Estimating the Coefficients

Note: These scatter plots here are NOT the actual ones from Advertising data.

What is the line of best fit? _____________________

• \(y =\)
• \(\hat{y} =\)
• \(e = y - \hat{y} =\)

We get the LS line (estimates for the model coefficients) by minimizing the sum of squares of residuals (RSS)

\[RSS = (y_1-\hat{y_1})^2+ (y_2-\hat{y_2})^2 +\dots + (y_n-\hat{y_n})^2 = \sum_{i=1}^n(y_i-\hat{y_i})^2\] OR

\[RSS = e_1^2+ e_2^2 +\dots + e_n^2 = \sum_{i=1}^ne_i^2\]

OR

\[RSS = (y_1-(\beta_0+\beta_1x_1))^2+ (y_2-(\beta_0+\beta_1x_2))^2 +\dots + (y_n-(\beta_0+\beta_1x_n))^2 = \sum_{i=1}^n(y_i-(\beta_0+\beta_1x_i))^2\]

Note: How do we minimize RSS? Derivatives are used to minimize RSS (outside the scope of our class).

After minimizing RSS we get our estimates \(\hat\beta_0\) and \(\hat\beta_1\) for the model coefficients:

\[\hat\beta_1 = \frac{\sum_{i=1}^n(x_i-\bar{x})(y_i-\bar{y})}{\sum_{i=1}^n(x_i-\bar{x})^2} = r\times \frac{s_y}{s_x}\]

Where \(r =\) Correlation coefficient, \(s_y =\) the standard deviation of \(y\) variable and \(s_x =\) the standard deviation of \(x\) variable

\[\hat\beta_0 = \bar{y} - \hat\beta_1\bar{x}\]

The standard errors associated with \(\hat{\beta_0}\) and \(\hat{\beta_1}\)

\[SE(\hat{\beta_1}) = \sqrt\frac{\sigma^2}{\sum_{i=1}^n(x_i-\bar{x})^2}\]

\[SE(\hat{\beta_0}) = \sqrt{\sigma^2 \left[\frac{1}{n}+\frac{\bar{x}^2}{\sum_{i=1}^n(x_i-\bar{x})^2}\right]}\]

In general, \(\sigma^2\) is not known, but can be estimated from the data. The estimate of \(\sigma\) is known as the residual standard error (RSE), and is given by the formula

\[RSE = \sqrt\frac{RSS}{n-2}\]

Example 2:

1. From Advertising data it is found that:

• The average cost for TV advertising is 147.0425
  
• The standard deviation of cost for TV advertising is 85.8542363
• The average sales is 14.0225
  
• The standard deviation of sales 5.2174566
• Correlation coefficient between the cost for TV advertising and sales is 0.7822244

Find the LS estimates for model coefficients (\(\hat{\beta_0}\), \(\hat{\beta_1}\)) when regressing sales onto TV.

2. Use the lm() function to find the LS estimates for model coefficients (\(\hat{\beta_0}\), \(\hat{\beta_1}\)) when regressing sales onto TV. Compare your answers with a)

3. State your final simple linear regression model

4. Predict the sales for a TV budgets of $20,000, $30,000 and $50,000

5. Now that we have the model, we can find RSE. Find RSE without using the function lm().

5. Estimate the \(SE(\hat\beta_1)\) without using the function lm().

Interpreating regression coefficients

• \(\hat{\beta_1}:\) The ___________ ____________ in \(Y\) for every ___________ ____________ in \(X\).

• \(\hat{\beta_0}:\) The ___________ value of \(Y\) when the value of \(X = ___\). In most cases, we will find no meaning in \(\hat{\beta_0}\).

Example 3: In the advertising data, the sales are recorded in thousands of units and the advertising costs are recorded in thousands of dollars. Interpret the model coefficient estimates you obtained in Example 2.

range(AdvertisingData$TV)

## [1]   0.7 296.4

Inference

Assessing the Accuracy of the Model

The quality of a linear regression fit is typically assessed using two related quantities:

Estimating the Regression Coefficients

As was the case in the simple linear regression setting, the regression coefficients _________ in the MLR model are unknown, and must be estimated. Given estimates _________ we can make predictions using the formula

Example 11: In the case of the advertising data;

1. find the MLR model using all the available predictors to predict the response Sales in R.

2. Write the MLR equation

3. Interpret each coefficient

Correlation matrix

A correlation matrix is a table of correlation coefficients for a set of variables used to determine if a relationship exists between the variables. The coefficient indicates both the strength of the relationship as well as the direction (positive vs. negative correlations)

Some Important Questions:

When we perform multiple linear regression, we usually are interested in answering a few important questions.

1. Is at least one of the predictors \(X_1,X_2, . . . , X_p\) useful in predicting the response?

2. Do all the predictors help to explain \(Y\) , or is only a subset of the predictors useful?

3. How well does the model fit the data?

4. Given a set of predictor values, what response value should we predict, and how accurate is our prediction?

We now address each of these questions in turn.

One: Is There a Relationship Between the Response and Predictors?

Recall that in the simple linear regression setting, in order to determine whether there is a relationship between the response and the predictor we can simply check whether ________.

In the multiple regression setting with \(p\) predictors, we need to ask whether all of the regression coefficients are zero, i.e. whether _______________.

As in the simple linear regression setting, we use a hypothesis test to answer this question. We test the null hypothesis,

This hypothesis test is performed by computing the F-statistic,

In our class, we will use R or a given R output to get the F-statistic.

Example 11: Conduct a hypothesis test to check if there is a relationship between Sales and the predictor variables Radio, TV and Newspaper. Use the MLR obtained by regressing Sales onto Radio, TV and Newspaper.

Two: Deciding on Important Variables

The task of determining which predictors are associated with the response, in order to fit a single model involving only those predictors, is referred to as ________. The variable selection problem is studied extensively in Chapter 6.

Three: Model Fit

1. Use \(R^2\)
2. Use $RSE = $

Four: Predictions

• Confidence interval for average \(Y\):

We can compute a confidence interval in order to determine how close ____ will be to ____. Confidence Interval will provide an interval estimate for the average \(Y\).

Example 12: Find a 95% CI for average sales when regressing Sales onto Radio, and TV, given that $100,000 is spent on TV advertising and $20,000 is spent on radio advertising in each city.

Given that interval $100,000 is spent on TV advertising and $20,000 is spent on radio advertising in each city, the 95% confidence interval for average sales is ________.

• Prediction interval for \(Y\) for a future \(X\):

Prediction Interval will provide an interval estimate for the \(Y\) given a new observation \(X\).

Example 13: Find a 95% PI for sales when regressing Sales onto Radio, and TV, given that $100,000 is spent on TV advertising and $20,000 is spent on radio advertising in a particular city.

Given that $100,000 is spent on TV advertising and $20,000 is spent on radio advertising in that particular city the 95% prediction interval for sales is _________.

Regression models with Qualitative Predictors

We use the Credit data set from the ISLR library to illustrate this idea.

Problem: Suppose that we wish to investigate differences in credit card balance between males and females, ignoring the other variables for the moment.

In this data set we have ___ qualitative variables. Namely, __________.

To answer the problem above, we use the variable Gender to predict the Balance.

If a qualitative predictor (also known as a factor) only has __________ levels, then incorporating it into a regression model is very simple.

We simply create an __________ or __________ variable that takes on two possible numerical values.

For example, based on the gender variable, we can create variable a new variable that takes the form

\[\begin{equation} x_i = \begin{cases} 1 & \text{if ith person is female} \\ 0 & \text{if ith person is male} \end{cases} \end{equation}\]

and use this variable as a predictor in the regression equation. This results in the model

\[\begin{equation} y_i = \beta_0 + \beta_1x_i + \epsilon_i = \begin{cases} \beta_0 + \beta_1+ \epsilon_i & \text{if ith person is female} \\ \beta_0 + \epsilon_i & \text{if ith person is male} \end{cases} \end{equation}\]

Use the output above and write the model

\[\begin{equation} \hat{Blanace} = 509.80 + 19.73 \times (Female) = \begin{cases} 509.80 + 19.73 & \text{if person is female} \\ 509.80 & \text{if person is male} \end{cases} \end{equation}\]

How to interpret the coefficients of dummy variables?

509.80: The average credit card debt for males is estimated to be $509.80.

19.73: The average difference in credit card balance between females and males is $19.73.

Example 14: Suppose that we wish to investigate differences in credit card balance using marital status, ignoring the other variables for the moment. Fit a model and interpret its coefficients.

model:

Interpretation:

Example 15: Suppose that we wish to investigate differences in credit card balance using ethnicity, ignoring the other variables for the moment. Fit a model and interpret its coefficients.

1. What is the baseline category for the Ethnicity variable in this model?

2. Write the equation of the model

\[\begin{equation} \hat{Blanace} = 531.00 -18.69 \times (Asian) - 12.5 \times (Caucasian) = \begin{cases} 531.00 -18.69 \times (1) - 12.5 \times (0) & \text{if person is Asian} \\ 531.00 -18.69 \times (0) - 12.5 \times (1) & \text{if person is Caucasian} \\ 531.00 -18.69 \times (0) - 12.5 \times (0) & \text{if person is African American} \end{cases} \end{equation}\]

3. Interpret the coefficients

4. Interpret the predicted values

Regression models with interactions

Consider the standard linear regression model with two variables,

According to this model, if we increase \(X_1\) by one unit, then Y will increase by an average of _____ units. Notice that the presence of \(X_2\) does not alter this statement—that is, regardless of the value of \(X_2\), a one-unit increase in \(X_1\) will lead to a -unit increase in Y . One way of extending this model to allow for ___ effects is to include a third predictor, called an interaction term, which is constructed by computing the product of \(X_1\) and \(X_2\). This results in the model

Exercise: Consider the Advertising example. write a linear model that uses radio, TV, and an interaction between the two to predict sales.

We can interpret ____ as the increase in the effectiveness of TV advertising for a one unit increase in radio advertising (or vice-versa).

Example 16:

Fit a linear regression model to Advertising dataset with an interaction term that uses radio, TV, and an interaction between the two to predict sales.

Write down your final model.

Perform a hypothesis test to check whether the interaction effect term is useful in the model.

Note: In this example, the p-values associated with TV, radio, and the interaction term all are statistically ________, and so it is obvious that all three variables should be included in the model.

However, it is sometimes the case that an interaction term has a very ______ p-value, but the associated main effects (in this case, TV and radio) do not. The hierarchical principle states that *if we include an interaction in a model, we should also include the main effects, even if the p-values associated with their coefficients are ___ significant.

In other words, if the interaction between \(X_1\) and \(X_2\) seems important, then we should include both \(X_1\) and \(X_2\) in the model even if their coefficient estimates have large p-values.

Non-linear Relationships

As discussed previously, the linear regression model assumes a linear relationship between the response and predictors. But in some cases, the true relationship between the response and the predictors may be nonlinear.

Example 17:

Consider the Auto data set and the relationship between mpg and horsepower.

It seems clear that this relationship is in fact _________: the data suggest a curved relationship, _________ shape in this example. A simple approach for incorporating non-linear associations in a linear model is to include transformed versions of the predictors in the model.

Let’s try out a model with a quadratic term:

Fit the model with a quadratic term in R

Potential Problems

When we fit a linear regression model to a particular data set, many problems may occur. Most common among these are the following:

1. Non-linearity of the response-predictor relationships.
2. Correlation of error terms.
3. Non-constant variance of error terms.
4. Outliers.
5. High-leverage points.
6. Collinearity.

1 Non-linearity of the Data

Residual plots are a useful graphical tool for identifying non-linearity. This is a plot of residuals, \(e_i = y_i − \hat{y}_i\), versus the predictor \(x_i\) or \(\hat{y}_i\).

Example 17:

Construct two residual plots for regressing mpg onto horsepower with and without a quadratic term.

library(cowplot)
model_no_quadratic <- lm(mpg ~ horsepower, data = Auto)
model_no_quadratic

## 
## Call:
## lm(formula = mpg ~ horsepower, data = Auto)
## 
## Coefficients:
## (Intercept)   horsepower  
##     39.9359      -0.1578

model_quadratic <- lm(mpg ~ horsepower + I(horsepower^2), data = Auto)
model_quadratic

## 
## Call:
## lm(formula = mpg ~ horsepower + I(horsepower^2), data = Auto)
## 
## Coefficients:
##     (Intercept)       horsepower  I(horsepower^2)  
##       56.900100        -0.466190         0.001231

p1<- ggplot(model_no_quadratic, aes(x = .fitted, y = .resid)) + geom_point(color = "red") + theme_bw() 

p2<- ggplot(model_quadratic, aes(x = .fitted, y = .resid)) + geom_point(color = "green") + theme_bw()

plot_grid(p1,p2)

The residuals in the first plot exhibit a clear U-shape, which provides a strong indication of non-linearity in the data.

In contrast, second plot displays the residual plot that results from the quadratic model, which contains a quadratic term. There appears to be little pattern in the residuals, suggesting that the quadratic term improves the fit to the data. (Still the residuals show some “fan out” pattern suggesting more improvements are needed to the model.)

Other transformations: If the residual plot indicates that there are non-linear associations in the data, then a simple approach is to use non-linear transformations of the predictors, such as ____________, in the regression model.

2. Correlation of error terms.

An important assumption of the linear regression model is that the error terms, \(\epsilon_1, \epsilon_2,\epsilon_3, \dots ,\epsilon_n\), are uncorrelated.

Here the errors seems to be uncorrelated.

3. Non-constant variance of error terms.

Another important assumption of the linear regression model is that the error terms have a constant variance, \(Var(\epsilon_i) = \sigma^2\).

One can identify non-constant variances in the errors, or heteroscedasticity, from the presence of a _______ shape in the residual plot.

When faced with this problem, one possible solution is to transform the response Y using a concave function such as log ____ or ____.

Here we use \(log(Y)\) transformation.

4. Outliers

An outlier is a point for which \(y_i\) is far from the value predicted by the model.

Residual plots can be used to identify outliers.

If we believe that an outlier has occurred due to an error in data collection or recording, then one solution is to simply remove the observation. However, care should be taken, since an outlier may instead indicate a deficiency with the model, such as a missing predictor.

5. High Leverage Points

We just saw that outliers are observations for which the response \(y_i\) is unusual given the predictor \(x_i\). In contrast, observations with high leverage have an unusual value for \(x_i\).

This problem is more pronounced in multiple regression settings with more than two predictors, because then there is no simple way to plot all dimensions of the data simultaneously.

In order to quantify an observation’s leverage, we compute the leverage statistic for the \(i^{th}\) \(x\) value.

\[h_i = \frac{1}{n} + \frac{(x_i - \bar{x})^2}{\sum_{j=1}^n(x_j - \bar{x})^2}\]

Note:

1. The leverage statistic \(h_i\) is always between \(1/n\) and 1
2. Average leverage for all the observations is always equal to \((p+1)/n\).
3. So if a given observation has a leverage statistic that greatly exceeds \((p+1)/n\), then we may suspect that the corresponding point has high leverage.

Example 18:

library(ISLR)
library(ggplot2)
data(Credit)

library(plotly)
p3 <- ggplot(Credit, aes(x = Limit, y = Rating)) + geom_point(color = "gold") + theme_bw()
ggplotly(p3)

1. Pick 2 data points that you might suspect to have high leverage

2. Use the leverage equation and the following information to calculate leverages for each point.

   \(n =\) 400, \(\bar{x} =\) 4735.6 and \({\sum_{j=1}^n(x_j - \bar{x})^2}\) = 2.125785^{9}

3. This data set has 10 predictors. Use your answers from 2) to determine whether the points you chosen in 1) are in fact high leverage points.

Solutions:

6. Collinearity

Collinearity refers to the situation in which two or more predictor variables are closely related to one another.

Example 19:

library(ISLR)
library(ggplot2)
data(Credit)

p4 <- ggplot(Credit, aes(x = Limit, y = Age)) + geom_point(color = "gold") + theme_bw()
p5 <- ggplot(Credit, aes(x = Limit, y = Rating)) + geom_point(color = "black") + theme_bw()

plot_grid(p4, p5)

The predictors age and limit are not collinear but the rating and limit show high collinearity.

Let’s consider two models. Model 1: a regression of balance on age and limit, and Model 2: a regression of balance on rating and limit.

model1 <- lm(Balance ~ Age + Limit, data = Credit)
summary(model1)

## 
## Call:
## lm(formula = Balance ~ Age + Limit, data = Credit)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -696.84 -150.78  -13.01  126.68  755.56 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -1.734e+02  4.383e+01  -3.957 9.01e-05 ***
## Age         -2.291e+00  6.725e-01  -3.407 0.000723 ***
## Limit        1.734e-01  5.026e-03  34.496  < 2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 230.5 on 397 degrees of freedom
## Multiple R-squared:  0.7498, Adjusted R-squared:  0.7486 
## F-statistic:   595 on 2 and 397 DF,  p-value: < 2.2e-16

model2 <- lm(Balance ~ Rating + Limit, data = Credit)
summary(model2)

## 
## Call:
## lm(formula = Balance ~ Rating + Limit, data = Credit)
## 
## Residuals:
##    Min     1Q Median     3Q    Max 
## -707.8 -135.9   -9.5  124.0  817.6 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -377.53680   45.25418  -8.343 1.21e-15 ***
## Rating         2.20167    0.95229   2.312   0.0213 *  
## Limit          0.02451    0.06383   0.384   0.7012    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 232.3 on 397 degrees of freedom
## Multiple R-squared:  0.7459, Adjusted R-squared:  0.7447 
## F-statistic: 582.8 on 2 and 397 DF,  p-value: < 2.2e-16

The first is a regression of balance on age and limit, and the second is a regression of balance on rating and limit. In the first regression, both age and limit are highly significant with very small pvalues. In the second, the collinearity between limit and rating has caused the p-value to increase to 0.701. In other words, the importance of the limit variable has been masked due to the presence of collinearity.

A simple way to detect collinearity is to look at the correlation matrix of the predictors.

library(dplyr)

## 
## Attaching package: 'dplyr'

## The following objects are masked from 'package:xts':
## 
##     first, last

## The following objects are masked from 'package:stats':
## 
##     filter, lag

## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union

Credit %>%
  select(Limit, Age, Rating) %>% 
  cor()

##            Limit       Age    Rating
## Limit  1.0000000 0.1008879 0.9968797
## Age    0.1008879 1.0000000 0.1031650
## Rating 0.9968797 0.1031650 1.0000000

It is also possible for collinearity to exist between three or more variables even if no pair of variables has a particularly high correlation. We call this situation multicollinearity.

a better way to assess multicollinearity is to compute the variance inflation factor (VIF).

• The smallest possible value for VIF is 1, which indicates the complete absence of collinearity.
• As a rule of thumb, a VIF value that exceeds 5 or 10 indicates a problematic amount of collinearity.

Find the VIF in the Credit data, a regression of balance on age, rating, and limit.

A regression of balance on age, rating, and limit indicates that the predictors have VIF values of 1.01, 160.67, and 160.59. As we suspected, there is considerable collinearity in the data!