Study Guides, Notes, & Quizzes

A multidimensional approach to conjoint analysis

Description
A multidimensional approach to conjoint analysis
Published
of 11
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/233382545 A Multidimensional Approach to ConjointAnalysis  Article   in  Applied Stochastic Models and Data Analysis · December 1998 DOI: 10.1002/(SICI)1099-0747(199812)14:43.0.CO;2-W CITATIONS 11 READS 43 3 authors:Some of the authors of this publication are also working on these related projects: Exploratory methods for multilayer networks   View projectCarlo LauroUniversity of Naples Federico II 87   PUBLICATIONS   1,415   CITATIONS   SEE PROFILE Giuseppe GiordanoUniversità degli Studi di Salerno 26   PUBLICATIONS   82   CITATIONS   SEE PROFILE Rosanna VerdeSecond University of Naples 84   PUBLICATIONS   484   CITATIONS   SEE PROFILE All content following this page was uploaded by Giuseppe Giordano on 29 December 2016. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the srcinal documentand are linked to publications on ResearchGate, letting you access and read them immediately.  * Correspondence to: C. N. Lauro, Dipartimento di Matematica e Statistica, Universita` degli Studi di Napoli,  && FedericoII ''  Complesso Monte Sant ' Angelo, Via Cinthia, 80126 Napoli, Italy. - The paper has been supported by a C.N.R. (no. 9301789) grant:  Analisi dei Dati e Statistica Computazionale , coord,C. N. Lauro. CCC 8755 } 0024/98/040265 } 10$17.50  Recei v ed June 1997  Copyright  1998 John Wiley & Sons, Ltd.  Re v ised June 1998 APPLIED STOCHASTIC MODELS AND DATA ANALYSIS Appl .  Stochastic Models Data Anal .  14 , 265 } 274 (1998) A MULTIDIMENSIONAL APPROACHTO CONJOINT ANALYSIS - C. N. LAURO, *  G. GIORDANO, R. VERDE  Dipartimento di Matematica e Statistica, Uni v ersita  `  degli Studi di Napoli,  **   Federico II  ++   Complesso Monte Sant  +   Angelo,Via Cinthia, 80126 Napoli, Italy SUMMARY In this paper we propose an alternative approach to conjoint analysis (CA) based on the principalcomponent analysis onto a reference sub-space (PCAR). Following this technique, the results of a CA areconsiderably enriched by some graphical interpretation tools and optimal synthesis of the individualestimated part-worth coe $ cients are also obtained with respect to hierarchized judges sub-popula-tions. Copyright  1998 John Wiley & Sons, Ltd. KEY WORDS:  conjoint analysis; principal component analysis onto a reference subspace 1. INTRODUCTIONConjoint analysis   deals with  preference judgements , expressed by  individuals  (hereafter called  judges ), about a set of   stimuli  (products or services). Such stimuli are described by several attributes .Conjoint analysis, in particular, aims at evaluating the relative importance of the levels of eachattribute in establishing the global preference associated to the di !  erent stimuli.In this paper we mainly consider the so-called  metric conjoint analysis  (CA) approach in whichthemultiplelinearregressionmodel is used inorder to estimatethe  part - worth coe  . cient or  utility of each level.For each judge a regression model is established. The dependent variable >   ( g " 1, 2 , G ) isa vector of judgements expressed as ranks on a set of   Q  stimuli; the explanatory variables arebinary indicator vectors associated with the levels of each attribute  X   ( k " 1, 2 , K ) based ona suitable experimental design. In order to reduce the number of stimuli, the experimental designis usually an orthogonal fractionated design.The basic CA model is expressed as >  " b   X  #2# b  X  #2# b  X  # e   (1)  where  e   is the error term. The regression coe $ cients  b   can be interpreted as individual & part - worth coe  . cients 1  of each level.Hereinafter we will use the following matrix notation: let X  be the design matrix of dimension Q  K , whose rows refer to  Q stimuli  and the columns are the levels of   p  attributes; each attributehas  K   ( h " 1, 2 , p ) levels, the total number of levels is equal to  K "    K  .We assume that the stimuli are characterized by just one level for each attribute, so that theattributes can be represented by indicator matrices.Thus, the design matrix  X  is a partitioned matrix consisting of   p  juxtaposed indicatormatrices  X   ( h " 1, 2 , p ) of general element  x  3  0, 1  , ( i " 1, 2 , Q ): X " [ X   X   2  X   2  X  ] (2)Let  Y  be the  preference  matrix of dimension  Q  G , whose columns consist of the judgementsgiven by  G  judges with respect to  Q stimuli . The judgements, expressed by ranking or ratings, areusually de " ned on an ordinal scale ranging from 1 to  Q ; usually the highest value corresponds tothe most preferred stimulus.According to the above notation, the CA model (1) can be rewritten as the followingmultivariate regression model: Y " XB # E  (3)where respectively,  B  is the matrix of the  G  individual part-worth coe $ cients associated to the K  attribute-levels and  E  the error term.The global preference of each stimulus can be computed adding up the part-worth coe $ cientsrelatedto the proper levels of the stimulus: XB  < u , where  u  is a  G -dimensionvector of unit elements.Usually the part-worth coe $ cients, estimated by individual models, are summarized by thearithmetic mean b    M " 1 G B  < u  (4)Sometimes such synthesis are obtained for homogeneous subsets of respondents.The problems in interpreting CA results mainly concern with the following aspects:(i) the synthesis of the  individual  part-worth coe $ cients;(ii) the validation of the additivity hypothesis;(iii) the classical regression hypothesis for interpreting results and, even if CA can be seen asa multivariate model, usual geometrical tools for interpreting results are not considered inliterature.In order to improve the interpretation of conjoint analysis results and overcome the mentionedproblems, in this paper we propose an alternative approach to CA in the context of multidimen-sional data analysis.With the aim of selecting additive hierarchical patterns of preferences,  " rstly we propose tosummarize the part-worth coe $ cients by performing a factorial decomposition, of the conjoint-analysis-estimated parameter matrix, through a principal component analysis. So doing theindividual part-worth coe $ cients are optimally aggregate with respect to each factorial axis bymeans of a suitable weighting system re # ecting the judges '  preferences.266  C. N. LAURO, G. GIORDANO AND R. VERDE Appl .  Stochastic Models Data Anal .  14 , 265 } 274 (1998)Copyright  1998 John Wiley & Sons, Ltd.  Moreover, aiming at improving the CA interpretation, the proposed approach considers therepresentation of the relationships existing between attribute-levels and individual preference judgements on a factorial subspace.In order to take into account the explanatory power of the adopted experimental design, wepropose a more re " ned geometrical approach based on  principal component analysis ontoa reference subspace  (PCAR).  It is worth noting that the criterion optimized in PCAR (i.e. variance explained by theexperimental design) is consistent with the one maximized in the CA model estimation. Thus, itcan be shown that the predicted preference structure is the same as in CA and the estimatedpart-worth coe $ cients are just related to the PCAR eigenstructure.Furthermore, the PCAR geometrical representations allow to enrich even more CA by jointplots of attribute-levels, judges and stimuli.An application of the proposed approach to real data (a co !  ee-preferences analysis) allows tohighlight its properties bringing more insight with respect to the results of Conjoint Analysis. Inthe application, some resampling-based tools are used for evaluating the stability of the results ina non-parametric context.2. OPTIMAL PART-WORTH DECOMPOSITION: FACTORIAL APPROACHESAccording to the metric conjoint analysis approach, the part-worth coe $ cients estimates B  < " ( X  X )  X  Y  are obtained by minimizing the expressionmin B  Y ! XB    (5)Due to the peculiar structure of   X , it can be easily seen that its rank is equal to ( K ! p ), andconsequently we cannot compute the inverse of the matrix X  X . Di !  erent alternatives can be usedas solution (5), for example ( X  X )   the Moore } Penrose-generalized inverse, or otherwise, consid-ering the inverse of   X  X , when  X ( Q  ( K ! p )) is a  full - rank  matrix obtained by dropping onecolumn  per  block of   X , and substracting it from the other columns of the same block.If we are interested in estimating only main e !  ects of each attribute, then we can consider thediagonal matrix    , obtained by setting to zero the o !  -diagonal elements of ( X  X ). Then, thepart-worth coe $ cients are the elements of the matrix  B  <  computed according to the followingexpression: B  < "     X  Y  (6)According to (6) the estimated preference matrix is given by the following expression: Y  < " X     X  Y  (7)The part-worth coe $ cients described by the elements of the columns of   B  <  correspond to themultiple regression coe $ cients of the individual preferences computed by the ordinary leastsquares.Note that the  non - metricconjoint analysis  approach   aims at computing part-worthcoe $ cientsin such a way that the rank order, reconstructed by the estimated model, is the same as theoriginal, or violates the ordering as little as possible. So, in our approach we can consideralternatively the part-worth coe $ cients computed under the monotone constraints on thedependent variables. CONJOINT ANALYSIS  267 Appl .  Stochastic Models Data Anal .  14 , 265 } 274 (1998)Copyright  1998 John Wiley & Sons, Ltd.  In order to obtain suitable synthesis of   part - worth  coe $ cients in a low-dimensional subspace,in terms of principal components of   B  < , we look for the axes of maximum variability of thepreference judgements as solutions of the following equation: B  <  B  <   "      (  " 1, 2 , m ) rank( B  < )) (8)under the orthonormality constraints:      " 1 and       " 0  ∀  O  ; where     and     are,respectively, the   th eigenvalue and eigenvector of the  B  <  B  < .Aggregate part-work coe $ cients with respect to the factorial axes are thus obtained as b  " B  <    (  " 1, 2 , m ) (9)They represent  optimal aggregate part - worth coe  . cients  being obtained for individual  part - worth  by means of a weighting system based on eigenvectors     (  " 1, 2 , m ). They allow todecompose the preference structure expressed by  B  <  with respect to the axes of maximumvariability in a hierarchicaland additive way. Obviously,this solution for the  b  is not unique andit depends on the particular choice of the inverse of ( X  X ).Nevertheless,this approach, based on a PCA synthesis, does not allow to take into account theexplicative power of attributes of the design matrix  X  explained by the factorial axes.In order to decompose the variance of the preference judgements explained by the attributes,expressed as  trace  ( Y  X ( X  X )  X  Y ), we consider the eigensolution, under the usual orthonormal-ity constraints, of the following equation:( Y  X ( X  X )  X  Y ) z  "   z   (  " 1, 2 , m ;  m " rank( X )) (10)In the expressions above the ( X  X )   is the Moore } Penrose-generalizedinverse matrix, while    and  z  are, respectively, the   th eigenvalue and eigenvector of ( Y  X ( X  X )  X  Y ).From a geometrical point of view, this approach is equivalent to the  principal componentanalysis onto a reference subspace  (PCAR) de " ned as the principal component analysis of the bestimage of   Y , and is  Y  < " P  Y , onto the sub-space  S   spanned by the columns of the matrix  X . Infact, according to the properties of the Moore } Penrose-generalized inverse,  P  " X ( X  X )   X   isan orthogonal projection operator onto  S  .An alternative solution can be obtained in relation to the approach proposed by D ' Ambra,Lauro,   when multiple reference sub-spaces  S    are considered. It consists in a PCA of the imagesof the column vectors of the preference-judgements matrix  Y , on the disjoint subspaces  S   ,spanned by the column vectors of the design matrix partitions  X   ( h " 1, 2 , p ).The maximum-inertia axes are obtained as solution of the PCAR characteristic equation,under orthonormality constraints:( Y  X ( X  X )  X  Y ) w  " Y       P   Y w  "   w   (  " 1, 2 , m ) (11)where:  P   " X  ( X   X  )  X    ( h " 1, 2 , p ) are orthogonal projectors associated to  S   .Similarly than in (9), the aggregate part-work coe $ cients can be obtained with respect to thefactorial axes  w  (  " 1, 2 , m ) as b     I  " B  < w  "     X  Y w  . (12)268  C. N. LAURO, G. GIORDANO AND R. VERDE Appl .  Stochastic Models Data Anal .  14 , 265 } 274 (1998)Copyright  1998 John Wiley & Sons, Ltd.
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks