This action might not be possible to undo. Are you sure you want to continue?

19(2) pp 213 – 226 CUSTOMER GENERALISED SATISFACTION MAXIMUM APPROACH MEASUREMENT ENTROPY MODELS:

AMJAD. D. AL-NASSER Department of Statistics, Faculty of Science Yarmouk University, Irbid Jordan amjadn@yu.edu.jo

ABSTRACT This paper presents the methodology of the Generalised Maximum Entropy (GME) approach for estimating linear models that contain latent variables such as customer satisfaction measurement models. The GME approach is a distribution free method and it provides better alternatives to the conventional method; Namely, Partial Least Squares (PLS), which used in the context of costumer satisfaction measurement. A simplified model that is used for the Swedish customer satis faction index (CSI) have been used to generate simulated data in order to study the performance of the GME and PLS. The results showed that the GME outperforms PLS in terms of mean square errors (MSE). A simulated data also used to compute the CSI using t he GME approach. KEYWORDS Generalised Maximum Entropy, Partial Least Squares, Costumer Satisfaction Models. 1. INTRODUCTION Much has been written in the past few years on Customer Satisfaction measurement models in order to study the relationship between satisfaction and market share, and the impact of customer switching barriers (Fornell 1992) in terms of customer satisfaction Index (CSI). A Customer Satisfaction Index quantifies the level of profitable satisfaction of a particular customer base and specifies the impact of that satisfaction on the chosen measure(s) of economic performance. Index can be generated for specific businesses or market segments or "rolled-up" into corporate or divisional measures of performance. The index is used to monitor performance improvement and to identify differences between markets or businesses. The CSI score provides a baseline for determining whether the marketplace is becoming more or less satisfied with the quality of products or services provided by individual industry or company. Traditional approaches in estimating CSI from especial linear structural relationship models have raised two important issues; the first concerns with the Maximum Likelihood (ML) approach 213

A proposed method can be used t o compute CSI based on statistical information about customer satisfaction measurements model.2. Helland(1990). Lohmoller (!989). Appendix A is describe the PLS algorithm.214 Customer satisfaction measurement models developed by Jöreskog (1973). After several versions in its development.1: The American Customer Satisfaction Framework . There are many authors who described PLS algorithms in their articles (Geladi and Kowalski (1986).1) which is a cross-industry measure of the satisfaction of customers in United States households with the quality of goods and services they purchase and use (Bryant 1995). namely. Perceived quality Customer Complaints Perceived Value Customer Expectation Customer Satisfaction Customer Loyalty Figuer. Wold (1980) presented the basic design for the implementation of PLS algorithm. a good examples on these models are the American customer satisfaction index (see Figure. However. Partial Least Square (PLS). In the literature. It differs from simple path analysis in that all variables are latent variables measured by multiple indicators. which can be depicted in a path diagram to analyse a set of relationships between variables. which estimates the parameters of the model by the maximum likelihood method using Davidon-Fletcher-Powell algorithm. 2. 1975) which he calls “soft modelling”. which have associated error terms in addition to the residual error factor associated with the latent variable. COSTUMER MODELS SATISFACTION MEASUREMENT Customer satisfaction model is a complete path model. The PLS method was developed by Wold (1973. which is an economic indicator. Bremeton(1990) and Garthwaite(1994) ). The other research issue concerns with the distribution free approach. Helland (1988). or “Nonlinear Iterative Partial Least Square” (NIPLAS). represents in Figuer. The Swedish CSI (Fornell 1992) and European’s CSI (Gronhlodt et al 2000) models are used PLS method. the PLS method is usually presented by two equivalent algorithms. and the European customer satisfaction index model. This paper will discuss the GME estimation approach in solving the customer satisfaction models.

e. The parameters. z is a vector of residuals or errors in equations. and B G is a matrix of coefficients of the effects of exogenous variables ( x on equations. The structural equation model (1) refers to relations among exogenous variables ( i.2 The European Customer Satisfaction Framework Many researchers from various disciplines have used Linear Structural Relationship (LISREL) as a tool for analysing customer satisfaction models. we observe a number of indicators called outer variables and described by two equations to represent the measurement . The inner varia bles are unobserved. and endogenous variables (i. and h xwhich is a vector of latent exogenous variables are related by a structural relation. h which is a vector of latent endogenous variables. The inner variables in this equation. a variables that is caused by one or more variable in the model). the first part is structural model (1) that represents a linear system for the inner relations between the unobserved inner variables. a variables that is not caused by another variable in the model).e. The second part is the measurement model (2) and (3) that represents the outer relation between observed and unobserved or latent and manifest variables.Al-Nasser 215 Image Customer expectation Perceived value price Customer satisfaction Loyalty Perceived quality of product Perceived quality of Figuer. The general and formal model of customer satisfaction can be written as a series of equations represented by three matrix equations Jöreskog (1973): (m x 1) = (m x m) y (p x m) * * * (m x 1) + G x n) * (m + e(p x 1) (n x 1) + (m x 1) (1) (2) (3) y x 1) = (p x x 1) = (q (m x 1) x (q x n) (n x 1) + d(q x 1) The structural equation models given in (1-3) have two parts. x’s) However. B is a matrix of coefficients of the effects of endogenous on endogenous variables. Instead.

L y is a matrix of coefficients. logical or empirical considerations. and these may be treated as hypotheses to be confirmed or disconfirmed and the rational of their specification in the model depend on methodological. and xis a q x 1 vector of measures of independent variables. The parameters. (ii) It is assumed that the inner variables ( . or loadings. G x n) . of h L xon the unobserved independent variables ( x Moreover.p k s uch that i =1 p = 1. The model given in (1-3) has many assumptions that may be perceived as restrictions. (m (q x 1) (p x 1) . The means of all variables are assumed to be zero. The measures in these two equations. and d a vector of errors of measurement of dis x. but the covariance matrices of e eand dneed to be diagonal.…. That is. MAXIMUM ENTROPY (GME) ESTIMATION 3. ) are not correlated with the error terms ( e. e(p x 1 ). E( ) = E( ) = E( ) = E(e) = E(d) = 0 E(ee `) = where 2 e and 2 d are 2 e. and are uncorrelated.p 2 . theoretical. Given information about the variables x recover the unknown parameters (m x 1) .216 Customer satisfaction measurement models model (2) and (3) which specify the relation between unobserved and observed. are z uncorrelated with the components of e and d The later are uncorrelated d. d). These i . which mean that the variables are expressed in the deviation scores. or latent and manifest variables. of y on unobserved dependent variables ( h ). d (q x 1) (m x the objective in this article is to y ( p x m ) . Moreover.x2 . GENERALIZED APPROACH Conventional work in information theory concerns with developing a consistent measure of the amount of uncertainty. x (q x n) and the disturbances by using the GME principle. and E( dd `) = 2 d diagonal matrices. e is a vector of errors of measurement x). yis a p x 1 vector of measures of dependent variables. z also.…. Suppose we have a set of events {x k 1 . xk }whose probabilities of occurrence are p 1 . That is. E(e `) = E( d `) = E( (iii) `) = 0 is nonsingular with zeros in its diagonal elements. or loadings. and L x is a q x n matrix of coefficients. as well. of y. and y m ) . these assumptions: (i) The elements of h and x and consequently those of h x. but they may be correlated with each other.

However. Hence the GME is to recover the unknown probabilities. the unknown parameters in customer satisfaction model are not in the form of probabilities and their sum does not represent the unity. which represents the distribution function of the random variable.2.m s=1 1 = s g = ij g f ij l ijl ij l = 1.m .p2 .1. Golan et al (1996) developed GME procedure for solving the problem of recovering information when the underling model is incompletely known and the data are limited.2. and the generalised inverse of . in order to recover the unknowns in the model we need to rewrite the unknowns in terms of probabilities values. k = 1. customer satisfaction and loyalty. n l =1 1 = . Therefore.….d) + .…. In this context we need to reparametrized the unknowns as expected values of discrete random variable with two or more sets of points.2. Using an axiomatic method to define a unique function to measure the uncertainty of a collection of events.Al-Nasser 217 probabilities are unknown but that is all we know concerning which event will occur.)-1 + e x (5) where I is the identity matrix. The GME principle stated that one chooses the distribution for which the information (the data) is just sufficient to determine the probability assignment.1 is y (I . . l L b f jk s = 1. partial or incomplete.….m . S S b jk = L z b jks jk s . The best approximation for the distribution is to choose p i that maximizes (4) with respect to the data Consistency constraints and the Normalization-additivity requirements. Al-Nasser et al (2000) developed the GME method for estimating Errors -In-Variables models and Abdullah et al (2000) used the same approach to study the functional relationship Between Image. k pi ln( pi ) (4) i= 1 The amount (–ln(p i )) is called the amount of self information of the event x . i = 1.….p k }. The average i of self-information is defined as the entropy. j = 1. RE-PARAMETERISATION In order to illustrate the use of GME in estimating the model given in (1-3) we rewrite this model as: y= y x-1 G(I x -1 ) (x .2. j = 1. Shannon (1948) defines the entropy or the information of entropy of the distribution (discrete events) with the corresponding probabilities P = {p as 1 . that is to say. which is the main characteristic of the probability density function. 3.…. H ( P) = where 0ln(0) = 0.

vy . maximising this function subject to the consistency and the add-up normalisation constraints can solve the problem.d .….p pe e =1 e= 1 Using these re-parameterisation expressions the model (5) can be rewritten as where x x y (b. p jc p jc C dy pjc c =1 1 = c z j = T v w jt jt .2. q qr r=1 r=1 e = p E vy wy . p = 1.wx .…. q = 1. j = 1. n = 1.2.…. j = 1.f. p = 1. qr qr R wx = 1. pe pe E wy = 1.….2. p.w) L jc y pj c d y pj c - -1 1 iks z ik s b iks * - Lx d x qi a qia q ia ijl g ij l f ij l qr x q vx wx qr qr + jt v w jt (6) jt + e v y wy pe pe x .dy .….2.m a =1 1 = ly = pj C Ly d y .2.v) will be chosen such that they The weight support of the disturbance parts (v are symmetric around zero for all j.d .w .f. i = 1. q .m jt t =1 t= 1 d = q R vx wx . However.w) = .w . q = 1. the GME system can be expressed as a non-linear programming problem subject to linear constraints.f.w .….dy . The model reformulation using the GME is given by: Maximize x y H(b. the choice of the support of the other parameters are chosen to span the possible parameter space for each parameter (Golan et al (1997) Golan et al(1996) and Al-Nasser and Abdullah (2000)).d . T w = 1.w) = yp = x x (b.2 REFORMULATION AND SOLUTION Given the re -parameterisation. 3.wy .dy . Its objective function can be stated in scalar summation notations.w .2.218 Customer satisfaction measurement models lx = qi A Lx d x . …. q and p.2. qi a q ia a A dx q ia = 1.

…. i = 1.wx . to solve this non-linear programming system a numerical method should be used. q . j = 1. …. n = 1.2.dx . The following diagram describes the GME algorithm in four steps.….….w .2.m . k = 1.….dy .….w . j = 1. p = 1. + . p.….2.2.2.….Al-Nasser 219 jks b - jks ln(b ) jk s ijl f ln( f ) ijl ij l qia d x ln(d x ) q ia q ia dy ln(d y ) pjc pjc pjc jt w ln(w ) jt jt qr wx ln(wx ) qr qr pe wy ln(wy ) pe pe Subject to (i) y (ii) 1 (iii) 1 (iv) (v) (vi) 1 (vii) (viii) p S = b f jk s x y (b.2.2.2. q qr r =1 E wy = 1.….2. 2 In this system we have ( p+ m 2 + nm + qn + pm + m + q+ p) equations including (Sm nmL + qnA + pmC + mT + qR + pE) unknowns.2.m = .….f.w) = .d .w . However. q = 1. q = 1. p pe e =1 where x y (b. i = 1.f.m s =1 L ijl l= 1 A dx dy q ia a =1 C p jc c =1 T w = .m .…. p = 1. j = 1. j = 1.w) as given in (6). n = 1.2.dy .m jt R t= 1 wx = 1.

and one = 1 1 + 2 2 + 3 3 + where 1 . The measurement models for variables are formative (Bagozzi and Fornell (1982)) and given by: .C et al (1999).2 Rewrite the model with the new reparametrizatio n as the data constraint Formulate the GME problem as non-linear programming problem in the following form Objective function = Shannon’s Entropy Function Step . and y for the variable. The inner structure is defined as 3 . 2 and is disturbance term. that consists of three exogenous variables 1 . The consistency constraints. 2 . proposed by Claes . The 3 are regression coefficients.1 Reparametrized the unknown parameters and the disturbance terms (if they are not in probabilities form) as a convex combination of expected value of a discrete random variable Step . A SIMULATION STUDY To illustrate the GME estimation method. which represents the new formulation of the model Step .3 With respect to the following constraints 1. and manifest variables are denoted as x for the variables. we conducted a simulation study using simplified model that is used for the Swedish customer satisfaction index. and endogenous variables .220 Customer satisfaction measurement models Generalized Maximum Entropy Algorithm Step . The Normalization constraints 2.4 Solve the non-linear programming by using numerical methods 4.

0. Given this structural model.Generate 100 random samples each of size 15.8).577E-4 1.The coefficients are initialled by (0.042E-4 3.The error terms dand eare generated from the Uniform distribution U(0.449E-4 2.679E-4 5.675E-4 4.788E-3 4.30.788E-2 4. 0.6).046E-3 3.009E-2 1.606E-2 3. TABLE-1 MSE of The Estimated Coefficients By Using The GME N MSE( 15 20 25 30 40 7. 0.8.For the formative model the x values were generated from symmetric Beta distribution with parameters (6.1.1. the simulation study was done under the following conditions: 1.827E-4 7.All pcoefficients are set to be 1/3.030E-2 1.406E-3 4.0. 6.266E-2 2.9. 4. The measurement y= 1 y= 2 y= 3 y= 4 1 2 3 4 + + + + e1 e2 e3 e4 where are coefficients and eare disturbance part.970E-4 3.470E-2 . 2.406E-2 4. the Beta varieties from RNBET and the GME system were solved by using successive quadratic programming method to solve a non-linear programming problem depending on NCONF based on the subroutine NLPQL. Under these conditions the results for the MSE are given as shown in Table (1) for the GME approach and in Table (2) for the PLS method.The coefficients are initialled by (1.1).348E-4 ˆ g 1 ) MSE( ˆ g 2 ) MSE( ˆ g 3 ) MSE( lˆ ) 6.1).965E-2 8.Using the Fortran power station environment programs linked to IMSL library.974E-3 3.032E-3 pˆ ) MSE( 6.915E-3 4. and the model for the variable is reflective and given by: dare disturbances. 0.081E-2 2. 1. while generated from the standard Normal distribution. 3.25. 5.493E-4 5.40 from the given model.111E-4 4.Al-Nasser 221 = p1 x1 + p2 x2 + p3 x3 + d1 = p4 x4 + p5 x5 + p6 x6 + d2 2 3 = p7 x7 + p8 x8 + p9 x9 + d 3 1 where pare regression coefficients. all Normal varieties were generated from the subroutine ANORIN. 7.20.

The GME estimated values are given in the following diagram: x x x x x x x x x x x 2 x 1 y y CS y y 3 .228E-1 2.037E-1 1. and it gives better estimate with a very small sample size.629E-1 1. the model described in this article for the Swedish customer satisfaction index used under conditions (1-7) given in the last section to generate a hypothetical data of size 12.528E-2 ˆ g 1 ) MSE( ˆ g 2 ) MSE( ˆ g 3 ) MSE( ) lˆ 15 20 25 30 40 Where 2.6287 1.421E-1 MSE( 1.857E-1 pˆ (the estimate mean of the coefficients in the measurement models for lˆ (the estimate mean of the coefficients the measurement model for the variables) and variable).105E-1 8.890E-2 2.474E-1 1.178E-1 9.3143 9. From the results it could be note that the GME outperform the PLS method.1 APPLICATION TO SIMULATED DATA In order to illustrate the GME algorithm in solving customer satisfaction models to compute CSI.842E-1 3.456E-1 4.874E-1 3. 4.854E-2 5.9715 1.370E-2 5.570E-1 1.148E-1 1.5772 1.716E-1 2.086E-1 0.222 Customer satisfaction measurement models TABLE 2 MSE Of The Estimated Coefficients By Using The PLS N MSE( p) ˆ 6.425E-2 7.845E-2 7.

Mi (C ) CS = ×100 Ma S ) -n Mi S(C ) (C I x S n S Where E(. Bagozzi. Bremerton G. England: Ellis Horwood limited. M. The performance of the GME approach investigated and compared with an existing technique from the literatures. 3. then the CSI computed as follows (Bryant E. D. B (1995)): E (C ) . partial least squares (PLS). Al-Nasser. Barbara. the GME can be considered as an alternative to the conventional method PLS to measure customer satisfaction index. American Customer Satisfaction Index: Methodology Report. Toronto: John Wiley. 5. P. B. Fornell. which provide a better approach as it is meant for situations with limited or incomplete data and it is more robust against departures from classical assumptions on statistical distributions.) denote the expected. 1982. & Fornell. MI 48109-1234.) and Max(. University of Michigan Business School.). ACKNOWLEDGMENT S This research has been supported by a grant from Yarmouk University REFERENCE S 1.. Therefore. Vol. the CSI using GME model is 82. w i are the weights. Total Quality Management 11(6): 826-829.03. Measurement and Evaluation . Max (CS ) = i i 4 w Max ( y ) i i i=1 i=1 where. E. Evaluating Functional Relationship Between Image. . 4. Those of corresponding manifest variables determine the minimum and the maximum values of CS latent variable: Min ( CS ) = 4 w Min ( y ) . An Arbor. Customer Satisfaction and Loyalty using General maximum Entropy. the minimum and the maximum value of the variable.Al-Nasser 223 CS represents the latent variable for customer satisfaction. A. 2000. Therefore. Chemometrics: Applications of Mathematics and Statistics to Laboratory Systems . respectively. and the GME approach outperform the PLS in terms of MSE. Abdullah. C. II. R. The Second Generation of Multivariate Analysis. (ed). In C. The CSI results indicate that the service quality regarding to the simulated data is Excellent. National Quality Research Center. & Nooreha. H. 42-60. 2. CONCLUDING REMARKS In this article we proposed the generalised maximum entropy (GME) estimation approach to the customer satisfaction models. R. 1990. 1995. Bryant. Theoritical Concept.Min(. It can be observed from the simulation results that PLS are unreliable when the sample size relatively small. for this example a uniform weights were used . Measurement and Meaning.

Uni Putra Malaysia. L. J. J. Lohmoeller. In Kmenta. Econometrics 79: 23-51. Scand. Applied Probability Trust. 27: 379-423. J & Ramsey. 18-19 July. P (Ed). 2000. Total Quality Management. H. D. J. H. S. W. Structural Equation Models in the Social Sciences . B & Wan Endut. 89(425): A 122-127. Dept. A National Customer Satisfaction Barometer: The Swedish Experience. & Karp. & Kristensen.224 Customer satisfaction measurement models 5. 1986. Al-Nasser. Helland. B. Partial least-Squares regression: A Tutorial. 1988. Bell System Technical Journal. J. New York: Springer-verlag. H. O (Eds). L. 18. Geladi. . Judge. Golan. 56:6-21. 17. 383-407. Barttlet. Statist. Applied Statistics. 17(2): 581-607. C. t 13. & Kowalski. & Westlund. In Krishnaiah. 2000. S. 1990. S. G. 12. Golan. In Gani. 1992. 8. Simula . 11(6):509-514. On the occation of his sixty-fifth birthday. 7.. J. Fornell C. The Nonlinear Itertive Partial Least Square (NIPALS). Gronholdt. Martensen. E (1948). J. M. 19. J. B. 1980. Claes. Judge. H. of Economic Dynamics and Control 20: 559-582. New York: Academic.. An Interpretation of Partial Least Squares. 1989.. I. 1994. A. P. 10. Perspectives in Probability and Statistitics: Papers in Honour of M. A maximum entropy approach to estimation and inference in dynamic models or counting fish in the sea using maximum entropy. 6. New York: Seminar Press. 14. Shannon C. Helland. Abdullah. Model Evaluation in Econometrics . R. Wold. A general Method of Estimating a Linear Structural Equation system. 117-142. Joreskog. 279287. J (Eds). B. New York: Academic. A. 85-112. 1973. 9. Mathematics.. The Relationship Between Customer Satisfaction and loyalty: cross-industry differences.1996. Garthwaite. P. 47-74. A. Nonlinear Iteritive Partial Least Squares (NIPALS) Modeling: Some Current Development. R. A. P. In Goldberg. H. Partial Least Squares Regression and Statistical model. & Duncan. Model Construction and Evaluation When Theoretical Knowledge is Scare: on the Theory and Application of Partial Least Squares. London: Academic. S. Wold. Wold. D. G. Soft Modeling by Latent Variables. 20. 26(4):435-446. 1975. 17: 97-114. 16. JAS . Proceedings of international conference on mathematics and its applications in the new millennium. Marketing. J (ed). 1973. Latent Variable Path Modeling with Partial Least Squares . G. 11. A mathematical Theory of Communications. On The Structure of Partial Least Squares Regression. 15. Statis . K. Estimation and Inference with Censored and Ordered Multinomial Response Data. I. Multivariate Analysis III. Hackl. 1999. J. Analytica Chimica Acta 185: 1-17. A. K. Robustness of partial Least Squares Method for Estimating Latent Variable Quality Structures. 1997. A. Comm. & Perloff.. On Robust Estimation of Error In Variables models by using Generalized Maximum Entropy.

.. The PLS algorithm has the following steps: (i) Define the starting values for the X residuals (e follows.ta Var(t / a ) ) Var(t (v) Find the new residuals ea = e a-1 – p a t a f = f a-1 – q a t a a .Al-Nasser 225 PARTIAL ALGORITHM APPENDIX. For a = 1.µx f = y . q are scalars with same scores. ta are N vectors of latent variables. + t ' p + E 1 1 2 2 A A A t'q + t' q + . pa are k vectors of loading variables. E is the a a residual matrix and f a the residual vector.2.µy 0 where N x N m = x ik i=1 .. A LEAST SQUARE Suppose we have the following structural equations with the following relation X= y= t ' p + t ' p + .2 . yis a vector of size N . wa = Cov(e a-1 . 0 ) and yresid ual (f 0 ) as e0 = x . + t' q + f 1 1 2 2 A A A where X is a matrix of size (N x K ).fa-1 ) (iii) Estimate the score for the next PLS component by ta = e `a-1 w a (iv) determine x loading and y loading by Least Squares with pa = Cov(e a. k = 1....1 .… do steps (ii)-(vi) below: (ii ) Calculate the loading weight.. K N yi N my = i and xis k vectors of size N.ta ) /a ) qa = Cov(e a..1 .

226 Customer satisfaction measurement models (vi) Compare the t values with the one from the preceding iteration. If they are equal (with certain error.00001) then exit with the results or else go to (ii). with calibration is done with one part and validation on the other part of the data. The number of factors is chosen so that the estimated error of prediction is minimised Wold (1978) discussed this method in context of PLS. However. Deciding the number of components to include in regression model is a tricky problem (Garthwaite. 1994). Helland (1988) noted that the number of factors to retain in final equation is usually determined by a cross-validation procedure: The data set is divided into G parts. . say 0.

- Explaining Cointegration Analysis Part II
- Meloeides
- 2 UR (1)
- ARDL Models-bounds testing.docx
- Tutorial on VAR
- SenSemRegress4s14
- Structural Equation Modeling
- Wiv
- Structural Equation Modeling
- Simplex Samples to Solve 2
- Gllamm Manual
- Gllamm Manual
- Anova
- Annotated SPSS Output.doc
- QMEX5
- tut-suhr.pdf
- Tutorial on VAR
- Fair R.C. Testing Macroeconometric Models (HUP, 1994)(ISBN 0674875036)(448s)_GK
- Algorithm - Logistic Regression
- Reynolds Number Friction Factor
- Output
- Dummy Dependent Variables Models
- Functional Form
- Awais QTB Final
- (10158) (4247) the Cost Function
- Discriminant Ana
- New Microsoft PowerPoint Presentation
- 2. Evaluasi Data
- Chap2,Prob34,Case3 Hart Venture
- Assignment 3.1
- dich vu

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd