Get a free consultation with a marketer marketolog@adbest.ru
The classification tree method is suitable for dependent variables of any type and a large number of variables. Based on differences in the response categories of objects, classification trees are constructed with nodes that show the percentage of the response of interest relative to the sample mean, at the root node of the tree. The tree allows us to find nodes where the parameters of interest are significantly below or above the average, and construct a payoff table. Thus, for example, we can understand that by showing ads only to men aged 35-46, for a given product, who live in Moscow, we will receive 80 percent more visits to the site, relative to the average number of visits for the overall sample..
The limitations of this method include the need to use sufficiently large data samples and correctly set the scaling intervals for quantitative variables in the settings.
This analysis allows us to identify and use only those criteria when selecting initial conditions that maximize results, while eliminating parameters that negatively impact results. In advertising, classification trees are used to separately assess the impact of user interests and audience socio-demographic characteristics on advertising response.
Below, in Figures 14 and 15, are examples of the payoff table for nodes that showed abnormally high values for site visit depth indicators and the classification tree itself.
Cost ofCHAID analysis
Figure 15. Classification tree by site pages
Our projects
