculation

Steps to Determine Best Fit Line

• Calculate squared residual errors to find the best fit line through least squares method, ensuring that residual sums equal zero confirms proper centering around data means.

• Squared values are necessary since summing raw residual values would yield zero without providing useful information about variability or fit quality.

Hypothesis Testing in Regression

Testing Significance of Variables

• Conduct hypothesis testing on coefficients (beta_1 and beta_0) to determine if they significantly contribute to explaining variations in Y.

• A test statistic greater than critical value leads to rejecting null hypotheses regarding significance levels at alpha/2 with degrees of freedom n−2.

• Results indicate whether predictor variables should be included in models based on their significant relationships with response variables.

Coefficient of Determination (R²)

Importance and Interpretation

• R² quantifies how well predictor variables explain variability in response variables; ranges from 0 (no predictive capability) to 1 (perfect prediction).

• Higher R² values indicate better predictive capabilities within models developed through regression analysis.

Validating Model Assumptions

Assessing Model Validity

• Ensure that all four assumptions are satisfied:

• Homogeneity of variance ensures consistent error across all levels.

• Independence checks that observations do not influence each other.

• Normal distribution verifies that errors follow normal patterns.

• Linearity confirms that relationships between predictors and responses are indeed linear throughout data points analyzed.