Advanced Analytics and Modeling
Advanced analytical techniques are often used to extract additional insight from data by identifying patterns that lie below the surface and drive business results. We employ particular techniques depending on the guidance being looked for; e.g., if targeting is the need, clustering, factor analysis and CHAID are useful; if optimization of a product bundle is the need, choice modeling and conjoint are useful.
Because we implement research projects using an integrated approach, the analyst is often the same person who designs the questionnaire, ensuring that surveys are designed for the right analysis and, importantly, that the analytic results can be understood and used by marketing managers and strategists to meet project objectives.
RONIN has solid experience in the following techniques and will use either one or, in some cases, a battery of these to match the needs of the project.
Custom structural equation modeling via LISREL and PLS
Structural equation modeling identifies relationships between complex attitudes and behaviors, such as satisfaction and loyalty, to give guidance on how to move them. Most of these analyses are built on two pillars, one that promotes our understanding of what market forces look like, and another that tells us what impacts them. Stated differently, structural equation models identify the meaningful components underlying abstract concepts such as loyalty and satisfaction, to enhance our understanding of them (what statisticians call the “measurement model”). These models also show us how these elements are related to each other, so we can identify what causes movement in the market and build programs to create movement we want to see.
Multiple regression (for driver analysis)
Multiple regression analysis identifies how several variables (such as attitudes and perceptions) work together to either predict or drive another. We typically use these analyses when we know that there are several variables that impact something (e.g., a brand choice is driven by perceptions of the brand as providing benefits such as value for the money, as providing high quality, or something else) and we want to understand which of those many elements has the greatest impact.
Factor analysis to uncover dimensions underlying evaluations and other survey responses
In the realm of market research, these variable reduction schemes identify underlying dimensions that guide respondents thoughts when, for example, they are evaluating a product or service. Please note that these analyses do not test whether the dimensions that surface relate to a specified outcome (e.g., an online purchase). Results of factor analyses can, however, be used as inputs to regression or higher-order predictive models.
Multiple discriminant function analysis
Discriminant analysis is useful for indentifying the attitudinal, behavioral and demographic (or firmographic) variables that most sharply distinguish one group from another. Groups can be defined as belonging to some category, or as responding in a particular way. In evaluating a test marketing program, for example, we may have two or more kinds of response (e.g., no response, contacted a sales rep, made a purchase). Discriminant analysis produces one or more functions (equations, really) that maximize our ability to categorize people into those different groups, based on variables for which we already have access (e.g., company size, industry, prior purchases, etc.). We can therefore use those pre-existing variables to predict response among a broader set of individuals or companies whose behavior has not yet been observed. It is particularly useful when we have large databases to work with and want to generalize from a small sample within the database whose behaviors are known, to a larger sample whose behaviors are not.
Cluster analysis for segmentation
Cluster analysis finds naturally occurring groups that differ along pre-specified dimensions. These dimensions may be based on demographics, firmographics, attitudes, behaviors or a combination of these. Used correctly, it can create and drive successful market segmentation and targeting.
Multidimensional scaling
Multidimensional scaling techniques allow us to reproduce the same perceptual space that customers use when evaluating different entities (such as products or brands), so that we can understand how they organize them conceptually. An informed examination of what that space looks like will help us identify the dimensions on which customers evaluate those entities, even if customers themselves cannot verbalize those dimensions themselves.
Applied data analysis statistical packages
Conjoint and Discrete Choice
These techniques identify buyer preferences for product features, the most desired set of features for a product, and what tradeoffs buyers are willing to make for their desired product. The techniques are thus effective tools for developing a successful product design and bundling of product or service offerings.
CHAID, Exhaustive CHAID, C & RT, QUEST
These are all tree-based tools that segment groups of respondents that share similar characteristics. CHAID (Chi-squared Automatic Interaction Detector) and Exhaustive CHAID are ideal for visualizing large data sets for consumer profiles and segments. C&RT (Classification and Regression Tree) and QUEST (Quick Unbiased Efficient Statistical Tree) provide similar results but, unlike CHAID techniques, produce trees with binary splits which are more appropriate for some types of research. All four techniques are effective variable reduction tools and precursors to other types of analyses, such as regression and higher-order predictive models.
Perceptual Mapping
This technique is particularly effective for exploring branding issues. Several brands can be compared and contrasted, on a number of different attributes, in one comprehensive picture. A perceptual map may indicate that several brands of laptops are perceived similarly in terms of price, performance, and wireless capabilities, but not in terms of reliability and warranty coverage. Another advantage of perceptual maps is that the data required to construct them is straightforward and typically not difficult to collect – consumers usually rate the product/service attributes on simple Likert-type scales (e.g., ranging from strongly agree to strongly disagree).