91Ó°ÊÓ

The least squares normal equations form the backbone of multiple linear regression analysis by helping to derive the estimated coefficients for the model. When dealing with multiple linear regression, we aim to find the best-fit line that minimizes the sum of the squared differences (errors) between observed and predicted values.

To achieve this, we set up a system of normal equations derived from the data points available. In our example, we have two predictors, depth \(x_1\) and moisture content \(x_2\), influencing the shear strength prediction \(y\). The equations are:

• \( n\hat{\beta}_{0} + \hat{\beta}_{1}\sum x_{i1} + \hat{\beta}_{2}\sum x_{i2} = \sum y_{i} \)
• \( \hat{\beta}_{0}\sum x_{i1} + \hat{\beta}_{1}\sum x_{i1}^{2} + \hat{\beta}_{2}\sum x_{i1}x_{i2} = \sum x_{i1}y_i \)
• \( \hat{\beta}_{0}\sum x_{i2} + \hat{\beta}_{1}\sum x_{i1}x_{i2} + \hat{\beta}_{2}\sum x_{i2}^{2} = \sum x_{i2}y_i \)

These equations are used to solve for the unknown coefficients \(\hat{\beta}_{0}\), \(\hat{\beta}_{1}\), and \(\hat{\beta}_{2}\) that define our predictive model.

In multiple linear regression, parameter estimation involves solving the normal equations to find the coefficients \(\hat{\beta}_{0}\), \(\hat{\beta}_{1}\), and \(\hat{\beta}_{2}\) that provide the best fit to the data. This estimation process seeks to minimize the error (difference) between observed and predicted values.

From the system of normal equations set up in the previous section, we substitute the given numeric values from the dataset:

• \( 10\hat{\beta}_{0} + 223\hat{\beta}_{1} + 553\hat{\beta}_{2} = 1916 \)
• \( 223\hat{\beta}_{0} + 5200.9\hat{\beta}_{1} + 12352\hat{\beta}_{2} = 43550.8 \)
• \( 553\hat{\beta}_{0} + 12352\hat{\beta}_{1} + 31729\hat{\beta}_{2} = 104736.8 \)

By utilizing algebraic methods such as matrix algebra or computational software, we solve these equations to find the coefficient values. These estimated parameters are crucial, as they indicate how much the response variable, in this case shear strength, changes with each unit change in the predictor variables.

Shear strength prediction involves using the fitted multiple linear regression model to predict the response variable, shear strength \(y\), based on predictor variables such as depth \(x_1\) and moisture content \(x_2\).

Once the parameters \(\hat{\beta}_{0}\), \(\hat{\beta}_{1}\), and \(\hat{\beta}_{2}\) are estimated, the model is constructed as follows:\[ y = \hat{\beta}_{0} + \hat{\beta}_{1}x_{1} + \hat{\beta}_{2}x_{2} \\]For instance, if you want to predict the shear strength when the depth is 18 feet and moisture content is 43%, you substitute \(x_1 = 18\) and \(x_2 = 43\) into the equation.
This yields the predicted value for shear strength. The prediction capability is vital in fields such as geotechnical engineering, where understanding soil behavior is crucial.

This exercise is a fantastic educational example in statistics, particularly in teaching the practical application of multiple linear regression.

Students learn how to:

• Set up normal equations from a dataset
• Estimate model parameters accurately
• Make predictions using the regression model
• Interpret the coefficients and their impact on the response variable

Such practical examples help solidify understanding by linking theoretical knowledge with real-world applications. Multiple linear regression is widely used in numerous fields, from economics to natural sciences, making it an essential statistical tool for students to master.