Estimation methods for choice-based conjoint analysis of consumer preferences

Conjoint analysis, a preference measurement method typical in marketing research, has gradually expanded to other disciplines. Choice-based conjoint analysis (CBC) is currently the most popular type. Very few alternative estimation approaches have been suggested since the introduction of the Hierarchical Bayes (HB) method for estimating CBC utility functions. Studies that compare the performance of more than one of the proposed approaches and the HB are almost non- existing. We compare the performance of four published optimization-based procedures and additionally we introduce a new one called CP. The CP is an estimation approach based on convex penalty minimization. In comparison with HB as the benchmark we use eight field data sets. We base the performance comparisons on holdout validation, i.e. predictive performance. Among the optimization based procedures CP performs best. We run simulations to compare the extent to which CP and HB can recover the true utilities. With the field data on the average, the CP and HB results are equally good. However, depending on the problem characteristics, one may perform better than the other. In terms of average performance, the other four methods were inferior to CP and HB.     

Unique finite conjoint measurement

This paper studies one of the most important types of measurement that has arisen from the social sciences, the additive conjoint measurement introduced by Debreu (1960) and Luce and Tukey (1964). It also considers the variant we call additive conjoint extensive measurement. Both types of measurement are based on qualitative comparisons between multiattribute alternatives in Cartesian products of sets. This paper initiates a study of their uniqueness for the case in which all sets are finite. It considers uniqueness up to similar positive affine transformations for additive conjoint measurement, and uniqueness up to similar proportionality transformations for additive conjoint extensive measurement. Both types of uniqueness are related to sets of ‘indifference’ comparisons that correspond to sets of linearly independent equations for the measurement representation. After we explicate necessary and sufficient conditions for uniqueness, we explore specific aspects of sets of unique solutions for two-factor (two-set) additive conjoint measurement and two-factor additive conjoint extensive measurement.     

Non-additive robust ordinal regression: A multiple criteria decision model based on the Choquet integral

Within the multicriteria aggregation–disaggregation framework, ordinal regression aims at inducing the parameters of a decision model, for example those of a utility function, which have to represent some holistic preference comparisons of a Decision Maker (DM). Usually, among the many utility functions representing the DM’s preference information, only one utility function is selected. Since such a choice is arbitrary to some extent, recently robust ordinal regression has been proposed with the purpose of taking into account all the sets of parameters compatible with the DM’s preference information. Until now, robust ordinal regression has been implemented to additive utility functions under the assumption of criteria independence. In this paper we propose a non-additive robust ordinal regression on a set of alternatives A, whose utility is evaluated in terms of the Choquet integral which permits to represent the interaction among criteria, modelled by the fuzzy measures, parameterizing our approach.     

Design and analysis of predictive sorting algorithms

The focus of this research is the class of sequential algorithms, called predictive sorting algorithms, for sorting a given set ofn elements using pairwise comparisons. The order in which these pairwise comparisons are made is defined by a fixed sequence of all unordered pairs of distinct integers {1, 2, ...,n} called a sort sequence. A predictive sorting algorithm associated with a sort sequence specifies pairwise comparisons of elements in the input set in the order defined by the sort sequence, except that the comparisons whose outcomes can be inferred from the preceding pairs of comparisons are not performed. In this paper predictive sorting algorithms are obtained, based on known sorting algorithms, and are shown to be required on the averageO(n logn) comparisons.     

On orthocomplemented subspaces in p-adic Banach spaces

For linear subspaces of finite-dimensional normed spaces over K, where K is a non-Archimedean complete valued field which is not spherically complete, we study orthocomplementation as related to strictness and the Hahn-Banach property. We prove that there exist finite-dimensional normed spaces which possess non-orthocomplemented, strict HB-subspaces.     

Comparative analysis of UTA multicriteria methods

This paper reviews the main variants of the utility additive (UTA) multicriteria method, and systematically compares their predictive performance on two sets of data. It analyses both those cases where the model provides a ranking with errors and those without errors. First, it shows that the reference projects should be chosen carefully in order to elicit as much information as possible from the decision maker: a set of projects satisfying a fractional factorial plan is recommended. Second, it discusses the use of alternative post-optimality methods for solving the problem of multiple solutions and their different predictive performances. Third, it presents the results of simulations based on utility functions involving interdependence between criteria, and shows that UTA handles this problem effectively by an adjustment of its coefficients. Finally, the influence of the model's parameters on the predictive performance is also investigated.     

Following the traces:: An introduction to conjoint measurement without transitivity and additivity

This paper presents a self-contained introduction to a general conjoint measurement framework for the analysis of nontransitive and/or incomplete binary relations on product sets. It is based on the use of several kinds of marginal traces on coordinates induced by the binary relation.This framework leads to defining three general families of models depending on the kind of trace that they use. Contrary to most conjoint measurement models, these models do not involve an addition operation. This allows for a simple axiomatic analysis at the cost of very weak uniqueness results.     

Simultaneous preference estimation and heterogeneity control for choice-based conjoint via support vector machines

Support vector machines (SVMs) have been successfully used to identify individuals’ preferences in conjoint analysis. One of the challenges of using SVMs in this context is to properly control for preference heterogeneity among individuals to construct robust partworths. In this work, we present a new technique that obtains all individual utility functions simultaneously in a single optimization problem based on three objectives: complexity reduction, model fit, and heterogeneity control. While complexity reduction and model fit are dealt using SVMs, heterogeneity is controlled by shrinking the individual-level partworths toward a population mean. The proposed approach is further extended to kernel-based machines, conferring flexibility to the model by allowing nonlinear utility functions. Experiments on simulated and real-world datasets show that the proposed approach in its linear form outperforms existing methods for choice-based conjoint analysis.     

Inequalities for integral means over symmetric sets

We prove that the integral of n functions over a symmetric set L in Rn, with additional properties, increases when the functions are replaced by their symmetric decreasing rearrangements. The result is known when L is a centrally symmetric convex set, and our result extends it to nonconvex sets. We deduce as consequences, inequalities for the average of a function whose level sets are of the same type as L, over measurable sets in Rn. The average of such a function on E is maximized by the average over the symmetric set E^*.     

About the generalized LM-inverse and the weighted Moore-Penrose inverse

The recursive method for computing the generalized LM-inverse of a constant rectangular matrix augmented by a column vector is proposed in Udwadia and Phohomsiri (2007) [16] and [17]. The corresponding algorithm for the sequential determination of the generalized LM-inverse is established in the present paper. We prove that the introduced algorithm for computing the generalized LM-inverse and the algorithm for the computation of the weighted Moore-Penrose inverse developed by Wang and Chen (1986) in [23] are equivalent algorithms. Both of the algorithms are implemented in the present paper using the package MATHEMATICA. Several rational test matrices and randomly generated constant matrices are tested and the CPU time is compared and discussed.     

Adaptive thinning for bivariate scattered data

This paper studies adaptive thinning strategies for approximating a large set of scattered data by piecewise linear functions over triangulated subsets. Our strategies depend on both the locations of the data points in the plane, and the values of the sampled function at these points—adaptive thinning. All our thinning strategies remove data points one by one, so as to minimize an estimate of the error that results by the removal of a point from the current set of points (this estimate is termed “anticipated error”). The thinning process generates subsets of “most significant” points, such that the piecewise linear interpolants over the Delaunay triangulations of these subsets approximate progressively the function values sampled at the original scattered points, and such that the approximation errors are small relative to the number of points in the subsets. We design various methods for computing the anticipated error at reasonable cost, and compare and test the performance of the methods. It is proved that for data sampled from a convex function, with the strategy of convex triangulation, the actual error is minimized by minimizing the best performing measure of anticipated error. It is also shown that for data sampled from certain quadratic polynomials, adaptive thinning is equivalent to thinning which depends only on the locations of the data points—nonadaptive thinning. Based on our numerical tests and comparisons, two practical adaptive thinning algorithms are proposed for thinning large data sets, one which is more accurate and another which is faster.     

On string topology of three manifolds

Let M be a closed, oriented and smooth manifold of dimension d. Let LM be the space of smooth loops in M. In [String topology, preprint math.GT/9911159] Chas and Sullivan introduced the loop product, a product of degree -d on the homology of LM. We aim at identifying the three manifolds with “nontrivial” loop product. This is an application of some existing powerful tools in three-dimensional topology such as the prime decomposition, torus decomposition, Seifert fiber space theorem, torus theorem.     

Axiomatic characterization of a general utility function and its particular cases in terms of conjoint measurement and rough-set decision rules

Utility or value functions play an important role of preference models in multiple-criteria decision making. We investigate the relationships between these models and the decision-rule preference model obtained from the Dominance-based Rough Set Approach. The relationships are established by means of special “cancellation properties” used in conjoint measurement as axioms for representation of aggregation procedures. We are considering a general utility function and three of its important special cases: associative operator, Sugeno integral and ordered weighted maximum. For each of these aggregation functions we give a representation theorem establishing equivalence between a very weak cancellation property, the specific utility function and a set of rough-set decision rules. Each result is illustrated by a simple example of multiple-criteria decision making. The results show that the decision rule model we propose has clear advantages over a general utility function and its particular cases.     

Countable Random Sets: Uniqueness in Law and Constructiveness

The first part of this article deals with theorems on uniqueness in law for σ-finite and constructive countable random sets, which in contrast to the usual assumptions may have points of accumulation. We discuss and compare two approaches on uniqueness theorems: first, the study of generators for σ-fields used in this context and, secondly, the analysis of hitting functions. The last section of this paper deals with the notion of constructiveness. We prove a measurable selection theorem and a decomposition theorem for constructive countable random sets, and study constructive countable random sets with independent increments.     

Multiutility representations for incomplete difference preorders

A difference preorder is a (possibly incomplete) preorder on a space of state changes (rather than the states themselves); it encodes information about preference intensity, in addition to ordinal preferences. We find necessary and sufficient conditions for a difference preorder to be representable by a family of cardinal utility functions which take values in linearly ordered abelian groups. We also discuss the sense in which this cardinal utility representation is unique up to affine transformations, and under what conditions it is real-valued. This has applications to interpersonal comparisons, social welfare, and decisions under uncertainty.     

Classification systems based on rough sets under the belief function framework

In this paper, we present two classification approaches based on Rough Sets (RS) that are able to learn decision rules from uncertain data. We assume that the uncertainty exists only in the decision attribute values of the Decision Table (DT) and is represented by the belief functions. The first technique, named Belief Rough Set Classifier (BRSC), is based only on the basic concepts of the Rough Sets (RS). The second, called Belief Rough Set Classifier, is more sophisticated. It is based on Generalization Distribution Table (BRSC-GDT), which is a hybridization of the Generalization Distribution Table and the Rough Sets (GDT-RS). The two classifiers aim at simplifying the Uncertain Decision Table (UDT) in order to generate significant decision rules for classification process. Furthermore, to improve the time complexity of the construction procedure of the two classifiers, we apply a heuristic method of attribute selection based on rough sets. To evaluate the performance of each classification approach, we carry experiments on a number of standard real-world databases by artificially introducing uncertainty in the decision attribute values. In addition, we test our classifiers on a naturally uncertain web usage database. We compare our belief rough set classifiers with traditional classification methods only for the certain case. Besides, we compare the results relative to the uncertain case with those given by another similar classifier, called the Belief Decision Tree (BDT), which also deals with uncertain decision attribute values.     

Decision theory and discrete mathematics

This paper discusses recent developments that emphasize the role of discrete mathematics in decision theory. It focuses on three topics which illustrate this role. The first is the transitive simple majorities problem of determining the maximum number of linear orders on n candidates that prevent the occurrence of cyclic majorities when voters' preferences are confined to those orders. The second involves the varieties of unique solutions to simple systems of n — 1 linearly independent homogeneous linear equations in n variables of types that arise from qualitative equivalence comparisons in the measurement of subjective probabilities and utility differences. The third topic describes a computer-efficient hierarchy of stochastic-dominance relations for comparisons of risky alternatives whose outcomes lie in a unidimensional set of evenly spaced points. Although the paper is primarily expository, new results and new conjectures are included.     

Lagrangean duals and exact solution to the capacitated p-center problem

In this work, we address the capacitated p-center problem (CpCP). We study two auxiliary problems, discuss their relation to CpCP, and analyze the lower bounds obtained with two different Lagrangean duals based on each of these auxiliary problems. We also compare two different strategies for solving exactly CpCP, based on binary search and sequential search, respectively. Various data sets from the literature have been used for evaluating the performance of the proposed algorithms.     

The loop product for 3-manifolds

Let M be a connected, closed, oriented and smooth manifold of dimension d. Let LM be the space of loops in M. Chas and Sullivan introduced the loop product, an associative product of degree ?d on the homology of LM. In this Note we aim at identifying 3-manifolds with “non-trivial” loop products. To cite this article: H. Abbaspour, C. R. Acad. Sci. Paris, Ser. I 338 (2004).     

Ensemble of a subset of <Emphasis Type="Italic">k</Emphasis>NN classifiers

Combining multiple classifiers, known as ensemble methods, can give substantial improvement in prediction performance of learning algorithms especially in the presence of non-informative features in the data sets. We propose an ensemble of subset of kNN classifiers, ESkNN, for classification task in two steps. Firstly, we choose classifiers based upon their individual performance using the out-of-sample accuracy. The selected classifiers are then combined sequentially starting from the best model and assessed for collective performance on a validation data set. We use bench mark data sets with their original and some added non-informative features for the evaluation of our method. The results are compared with usual kNN, bagged kNN, random kNN, multiple feature subset method, random forest and support vector machines. Our experimental comparisons on benchmark classification problems and simulated data sets reveal that the proposed ensemble gives better classification performance than the usual kNN and its ensembles, and performs comparable to random forest and support vector machines.