A Framework for Choice Modeling in Usage Context ...

Proceedings of the ASME 2010 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2010 Proceedings of 2010 ASME International Design Engineering Conferences & Computers and August Technical 15-18, 2010, Montreal, Quebec, Canada Information in Engineering Conference IDETC/CIE 2010 August 15-18, 2010, Montreal, Quebec, Canada

DETC2010-28 DETC2010-28490 A FRAMEWORK FOR CHOICE MODELING IN USAGE CONTEXT-BASED DESIGN

Lin He Graduate Student

Christopher Hoyle Postdoctoral Scholar Integrated DEsign Automation Laboratory Department of Mechanical Engineering Northwestern University

Jiliang Wang Graduate Student

Wei Chen* Professor

Bernard Yannou Professor

Laboratoire Genie Industriel Ecole Centrale Paris

E EW EY F Jn M S U X Y

ABSTRACT Usage Context-Based Design (UCBD) is an area of growing interest within the design community. A framework and a step-by-step procedure for implementing consumer choice modeling in UCBD are presented in this work. To implement the proposed approach, methods for common usage identification, data collection, linking performance with usage context, and choice model estimation are developed. For data collection, a method of try-it-out choice experiments is presented. This method is necessary to account for the different choices respondents make conditional on the given usage context, which allows us to examine the influence of product design, customer profile, usage context attributes, and their interactions, on the choice process. Methods of data analysis are used to understand the collected choice data, as well as to understand clusters of similar customers and similar usage contexts. The choice modeling framework, which considers the influence of usage context on both the product performance, choice set and the consumer preferences, is presented as the key element of a quantitative usage contextbased design process. In this framework, product performance is modeled as a function of both the product design and the usage context. Additionally, usage context enters into an individual customer’s utility function directly to capture its influence on product preferences. The entire process is illustrated with a case study of the design of a jigsaw.

Performance-related usage context attributes Usage importance indices Product choice set Non-engineering attributes Customer attributes Usage context scenario Product design variables Engineering performance

1. INTRODUCTION Usage Context-Based Design (UCBD) has become an area of growing interest in engineering design research. The usage context of a product refers to “all factors characterizing the application and environment in which a product is used that may significantly impact customer preferences for product attributes” [1]. In other words, usage context is the set of scenarios in which a product (or service) is to be used, including the environments in which the product is used, the types of tasks the product performs, and the conditions under which the product will operate. It is proposed in this work that usage context should also be a part of the primary descriptors in the definition of a customer profile, in addition to customer’s socio-economic status, anthropomorphic attributes, and previous product experience [2]. Because a product will perform or be viewed differently for various usage contexts, their impacts on customers’ preferences and choice behaviors need to be studied. Even though previous works in marketing [3] and engineering [4] have illustrated the significance of

NOMENCLATURE Customer desired product attributes A *

Usage context attributes Preference-related usage context attributes

Corresponding Author: Email: [email protected]

Copyright © 2010 by ASME

1 Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/13/2014 Terms of Use: http://asme.org/terms

1


provides an illustration of a case study of jigsaw design. Conclusion and future work are summarized in Section 6.

usage context in a consumer’s choice process, a general framework for quantitatively incorporating intended product usages to predict product performance and consumer’s choice, does not exist; which is the focus of this study. In our previous work [5], we laid out the general principles and taxonomy for UCBD and the need for a quantitative approach for capturing the influence of usage context. The key element of the approach is to build a demand model that provides a link between engineering design and the preferences of potential customers for a product [6-10]. Discrete Choice Analysis is a commonly used approach to demand modeling [11-13] which provides an estimate of product demand based on a parametric model of product and consumer attributes to describe how consumers make trade-off among multiple product attributes when selecting a product. Previous works in demand modeling for design have assumed that product performance is the same across all users and all usage contexts (e.g., cordless power tools [8,10], bathroom scales [9,14], electric motors [6,15]). Also, these works have assumed that the set of competing products considered by different consumers are similar, not considering the specific usage context scenarios and their impact on the choice set. This paper builds upon our previous work [5] by offering a detailed framework and procedure of choice modeling which considers that usage context can have an influence on the performance of the product, the choice set, and preferences for the product. An example of a product in which usage context influences both performances and preferences is a hybrid electric vehicle (HEV), which operates on both an electric motor and internal combustion engine. A HEV exhibits a greater advantage in fuel economy over a conventional vehicle in urban driving situations compared to highway usage. Therefore, the fuel economy performance a consumer experiences with hybrid technology will be a function of how the vehicle is used. In the remaining part of the paper, more details of defining usage context and studying its impact on consumer choice will be illustrated by a jigsaw design. The key aspect of our approach is the ability to account for differences in product performance and customer preferences between different individuals under different usage scenarios. In order to understand consumer preferences and generate the data needed to estimate the choice model, a method for designing choice experiments [16] incorporating usage context is first required. Secondly, the conventional choice model needs to be augmented to consider the diverse usages of customers and the usage-specific performance. In addition, the choice model is also a function of product performance attributes not influenced by usage, such as automobile engine reliability or price, and the customer demographic, attributes, such as gender or income bracket. The integrated choice model enables investigation of influence of design decisions on the potential market share for a product, and further can be incorporated into an optimization framework for determination of the preferred design [12] for a given population with specific usage contexts. The rest of this paper is organized as follows: a literature review of usage context research is provided in Section 2, followed by taxonomy of UCBD in Section 3. Section 4 describes the proposed framework and a step-by-step procedure for choice modeling in UCBD, while Section 5

2. LITERATURE ON STUDYING USAGE CONTEXT 2.1. Usage context literature in market research The marketing researchers have been long been interested in understanding and conceptualizing the underlying factors behind customer behavior and therefore are among the first to recognize the power of situational (usage contextual) influence on behavior [17-19]. Belk [20] laid out the definition of use situation as follows: “all those factors particular to a time and place of observation which do not follow from a knowledge of personal (intra-individual) and stimulus (choice alternative) attributes, and which have a demonstrable and systematic effect on current behavior.” Belk later proposed a revised stimulus-organism-response (S-O-R) paradigm [21] in which the stimulus is divided into an object and a situation, or usage context in our terminology. Relating to Belk’s S-O-R paradigm, we propose here an UCBD influence diagram as illustrated in Figure 1. STIMULUS

ORGANISM

RESPONSE

Situation Usage Context U Object

Person

Behavior

Customer S

Choice C

Product X

Figure 1: UCBD Influence Diagram based on Belk’s S-O-R Paradigm

In the context of UCBD, object refers to product and situation refers to usage context. Both usage context and product act as stimulus to a customer which leads to his/her choice behavior. Besides the conceptualization, Belk’s categorization of five groups situational characteristics (named as usage context attributes E in this work) [21] serves as the foundation for developing and classifying the usage context attributes for choice modeling; more details on this subject will be provided in Section 3. The need for considering situational (usage contextual) variables in market segmentation was recognized in the 1980s. Dickson [22] pointed out that usage situation is overlooked in market segmentation and presented a person-situation segmentation framework in which the market is explicitly segmented by groups of consumers within usage situations. The work by Christensen et al. [23] recommends stopping the common practice of segmenting customers based on their demographics and replacing it with ways that reflect how customers actually live their lives. The “substitution in use” (SIU) approach by Stefflre [24] was developed based on the premise that consumers think about product category instances within their functional roles in various possible usage contexts. As a further validation of this premise, in a more recent case study of snack foods, Ratneshwar and Shocker [3] showed that products which do not belong to the same category could be considered as comparable in certain usage context, which brings up the need for constructing different choice set alternatives based on customer profile and usage Copyright © 2010 by ASME


2


context. More recently, De la fuente and Guillen [25] analyzed consumer perceptions with regard to the suitability of household cleaning products to anticipated usage contexts, as well as their influences on purchase behavior. While their approach demonstrated the influence of usage suitability on consumer choice, the linkage between usage context and product performance as well as product design is absent. In the case of multiple usage context scenarios, Berkowitz, Ginter and Talarzyk [26] suggested to aggregate individual’s usage situation demand weighted by the situation’s frequency of occurrence or importance, which provides a guideline for handling the multiple-usage case in this work (Section 4).

Usage context attributes – E Usage context attributes E refer to the characteristics or attributes used to describe the usage context. Associated with this taxonomy is the definition of “usage context”. Belk [20] stated that use situation includes all factors that influence the customer behavior at a given time and place, except for the customer profile and product attributes. Unlike Belk, Green et al. [4] narrows down the scope of “usage context” to two major aspects, the application context and the environment context, and limits the influence of usage context to customer preferences only. Usage context in real life varies significantly across product categories. In our view, its influence on customer behavior includes the impact on product performance, choice set, and customer preference. Hence, we define the usage context in our work as “all aspects related to use of a product that have influence on customer choice behavior.” Here we emphasize two things: first, usage context covers all aspects related to the use of a product, but excludes customer profile and product attributes, which will be defined later on in this section; second, usage context influences customer behavior through product performance, choice set, as well as customer preference. Following Belk’s classification [21], usage situation (context) can be categorized into five types: physical surroundings, social surroundings, temporal perspective, task definition and antecedent states. In Table 1, we use the jigsaw example to illustrate how the usage context attributes can be defined by following these five basic categories. It should be noted that based on Belk’s classification, the scope of the usage context attributes is beyond the act of using the product, but also includes the context of purchase.

2.2. Usage context literature in engineering design While the study of usage context in consumer behavior has been prevalent for years, it had not been applied to engineering design until 1990s. In Ulrich and Eppinger’s product design and development book [27], the need for designers to envision a product’s “use environment” in identifying customer needs is emphasized. Methods have been suggested to observe a product in use as a way of gathering raw data from customers. More recently, Green et al published three successive papers [4,28-29] on a frontier design method for product usage context, which is defined as a combination of application and environment in which a product will be used. A broader concept of product design context is constructed including three contexts that influence customer preferences: usage context, customer context and market context. Their work supports the idea that context can be differentiated based upon functional attributes, indicating a link between engineering parameters and perceived usefulness, which occurs under the influence of different usage contexts. While Green’s work introduced usage context-based design, their findings on usage context are mainly focused on qualitative analysis to support concept generation. However, as proposed in our previous work [5], the benefits of understanding usage context have great potential in analytical design process as well. Through a choice model, we can understand the impact that usage context has on customer preferences, and therefore optimize product design to maximize the customer demand, or the profit contributed by the product. In this work, we further enrich the Usage Context-Based Design (UCBD) approach in our earlier work by developing a choice modeling framework and the detailed implementation procedure, as presented in the following sections.

Table 1: Five Categories of Usage Context Usage Context Type Jigsaw Example Physical surroundings Location of cutting, accessibility of an outlet, availability of workbench. Social surroundings Presence of children. Temporal perspective Purchase history, time since last purchase. Task Definition Material properties, thickness, minimal linear speed, maximal vibration level, noise and safety conditions. Antecedent states Set of saw tools already in possession, new life conditions or projects, cash at disposal.

1.

3. TAXONOMY IN CHOICE MODELING FOR UCBD As shown in the literature review section, previous works in marketing research and product design fields have employed different definitions and terminologies of usage context related variables. For instance, usage context is also called use situation; a usage context attribute is also referred to as a situational variable. To avoid confusion, we devote this section to lay out our taxonomy in UCBD. The list is consolidated based on the one presented in our previous work [5] by following the established classification in the market research domain and the specific needs associated with choice modeling. To illustrate the concepts, a jigsaw design problem is used as an example throughout this section.

2.

Physical surroundings are the most apparent characteristics of a usage. These characteristics include geographical location, weather condition, lighting, and other physical characteristics of a usage. In the case of the use of a jigsaw for cutting some board, the location where the operation must take place (indoor/outdoor), the accessibility of a power outlet, the availability of a workbench are typical examples of physical surroundings. Social surroundings provide additional information about the social situation of a usage. Whether another person is present, his/her influence on the user, and other social characteristics belong to this category. For instance in cutting a wood branch, one can prefer a jigsaw to a chain saw or a circular saw often used under these conditions, because of the presence of children nearby for reasons of safety and noise. Copyright © 2010 by ASME


3


3.

4.

5.

Usage context scenarios U refer to the most common combinations of usage context attributes E describing common usage scenarios, which can be identified through survey and using data analysis techniques such as cluster analysis. Identifying common usages can significantly simplify the data collection process compared to surveying all factorial combinations of usage context attributes E. In addition to the situation that each customer has one primary usage scenario, there are cases that multiple-usage scenarios need to be considered. The idea of usage importance indices, denoted as F, emerges from the need for considering multipleusage scenarios, where a single product is used under a series of different usage scenarios. In this case, multiple usage scenarios are weighted by their usage importance indices in the range of [0 1]. Eq. (1) shows that the usage context attributes E and usage importance indices F together define the usage context scenario U. U = (E, F ) (1) The usage importance indices can be either specified by a user, or determined based on the observations of user choices under multiple-usage scenarios. In the former case, a user is asked to provide the best estimate of the importance of a particular usage in the survey. In the latter situation, the survey questionnaire is designed to identify the importance indices through choice model estimation.

Temporal perspective refers to those aspects of the purchasing situation or to those of a given usage which are specific for a given range of time. It is common that customers show their brand loyalty when they, or their family members, have positive experiences with a particular brand. Those features are critical in the behavior, but cannot be treated as customer attributes, as they may change over time. Purchase history and time since last purchase both belong to temporal perspective. Task definition covers all features that explain the purpose of the purchase. For instance, one must consider the cutting objects (wood, steel, etc.), the properties of material (type or density of wood), the thickness of the board to cut, the minimal linear speed that is acceptable, the maximal vibration level that is tolerable, the noise and safety conditions. Antecedent states are a dimension of usage which is antecedent to the purchase. The factors for a new jigsaw acquisition may be the set of saw tools one already possesses (circular, chain, panel, bow, miter, etc.), a new life condition or project (moving from a flat to a house, or a house remodeling), and the cash at disposal.

One thing to note here is that a clear distinction is hard to find between customer profile and usage context as separate sources of influence on customers’ choice. In some cases, customer attributes may also seem like a usage context attribute, or vice versa. For example a customers’ purchase history can be regarded either as a customer attribute or as a usage context. As a guideline, we refer to customer attributes as those stable or permanent characteristics of a customer, while those temporal, transitory characteristics of a customer belong to the area of usage context. Therefore, purchase history is treated as a usage context attributes in this framework. Relating to the interest in choice modeling, we can divide usage context attributes E into performance-related and preference-related, according to the way in which they impact customer behavior. While performance-related attributes EY influence product performance Y, preference-related attributes EW have an impact on choice set and customer preference. In some cases, performance-related and preference-related usage context attributes are not mutually exclusive. For example, in using a jigsaw, the thickness of a board impacts the advance speed (performance) but also changes customers’ preferences for the saw design: the thicker the board, the more customer care about the vibration and noise level. Whether a usage context attribute is related to performance or not can be determined by prior knowledge of experienced users or by the observations of products being used; whether a usage context attribute belongs to the preferences-related type is identified through the choice model estimation process. Prior knowledge of usage context attribute’s influence on preference can be used to reduce the complexity of estimating a choice model. The usage context attributes E can be either continuous or discrete. Different settings of attributes E are referred as the levels of usage context attributes.

Customer attributes – S The customer profile S includes all stable or permanent aspects of customer attributes impacting customer choice behavior, for example, gender, age, income bracket, etc. In the choice modeling of usage context-based design, customer attributes S may have a direct impact on customers’ preference and therefore may influence their choices. Product design variables – X Product design variables describe the engineering decisions involved in product design. In the jigsaw case, blade tooth height, stroke frequency, step distance between two teeth, etc. all belong to the product design variables X. Customer desired product attributes – A Customer desired product attributes A are defined as key product characteristics that influence consumers’ choice in selecting a product. In a market survey, consumers are usually asked to rate these customer desired product attributes. They include not only engineering performances Y, but also nonengineering attributes M. Engineering performance Y refers to all performancerelated engineering attributes. Since Y plays a critical role in the engineering design process, engineering performance Y is our focus in this work. In the jigsaw example and other similar cases, engineering performance Y is further divided into performance of service results and performance of service delivery or transformations [5]. The performance of service results Yr represents the measures of the end performances of the resulting service, such as cutting precision, planarity, etc. On the other hand, the performance of service delivery or transformations Yt represents the measures of the performances related to the delivery of the service, such as linear speed, noise, etc.

Usage context scenarios - U



4


Non-engineering attributes M include all non-engineering aspects of customer desires attributes, for example, price, brand, aesthetics and other common marketing measures. Price is one of the most influential non-engineering attributes M in customers’ choice. In practice, price can enter the utility function as a single term, or can be scaled by income or log income to reflect the connection between income and price sensitivity, as shown in the case study.

potential usage context attributes. In our case study shown in Section 4, a user survey is used to collect the information of primary usage context attributes E and cluster analysis is further applied to identify the common usage scenarios U, which represent the most common combinations of attributes E. For problems with a large number of usage context attributes E, the cluster analysis becomes essential to focus the study on a set of common usages U.

Product choice set – Jn The product choice set Jn is defined as a group of product alternatives customers consider during their choice procedure. Simonson [30] showed that choices are made in context of consideration set. Since only differences in utility matter due to the nature of choice models, the selection of product consideration set exhibits great impact on customer choice. Methods for determining the appropriate choice set considering usage context are described in Section 4 (Phase II: Data Collection).

Phase II: Data Collection The choice data for building the choice model is collected in the second phase. To better understand the importance of usage context attributes and how customers make tradeoffs in the decision-making process, a user survey is conducted in which customers are asked to make a choice among several available alternatives under given usage scenarios. Since the number of products available is usually much larger than the number of products a customer can use and compare in a choice experiment within a reasonable amount of time, an optimal experimental design can be applied to reduce the number of products in the choice set to a feasible level. In the illustrative case study in this paper, a nested design of experiments on (Jn|S,U) is applied to find the optimal set of choice alternatives for respondents based on their customer profile and usage context information. Methods, either physics-based or behavior-based, can be applied here to identify the feasible choice alternatives for each respondent with given usage, which limit the number of potential products in the choice set. Moreover, the D-optimal experiment design algorithm for human appraisal surveys [16] can be used to pick out the products to include in the choice set which will provide the best model estimation. When using a survey in which each respondent is surveyed for more than one usage scenarios (but only single primary usage scenario at a time in the choice experiment), it is assumed that all respondents have some level of experience with the product and are able to differentiate between the different usage scenarios described in the survey questionnaire. A try-it-out survey is highly recommended, in which customers are asked to use the products under certain usage contexts, rate the performance, and make a choice of one of the products. A sample questionnaire of the try-it-out survey is shown in Appendix A for the case study. There are many advantages of conducting a try-it-out survey: first, hands-on experience is very important as it simulates a real purchase process; second, customers get a chance to experience the product under certain usage contexts before they choose, which ensures the relevance of usage context; and third, assessments of the product performance during the usage are given by the customer which reflect their perceived product performance during the choice process. On the other hand, the try-it-out survey often requires more resources than a paper survey, where photos or images are commonly used to present the products. An alternate approach for data collection corresponds to a two-stage, consider-then-choose decision model [31], in which respondents first narrow down the number of available products to a reasonable-sized consideration set by following a set of compensatory or non-compensatory decision rules. In the second stage, they are asked to compare the products’ features and make a choice. As a result of the two-stage data

4. FRAMEWORK FOR CHOICE MODELING IN UCBD In order to capture the impact of usage context attributes and utilize usage context information in a design process, a framework for choice modeling in usage context-based design is presented in this work. In this section, we focus on the procedure for implementing choice modeling in UCBD and discuss the potential issues involved in each phase. Our discussion follows the sequence of the four major phases for implementing choice modeling. Phase I Collect usage context information and identify usage context attributes E and common usage scenarios U (common usage identification) Phase II Design a customer choice experiment in which a unique choice set is selected based upon customer attributes S and usage context scenario U. In the experiment, choice data, customer attributes S and their usage context scenario U are collected. (data collection) Phase III Create a physics-based model or a humanappraisal-survey-based regression model for predicting engineering performance Y as a function of X and E. (linking performance with usage context attributes) Phase IV Create the choice model for market share and demand estimation (choice model estimation) Phase I: Common Usages Identification A successful product design requires an understanding of customers’ needs so that the products provided will match customers’ interest. Similarly, in the proposed framework, we start with identifying the common usages among target customers. Three ways that are widely used in identifying customer needs, namely focus group, one-on-one interview of an experienced user, and observing the product being used, are also applicable here in understanding common usages. Each of them has their own advantages and disadvantages, as detailed in [27]. Five categories of usage context listed in Table 1 can be used as a checklist in the process of determining the



5


collection, two separate models will be built for predicting the choice set and choice, respectively. However, assumptions of users’ familiarity with all usage contexts are again involved in designing a two-stage survey in which each user is asked to make a choice beyond his (her) primary usage scenario. This raises open research issues in extending the current framework to two-stage consumer decision models.

Phase IV: Choice Model Estimation Using the customer survey data collected in Phase III, a choice model for UCBD can be created and used for choice (demand) estimation. The information flow within the proposed choice modeling framework for UCBD is shown in Figure 2. At the left side of the flow, we have product design variables X, usage context scenarios U = ( E , F ) , and customer profile S. The usage context and customer profile define the target market. Product design variables X and performance related usage context attributes EY are taken as inputs for physics-based models and/or human appraisal experiments to predict engineering performance Y. The relationship between non-engineering attributes M and design variables X also need to be established. The customer choice utility W is ultimately a function of customer desired product attributes A, preference-related usage context attributes EW, and customer profile S. The deterministic portion of the choice utility for customer i , product j is shown as follows: Wij = g ( β : A, EW , S )=g ( β : M , Y , EW , S ) (2)

Phase III: Linking Performance with Usage Context This is a unique phase for UCBD applications in which a prediction model needs to be established to link engineering performance Y with design variables X and usage context attributes E. As shown in Figure 2, both design variables X and usage context attributes E have an impact on engineering performances Y, i.e., Y = f ( X , EY ) where EY stands for , performance-related attributes that influence product performance Y. A prediction model of engineering performance is critical in the choice modeling for UCBD, as it reflects how usage context influences engineering performance. The engineering performances Y can be in the form of a quantitative performance measures, such as hp, rpm, or a rating of perceived performance by the user, such as on a scale of 1-10. Two modeling approaches can be considered: a physics-based model and a human-appraisal-survey-based regression model. The physics-based model is constructed based on the physical relations following the laws of physics. Taking the jigsaw design as an example, a system of equations can be derived to calculate the forces on the user’s wrist to assess the user’s comfort level during the cutting process as a function of wood type etc. [32]. The second approach utilizes rating data given by customers in a human appraisal survey and builds a regression model to predict the ratings of performances Y as a function of product designs (X) and usage attributes E. While the physics-based model saves the time and cost of a survey and may provide a good prediction of some engineering performance, its applicability is relatively limited because many engineering performances, such as vibration levels during the operation, cannot be expressed by physical equations. A human appraisal survey, on the other hand, can be used to provide prediction of both quantitative and qualitative performance perceived by the consumers. Such surveys can be integrated into the try-it-out survey for choice modeling, as described in Phase II.

Design Variables X Target Market Usage Context Scenario U (E,F)

EY

Physics-based models and/or Human appraisal experiments

where β denotes for the model coefficients to be estimated. Using the data collected in Phase II, also depending on the expected level of details and the assumptions made, various choice modeling techniques, such as multinomial logit, nested logit, and mixed logit [33] can be used to identify the model coefficients in the choice utility function W. For the multinomial logit model used in our case study, the random error term in the choice utility function is assumed to be independently, identically distributed extreme value. With multinomial logit modeling, the heterogeneity of consumer preferences is captured by including both EW and S in the choice utility (also called systematic heterogeneity). Inclusion of EW and S explicitly in the choice model enables a better estimate of individual-level choice probability and allows for choice predictions to be made for a new target market with a different demographic and usage context distributions than the survey market used for model estimation. Other types of consumer heterogeneity, e.g., random heterogeneity, can be captured by treating the model coefficients as random variables [34].

Event

Customer Desired Attributes A Non-engineering Attributes M Engineering Performance Y=f(X,EY)

EW

Entity

Utility W=g(β:A,EW,S)

Choice Share Pni =

Customer Profile S Choice Set Jn

exp(Wni )

∑ exp(W

ni )

j

Figure 2: Information Flow within Choice Modeling in UCBD Copyright © 2010 by ASME


6


Using the choice utility function created and the choice set Jn defined based on the target market, choice share Pr (probability of choice) of a given design alternative can be predicted following the same information flow shown in Figure 2. According to the utility maximization theory, customers choose the product with the highest utility value. The choice probability for product i and customer n based on multinomial logit model is shown in the following closed form expression: exp(Wni ) Pni = (3) exp(Wni )

Please answer the following question by choosing the best description of your primary saw usage scenario. 1. The woods you are cutting are: □. soft □. medium □. hard 2. The working environment of your cutting is: □. indoor □. outdoor … Figure 3: Sample User Survey Questionnaire for Phase I

In this case study, we select wood type and working environment as two usage context attributes E for demonstration purpose; wood type is considered as a performance-related attribute EY that influences product performance Y, while both wood type and working environment are treated as preference-related attributes EW with an impact on customer preference. The wood type variable is coded as 1 for soft, 2 for medium, and 3 for hard, while the working environment variable is coded as 0 for indoor and 1 for outdoor. Based on the simulated usage data, a cluster analysis is performed and four common usages, indoor cutting for soft wood, outdoor cutting for medium wood, indoor cutting for medium wood, and outdoor cutting for hard wood are identified, as shown in Table 2.

∑ j

Market demand for each design alternative in the choice set can be forecasted by aggregating the individual choices. Furthermore, using the choice model created based on predominately single-usage surveys, the choice prediction can be expanded to multiple-usage cases using the following equation:

Wij = g ( β : A,U , S ) = ∑ g ( β : M , Y k , EW k , S ) ⋅ F k k

(4)

where Y k = f ( EY k , X ) where F k is an importance measure indicating how important the usage scenario U k is for the customer, i.e., U k = ( EW k , F k ) ; k is indicator of different usage scenarios. The above expansion is based on the assumption that the terms resulting from each usage scenario are independent from each other, and the utility function for the multiple-usage case can be treated as the weighted sum of individual usages, as Berkowitz suggested in [26].

# 1 2 3 4

5. CASE STUDY OF JIGSAW In this section, a jigsaw problem is used as a case study to demonstrate the implementation of the proposed choice modeling framework for UCBD. The jigsaw is a common power tool. Under different usage scenarios, the performance of the saw as well as the customers’ preferences for the saw change. The choice set considered in the user survey is formed by a few representative jigsaw products in the market. The four phases of choice modeling in UCBD are illustrated with the hypothetical saw design using simulated data and a few representative attributes for demonstration. A choice model is built and estimated on hypothetical data. Results are discussed which leads to several interesting implications.

Table 2: Cluster Analysis of Usage Scenarios Wood Working type environment Usage scenario U EY EW 1 0 indoor cutting for soft wood 2 1 outdoor cutting for medium wood 2 0 indoor cutting for medium wood 3 1 outdoor cutting for hard wood

•

Phase II: Data collection In this study, we assume that there are eight products available in the market. In each experiment, the respondents compare the four products that are most relevant for their usage, and make a choice of one. Table 3 shows a sample DOE of a customer survey for 2 respondents. Here, each respondent is assigned with one choice experiment under each of the usage scenarios identified in Phase I (see Table 3). Four out of eight jigsaws are selected as choice alternatives in the choice set Jn in each choice experiment. Therefore, each respondent has 16 experimental runs. Three customer attributes are included: gender (0 for male and 1 for female), skill level (1 for elementary user, 2 for experienced user, and 3 for professional user), and income (annual income in $1000). For example, respondent 1 is assigned with four choice experiments under usage scenario 1, 2, 3, 4, respectively. Under usage scenario 1, products 1, 3, 4, and 5 are chosen as choice alternatives, as shown in the last column of Table 4, while products 2, 5, 7, and 8 are chosen to form a choice set for respondent 1 under usage scenario 2. Beyond the DOE approach, the choice set can also be selected based on products’ suitability. The suitability of a product depends on customer profile S and usage context attributes E. It can be assessed using physics-based models. For example, a product resulting in over-the-limit wrist force is considered not suitable.

•

Phase I: Common usages identification Phase I is completed with the following three tasks: collect usage information, perform cluster analysis, and identify common usages. We start with a user survey in which questions about primary usages are asked. As described in Section 4, five categories of usage context can be used as a guideline for determining the usage context attributes. Figure 3 shows a small part of the sample user survey questionnaire as an example. A few typical usage context questions for a jigsaw user would include wood type, working environment, etc.



7


exp # 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Table 3: Sample DOE of Customer Survey Skill Income Usage Product Gender Level (k) Scenario in Jn 1 3 1 4 5 2 5 2 7 8 0 1 70 2 5 3 6 7 1 3 4 4 6 2 3 1 4 5 1 5 2 7 8 1 2 100 2 5 3 6 8 2 3 4 4 7

scenarios, but only single usage scenario at a time. This is based on the assumption that the respondents have some level of experience with the product under each usage scenario and are able to differentiate between the different usage scenarios described in the survey questionnaire. It should be pointed out that the DOE shown in Table 4 is not unique, and can be designed based on the number of respondents who are available. For example, when there are sufficient respondents, less choice experiments should be used for each respondent. Pairing the usage scenarios to customers’ primary usages is expected to yield the best understanding of the influence of usage context attributes. If the two-stage (consumer) decision making described in Phase II of Section 4 is considered, the DOE should be redesigned together with the survey questionnaire. The respondent will be first presented with all products available in the market and asked to choose the ones that he/she will consider for purchase, given the primary usage scenario. A similar try-it-out survey follows to collect customers’ choice. In the choice modeling phase, a separate model will be built for predicting the choice set for each customer. •

Phase III: Linking performances with usage contexts In this study, the link between product variables X, performance related usage context attributes EY and engineering performance Y, is established using a series of physics based equations based the functional principles of the jigsaw [5,32]. Among the two engineering performance Y considered in this study, the advance speed Sa is calculated as follows: Sa =

2Hd f ⋅ A s

(5)

where H d is the blade tooth height, f is the stroke frequency, A is the blade translation, and s is the step distance between two teeth. All variables in the equation ( H d , f , A and s ) are product design variables X; usage context doesn’t influence this particular performance. Based on the physical relation, user’s comfort level Pcomfort is calculated as follows:

The hypothetical data are simulated with 500 respondents, 4 choices alternatives and 4 usage scenarios (8,000 observations in total). The suggested questionnaire for respondent 1 in customer survey is shown in Appendix A. As each choice experiment has a different choice set, the products listed in the questionnaire might be different for each respondent.

Pcomfort = 1 −

Table 4: Attributes included in Jigsaw Case Study Customer Profile S Values Income uniform dist. , [50k, 150k] S1 Gender {male, female} S2 Skill level {1, 2, 3} S3 Usage Context Attributes E Working environment {indoor, outdoor} E1 Cutting board woodtype {soft, medium, hard} E2 Customer Desired Product Attributes A Price M Sa (advance speed) Yr Pcomfort (%) Yt

Mw M w − max

(6)

where M w is the wrist torque, and M w − max is the maximal wrist torque that can be delivered by the user. Since usage context attribute wood type has an impact on M w , the wrist torque is a function of both product design variables X and usage context attributes EY, while the maximal wrist torque depends on customer attributes S. •

Phase IV: Choice model estimation Assuming that each customer has one primary usage of the jigsaw (the so-called single-usage case), the choice utility function of Eq. (2) becomes: Wij = f ( S , E , A ) = α 0 + α j S + β A A + β A⋅S A ⋅ S + β A⋅ E A ⋅ E (7) Substituting all variables listed in Table 4 into Eq. (7), we have: Wij = α 0 + α S2 S2 + α E2 E2 + βYr Yr + βYt Yt (8) + β1 M / S1 + β 2 S2Yt + β3 S3Yr + β 4 E1Yt + β 5 E2Yr

Table 4 presents three categories of attributes considered for choice modeling, including three customer profile attributes S (income, gender, and skill level), two usage context attributes E (working environment, cutting board wood type, and skill level), and three customer desired product attributes A (price, advance speed, and comfort). For demonstration purpose, in this case study, each respondent is surveyed for more than one single-usage



8


Table 7: Predicted Choice Probability under Different Usage Scenarios for Eight Products Pr (%) 1 2 3 4 5 6 7 8 0.4 75.2 24.2 0.0 0.0 0.2 0.0 0.0 Single-usage

A multinomial logit model is estimated using STATA [35]. The goodness of fit, measured by the rho square is 0.82 with a log likelihood of -500.76. The parameter estimator results are shown in Table 5. The coefficients estimators, standard errors, and the significance of their p values are provided in Table 6. The price M is divided by the income S1, as customer with higher income are expected to be less sensitive to the price. The sign of M/S1 coefficient shows that price has a negative impact on the utility function. The coefficients of Yr and Yt, are both significant, showing that both performances are important in users’ choice. The coefficient for S2*Yt is significantly positive, which indicates that the female users tend to care more about the comfort than male users do. Similarly, the coefficient for S3*Yr is significantly positive, meaning that skilled users care more about the advance speed during cutting, compared with amateur users. As for the interactions between performance Yr and usage context variable E1 (indoor / outdoor) and performance Yt and usage context variable E2 (wood type), both coefficients are statistically significant, which indicates that both E1 and E2 belong to the category of preference-related usage context variable EW. Moreover, the negative sign suggests that Yt (comfort) is less important when users are cutting outdoor (E1=1), while the positive sign indicates that advance speed is more critical when users are cutting hard wood (E2=3).

Multiple-usage

Yr Yt M/S1 S2*Yt S3*Yr E1*Yt E2*Yr

Coef. 5.39 27.30 -35.86 4.13 7.42 -4.94 4.06

Std.Err. 1.42 1.98 1.76 1.51 0.49 1.62 0.49

44.1

55.4

0.0

0.0

0.1

0.0

0.0

In the single-usage case, the most preferred product 2 has a choice probability of 75.2%. However, in the multiple-usage case for the same user, product 3 has the highest choice probability of 55.4%, while the choice probability of product 2 (44.1%) is the second highest. It is interesting to note that the preference rank order of products may change when the usage scenario is different. Similar approach can be applied to forecast the choice probability of a group of target customers each with different usage scenarios by aggregating individual’s choice probability over a target population.

6. CONCLUSION In this work, we have laid out a framework to choice modeling in usage context-based design (UCBD). UCBD is an area of growing interest within the design community. Previous works have illustrated the importance of considering usage context in design, but did not present a systematic and quantitative approach to choice modeling. The main contribution of this work is the development of a choice modeling framework and a step-by-step procedure for data collection, data analysis, and model estimation to assess quantitatively the impact of usage context on product performance, the choice set, and consumer preferences. A consolidated taxonomy for the UCBD is first defined in this work by following the established classification in the market research domain and the needs associated with choice modeling. The step-by-step procedure for creating the choice mode in UCBD is then presented. For data collection, a method of try-it-out choice experiments is developed. This method is necessary to account for the different choices respondents may make conditional on the given usage context. In these experiments, each respondent is presented a set of products to choose from for a variety of different usage scenarios. This approach allows us to examine simultaneously the influence of product design, demographic, usage context attributes, and their interactions, on the choice process. Methods of data analysis are used to understand the collected choice data, as well as to understand clusters of similar customers and similar usage scenarios. This is necessary for product design so that products can be targeted for specific usage context scenarios and customer groups. Finally, the development of a choice modeling framework which considers the influence of usage context on both the product performance and the consumer preferences is presented as the key element of a quantitative usage context-based design process. In this framework, product performance is modeled as a function of both the product design and the usage context. Additionally, usage context enters into an individual’s choice utility function directly to capture the influence of usage context on product preferences. The usage context choice modeling approach in this work represents a significant expansion to traditional choice modeling approaches in the design literature. The entire process has been illustrated with a case study of the design of a jigsaw. Results illustrate that both jigsaw

Table 5: Multinomial Logit Model Estimation Results Variables

0.4

P>|z| 0.00 0.00 0.00 0.01 0.00 0.00 0.00

With the estimated choice model, future demand of a target market (including target customers and target usages) can be projected. Here we take the prediction of a single user’ choice probability as an example to illustrate the difference between single usage and multiple usage scenarios. Considering the following two scenarios as shown in Table 6 for a female user with $70k annual income and skill level 3: 1) Single-usage: she uses the product solely under single usage 1, indoor cutting for soft wood; 2) Multiple-usage: she uses the product under usage 1, indoor cutting for soft wood, with 30% relative importance and usage 4, outdoor cutting for hard wood, for 70% relative importance. Table 6: Usage Importance Index F for Choice Prediction Usage 1 Usage 2 Usage 3 Usage 4 100% 0% 0% 0% Single-usage 30% 0% 0% 70% Multiple-usage

The weighted-sum prediction formulation in Eq.(4) is used to evaluate the utilities of the eight products under both the single-usage scenario and the multiple-usage scenario. The resulting choice probabilities Pr are summarized in Table 7.



9


[12] Wassenaar, H. J., Kumar, D., and Chen, W., 2006, "Discrete Choice Demand Modeling for Decision-Based Design," Decision Making in Engineering Design, K. Lewis, W. Chen, and L. Schmidt, eds., ASME Press, New York. [13] Ben-Akiva, M., and Lerman, S. R., 1985, Discrete Choice Analysis: Theory and Application to Travel Demand, MIT Press, Cambridge, MA. [14] Michalek, J. J., Ceryan, O., Papalambros, P. Y., and Koren, Y., 2005, "Manufacturing Investment and Allocation in Product Line Design Decision Making," Proceedings of the 2004 DETC: ASME Design Engineering Technical Conferences and Computers in Engineering Conference, Long Beach, CA. [15] Kumar, D., Chen, W., and Simpson, T. W., 2006, "A Market-Driven Approach to the Design of Platformbased Product Families," AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Portsmouth, VA. [16] Hoyle, C., Chen, W., Ankenman, B., and Wang, N., 2009, "Optimal Experimental Design of Human Appraisals for Modeling Consumer Preferences in Engineering Design," Journal of Mechanical Design, 131(7), 98-107. [17] Robertson, T., and Ward, S., 1973, "Consumer Behavior Research: Promise and Prospects," Consumer behavior: theoretical sources, 3–42. [18] Lavidge, R., 1966, "The cotton candy concept: Intraindividual variability," Lee Adler & Irving Crespi. Attitude Research at Sea, 39-50. [19] Engel, J., Kollat, D., and Blackwell, R., 1969, "Personality measures and market segmentation: Evidence favors interaction view," Business Horizons, 12(3), 61-70. [20] Belk, R., 1974, "An exploratory assessment of situational effects in buyer behavior," Journal of Marketing Research, 156-163. [21] Belk, R., 1975, "Situational variables and consumer behavior," Journal of Consumer Research, 157-164. [22] Dickson, P., 1982, "Person-situation: Segmentation's missing link," The Journal of Marketing, 46(4), 56-64. [23] Christensen, C., Cook, S., and Hall, T., 2005, "Marketing malpractice: the cause and the cure," Harvard business review, 83(12), 74. [24] Stefflre, V., 1971, "New Products and New Enterprises: A Report of an Experiment in Applied Social Science," Irvine, CA: University of California. [25] De la Fuente, J., and Guillén, M., 2005, "Identifying the influence of product design and usage situation on consumer choice," International Journal of Market Research, 47(6), 667. [26] Berkowitz, E., Ginter, J., and Talarzyk, W., 1977, "An investigation of the Effects of Specific Usage Situations on the Prediction of Consumer Choice Behavior," Educators' Proceedings, Chicago. [27] Ulrich, K., and Eppinger, S., 2003, Product design and development, Irwin Professional Pub. [28] Green, M. G., Palani, R. P. K., and Wood, K. L., 2004, "Product Usage Context: Improving Customer Needs Gathering and Design Target Setting," 2004 ASME

performances and preferences for jigsaw design variables are influenced by usage context. An actual application of the method using human respondents is needed in the future to ensure methods for presenting different choice scenarios to respondents in a choice experiment are efficient and that the resulting choice model captures the influence of usage context accurately. Expansion to two-stage (consumer) decision model, as well as multiple usages and application in product family design are also interesting topics for future research.

ACKNOWLEDGEMENT Grant support from National Science Foundation (CMMI0700585 and DUE-0920047) is greatly appreciated. REFERENCES [1] Green, M. G., Tan, J., Linsey, J. S., Seepersad, C. C., and Wood, K. L., 2005, "Effects of Product Usage Context on Consumer Product Preferences," 2005 IDETC/CIE Conference, Long Beach, CA. [2] Kumar, D., Hoyle, C., Chen, W., Wang, N., Gomez-Levi, G., and Koppelman, F., 2009, "A Hierarchical Choice Modelling Approach for Incorporating Customer Preferences in Vehicle Package Design," International Journal of Product Development, 8(3), 228-251. [3] Ratneshwar, S., and Shocker, A., 1991, "Substitution in use and the role of usage context in product category structures," Journal of Marketing Research, 28(3), 281295. [4] Green, M. G., Linsey, J. S., Seepersad, C. C., Wood, K. L., and Jensen, D. J., 2006, "Frontier Design: A Product Usage Context Method," 2006 ASME Design Engineering Technical Conference, Philadelphia, PA, DETC/DTM 2006-99608. [5] Yannou, B., Wang, J., Rianantsoa, N., Hoyle, C., Drayer, M., Chen, W., Alizon, F., and Mathieu, J.-P., 2009, "Usage Coverage Model For Choice Modeling: Principles And Taxonomy," 2009 ASME Design Engineering Technical Conferences, San Diego, CA. [6] Wassenaar, H. J., and Chen, W., 2003, "An Approach to Decision-Based Design with Discrete Choice Analysis for Demand Modeling," Transactions of the ASME: Journal of Mechanical Design, 125(3), 490-497. [7] Wassenaar, H. J., Chen, W., Cheng, J., and Sudjianto, A., 2005, "Enhancing Discrete Choice Demand Modeling for Decision-Based Design," Journal of Mechanical Design, 127(4), 514-523. [8] Li, H., and Azarm, S., 2000, "Product Design Selection under Uncertainty and with Competitive Advantage," Transactions of ASME: Journal of Mechanical Design, 122(4), 411-418. [9] Michalek, J. J., Feinberg, F. M., and Papalambros, P. Y., 2005, "Linking Marketing and Engineering Product Design Decisions via Analytical Target Cascading," Journal of Product Innovation Management, 22(1), 4262. [10] Besharati, B., Azarm, S., and Farhang-Mehr, A., 2002, "A Customer-Based Expected Utility Metric for Product Design Selection," Proceedings of ASME 2002 IDETC Conference, September, Montreal, Canada. [11] Cook, H. E., 1997, Product Management: Value, Quality, Cost, Price, Profit and Organization, Chapman & Hall, London.



10


Design Engineering Technical Conference, Salt Lake City, UT, DETC/DTM2004-57498. [29] Green, M. G., Tan, J., Linsey, J. S., Seepersad, C. C., and Wood, K. L., 2005, "Effects of Product Usage Context on Consumer Product Preferences," 2005 IDETC/CIE Conference, Long Beach, CA, DETC2005-85438. [30] Simonson, I., and Tversky, A., 1992, "Choice in context: Tradeoff contrast and extremeness aversion," Journal of Marketing Research, 29(3), 281-295. [31] Hauser, J., Toubia, O., Evgeniou, T., Befurt, R., and Silinskaia, D., 2009, "Disjunctions of Conjunctions, Cognitive Simplicity and Consideration Sets."

[32] Yannou, B., Wang, J., and Yvars, P., 2010, "Computation of the usage contexts coverage of a jigsaw with CSP techniques," 2010 ASME Design Engineering Technical Conferences, Montreal, Canada. [33] Train, K. E., 2003, Discrete Choice Methods with Simulation, Cambridge University Press, Cambridge. [34] Hoyle, C., Chen, W., Wang, N., and Gomez Levi, G., 2010, "Understanding and Modeling Heterogeneity of Human Preferences for Engineering Design," Journal of Engineering Design. [35] Corporation, S., 1996-2009, "StataSE 9.2," College Station, Texas.

Appendix A: Sample Survey Questionnaire for Try-It-Out Survey (User 1, Usage Scenario 1) ************************************************************************************** Assume that you are in the market for a new saw. The four choices you have are shown as follows: Product 1 3 4 5 Picture

Type Amperage Speed Range High Height Weight Price … 1.

Jig Saw 5A 3000 SPM 10 in. 5.8 Lbs $39.97

Jig Saw 6.4 A 2800 SPM 4.6 in. 10 Lbs $97.99

Jig Saw 5.5 A 3200 SPM 13.63 in. 9.39 Lbs $69.00

Jig Saw 6.5 A 3100 SPM 4.5 in. 10 Lbs $119.00

Given the primary usage of cutting soft wood indoor, please try these products out and rate their performance on a scale from 1 to 5 (5 being the highest) in the following table: Product 1 3 4 5 Advance speed Comfort …

2.

3.

4.

Please make a choice among these four products (which product would you like to purchase? You may not make a selection if you are not happy with any of these products). ____________ Please tell us a little bit about yourself: o Are you: . male . female □ □ o What is your skill level in terms of saw usage? . Beginner . Intermediate . Experienced □ □ □ o Which one of the following groups best describes your household’s total annual income before taxes? . Under $50,000 . $50,000-59,999 . $60,000-69,999 □ □ □ . $70,000-$79,999 . $80,000-89,999 . $90,000-99,999 □ □ □ . . . □ $100,000-$109,999 □ $110,000-119,999 □ $120,000-129,999 . $130,000-$139,999 . $140,000-149,999 . $150,000 or more □ □ □ … Please tell us about your saw usage (up to three usages scenarios): Usage Scenario 1 (Primary Usage): Copyright © 2010 by ASME


11


o o

Do you use saws: . Outdoor □ Do you use saws to cut: . Soft wood □

. Indoor □ . Medium wood □

. Hard wood □

… Usage Scenario 2: o Do you use saws: . Outdoor . Indoor □ □ o Do you use saws to cut: . Soft wood . Medium wood . Hard wood □ □ □ … Usage Scenario 3: o Do you use saws: . Outdoor . Indoor □ □ o Do you use saws to cut: . Soft wood . Medium wood . Hard wood □ □ □ … Please tell us about the importance of these three usages in percentage: Usage Scenario 1: % Usage Scenario 2: % Usage Scenario 3: %



12


A Framework for Choice Modeling in Usage Context ...

A Framework for Choice Modeling in Usage Context ...

Suggest Documents

a framework for context modeling in adaptive web ... - CiteSeerX

a framework for context modeling in adaptive web ... - Semantic Scholar

Extended Framework for Modeling Choice Behavior

A context-sensitive conceptual framework for activity modeling

A Comprehensive Context Modeling Framework for ... - Springer Link

A context-sensitive conceptual framework for activity modeling

A framework for modeling evidence-based, context-influenced reasoning

CCMF, Computational Context Modeling Framework

a context-based multi-perspective modeling and simulation framework

A Good Choice for Choice Modeling - Multivariate Solutions

A Good Choice for Choice Modeling - Multivariate Solutions

A framework for context information management

A Mobile Context-aware Framework for Managing

A Framework for Context-Aware Service ... - OpenScience

A Computational Framework for Context-aware ... - CiteSeerX

A Context-Aware Service Oriented Framework for

DynamicSchema: A Lightweight Persistency Framework for Context ...

Towards a Context-aware Framework for

A Context Framework for Ambient Intelligence - SRDC

A Framework of Mobile Context Management for Supporting Context ...

A Framework for Managing Services in a Virtual Community Context

A Context Gathering Framework for Context-Aware ... - Semantic Scholar

Context-Aware Usage Control for Android

context aware framework in iot: a survey