Recently we came across a supplier Quality evaluation that struck home the problem of using pricing as a Quality attribute. Supplier Quality evaluations have become standard communication devices in our increasingly Quality conscious environment. This firm was given an excellent rating in all aspects of performance other than in pricing. It was noted that the company had not reduced price as had been requested by the customer during the past year and therefore, had a poor "price Quality." This unacceptable rating in Pricing resulted in the supplier being given merely an acceptable or good overall Quality rating.
Clearly price is a part of the customer's expectations and it is justified to both measure customer satisfaction with price as well as communicate dissatisfaction. Many customer satisfaction studies include price in their evaluation attributes. Unfortunately, these values are often combined with the other attributes to obtain an overall customer satisfaction rating. However, price is different from the other attributes of Quality. It plays a multiple role representing: (1) a cost to the customer, (2) the source of value to the supplier, and (3) a communicator of value.
Price implies a level of value that differentiates the product from competition. Premium prices allow services to be rendered to the customer. Reduced prices often result in a different positioning of the product with a redefinition of Quality. What constitutes a good price in the long run may not run coincide with the hopes of the customer. The very survival of the customer's firm may depend on the product and the maintenance of a satisfactory price. Focusing on price holds an additional danger for any Quality partnership. Reduced overall costs to the customer should follow from improved Quality, but reduced price may not. The goal is to make both the customer and the supplier better off. Focusing on price reduces continuous improvement to the function of "beating down" the supplier.
This discussion is an abstract from a short article published by E. B. Lieb and W. H. Strawhacker in the Quality Digest.
The appropriate price for a product is a complex decision. The value that customers put on an offering should be a major constituent for setting prices. It ultimately determines the price sensitivity of the market. Customer value should be measured rather than assumed. Our experience has indicated that product management and sales often (if not usually) have a different perspective of value than the customer. It is critical that an independent assessment, when feasible, be carried out.
Irv Gross at the Institute for the Study of Business Markets (Pennsylvania State University) speaks to two types of values: (1) a normative economic value or value-in-use and (2) perceived value. The value-in-use corresponds to engineering economic cost calculations of using the product. It is applicable to a broad range of industrial and consumer products where cost reduction is the dominate objective for the purchase. Value-in-use application models are usually developed based on interviews with cooperative customers and development partners. The overall market value-in-use models are constructed by combining these application models with industry statistics.
Perceived value represents how customers evaluate the offering. It is a personal measure of value. Many perceived value studies are undertaken as quantitative research with the goal of developing a market model. However, we have found that valuable insight can be obtained with very small samples using these techniques. We have measured perceived value during focus groups as well as with industry experts. There are five general methods for measuring perceived value:
| 1. Explicit Evaluation - The simplest method is to ask customers their purchase intent for a number of concepts. This is feasible with a small number of concepts and a fairly narrow range of prices. The trick in this method is to make the evaluation exercise as similar as possible to the purchase situation. |
| 2. "Choice Modeling" - If the pricing of competitive products is a critical issue, simple explicit evaluation will not be adequate. More complex experiments can be used, however. "Choice Modeling" involves requesting the customer respond to a number of sets of product prices. The experiments are statistically designed to capture the sensitivity of the customer to price changes in multiple products. |
| 3. Compositional Conjoint and Profiling - If the characteristics, features, or attributes of the products need to be selected as well as price, than "Choice Modeling" quickly becomes too involved. Methods exist to explicitly measure the value of attribute levels. Compositional conjoint is the simplest of these and involves having the potential customer rank the attribute levels along with discount information. Profiling or "Simalto" extends this approach to giving a series of purchasing situations. |
| 4. Full Profile Conjoint - However, compositional conjoint and profiling assumes that the customer can give an explicit value to the attributes and features. We can not expect that all attributes will have an explicit value. For example, brand and company names and packaging attributes are often difficult to evaluate explicitly. Value for these attributes must be derived from behavior. Full profile conjoint experiments provide this information by requesting customers to rank alternative product descriptions. The set of products are statistically chosen to capture the value of these attributes. |
| 5. Importance and Performance Scales - Ratings of the attribute importance and performance are typically available from customer satisfaction and product positioning studies. It is very tempting to use this data to model price sensitivity. Unfortunately, these models are notoriously inaccurate in predicting price sensitivity. |
Details and limitations of each of these approaches have been summarized in a Perceived Value Methods Table
Estimating the relationship between price and sales is still one of the most frequent business analysis chores. Data is collected as changes in the intention to purchase with changes in price. This can be done either on established products or during concept testing. In order to merge this data and obtain an overall market description of price and sales, some functional relationship is imposed. The functional relationship that we choose to use should reflect expected behavior and allow reasonable extrapolation to both lower and higher prices.
There are four commonly used types of expressions describing the relationship between price and sales. These include: (1) the "power law" expression favored by economists since it predicts a constant price elasticity, (2) the polynomial favored by statisticians for its simplicity, (3) exponential and (4) the log normal. Psychological research has indicated that perceptions tend to have a geometric or logarithmic relation-ship with stimuli. As such one would expect that purchase behavior would be a function of the logarithm of price. The exponential and the log normal both capture this logarithmic nature of price.
We prefer the log-normal function since it also captures the expected limiting behavior of price distributions. Below are typical examples of these four distributions fitted to normalized hypothetical data. We expect that the price/share curve should be downward sloping, approaching zero at high prices and one at low prices. At no time should the quantity be larger than one. As we can see, only a distribution function like the log-normal maintains these conditions. This is particularly important when it is necessary to extrapolate beyond the range of specific tested price points.
Fitting data to functions like the log-normal distribution is not particularly difficult with spreadsheets, like Microsoft EXCEL or with advanced statistical packages. Fitting can be done either by converting the price and quantity data to the appropriate form or by using the "Solver" capabilities in EXCEL to fit the data to the functions.
Conjoint for Everyone
Full profile conjoints are wonderful procedures to measure attribute value. But they are complex to design, hard to execute and difficult to analyze. Several packages are available to help in the design and analysis of these procedures. Packages are available through Bretton-Clark, SPSS, and SAS. I've reviewed the Bretton-Clark software on the PC and the mini-computer version of the SPSS Categories module. These are market modeling packages designed to develop market simulators based on the application of specific marketing research methodologies. The packages are designed to facilitate design, analysis and model development.
Of the two, the series of "Conjoint" modules by Bretton-Clark is probably the most flexible analysis and design package of this type. It allows for the design of fairly complex research studies and the use of alternative model assumptions. Stand alone simulators can be developed using these packages. The principal strength and weakness of this package both tend to come from its user friendly nature. Analysts with limited understanding of the limitations of the method can use the tools to develop market models. This opens up the potential for inappropriate application of the tools. SPSS provides similar, but more limited, capabilities than the Bretton-Clark packages.
As a stand alone system, the Bretton-Clark package appears to be the best available. The SPSS and the SAS packages would be of interest to those organizations already using them for statistical analysis. It should be noted that the complete set of Bretton-Clark modules is fairly expensive. We recommend their use by organizations who intend to do significant amount of modeling using full profile conjoint. The Conjoint Designer, Analyzer and Segmenter are published by Bretton-Clark, Morristown, NJ and is distributed through SciTech International. SPSS Categories is available from SPSS, Inc., Evanston, IL.
We serve marketing research firms by providing analytical support. While normally, we are involved in the design of the studies we will analyze, occasionally we are asked to analyze studies which have already been fielded Recently we have been asked to undertake the analyses of two separate pricing studies whose designs were gravely flawed. Major design flaws in full profile conjoint or choice modeling can result in a dangerous situation. The resulting models from these studies may not reflect the opinion of the respondents.
Both full profile conjoint and choice modeling use regression to produce decision models. Effective regression analysis requires independence between the design variables to produce meaningful results. There must be little or no intercorrelation between attributes. The design of the "cards" for full profile conjoint or the exercises for choice models should provide this independence. While complex designs may require some intercorrelation to reduce the task, it is important that the intercorrelation should be below 0.2. We have seen intercorrelations as high as 0.56 in studies that have already been fielded.
Correlations this high produce unreliable results. Once the study is fielded there is nothing that can be done about it. It is critical that all full profile conjoint and choice models designs be tested before the study is fielded, no matter where the design came from. Correlations can be obtained from the design on an EXCEL spreadsheet using the Data Analysis option (under the Tools menu), or using one of the standard statistical packages. At a minimum, we recommend that the correlation matrix for all designs should be provided to clients for approval before fielding.
I hope that the articles in this newsletter stimulate discussion and provide some new insights into planning for the future. Future newsletters are planned to include articles on: (1) modeling industries, (2) mining customer satisfaction data, (3) market segmentation and (4) using the Internet. I appreciate any comments and contributions from you. You can get in touch with me at: Custom Decision Support, Inc., Phone (610) 793-3520. Fax (610) 793-2531 or E-Mail at genelieb@lieb.com.