Survey data lends itself well to a process known as key driver analysis. Identifying and analyzing key drivers can help marketers and advertisers find answers to questions such as: What drives my customer to switch to another brand? What contributes to a consumer's propensity to purchase my product? Which consumer group is most satisfied with our services?
As in any research, the initial step is identifying questions the survey is designed to answer. The marketer must determine if the research objective is predictive, explanatory or descriptive (rare for a survey). What if both types of objectives are important?
Time Required: One Week
It's All About Relationship
A wide array of dependent and independent variables can be studied through key driver analysis and, typically, the analysis is directed at one or more dependent variables and multiple independent variables. It is the statistically significant effect of the independent variable on the dependent variable that is the focus of the research. On the one hand, there is a strategic characteristic (like market share) of interest to the client. On the other, there is a set of performance indicators or descriptive attributes that are believed to be related to the strategic characteristic in some way.
The relevant variables chosen and the analytical method selected for key driver analysis are largely a function of the research objective: explanation, prediction, description.
If an explanation is the goal, the independent variables selected are believed to influence variation observed in the dependent variable. The independent variables should be actionable, too. For example, overall satisfaction with customer service (the dependent variable) is likely related to wait-time, the simplicity of returns, and refund policy (all independent variables and responsive to change, or action).
If prediction is the research objective, independent variables are sought that show promise for predicting an outcome. In this instance, the independent variables don't have to be actionable. The goal of predictive research isn't to change the dependent variable, but to predict something about it. For instance, key driver analysis might be designed to predict recidivism after participation in a smoking prevention program, but the researchers might also examine a different set of independent variables believed to improve the success rate of their smoking cessation program.
It Is Survey Friendly
Brand attributes often fall into one of three categories: Satisfaction, agreement, or performance ratings. A variety of scales may be used to record survey respondents ratings or ranking of attributes in these categories. The most common rating scale is the Likert, which is easily applied to satisfaction and agreement statements. When survey respondents rate many attributes of a product or service or attributes across several brands, they can check a box for "yes," with the resultant data coded 1/0. This binary data is easily converted for purposes of statistical analysis.
Different Market Segments
Market segmentation research indicates that different key drivers may be important in different markets and that some key drivers may be important across all market segments. Key driver analysis can simplify survey design since an attribute can be asked only once in a survey, but the resultant data can be filtered into different "cuts" or tranches that reflect discrete consumer groups. For example, cuts can reflect demographics, age, gender, socioeconomic status, income, or educational attainment levels.
A variety of analytical techniques can be used to perform a key driver analysis. Some dependent variables are categorical, not scaled, and so cannot be analyzed by linear regression. Instead, linear discriminant analysis or logistic regression are used. Categorical variables can be used in surveys with both predictive and explanation objectives. Customer satisfaction or loyalty surveys often employ categorical values that indicate, for instance, the status of the customer relationship (active/non-active).
Linearity: One More Thing to Consider
A key driver is an attribute with a statistically significant relationship to a desired outcome or strategic characteristic. The independent variable is considered to be linear if it has a straight-line relationship with the dependent variable. An example would be price elasticity - as the price of product changes, a linear pattern of sales volume occurs in response to these changes. Unless a very high level of predictive validity is required, in a well-designed study, linear data can fairly represent non-linear data, without having to resort to more advanced techniques.
Many software packages are designed to carry out the statistical processes needed for key driver analysis. Quirk's magazine publishes software reviews.
The two listed here span the range of available options from the most basic applications designed to work as Microsoft Excel Add-ins to comprehensive platforms such as SPSS.
ALLSTAT is an inexpensive data analysis and statistical solution for Microsoft Excel.
Because key driver analysis is efficient and scalable, it helps to maintain the budgetary and resource boundaries of survey design and analysis. Existing brand drivers - say, that are familiar to clients who annually take a survey - can be used within existing survey frameworks; surveys that employ key driver analysis don't need to be made longer or more complicated. Client-facing questionnaires need not change noticeably to accommodate key driver analysis. A story that uses key driver analysis is understandable and lends itself to a visual display of the data for presentation.
Quirk's Market Research Review publishes articles on a wide range of market research topics. Their series on Data Use and Research Techniques and Trends are particularly useful for researchers interested in the nuts and bolts of survey research.
- Quirk's Article #20010104 - A Survey of Analysis Methods by Rajan Sambandam (of the Response Center in Fort Washington, PA)
- Quirk's Article #20010297 - Key Driver Analysis by Micheal Lieberman (of Multivariate Solutions, New York