9
XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006 1 ICIEOM 2006 ABEPRO Designing choice menus for mass customization Flávio Sanson Fogliatto (DEPROT/UFRGS) [email protected] Giovani J. C. da Silveira (University of Calgary/Canadá) [email protected] Abstract This paper proposes a method for designing choice menus for mass customization. The method is based on the analysis of stated preferences on product or service attributes obtained through panel studies. The method is presented, followed by a real world case application in a natural gas distribution company. The application indicated the method was able to elicit stated preferences on a broad range of attributes enabling the design of choice menus for alternative customer segments, balancing the trade-off between flexibility and value that is in the core of choice menu design. Keywords: Mass customization, Choice models, Conjoint analysis, Stated preference. 1. Introduction Mass customization (MC) has been defined as the ability to produce individually designed products and services at near mass production cost (DAVIS, 1987). It is enabled by a series of advanced technologies and practices including flexible manufacturing systems, computer aided design, and lean manufacturing (Da SILVEIRA et al., 2001). Over the last decade, MC has evolved from being a visionary idea to become a widespread strategy in manufacturing and service industries. Research on MC has also progressed from an initial focus on the manufacturing capabilities to produce variety at low cost to a broader emphasis on supply chain coordination and customer involvement in the conception of MC products and services. In this expanded view of MC, facilitating customer involvement in the process of specification and design of a personalized product has become one major determinant of a successful customization strategy (DURAY, 2002). Due to the limitation of traditional techniques such as surveys and interviews to elicit individual customer preference in an efficient and reliable manner, firms have been increasingly using choice menus (LIECHTY et al., 2001). Choice menus consist of producer-user interfaces that enable customers to select product attributes and features in a consistent and economical way (OLIVA, 2002). Notwithstanding the increasing presence of choice menus in business and consumer industries ranging from personal computers to financial services (SLYWOTZKY, 2000), few studies have focused on problems associated to their design and configuration. In particular, the design of choice menus must balance a trade-off between flexibility and value to customers, as complexity aversion implies that the value of a menu often decreases with cardinality. Despite the documented cases of companies that have been challenged by this particular trade-off (WIND & RANGASWAMY, 2001), there has been limited research on methods to specify the set of options to use in choice menus. This paper proposes a method for choice menu design in a MC context. The method, based on the analysis of customers’ stated preferences regarding product or service attributes, has some important features. First, it is based on the use of well-known market research techniques, such as focus groups, questionnaires and SP modeling, not demanding any special training from analysts. Second, it uses cluster analysis and experimental design techniques to guide

Mass customization design_choice_menu

Embed Size (px)

DESCRIPTION

 

Citation preview

Page 1: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

1 ICIEOM 2006 ABEPRO

Designing choice menus for mass customization

Flávio Sanson Fogliatto (DEPROT/UFRGS) [email protected] Giovani J. C. da Silveira (University of Calgary/Canadá) [email protected]

Abstract

This paper proposes a method for designing choice menus for mass customization. The method is based on the analysis of stated preferences on product or service attributes obtained through panel studies. The method is presented, followed by a real world case application in a natural gas distribution company. The application indicated the method was able to elicit stated preferences on a broad range of attributes enabling the design of choice menus for alternative customer segments, balancing the trade-off between flexibility and value that is in the core of choice menu design. Keywords: Mass customization, Choice models, Conjoint analysis, Stated preference.

1. Introduction

Mass customization (MC) has been defined as the ability to produce individually designed products and services at near mass production cost (DAVIS, 1987). It is enabled by a series of advanced technologies and practices including flexible manufacturing systems, computer aided design, and lean manufacturing (Da SILVEIRA et al., 2001). Over the last decade, MC has evolved from being a visionary idea to become a widespread strategy in manufacturing and service industries. Research on MC has also progressed from an initial focus on the manufacturing capabilities to produce variety at low cost to a broader emphasis on supply chain coordination and customer involvement in the conception of MC products and services.

In this expanded view of MC, facilitating customer involvement in the process of specification and design of a personalized product has become one major determinant of a successful customization strategy (DURAY, 2002). Due to the limitation of traditional techniques such as surveys and interviews to elicit individual customer preference in an efficient and reliable manner, firms have been increasingly using choice menus (LIECHTY et al., 2001). Choice menus consist of producer-user interfaces that enable customers to select product attributes and features in a consistent and economical way (OLIVA, 2002).

Notwithstanding the increasing presence of choice menus in business and consumer industries ranging from personal computers to financial services (SLYWOTZKY, 2000), few studies have focused on problems associated to their design and configuration. In particular, the design of choice menus must balance a trade-off between flexibility and value to customers, as complexity aversion implies that the value of a menu often decreases with cardinality. Despite the documented cases of companies that have been challenged by this particular trade-off (WIND & RANGASWAMY, 2001), there has been limited research on methods to specify the set of options to use in choice menus.

This paper proposes a method for choice menu design in a MC context. The method, based on the analysis of customers’ stated preferences regarding product or service attributes, has some important features. First, it is based on the use of well-known market research techniques, such as focus groups, questionnaires and SP modeling, not demanding any special training from analysts. Second, it uses cluster analysis and experimental design techniques to guide

Page 2: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

2 ICIEOM 2006 ABEPRO

data collection from customers, leading to databases that are both cost-efficient and representative. Third, it proposes the use of SP modeling using logistic regression, which is both easy to interpret and available from common statistical packages.

2. Research background

This literature review is divided in two parts. Initially, we present the current relevant research on choice menus. Then, we introduce basic aspects of stated preference modeling.

The ability of customers to co-design products or services based on individual preferences is one of the most distinctive features of MC. Over recent years, firms have been increasingly offering choice menus to allow customers to design solutions by selecting items that can best fulfill their needs (LIECHTY et al., 2001). Choice menus, also called choiceboards or design palettes, involve a broad range of customer-supplier interfaces, from simple menus offering product options and features to intelligent aid, mostly web-based systems assisting in the process of designing, comparing and cost-estimating mass customized orders.

From a research perspective, the major challenge with choice menus is the specification of a set of options to balance flexibility and value to customers. On one hand, larger menus provide more flexibility as they offer more options. On the other hand, complexity-aversion implies that the value of a menu decreases with cardinality (SONSINO & MANDELBAUM, 2002). Problems with flexibility-value trade-offs are compounded by the fact that both flexibility demands and complexity-aversion levels will often vary across customer groups. Demand for flexibility increases with increased uncertainty about future tastes; complexity aversion increases with the desire to minimize the risk of making wrong decisions (STODDER, 1997). Thus, as pointed out by DeShazo and Fermo (2002), designers must build choice sets that minimize the detrimental effects of choice complexity to a customer category. In practice, information about customer preferences must be used to tailor the design of the choice menu itself, customizing the set of options presented to buyers and promoting sales (SLYWOTZKY, 2000).

The stated preference (SP) method is an applied conjoint paradigm that quantifies respondents’ choices regarding hypothetical market situations (UNTERSCHULTZ et al., 1997). Preferences are given to commodity alternatives decomposed into separable attributes, each of which can be examined for their individual influence on choice. This approach is derived from Lancaster’s (1966) theory of characteristics stating that utility is derived not from goods themselves, but from the attributes or characteristics of goods.

The SP method is based on random utility theory (RUT) choice models. The theory is derived from the observation that an individual can make different judgments from one occasion to the next. Therefore, utility is expressed as a sum of observable and non-observable (random) components (HENSHER et al., 1999):

ininin VU ε+= , (1)

where inU is respondent n’s utility of choosing alternative or scenario i, inV is the systematic,

observable component of utility and inε is the random component. The utility inV of

alternative i is a function of its attributes, which is often assumed to have linear form,

inkkininin xxxV ββββ ++++= ...22110 , (2)

where inV is respondent n’s systematic utility of alternative i, inkx are the attributes of

alternative i for respondent n and 0β to kβ are the coefficients to be estimated.

Page 3: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

3 ICIEOM 2006 ABEPRO

3. Method

The method proposed in this paper is illustrated in Figure 1. There are three phases, each with a number of operational steps and decision points detailed next. Although the method may be applied in the design of product and service choice menus, we hereafter refer to products only.

Figure 1 – Phases and operational steps of the method proposed

The main objective in Phase 1 is to obtain a comprehensive list of product attributes relevant to customers. Such attributes will be treated as candidate choice menu variables. We start this phase by analyzing the product market in terms of customer diversity. When customers have theoretically distinct customization demands it may be advisable to qualitatively cluster them, using expert opinion, prior to data collection; otherwise, a representative random sample of unclusterized product customers may be used in the focus group sessions.

Data collection using focus groups is typically accomplished in six steps: (i) focus group planning, (ii) participants selection, (iii) definition of questions and session moderation guidelines, (iv) definition of sessions logistics, (v) choice of moderator and (vi) data collection and analysis. For details on each step, see Greenbaum (2000).

Information in Phase 1 may be organized in a table, with columns headings corresponding to a description of the ad hoc clusters (in case they arise in the analysis), and row elements given by the attributes elicited by individuals in focus groups. Ad hoc clusters are identified by IiCi ′=′ �1, ,

and attributes by JjAij�1, = . It is expected that attributes in different clusters coincide.

Objectives in Phase 2 are (i) to obtain importance weights ijw to attributes ijA listed in Phase

1 and (ii) group customers using formal clustering techniques. To accomplish objective (i) we propose a quantitative research using a questionnaire for data collection; to accomplish objective (ii) a cluster analysis may be performed.

Using the proper sampling technique is fundamental in the research design. If the sample is intended to be a smaller scale representation of a population of interest, the researcher should know beforehand whether customers differ substantially regarding customization demands. In case there is evidence to the existence of clusters of customers, their proportion in the population should be estimated, to enable a probabilistic sampling of that population. Otherwise, a non-probabilistic sampling strategy may be appropriate (for details on sampling strategies and samples size determination see LEVY & LEMESHOW, 1999).

Questionnaires must be elaborated such that the objective of data collection is clear to

Page 4: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

4 ICIEOM 2006 ABEPRO

respondents upfront; for that matter, a title and introduction text must always be used. If possible, the analysis should group similar customization attributes and provide a subtitle to each group. As the number of attributes increases, grouping of items becomes crucial.

To gather importance weights ijw from respondents there are two possible approaches. In

case 15>j and attribute descriptions are complex, a good strategy is to ask respondents to rank attributes within each group. An item ranked in the k-th position will be given an importance weight k1 (similar to situations where a ratio scale is used, importance of items becomes directly comparable within a group). If the number of attributes is small, weights may be elicited using an importance scale, such as the Likert 5-point scale. Weights from all respondents are then added up to obtain the final importance weight for a given attribute.

Once importance weights are available to each attribute, cluster analysis is performed using ijw

as classificatory variables. The objective is to formally identify clusters of customers with similar demands in terms of product customization attributes and optimize data collection in Phase 3. If ad-hoc clusters of customers were previously identified in Phase 1, formal cluster analysis will allow validation of such clustering. We denote the new clusters by IiCi

�1, = , and point out

that I ′ may differ from I , and that the final clustering of customers should arise from careful analysis of the groupings in iC′ and iC , since they were generated using different information.

The main objective in Phase 3 is to obtain preference models relating stated preference (SP) and product attributes. Design of experiments (DOE) in conjunction with SP modeling is used for that purpose. DOE guarantees a data collection both efficient and economically feasible. The SP method allows data modeling where the influence of individual attributes as well as their interactions on customer preference may be assessed. Once SP models are at hand choice menus may be defined for the product, both in terms of attributes to be customized as in terms of choice levels within each attribute.

We stress the importance of building models where the significance of interactions between attributes is assessed. SP data are usually summarized using main effect models, which demand a low cost data collection. However, customization attributes of a product may not affect independently customer preference and therefore the significance of interaction terms should be verified. Clustering of customers in Phase 2 allows us to use more complex, yet still economically feasible data collection designs enabling SP modeling of both main effects and interactions.

The SP method is used here to compare a control (or reference) scenario against several alternative scenarios. The control scenario expresses the standardized version of the product under analysis. In the alternative scenarios, product attributes are made flexible to create different customized products. To make an attribute flexible implies in setting it to a given level. Therefore, an SP customization model will indicate not only the relative importance of attributes but also the market benefits from offering each attribute at different levels.

As previously mentioned, data collection for SP modeling must be planned separately within clusters of customers. Recall that customers in a given cluster value similarly the same customization attributes. Therefore, creating SP scenarios specifically for each cluster enables to include only attributes that are highly valued by them, reducing the number of attributes and consequently the number of alternative scenarios in the SP study.

We recommend using factorial designs to organize the SP data collection process. In a factorial design applied to collect SP data from a given cluster of customers, attributes are varied within pre-defined levels to generate alternative scenarios such that the total number of scenarios

Page 5: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

5 ICIEOM 2006 ABEPRO

analyzed will be given by ∏ == J

j jkN1

, where jk denotes the number of levels of attribute j.

Clearly, a large number of attributes and levels will lead to a large data collection, which is usually undesirable. Therefore, we should restrict the SP study to include attributes with large values of wij (excluding attributes not included in the study from the choice menu). Similarly, the number of levels of an attribute should be greater than two only if there is strong evidence of a non-linear relationship between attribute levels and the response (i.e. the stated preference). Fractioning and/or blocking the experimental data collection matrix are usually necessary in SP studies and are likely to be needed when applying the method we propose. If clustering of customers in Phase 2 leads to clearly defined clusters, blocking the data collection within clusters is the recommended course of action; otherwise, fractioning should be preferred. Once the experimental matrix is defined, data collection may take place following the guidelines of SP studies such as Bateman et al. (2002).

Data modeling should lead to preference models for each cluster, following the guidelines in section 2. Eq. (2) gives the respondent’s preference to a given scenario, which is the usual outcome of an SP study. Here, respondents are customers grouped in clusters and it is assumed that they share the same model. Using such models, it is possible to arrive to choice menus for the product under study. This is done upon inspection of their regression coefficients and preference values, as follows. Coefficients that are significant at (1 – p)% are ranked in importance according to their p-values (typical choices are p < 0.05 or p < 0.1); they indicate the attributes to be included in the choice menu. Attributes that appear significant exclusively in interaction terms should also be considered for inclusion in the menu. The number of levels of an attribute in a menu should be defined considering the attribute’s rank position and the practical relevance of offering a large number of levels to customers.

Menus may also have their attributes and levels defined based on preference threshold values chosen by the analyst, performing simulations with the preference models. For example, it may be possible that only certain levels of an attribute yield predicted preference values above the threshold; these are thus the levels to be included in the choice menu. It may also happen that even excluding an attribute from the model one is able to determine combinations of levels for the remaining attributes that yield predicted preference values above the threshold. If parsimony is sought, such attributes could be also excluded from the choice menu.

4. Case application

The method proposed in this paper was applied in a case study where the main objective was to determine product, service and technological attributes valued by potential natural gas (NG) customers of a Brazilian distribution company, as well as their proper customization levels. Although available in several regions in central Brazil, only recently the NG distribution network reached the south of the country, where the present study took place. A wide variety of industrial, commercial, automotive and residential NG applications are possible, and customer needs regarding the commodity tend to vary according to end use. Following the steps proposed in section 3 we were able to identify clusters of NG customers in the geographical region of interest and create choice menus for them, as detailed next.

We start describing the steps in Phase 1. To obtain the list of NG attributes from focus groups in the case study, ad-hoc clustering of customers was mandatory for two main reasons. First, there are a large number of uses for NG reported in the literature (for example, as a vehicular fuel in cars and buses as well as in the air-conditioning of hospitals). Although the product is essentially the same, some of its attributes are more or less valued according to the final use. Therefore, treating all potential NG customers as members of the same population would lead

Page 6: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

6 ICIEOM 2006 ABEPRO

to a list of attributes excessively broad and non-representative. Second, data collection tends to be time consuming if applied to a large number of focus groups. Therefore, we must deviate from the ideal case where a list of customization attributes is obtained from each type of NG customer, and collect data from groups of customers that use the commodity similarly.

The first clustering of NG customers was obtained following a three-step approach. We first identified economic sectors with a reported history of NG use in the literature. We then verified both the presence and the economical representation of such sectors in the geographical region of interest. Finally, ad-hoc clustering of sectors was obtained based on expert opinion from members of the NG distributor technical staff. A total of 39 relevant economic sectors, all potential users of NG, were identified in the region. Their typical NG applications were listed and served as basis to clustering. As partially shown in Table 2 (first and second columns), eleven clusters of customers (identified as economic sectors) were formed. Focus groups were limited to ten participants. In a given focus group, the number of cluster sector representatives was determined based on their economical relevance and NG usage potential.

Economic sectors Ad-hoc cluster ing ( iC′ - Phase 1) Formal cluster ing ( iC - Phase 2) Supermarkets 1 5 Meat markets 1 1 Open malls 2 1 Hotels 2 5 Hospitals 2 5 Food manufacturers 3 2 Rubber processors 3 5

Tanneries 11 5

Table 1 – Sample of ad-hoc and formal clusters in the case study

Information gathering from the 11 focus groups was completed in approximately 4 months. Sessions varied in length from 60 to 90 minutes and were moderated by one individual and two assistants. Invited members from sectors in each cluster voluntarily participated in the meetings. They were requested to list important attributes related to the product, the distribution network (and services provided by the distributor) and the technological aspects of using NG (in particular those related to equipment conversion and its maintenance). At the end of each section, participants were requested to rank attributes in importance within each category. Although respondents were aware that attributes listed should be preferably customizable, there were exceptions (for example, supply longevity). Table 2 displays a list of the highest ranked attributes in each category over all groups.

Product Network Equipment Price Time to supply in large scale Safety Operative performance Network capillarity Compliance to legislation Emission of pollutants Contractual conditions Specific technical solutions Storage space Supply regularity Maintainability Reliable measurement system Supply diversity Technical literature availabilityNG adoption projects provided by distributor

Standardized valves and connectors

Technical training prior to equipment use

Supply longevity Network maintenance Convertibility Cost to install NG tubulation Multi-fuel capacity (NG / LPG) Network pressure Operational cost

Table 2 – Some NG attributes as elicited by consumers in focus groups

Ranking of attributes varied substantially among clusters. For example, customers in 1C′

Page 7: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

7 ICIEOM 2006 ABEPRO

ranked price, supply regularity and operative performance as most important attributes, while customers in 11C′ elected time to supply in large scale, price and supply longevity. That exemplifies the importance of pre-clustering customers prior to focus group data collection.

Phase 2 starts with quantitative research to obtain importance weights ijw to attributes in Table 2. A

questionnaire was prepared for that, with 18 attributes (six in each category) to be ranked in importance by respondents. We used a questionnaire where items are ranked in importance. Questionnaires were customized for clusters of clients presenting the attributes most valued by them in Phase 1, but applied separately to customers in the 39 economic sectors previously identified.

The average sample size per economic sector was 16 questionnaires, estimated using proper sample size estimation techniques. This average value was adjusted to include the following information regarding each sector: (i) relative percentage of electricity consumption, (ii) relative number of participants in sector (in %) and (iii) NG adoption potential (given as a probability). The smallest and largest adjusted samples sizes were 5 and 50, respectively. A total of 450 questionnaires were personally applied by a team of administrators. As previously mentioned, questionnaire respondents were asked to rank attributes in importance regarding their customization requirements. To convert ranks into scores, we used the reciprocal of the attribute rank position as explained in section 3.

A formal cluster analysis was carried out using the 18 attributes as clustering variables. Responses from a same economic sector were added and normalized to generate a ( )1839× data matrix used in the analysis. Cluster analysis was carried out in two steps. First, the proper number of clusters to be used was identified using a hierarchical approach. Once the ideal number of clusters was identified, sectors to integrate each cluster were determined using a k-means partitioning algorithm. The final assignment of sectors to clusters is partially presented in the last column of Table 2. As expected, clustering in Phase 2 yielded different results from clustering in Phase 1. We entered Phase 3 of the method using the Ci clusters.

We now describe the steps in Phase 3. Twelve experiments were planned to collect SP data from clusters. Each experiment was comprised of six attributes, corresponding to the two most important in each category, explored at different levels. Price, an attribute included in all clusters, was explored at four levels; all remaining attributes were investigated in two levels. Thus, the total number of scenario alternatives investigated in each experiment was 128 (25×41). Further, it was decided that 16 scenarios would be presented to each respondent, to avoid fatigue. Therefore, experiments were divided into 8 blocks with 16 scenarios each. Attributes were coded A to F and attribute levels were coded -1, -0.3, +0.3, +1.

Each of the eight blocks was presented as a questionnaire to respondents and replicated ten times. As discussed earlier, in a customization SP study alternatives may be compared to a control scenario where the product is presented in a standardized format; the alternatives therefore make the product flexible regarding attributes of interest. However, in our study, NG was not yet available to customers. Therefore, the control scenario was established to be the alternative energy sources, e.g. PLG, oil, wood, etc. (and their corresponding characteristics) used at that moment by customers from each cluster. Alternative scenarios presented levels of NG attributes in each cluster in comparison to that control scenario.

Due to the large number of clusters in this case study, we restrict ourselves to present results from cluster 5 that grouped a large number of industrial and commercial sectors. The preference loglinear model for cluster 5 is:

Page 8: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

8 ICIEOM 2006 ABEPRO

( )( )

)002(.232.0)000(.300.0)002(.170.0)000(.770.0)000(.214.0

)000(.267.0)000(.344.0)001(.176.0)000(.241.0488.11ln

64632165

4321

XXXXXXXX

XXXXyy

+++++++++−=−

(3)

where y denotes the response (i.e. preference of a given scenario for each cluster) and iX (i = 1,…, 6) are the

attributes investigated (detailed in Table 3). P-values are given within parentheses after their corresponding model term. Transformation in response y led to a linear model, with coefficients determined using the statistical package SPSS. The model fit given by its coefficient of determination was 704.02 =R .

Coefficients in eq. (3) were directly comparable in view of the attributes level coding. As expected, Price appeared as the most important attribute, followed by Emission of pollutants and Convertibility. Three interactions were significant: Supply regularity × Cost to install NG tubulation, Emission of pollutants × Price and Convertibility × Price. Since all interaction coefficients were positive, they were easy to interpret. Consider interaction 21XX , for example, and suppose a menu where all

attributes are set at their most favorable levels, including 1X and 2X . Customer preference in that case would be 77.3%; ignoring the interaction the preference value decreases to 74.2%. It is also noteworthy that some interactions had larger coefficients than individual attributes. Ignoring such interactions, as in classical SP data analysis, would lead to a less representative model.

Attr ibute Descr iption Levels X1 = Supply regularity Degree to which customers may be exposed to

interruptions in NG supply (-1) Irregularity possible (+1) Irregularity not possible

X2 = Cost to install NG tubulation

Cost incurred by customers to extend tubulation from distributor’s network to point of use

(-1) High; (+1) Low

X3 = Emission of pollutants

NG emission of pollutants into environment in comparison to energy sources to be replaced

(-1) Pollutes more (+1) Pollutes less

X4 = Convertibility Degree of equipment convertibility to NG (-1) Low; (+1) High X5 = Operational safety Level of equipment operational safety using NG in

comparison to energy sources to be replaced (-1) Safety decreases (+1) Safety increases

X6 = Price Indicates expenses with GN in comparison to the energy sources it is intended to replace

(-1) 10% larger; (-0.3) About the same; (+0.3) 10% smaller (+1) 10% to 30% smaller

5. Conclusions

This paper proposed a method to design choice menus in a MC context. The method incorporates stated preferences to define which attributes of a product or service should be offered at different levels to be selected by individual customers. The method aims mainly at designing choice menus with an appropriate number of options to balance the flexibility-complexity trade-off indicated by Stodder (1997), among others. This appears to be one of the first methods to support the design of choice menus for MC.

Due to its originality and stated objectives, this study has limitations, most of which can be addressed by further studies. First, the method did not incorporate approaches to update choice menus through combining revealed preferences with the stated preferences assessed in the panel study. Second, the idea of incorporating interactive terms to assess the moderated effect of one attribute on the regression estimate of another attribute may have broad implications for modularity and bundled choice, but these implications were not sufficiently explored in this paper.

Referências

BATEMAN, I .J.; CARSON, R.T. & DAY, M. Economic Valuation with Stated Preference Techniques: A Manual. Cheltenham: Edward Elgar, 2000.

Da SILVEIRA, G., BORENSTEIN, D. & FOGLIATTO, F.S. Mass customization: literature review and research directions. International J. of Production Economics, Vol. 72, 1-13, 2001.

DAVIS, S. Future Perfect. Reading (MA): Addison-Wesley, 1987

Page 9: Mass customization design_choice_menu

XII ICIEOM - Fortaleza, CE, Brasil, October 9 - 11, 2006

9 ICIEOM 2006 ABEPRO

DeSHAZO, J.R. & FERMO, G. Designing choice sets for stated preference methods: the effects of complexity on choice consistency. J. of Environmental Economics and Management, Vol. 44, 123-143, 2002.

DURAY, R. Mass customization origins: mass or custom manufacturing? Int. J. Oper. & Prod. Manag., Vol. 22, 314-328, 2002.

GREENBAUM, T.L. Moderating Focus Groups: A Practical Guide. Thousand Oaks: Sage, 2000.

HENSHER, D., LOUVIERE, J. & SWAIT, J. Combining sources of preference data. J. Econometrics, Vol. 89, 197-221, 1999.

LANCASTER, K.J. A new approach to consumer theory. J. of Political Economy, Vol. 74, 132-157, 1966.

LEVY, P.S. & LEMESHOW, S. Sampling of Population – Methods and Applications. New York: Wiley, 1999.

LIECHTY, J.; RAMASWAMY, V. & COHEN, S.H. Choice menus for mass customization: an experimental approach for analyzing customer demand. J. of Marketing Research, Vol. 38, 183-196, 2001.

OLIVA, R.A. Way beyond web sites. Marketing Management, Vol. 11, n. 6, 46-48, 2002.

SLYWOTSKY, A.J. The age of the choiceboard. Harvard Business Review, Vol. 78, n.1, 40-41, 2000.

SONSINO, D. & MANDELBAUM, M. On preference for flexibility and complexity aversion: experimental evidence. Theory and Decision, Vol. 51, 197-216, 2001.

STODDER, J. Complexity aversion: simplification in the Herrnstein and Allais behaviors. Eastern Economic J., Vol. 23, 1-15, 1997.

UNTERSCHULTZ, J.; QUAGRAINIE, K.K. & VINCENT, M. Evaluating Quebec's preference for Alberta beef versus US beef. Agribusiness, Vol. 13, 457-468, 1997.

WIND, J. & RAMASWAMY, A. Customerization: the next revolution in mass customization. J. Interactive Marketing, Vol. 15, 13-32, 2001.