This article explains faster ways to do driver analysis, which helps identify what factors matter most but can be very slow because it tests many combinations of variables. It focuses on a method called component order of addition (COA), introduced by Dave Lyon, that greatly reduces the number of calculations needed. The author compares COA to other shortcuts and finds that COA is both fast and very accurate. It also works well for different types of importance measures. Overall, COA is a practical way to get results similar to complex methods without the heavy computational cost.

The Q-sort method is a simple, effective alternative to traditional rating scales and an easier option than MaxDiff (since it doesn’t involve an experimental design) for evaluating preferences. It combats straightlining by forcing respondents to use ratings (e.g., 1 to 5) in a prescribed way across items, producing quasi-normal rating distributions. Efficient and intuitive, Q-sort works best with 8-18 items. Enhancements include anchored Q-sort (adding absolute evaluations) and advanced analysis via HB-MNL or LC-MNL, which enable stronger individual estimates and probabilistic modeling. Though not as generally robust, Q-sort delivers near-MaxDiff quality insights with fewer clicks and reduced cognitive burden for respondents.

This paper covers rating scales commonly used in driver analysis, both for the overall measure and for the attribute ratings. Where possible we base our recommendations on research we (e.g., research we've done on customer satisfaction and loyalty scales, brand image scales) and others (e.g., purchase intention ratings) have done plus what we see commonly used by our clients.

Situational choice experiments (SCE) resemble choice-based conjoint experiments, but (a) they have different experimental design requirements and (b) they ask for a different type of response from survey respondents. These differences lead to a statistical model that differs from the conditional multinomial logit typically used in conjoint experiments. After a brief review and taxonomy of choice models and choice experiments, we illustrate the process of executing a SCE, tracing the process from design to data formatting to analysis and reporting.

A type of choice modeling we do that's unique to pharmaceutical research is the patient chart study. These involve modeling physicians' prescribing decisions using data taken from patients' medical charts. Patient charts provide rich data about individual patients, including their demographics and "diseaseographics" (facts about their disease, its treatment and progression, and about concomitant conditions). Add in data about each patient's insurance coverage and about the physicians and their "practiceographics," and we have potentially a wealth of information to use to understand prescribing decisions. This paper describes three choice analysis options for patient chart databases, plus a suggestion for combining this kind of revealed preference (RP) data with experimentally designed stated preference (SP) data.

In a driver analysis we seek to measure the importance of various attributes in predicting some overall performance measure like an overall brand rating or a customer satisfaction rating. Unfortunately, a pervasive data problem called collinearity afflicts and ruins traditional methods of attribute importance measurement like correlation analysis and regression analysis. Some newer methods address the data problem and allow us to do driver analysis well.