Professionals often begin their work with a design: an outline or sketch of the job to be done. Architects sketch designs for buildings. Fashion designers outline the cut and style of apparel. Psychologists design research. Research designs are plans for the execution of scientific research projects.

When conducting research, psychologists choose designs that achieve specific research goals. Let’s first look at three goals for research, and then examine the research designs that enable psychologists to achieve them.

In psychology, or any science, a researcher usually has one of three goals: description, prediction, or causal explanation.

39

DESCRIPTION. When researchers begin to study a topic, they seek careful, systematic descriptions. Descriptions establish basic facts about the topic under study. These facts, once obtained, guide subsequent theory and research.

As an example, suppose your research topic is social anxiety, which is the tendency to become highly apprehensive, worried, and self-conscious in social settings. To get started, you need basic descriptive information; your overall goal is to establish facts about social anxiety. Those facts might be answers to questions such as:

These facts alone do not enable you to accomplish much; you cannot tell who is likely to become anxious, or why, or how to help alleviate their anxiety. Nonetheless, descriptive information that answers these questions would provide a foundation for future research.

PREDICTION. When pursuing the second goal, prediction, scientists try to forecast the occurrence of events. They want to know if information available at one point in time can be used to estimate an outcome that occurs later.

When studying something, there are two types of predictions: (1) What predicts its occurrence, and (2) once it occurs, what does it, in turn, predict. We will illustrate with the social anxiety example. Two types of prediction questions are:

CAUSAL EXPLANATION. The third research goal is the ultimate goal of science: explanation (Salmon, 1989). Scientists try to explain why things happen: Why do solids melt at high temperatures? Why do offspring resemble their parents? Why do people experience social anxiety?

To answer why questions, scientists seek causal explanations. Causal explanations identify factors that directly influence outcomes; they show “how an outcome depends on other variables or factors” (Woodward, 2003, p. 6). Consider some possibilities in the case of social anxiety (i.e., where the “outcomes” are people’s varying levels of social anxiety):

40

A key aspect of causal explanation is manipulation. If a factor is manipulated and, as a result, an outcome varies, then the manipulated factor causally influenced the outcome (Woodward, 2003).

Causal explanations are valuable for two reasons. First, they advance the central goal of science: understanding of how the world works. Second, they open the door to practical applications. If you know, for example, that certain types of thoughts cause people to become anxious, then you can apply that knowledge to psychological therapy. Therapists can help people to eliminate those thoughts and replace them with ones that are less anxiety-inducing.

Importantly, prediction does not accomplish the goal of causal explanation. If, for instance, on a sunny summer day, you see people living near a beach putting boards over the windows of their homes and businesses, you can make a prediction: There is likely to be a tropical storm or hurricane. The people’s behavior predicts the storm but, of course, does not cause it. (We return to the distinction between prediction and causal explanation later in the chapter, when discussing correlational designs.)

Now that you have learned about the three goals for research, let’s turn to three research designs. The matching number is no coincidence. Each of the three designs— surveys, correlational studies, and experiments— accomplishes one of the three research goals: description, prediction, and explanation.

Researchers often want information about a large group of people. Before a national election, for example, they want to know the political opinions of voters. Getting the information might seem easy enough: Just ask the voters their opinions. But the problem is that there are so many voters. Who has time to talk to all of them?

It turns out that you don’t need to talk to all of them. Instead, you can use the survey method. A survey method is a research design in which researchers obtain descriptive information about a large group of people by getting information from a select subgroup of them (Groves et al., 2009). By selecting the subgroup carefully (as we explain below), one can be confident that the subgroup’s views are similar to those of the group as a whole.

41

Surveys can be used to collect various types of evidence (Punch, 2003); you could, for instance, survey a subgroup of people and collect biological information (e.g., their height, weight, or blood pressure). In practice, however, most survey research presents people with simple yes/no or multiple-choice questions known as survey items. A survey (i.e., the full set of items) can address any topic and be of almost any length. The “5000 Question Survey” contains a vast array of items, including the question, “When you aren’t filling out 5000 question surveys like this one, what are you doing?” (http://5000questionsur.livejournal.com/483.html). A survey designed to measure people’s satisfaction with their life contains only one item: “I would describe my satisfaction with my overall life as ________ ” (Seligson, Huebner, & Valois, 2003).

CULTURAL OPPORTUNITIES

42

POPULATION AND SAMPLE. When survey researchers conduct a study, the overall, large group of people of interest to them is the study’s population. In our earlier example, the nation’s entire group of eligible voters is the population.

The select subgroup of people contacted by the researcher is called the sample (Figure 2.1). Researchers use information from the sample to draw conclusions about the population as a whole.

In many surveys, the sample is only a small fraction of the population. Nonetheless, information from the sample proves to describe the population accurately. For instance, in presidential elections, pollsters commonly sample only a couple thousand people from a voting population of 125 million. Yet, in the last two elections, the samples predicted the outcomes precisely (Graefe et al., 2013; Panagopoulos, 2009). The key to this precision is the selection of the sample.

RANDOM SAMPLES. How do researchers decide which people from the population go into the sample? Hint: It’s a trick question.

Researchers do not personally decide who goes into a sample. Instead, the sample is chosen randomly. In a random sample, each individual’s presence in (or absence from) the sample is determined by a chance process (similar to flipping a coin or rolling dice). The chance process usually gives every member of the population an equal chance of being in the sample. For instance, to obtain a random sample of 100 of the 5000 students at a college, you would employ a chance process that gives each person a 1-in-50 chance of being selected.

Random sampling helps researchers obtain a sample that is representative. A representative sample is one whose qualities (gender, ethnicity, income, attitudes, personalities, etc.) match those of the overall population. If, for example, you are surveying students at your college where 13% are Latin American, 12% are African American, and 12% are Asian American, then a representative sample would contain approximately 13% Latin Americans, 12% African Americans, and 12% Asian Americans. A sample that is chosen randomly generally will be representative of the overall population from which it is drawn.

Even a large sample may produce inaccurate results if it is not chosen randomly. Suppose your instructor gives out a course evaluation on a day when 80 of 100 students enrolled in the course are in attendance. The sample is relatively large (80% of the population of 100 students) but is not random. The results thus may misrepresent the opinions of the class as a whole. For instance, the 20 students not in attendance might not like the course as much as the others (maybe that’s why they skipped class), but none of them are in the sample. The sample results thus could be biased; they might overestimate the population’s opinions of the course. The bias could be eliminated by sampling students randomly from the population of 100. (Among its many problems, the dorm study presented earlier failed to sample randomly.)

43

Survey research describes a population at a given point in time, thus achieving the first of the three goals for research. But to achieve the second goal, prediction, one needs a second research strategy: correlational designs.

In a correlational study (or a correlational design), researchers measure two or more quantities and determine whether higher (or lower) levels of one are associated with higher (or lower) levels of another. The measured quantities are called variables. A variable is any property that fluctuates. The property may fluctuate from person to person or from one time to another for any given person (e.g., your mood varies at different times of day). The results of a correlational study indicate the degree to which two or more variables are related.

Correlational studies commonly are conducted with large numbers of people. Researchers measure two or more variables in each individual and then ask whether, in the group as a whole, the variables are correlated. Sometimes the two variables are measured at the same point in time; for example, a researcher might measure people’s current (1) annual income and (2) level of happiness in life to see whether the two are correlated (Diener, Tay, & Oishi, 2013). In other studies, the variables are measured at different time points; for example, researchers have correlated (1) personality qualities measured in childhood with (2) professional and personal success achieved in adulthood (Shiner & Masten, 2012). (Note that, in practice, most correlational studies are not conducted with participants chosen randomly from a large population, as in survey research.)

Psychologists achieve the scientific goal of prediction, discussed earlier, through the use of correlational designs. By measuring two variables, they can determine whether one variable predicts the other. Recall the prediction question “Do higher levels of social anxiety in childhood predict higher levels of social anxiety in adulthood?” When answering this question in a correlational study, the two variables would be:

44

A researcher would measure social anxiety in a group of children and then measure social anxiety again in those same people when they reach adulthood. At that point, the research would determine whether the two variables are related, that is, if Variable 1 predicts Variable 2.

Now the question is how, exactly, to determine whether two variables are related. One option is to draw a graph. A scatterplot is a graph that displays the relation between two variables. Data points representing the measurements of the two variables “scatter” about the visual display.

Figure 2.2 shows a scatterplot for two variables: amount of rainfall and percentage of people using umbrellas. In this hypothetical study, a researcher measured both variables on each of 20 days. Each point represents the measurement of the two variables on a given day.

POSITIVE AND NEGATIVE CORRELATIONS. If you look at the graph (Figure 2.2), the results are clear. When there’s more rain, more people use umbrellas. This type of relation, in which higher levels of one variable co-occur with higher levels of the other variable, is called a positive correlation. Relations that go in the opposite direction, with higher amounts of one variable co-occurring with lower amounts of the other, are called negative correlations. You would expect, for example, that rainfall and the use of sunscreen would be correlated negatively. If the points in a scatterplot do not systematically go up or down—for instance, if they roughly form a circle—then the two variables are uncorrelated, that is, neither positively nor negatively correlated with each other.

The second option for determining whether two variables are related is to compute a number that summarizes the relation between them. The correlation coefficient (or simply correlation) is a numerical value that represents the strength of the relation between any two variables. It usually is symbolized with the letter r. The correlation is based on a formula (which can be found in this book’s Statistics Appendix) that limits the minimum and maximum correlation to −1.0 and +1.0. Correlations above zero are called positive correlations and those below zero are called negative correlations. Let’s illustrate what these numbers mean using some examples:

45

Figure 2.3 displays some scatterplots and the correlation coefficients associated with each. In our rainfall and umbrellas example from Figure 2.2, the correlation is +.91. That number confirms the impression you got from the graph: Precipitation and umbrella use are strongly positively correlated.

CORRELATION … AND CAUSATION? Does a positive correlation between two variables prove that one variable caused the other? No. In fact, “No, no, a thousand times no.” There could be other reasons for the correlation—such as coincidence. Here’s an example.

In California, the average amount of rainfall in a given month correlates positively with the number of letters in the name of the month (Figure 2.4). Short-letter months get less rain. Yet neither letters nor rainfall causes one another; the correlation is sheer coincidence. California rainfall is less common in spring and summer than fall and winter, and the spring and summer month names (e.g., May, June) happen to be shorter.

46

In our example, the conclusion is obvious; names of months can’t cause rainfall. But often the question of causality is trickier. Consider these two examples:

In general, when two variables X and Y are correlated, there are four possibilities: (1) X causes Y; (2) Y causes X; (3) some third variable affects both X and Y, and thus neither X nor Y is a cause; (4) the relation between X and Y is sheer coincidence (Figure 2.5). A correlational research design can’t distinguish one possibility from another. For this, one needs an experiment.

47

When psychologists want to identify the causes of events, they conduct experiments. In everyday language, to “experiment” means to try something new. In psychology, however, the word has a more specific meaning. An experiment is a research design in which a variable is manipulated.

Earlier, we said that manipulation is key to causal explanation (Woodward, 2003). Experiments, which manipulate variables, thus are the psychologist’s tool for achieving the goal of causal explanation.

In experiments, a researcher manipulates one or more variables to determine whether the manipulation affects other variables. The researcher’s plan for manipulating variables is the experimental design.

EXPERIMENTAL DESIGN BASICS. In the terminology of experimental design, there are two types of variables: independent and dependent variables. It’s important to remember which is which:

48

To manipulate the independent variable, the researcher creates different settings in which the dependent variable is observed. The settings are identical in all ways but one: the level of the independent variable changes. These different settings, which contain different levels of the independent variable, are called the study’s experimental conditions. Memorization times of one minute and two minutes, in our example above, would be that study’s two experimental conditions. (Table 2.1 summarizes the three research goals and three research designs we have discussed.)

The purpose of an experiment usually is to test a hypothesis. A hypothesis is a prediction about the result of a study. In our example, the researcher might predict that the number of words remembered will be higher in the two-minute memorization time condition than the one-minute condition. This prediction is the experimenter’s hypothesis.

Many experiments contain one experimental condition that receives a special name: the control group. A control group is an experimental condition that eliminates the factor or factors that vary in the other experimental conditions. In other words, the control group brings the level of the independent variable down to zero: It eliminates it.

That was a lot of terminology: independent variable, dependent variable, experimental conditions, hypothesis, control group. Let’s make these terms concrete with another example. Imagine you are interested in therapies for depression. In particular, you think that the frequency of therapy sessions (or the “intensity” of therapy; Moos & Moos, 2003) influences therapy’s effectiveness; you expect that more frequent meetings will increase effectiveness. To test your idea, you plan a study in which different people receive therapy at different frequencies—for example, once every two weeks, once a week, and twice a week. Now let’s phrase this in the terminology you just learned (Figure 2.6):

If you are tracking the details, you’re now asking, “What about the control group?” As a control group, you could create a fourth condition in which the level of the independent variable is brought down to zero: People receive no therapy. To see its value, imagine that you had run only the other three conditions and people in all three had displayed similar levels of depression at the end of therapy. What could you conclude? Maybe all three therapies were very effective, but to the same degree. Or maybe none of the three therapies worked at all. Without the control group, there is no way of knowing. But with the control group, you can see whether therapy, in general, had any effect.

49

THE POWER OF RANDOM ASSIGNMENT. Suppose you did conduct the depression study we just described. How would you decide which participant is assigned to which condition?

The answer is the same as that for random sampling: You wouldn’t decide. Nobody would decide. Each participant would be allocated to an experimental condition through a random procedure. Random assignment is a process in which participants are allocated to conditions of an experiment on an entirely chance basis.

An experiment, then, has two defining characteristics: (1) more than one experimental condition, and (2) the random assignment of research participants to experimental conditions.

Random assignment serves a critical function. It reduces the possibility of a potential experimental confound, a factor other than the independent variable that might create differences between experimental conditions. One potential confound is differences between people. If different types of people are assigned to the different conditions of an experiment, one cannot determine the true cause of differences between the conditions, which may have resulted from (a) the experimental conditions or (b) the people assigned to the experimental conditions (see Research Toolkit). Random assignment solves this problem. It assures that, on average, the characteristics of the people assigned to the different conditions of an experiment will not differ at the outset of the study.

To see how this works, imagine running an experiment on memory with three experimental conditions and 90 participants. Some of the 90 might have exceptionally good memory and others might have unusually poor memory. Will these differences between people affect your results? If you use random assignment, they probably won’t. By chance—thanks to random assignment—a few of the people with exceptionally good and poor memory are likely to end up in each of your three experimental conditions. As a result, on average, the three experimental groups will not differ at the outset of the study. If the groups differ at the end of the study, when the dependent variable is measured, you thus can be confident that the differences are due to your experimental manipulation.

RESEARCH TOOLKIT

51

Now that you have learned about random assignment and the elimination of confounds, you can answer the question in this chapter’s opening story. Recall that men walking on a frightening, physically arousing bridge showed more sexual attraction toward a woman than did men walking on a safe, solid bridge. Yet the researchers did not find the results to be convincing. Why not? They knew their study lacked random assignment and thus might have a confound. The men had not been randomly assigned to one bridge or the other; they had personally chosen to walk across a particular bridge. As a result, the variable (type of bridge) may have been confounded with men’s personality styles. For example, men who opted for the frightening bridge may have been more bold, adventurous, and thrill-seeking—and maybe that is why they differed from others in sexual imagery and the tendency to call the female experimenter. Subsequent studies using random assignment (Dutton & Aron, 1974; Meston & Frohlich, 2003) demonstrated that fearful arousal can, in fact, increase sexual attraction. Those studies were much more convincing than research findings from a study lacking random assignment.