Eliminating predictors

You do not want to exclude the lagged coupon effect yet. The company needs a really sound model to improve it's future marketing decisions and to effectively run their campaign. To be 100% sure, you perform backward selection of predictors by using the function stepAIC(), loaded from the add-on package MASS.

The stepAIC() function builds all possible combinations of predictors and determines which has the lowest AIC. The argument direction = "backward" starts the selection process with the extended.model and sequentially removes terms in an effort to lower the AIC. The argument trace = FALSE suppresses information to be printed during the running of the selection process. The final model, resulting in the minimum AIC value, is summarized by the function summary().

This exercise is part of the course

Building Response Models in R

Exercise instructions

Load the add-on package MASS by using the function library().
Perform backward elimination of predictors on the extended.model object by using the function stepAIC(). Assign the result to an object named final.model.
Summarize the final.model object by using the function summary().

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Load the MASS package


# Backward elemination
final.model <- ___(___, direction = ___, trace = ___)

# Summarize the final model

Edit and Run Code

This exercise is part of the course

Building Response Models in R

IntermediateSkill Level

4.9+

Start Course for Free

The first chapter introduces you to the basic principles and concepts of market response models. Here, you will learn how to build simple response models for product sales. In addition, you will learn about the theoretical and practical differences between linear and non-linear models for sales responses.

Exercise 1: Fundamentals of market response models Exercise 2: Retail sales Exercise 3: Understanding sales Exercise 4: Linear response models Exercise 5: A linear response model for sales Exercise 6: Making predictions Exercise 7: Predictive performance Exercise 8: Nonlinear response models Exercise 9: Linearizing nonlinear functions Exercise 10: What 's the value added?

An effective marketing strategy combines all the tools available to communicate the benefits of a product. The key is crafting the right mix of these tools to achieve sales increases and market share goals. In the second chapter, you will learn how to incorporate the effects of advertising and promotion in your sales-response model and how to identify the marketing strategy that is most likely to succeed.

Exercise 1: Model extension part 1: Dummy variables Exercise 2: Understanding dummy variables Exercise 3: The effect of display on sales Exercise 4: The effect of multiple dummies on sales Exercise 5: What about price?Exercise 6: Model extensions part 2: Dynamic variables Exercise 7: How to lag?Exercise 8: Adding lagged price effects Exercise 9: More lags Exercise 10: What's the value added?Exercise 11: How many extensions are needed?Exercise 12: Summarizing the model Exercise 13: Unnecessary predictors Exercise 14: Dropping predictors Exercise 15: Eliminating predictors

Current Exercise

A company can only be successful in the market if its products have a competitive advantage over those of its rivals. To develop an effective marketing strategy in a competitive environment, it is essential to understand the interrelationship between marketing activity and customer behavior. In this chapter, you will learn how to explain the effects of temporary price changes on customer brand choice by employing logistic and probit response models.

Exercise 1: Models for individual demand Exercise 2: Customer purchases Exercise 3: Summarizing the data Exercise 4: Competition Exercise 5: A linear probability model for beer demand Exercise 6: Logistic response models Exercise 7: A logistic model for beer demand Exercise 8: Bounded predictions Exercise 9: Average marginal effects Exercise 10: Effect plots Exercise 11: Probit response models Exercise 12: A probit model for beer demand Exercise 13: Logistic vs. probit Exercise 14: Model comparison

The main goal of response modeling is to enable marketers to not only see a payoff for their actions today, but also tomorrow. In order to view this future payoff, a simple but reliable statistical model is required. In this last chapter, you will learn how to evaluate the predictive performance of logistic response models.

Exercise 1: Model selection Exercise 2: Extending the logistic response model Exercise 3: Summarizing the model Exercise 4: The deviance principle Exercise 5: Eliminating predictors Exercise 6: Predictive performance Exercise 7: Classifications Exercise 8: Model confusion Exercise 9: ROC curves Exercise 10: Model validation Exercise 11: Subsetting Exercise 12: Model training Exercise 13: Out-of-sample testing Exercise 14: Wrapping it up