1.4 Introduction to Hypothesis Testing

When John Snow suggested that the pump handle be removed from the Broad Street well, he was testing his idea that an independent variable (contaminated well water) led to a dependent variable (deaths from cholera). Behavioral scientists use research to test ideas through a specific statistics-based process called hypothesis testing. More formally, hypothesis testing is the process of drawing conclusions about whether a particular relation between variables is supported by the evidence. Typically, we examine data from a sample to draw conclusions about a population. There are many ways to conduct research. In this section, we discuss the process of determining the variables, two ways to approach research, and two experimental designs.

Determining what breed of dog you most resemble might seem silly; however, adopting a dog is a very important decision. Can an online quiz such as the Animal Planet “Dog Breed Selector” help (http://animal.discovery.com/breed-selector/dog-breeds.html, 2013)? We could conduct a study by having 30 people choose a type of dog to adopt and have another 30 people let the selector dictate their choice. We would then have to decide how to measure the outcome.

An operational definition specifies the operations or procedures used to measure or manipulate a variable. We could operationalize a good outcome with a new dog in several ways. Did you keep the dog for more than a year? On a rating scale of satisfaction with your pet, did you get a high score? Does a veterinarian give a high rating to your dog’s health?

Do you think a quiz would lead you to make a better choice in dogs? You might hypothesize that the quiz would lead to better choices because it makes you think about important factors in dog ownership, such as outdoor space, leisure time, and your tolerance for dog hair. You already carry many hypotheses like these in your head. You just haven’t bothered to test most of them yet, at least not formally. For example, perhaps you believe that North Americans use banking machines (ATMs) faster than Europeans do, or that smokers simply lack the willpower to stop. Maybe you are convinced that the parking problem on your campus is part of a conspiracy by administrators to make your life more difficult.

In each of these cases, as shown in the accompanying table, we frame a hypothesis in terms of an independent variable and a dependent variable. The best way to learn about operationalizing a variable is to experience it for yourself. So propose a way to measure each of the variables identified in Table 1-2. We’ve given you a start with “continent”—North America versus Europe (easy to operationalize)—and how bad the parking problem is (more difficult to operationalize).

Conducting Experiments to Control for Confounding Variables

Once we have decided how to operationalize the variables, we can conduct a study and collect data. There are several ways to approach research, including experiments and correlational research. A correlation is an association between two or more variables. In Snow’s cholera research, it was the idea of a systematic co-relation between two variables (the proximity to the Broad Street well and the number of deaths) that saved so many lives. A correlation is one way to test a hypothesis, but it is not the only way. Researchers usually prefer to conduct an experiment rather than a correlational study because it is easier to interpret the results of experiments.

The hallmark of experimental research is random assignment. With random assignment, every participant in the study has an equal chance of being assigned to any of the groups, or experimental conditions, in the study. And an experiment is a study in which participants are randomly assigned to a condition or level of one or more independent variables. Random assignment means that neither the participants nor the researchers get to choose the condition. Experiments are the gold standard of hypothesis testing because they are the best way to control confounding variables. Controlling confounding variables allows researchers to infer a cause–effect relation between variables, rather than merely a systematic association between variables. Even when researchers cannot conduct a true experiment, they include as many of the characteristics of an experiment as possible. The critical feature that makes a study worthy of the descriptor experiment is random assignment to groups.

Experiments achieve equality between groups by randomly assigning participants to different levels, or conditions, of the independent variable. Random assignment controls the effects of personality traits, life experiences, personal biases, and other potential confounds by distributing them across each condition of the experiment to an equivalent degree.

EXAMPLE 1.1

It is difficult to control confounding variables, so let’s see how random assignment helps to do that. You might wonder whether the hours you spend playing Angry Birds or Call of Duty are useful. A team of physicians and a psychologist investigated whether video game playing (the independent variable) leads to superior surgical skills (the dependent variable). They reported that surgeons with more video game playing experience were faster and more accurate, on average, when conducting training drills that mimic laparoscopic surgery (a surgical technique that uses a small incision, a small video camera, and a video monitor) than surgeons with no video game playing experience (Rosser et al., 2007).

In the video game and surgery study, the researchers did not randomly assign surgeons to play video games or not. Rather, they asked surgeons to report their video game playing histories and then measured their laparoscopic surgical skills. Can you spot the confounding variable? People may choose to play video games because they already have the fine motor skills and eye–hand coordination necessary for surgery, and they enjoy using their skills by playing video games. If that is the case, then, of course, those who play video games will tend to have better surgical skills—they already did before they took up video games!

MASTERING THE CONCEPT

1.5: When possible, researchers prefer to use an experiment rather than a correlational study. Experiments use random assignment, which is the only way to determine whether one variable causes another.

It would be much more useful to set up an experiment that randomly assigns surgeons to one of the two levels of the independent variable: (1) play video games or (2) do not play video games. Random assignment assures us that our two groups are roughly equal, on average, on all the variables that might contribute to excellent surgical skills, such as fine motor skills, eye–hand coordination, and experience playing other video games. Random assignment diminishes the effects of all these potential confounds. Consequently, random assignment to groups increases our confidence that the two groups were similar, on average, on aptitude for laparoscopic surgery prior to this experiment. (Figure 1-2 visually clarifies the difference between self-selection and random assignment. We explore more specifically how random assignment is implemented in Chapter 5.) If we use random assignment and the “play video games” group has better average laparoscopic surgical skills after the experiment than the “do not play video games group,” then we are more confident in the conclusion that playing video games caused the better surgical skills.

12

Figure 1-2

Self-Selected into or Randomly Assigned to One of Two Groups: Video Game Players Versus Non–Video Game Players This figure visually clarifies the difference between self-selection and random assignment. The design of the first study does not answer the question “Does playing video games improve laparoscopic surgical skills?”

Indeed, many researchers have used experimental designs to explore the causal effects of video game playing. They have found both positive effects, such as improved spatial skills following action games (Feng, Spence, & Pratt, 2007), and negative effects, such as increased hostility after playing violent games with lots of blood (Bartlett, Harris, & Bruey, 2008).

Between-Groups Design versus Within-Groups Design

Experimenters can create meaningful comparison groups in several ways. However, most studies have either a between-groups research design or a within-groups (also called a repeated-measures) research design.

A between-groups research design is an experiment in which participants experience one and only one level of the independent variable. In some between-groups studies, the different levels of the independent variable serve as the only relevant distinction between two (or more) groups that otherwise have been made equivalent through random assignment. An experiment that compares a control group (such as people randomly assigned not to play video games) with an experimental group (such as people randomly assigned to play video games) is an example of a between-groups design.

A within-groups research design is an experiment in which all participants in the study experience the different levels of the independent variable. An experiment that compares the same group of people before and after they experience a level of an independent variable, such as video game playing, is an example of a within-groups design. The word within emphasizes that if you experience one condition of a study, then you remain within the study until you experience all conditions of the study.

Many applied questions in the behavioral sciences are best studied using a within-groups design. This is particularly true of long-term (often called longitudinal) studies that examine how individuals and organizations change over time, or studies involving a naturally occurring event that cannot be duplicated in the laboratory. For example, we obviously cannot randomly assign people to either experience or not experience a hurricane. However, we could use nature’s predictability to anticipate hurricane season, collect “before” data, and then collect data once again “after” people experience a hurricane. Such a before/after study is one version of a within-groups design.

Figure 1-3

Correlation Between Aggression and Playing Video Games This graph depicts a relation between aggression and hours spent playing video games for a study of 10 fictional participants. The more one plays video games, the higher one’s level of aggression tends to be.

Correlational Research

Often, we cannot conduct an experiment because it is unethical or impractical to randomly assign participants to conditions. Snow’s cholera research, for example, did not use random assignment; he could not randomly assign some people to drink water from the Broad Street well. His research design was correlational, not experimental.

In correlational studies, we do not manipulate either variable. We merely assess the two variables as they exist. For example, it would be difficult to randomly assign people to either play or not play video games over several years. However, we could observe people over time to see the effects of their actual video game usage. Möller and Krahé (2009) studied German teenagers over a period of 30 months and found that the amount of video game playing when the study started was related to aggression 30 months later. Although these researchers found that video game playing and aggression are related, they do not have evidence that playing video games causes aggression (as shown in Figure 1-3). As we will discuss in Chapter 13, there are always alternative explanations in a correlational study; don’t be too eager to infer causality just because two variables are correlated.