Learning From Data

An Introduction To Statistical Reasoning

Arthur Glenberg, Matthew Andrzejewski

• Verlag: Routledge
• Jahr: 2007
• ISBN:
  9781136676628
• Auflage: 3
• Thema: Education
• Thema: Education Theory & Practice

Learning from Data focuses on how to interpret psychological data and statistical results. The authors review the basics of statistical reasoning to helpstudents better understand relevant data that affecttheir everyday lives.

Numerous examples based on current research and events are featured throughout.To facilitate learning, authors Glenberg and Andrzejewski:

• Devote extra attention to explaining the more difficult concepts and the logic behind them
• Use repetition to enhance students' memories with multiple examples, reintroductions of the major concepts, and a focus on these concepts in the problems
• Employ a six-step procedure for describing all statistical tests from the simplest to the most complex
• Provide end-of-chapter tables to summarize the hypothesis testing procedures introduced
• Emphasizes how to choose the best procedure in the examples, problems and endpapers
• Focus on power with a separate chapter and power analyses procedures in each chapter
• Provide detailed explanations of factorial designs, interactions, and ANOVA to help students understand the statistics used in professional journal articles.

The third edition has a user-friendly approach:

• Designed to be used seamlessly with Excel, all of the in-text analyses are conducted in Excel, while the book's downloadable resources contain files for conducting analyses in Excel, as well as text files that can be analyzed in SPSS, SAS, and Systat
• Two large, real data sets integrated throughout illustrate important concepts
• Many new end-of-chapter problems (definitions, computational, and reasoning) and many more on the companion CD
• Online Instructor's Resources includes answers to all the exercises in the book and multiple-choice test questions with answers
• Boxed media reports illustrate key concepts and their relevance to realworld issues
• The inclusion of effect size in all discussions of power accurately reflects the contemporary issues of power, effect size, and significance.

Learning From Data, Third Edition is intended as a text for undergraduate or beginning graduate statistics courses in psychology, education, and other applied social and health sciences.

CHAPTER 1

Why Statistics?

Variability

Sources of Variability

Variables and Constants

Populations and Samples

Statistical Populations

The Problem of Large Populations

Samples

Descriptive and Inferential Statistical Procedures

Descriptive Statistical Procedures

Inferential Statistical Procedures

Measurement

Considering Measurement in a Social and Political Context

Differences Among Measurement Rules

Properties of Numbers Used as Measurements

Types of Measurement Scales

Importance of Scale Types

Using Computers to Learn From Data

What Statistical Analysis Programs Can Do for You

What the Programs Cannot Do for You

Summary

Exercises

Terms

Questions

There are many ways to learn about the world and the people who populate it. Learning can result from critical thinking, asking an authority, or even from a religious experience. However, collecting data (that is, measuring observations) is the surest way to learn about how the world really is.

Unfortunately, data in the behavioral sciences are messy. Initial examination of data reveals no clear facts about the world. Instead, the data appear to be nothing but an incoherent jumble of numbers. To learn about the world from data, you must first learn how to make sense out of data, and that is what this textbook will teach you. Statistical procedures are tools for learning about the world by learning from data.

To help you to understand the power and usefulness of statistical procedures, we will explore two real (and important!) data sets throughout the course of the book. One of the data sets is courtesy of Professor Timothy Baker at the University of Wisconsin Center for Tobacco Research and Intervention (which we will call the Smoking Study). The data were collected to investigate several questions about smoking, addiction, withdrawal, and how best to quit smoking. The data set consists of a sample of 608 people who wanted to quit smoking. These people were randomly assigned (see Chapter 14 for the benefits of random assignment) to three groups. The participants in one group were given the drug bupropion SR (Zyban) along with nicotine replacement gum. In a second group, the participants were given the bupropion along with a placebo gum that did not contain any active ingredients. The final group received both a placebo drug and a placebo gum. The major question of interest is whether people are more successful in quitting smoking when the the active gum is added to the bupropion. These data are exciting for a couple of reasons. First, given the tremendous social cost of cigarette smoking, we as a society need to figure out how to help people overcome this addiction, and these data do just that. Second, the study included measurements of about 30 other variables to help answer ancillary questions. For example, there are data on how long people have smoked and how much they smoked; data on health factors and drug use; and demographic data such as gender, ethnicity, age, education, and height. These variables are described more fully within the Excel and SPSS data files on the CD that comes with this book and in Appendix A. The statistical tools you will learn about will give you the opportunity to explore these data to the fullest extent possible. You can ask important questions—some that may never have been asked before—such as whether drug use affects people’s ability to quite smoking, and you can get the answers. In addition, these data will be used to illustrate various statistical procedures, and they will be used in the end-of-chapter exercises.

The second data set is courtesy of Professors Janet Hyde and Marilyn Essex of the University of Wisconsin-Madison. The data set is a subset of the data from the Wisconsin Maternity Leave and Health Project and the Wisconsin Study of Families and Work (we will refer to it as the Maternity Study). This project was designed to answer questions about how having a baby affects family dynamics such as marital satisfaction, and how various factors affect child development. The data set consists of measurements of 26 variables for 244 families. Some of these variables are demographic, such as age, education, and family income. Marital satisfaction was measured separately for mothers and fathers both before the child was born (during the 5th month of pregnancy) and at three times after the birth (1, 4, and 12 months postpartum). There are also data on how much the mother worked outside the house and how equally household tasks were divided among the mothers and fathers. Finally, there are eight measures of the quality of mother-child interactions at 12 months after birth, and three measures of child temperament (for example, hyperactivity) measured when the child was 4.5 years old. These variables are described more fully on the CD that comes with this book and in Appendix B. As with the smoking data, you are free to use these data to answer important questions, such as whether the amount of time that a mother works affects child development.

This chapter introduces a number of topics that are basic to statistical analyses. We begin with a discussion of variability, the cause of messy data, and move on to the distinctions between population and sample, descriptive and inferential statistics, and types of measurement found in the behavioral sciences.

VARIABILITY

The first step in learning how to learn from data is to understand why data are messy. A concrete example is useful. Consider the CESD (Center for Epidemiologic Studies Depression) scores from the Smoking Study (see Appendix A). Each participant rated 20 questions such as “I felt lonely” using a rating of 0 (rarely or none of the time during the past week) to 3 (most of the time during the past week). The score is the sum of the ratings for the participant. For the 601 participants for whom we have CESD scores, the scores range from 0 to 23. About a quarter of the scores are below 2, but another quarter are above 9. These data are messy in the sense that the scores are very different from one another.

Variability is the statistical term for the degree to which scores (such as the depression scores) differ from one another.

Chapter 3 presents statistical procedures for precisely measuring the variability in a set of scores. For now, only an intuitive understanding of variability is needed. When the scores differ from one another by quite a lot (such as the depression scores), variability is high. When the scores have similar values, variability is low. When all the scores are the same, there is no variability.

Sources of Variability

It is easy enough to see that the CESD data are variable, but why are they variable? In general, variability arises from several sources. One source of variability is individual differences: Some smokers are more depressed than others; some have difficulty reading and understanding the items on the test; some smokers’ answers on the inventory are more honest than the answers of other smokers. There are as many potential sources of variability due to individual differences as there are reasons for why one person differs from another in intelligence, personality, performance, and physical characteristics.

Another source of variability is the procedure used in collecting the data. Perhaps some of the smokers were more rushed than others; perhaps some were tested at the end of the day and were more tired than others. Any change in the procedures used for collecting the data can introduce variability. Finally, some variability may be due to conditions imposed on the participants, such as whether they are taking the placebo gum.

Variables and Constants

Variability does not occur only in textbook examples; it is characteristic of all data in the behavioral sciences. Whenever a behavioral scientist collects data, whether on the incidence of depression, the effectiveness of a psychotherapeutic technique, or the reaction time to respond to a stimulus, the data will be variable; that is, not all the scores collected will be the same. In fact, because data are variable, collecting data is sometimes referred to as measuring a variable (or a random variable).

A variable is a measurement that changes from one observation to the next.

CESD is a variable because it changes from one smoker (observation) to the next. “Effectiveness of a psychotherapeutic technique” is another example of a variable, because a given technique will be more effective for some people than for others.

Variables should be distinguished from constants.

Constants are measurements that stay the same from one observation to the next.

The boiling point of pure water at sea level is an example of a constant. It is always 100 degrees Centigrade. Whether you use a little water or a lot of water, whether the water is encouraged to boil faster or not, no matter who is making the observation (as long as the observer is careful!), the water always boils at the same temperature. Another constant is Newton’s gravitational constant, the rate of acceleration of an object in a gravitational field (whether the object is large or small, solid or liquid, and so on).

Many of the observations made in the physical sciences are observations of constants. Because of this, it is easy for the beginning student in the physical sciences to learn from data. A single careful observation of a constant tells the whole story.

You may be surprised to learn that there is not one constant in all of the behavioral sciences. There is no such thing as the effectiveness of a psychotherapeutic technique, or the depression score, because measurements of these variables change from person to person. In fact, because what is known in the behavioral sciences is always based on measuring variables, even the beginning student must have some familiarity with statistical procedures to appreciate the body of knowledge that comprises the behavioral sciences and the limitations inherent in that body of knowledge. In case you were wondering, this is why you are taking an introductory statistics course, and your friends majoring in the physical sciences are not.

The concept of variability is absolutely basic to statistical reasoning, and it will motivate all discussions of learning from data. In fact, the remainder of this chapter introduces concepts that have been developed to help cope with variability.

POPULATIONS AND SAMPLES

The psychologists studying addiction might be interested in the CESD scores of the specific smokers from whom they collected data. However, it is likely that they are interested in more than just those individuals. For example, they may be interested in the incidence of depression among all smokers in Wisconsin, or all smokers in the United States, or even all smokers in the world. Because depression is a variable that changes from person to person, the specific observations cannot reveal everything the researchers might want to know about all of these depression scores.

Statistical Populations

A statistical population is a collection or set of measurements of a variable that share some common characteristic.

One example of a population is the set of CESD scores of all smokers in Wisconsin. These scores are measurements of a variable (CESD), and they have the common characteristic of being from a particular group of people: smokers in Wisconsin. A different statistical population consists of the CESD scores for smokers in the United States. And, a very different population consists of the marital satisfaction scores for new mothers who work fulltime outside of the home. The point is t...