Introducing a Third Variable - Interactive Causal Relationships
Introducing a Third Variable - Interactive Causal Relationships
The introduction of a third variable also might suggest some interactions. Sales (S) might be influenced by distribution (D), but only if the packagingand display (P) is appealing and capable of attracting the shoppers' attention.
Interactions, which are present when associations between two variables are affected by the presence or absence of a third variable, were explored in the context of factorial experimental designs (recall Figures 10-1 and 14-5). In the context of cross-tabulations this possibility is explored fat development cross-tabs for subgroups defined by the third variable.
Suppose that a segmentation study is attempting to identify the heaij users of a certain library. In Figure 15-3 it is shown that 28 percent of those below age 40 use it, while only 20 percent of those over age 40 use it. 7: explore the relationship further the variable of sex is introduced and the analysis is repeated for men and for women. The age and usage relationship is much stronger for men than it is for women. Thus, the addition of the sex variable has permitted the analysis to be more refined. As Figure 15-1 shows, the two lines are not parallel, since an interaction is present. This interactive causal relationship is depicted as follows:
In another case the relationship appeared to be zero until a third van-able was introduced. In two age groups, 64 percent of the sample listened to classical music. However, among the college-educated people, the older group tended to listen more, while among noncollege-educated people, the younger group tended to have a higher percentage of listeners. Again, the addition of the interactive variable, education, helped refine the analysis
AN INTERACTIVE EFFECT ON LISTENING HABITS
Age-
Listening habits
Education-
The conclusion occasionally can be clouded by the existence of a portion of the sample that is not responsive to one of the variables. For example, in a study of the impact of advertising exposure, it was found that product users, the bulk of the sample, were not responsive to the advertising. When only nonusers were examined, this hypothesized effect emerged:
AN INTERACTIVE EFFECT ON IMPACT ON AUDIENCE Advertising -
Impact on audience
Product Usage -
In taste tests, there may be a substantial group who simply are not taste-sensitive. When these people are removed from the analysis, the conclusions can be more pronounced.
The soundest way to identify interactions is to employ theory, previous findings, and common sense. However, when there are many variables involved and theory is underdeveloped, it is useful to identify interactions that are present in the study without performing all conceivable cross-tabs
(sometimes literally infeasible). An appropriate method is AID, to be described in Chapter 20.
Again, a combination of relationships can exist. For example, an interactive effect might be found in addition to an independent effect. Perhaps distribution (D) will influence sales (S), but its influence will be larger when the packaging and display (P) are effective.
The introduction of a third variable also might suggest some interactions. Sales (S) might be influenced by distribution (D), but only if the packagingand display (P) is appealing and capable of attracting the shoppers' attention.
Interactions, which are present when associations between two variables are affected by the presence or absence of a third variable, were explored in the context of factorial experimental designs (recall Figures 10-1 and 14-5). In the context of cross-tabulations this possibility is explored fat development cross-tabs for subgroups defined by the third variable.
Suppose that a segmentation study is attempting to identify the heaij users of a certain library. In Figure 15-3 it is shown that 28 percent of those below age 40 use it, while only 20 percent of those over age 40 use it. 7: explore the relationship further the variable of sex is introduced and the analysis is repeated for men and for women. The age and usage relationship is much stronger for men than it is for women. Thus, the addition of the sex variable has permitted the analysis to be more refined. As Figure 15-1 shows, the two lines are not parallel, since an interaction is present. This interactive causal relationship is depicted as follows:
In another case the relationship appeared to be zero until a third van-able was introduced. In two age groups, 64 percent of the sample listened to classical music. However, among the college-educated people, the older group tended to listen more, while among noncollege-educated people, the younger group tended to have a higher percentage of listeners. Again, the addition of the interactive variable, education, helped refine the analysis
AN INTERACTIVE EFFECT ON LISTENING HABITS
Age-
Listening habits
Education-
The conclusion occasionally can be clouded by the existence of a portion of the sample that is not responsive to one of the variables. For example, in a study of the impact of advertising exposure, it was found that product users, the bulk of the sample, were not responsive to the advertising. When only nonusers were examined, this hypothesized effect emerged:
AN INTERACTIVE EFFECT ON IMPACT ON AUDIENCE Advertising -
Impact on audience
Product Usage -
In taste tests, there may be a substantial group who simply are not taste-sensitive. When these people are removed from the analysis, the conclusions can be more pronounced.
The soundest way to identify interactions is to employ theory, previous findings, and common sense. However, when there are many variables involved and theory is underdeveloped, it is useful to identify interactions that are present in the study without performing all conceivable cross-tabs
(sometimes literally infeasible). An appropriate method is AID, to be described in Chapter 20.
Again, a combination of relationships can exist. For example, an interactive effect might be found in addition to an independent effect. Perhaps distribution (D) will influence sales (S), but its influence will be larger when the packaging and display (P) are effective.
Comments
Post a Comment