Introducing a Third Variable
Introducing a Third Variable
In the HMO study it was found that the greatest interest in the proposed HMO was among the low-income respondents (see Figure 13-3 in Chapter 13). An appropriate question concerns why this association was found. Perhaps it was because higher-income people are more satisfied with their existing medical programs. They may be satisfied because they can afford good health care or because they eat better and are healthier. Or perhaps the interest in the HMO by the low-income respondents might be because the low-income people are students, who have needs and characteristics other than income that make them want the HMO. In fact, maybe the association between income and intentions to enroll has nothing to do with income but was just due to occupation.
Data analysis, particularly in descriptive studies, should be guided by such inquiries. Each analysis and potential analysis should be exposed to penetrating questions. The process involves mental effort, attempting to generate and explore hypotheses that are as rich and sophisticated as possible. It should not be satisfied with findings that are naive and simple if theory, common sense, and directed analysis can show otherwise.
Hopefully as much of this mental effort as possible will be exerted Ic r_£. before the data analysis starts. Recall the discussion in Chapter 2 about developing research questions with accompanying hypotheses that are as specific as possible. It is rare in a descriptive study, however, to have all the hypotheses (or even all the research questions) fully developed before the data analysis stage. Thus, the process needs to continue throughout the data analysis. Data analysis is not a set of computer techniques that pro1.:: I a nice output if someone pushes a button.
A primary goal of this chapter is to open the reader's mind to the possible logical explanation of an association (or the lack of an association found between two variables. Hopefully a richer and more sophisticated thought process will emerge. Methodologically, the chapter will introduce a third variable into the analysis. In Chapter 13, associations between two variables were discussed. In this chapter, relationships among three variables will be considered. Thus, this is the first of the "multivariate analysis chapters. Many of the concepts and issues of multivariate analysis can be considered best in the three-variables context. Conceptually, this chapter will deal with causal relationships and with the problems of inferring causality in a descriptive study where experimental controls have not bee-employed.
Much of marketing research is concerned with identifying and understanding causal relationships. What influence will a price change have cr. attendance? Do lifestyle patterns affect the use of regional shopping centers? The problem of research design and data analysis is to provide e dence relevant to making judgments about causal relationships.
As the discussions on causal inference presented in Chapter 10 indicated, there are three types of evidence relevant to evaluating causal relationships:
• Evidence that a strong association exists between an action and an observed outcome
• Evidence that the action preceded the outcome
• Evidence that there is no strong competing explanation for the relationship
The strongest type of evidence comes from experimental design studies where competing explanations for associations are reduced or eliminated by the design. The question is: how can competing explanations be identifier: and explored when experimental controls are not present? The first step in addressing that question is to provide an overview of the type of competing explanations that should be considered.
In the HMO study it was found that the greatest interest in the proposed HMO was among the low-income respondents (see Figure 13-3 in Chapter 13). An appropriate question concerns why this association was found. Perhaps it was because higher-income people are more satisfied with their existing medical programs. They may be satisfied because they can afford good health care or because they eat better and are healthier. Or perhaps the interest in the HMO by the low-income respondents might be because the low-income people are students, who have needs and characteristics other than income that make them want the HMO. In fact, maybe the association between income and intentions to enroll has nothing to do with income but was just due to occupation.
Data analysis, particularly in descriptive studies, should be guided by such inquiries. Each analysis and potential analysis should be exposed to penetrating questions. The process involves mental effort, attempting to generate and explore hypotheses that are as rich and sophisticated as possible. It should not be satisfied with findings that are naive and simple if theory, common sense, and directed analysis can show otherwise.
Hopefully as much of this mental effort as possible will be exerted Ic r_£. before the data analysis starts. Recall the discussion in Chapter 2 about developing research questions with accompanying hypotheses that are as specific as possible. It is rare in a descriptive study, however, to have all the hypotheses (or even all the research questions) fully developed before the data analysis stage. Thus, the process needs to continue throughout the data analysis. Data analysis is not a set of computer techniques that pro1.:: I a nice output if someone pushes a button.
A primary goal of this chapter is to open the reader's mind to the possible logical explanation of an association (or the lack of an association found between two variables. Hopefully a richer and more sophisticated thought process will emerge. Methodologically, the chapter will introduce a third variable into the analysis. In Chapter 13, associations between two variables were discussed. In this chapter, relationships among three variables will be considered. Thus, this is the first of the "multivariate analysis chapters. Many of the concepts and issues of multivariate analysis can be considered best in the three-variables context. Conceptually, this chapter will deal with causal relationships and with the problems of inferring causality in a descriptive study where experimental controls have not bee-employed.
Much of marketing research is concerned with identifying and understanding causal relationships. What influence will a price change have cr. attendance? Do lifestyle patterns affect the use of regional shopping centers? The problem of research design and data analysis is to provide e dence relevant to making judgments about causal relationships.
As the discussions on causal inference presented in Chapter 10 indicated, there are three types of evidence relevant to evaluating causal relationships:
• Evidence that a strong association exists between an action and an observed outcome
• Evidence that the action preceded the outcome
• Evidence that there is no strong competing explanation for the relationship
The strongest type of evidence comes from experimental design studies where competing explanations for associations are reduced or eliminated by the design. The question is: how can competing explanations be identifier: and explored when experimental controls are not present? The first step in addressing that question is to provide an overview of the type of competing explanations that should be considered.
Comments
Post a Comment