Peter Hackl

Robustness of partial least-squares method for estimating latent variable quality structures

Latent variable structural models and the partial least-squares (PLS) estimation procedure have found increased interest since being used in the context of customer satisfaction measurement. The well-known property that the estimates of the inner structure model are inconsistent implies biased estimates for finite sample sizes. A simplified version of the structural model that is used for the Swedish Customer Satisfaction Index (SCSI) system has been used to generate simulated data and to study the PLS algorithm in the presence of three inadequacies: (i) skew instead of symmetric distributions for manifest variables; (ii) multi-collinearity within blocks of manifest and between latent variables; and (iii) misspecification of the structural model (omission of regressors). The simulation results show that the PLS method is quite robust against these inadequacies. The bias that is caused by the inconsistency of PLS estimates is substantially increased only for extremely skewed distributions and for the erroneous omission of a highly relevant latent regressor variable. The estimated scores of the latent variables are always in very good agreement with the true values and seem to be unaffected by the inadequacies under investigation.

On measurement of intangible assets: A study of robustness of partial least squares

The customer asset is an important intangible. Its value depends, for example, on the customer satisfaction level. Thus, it is important to monitor that level, and to identify cost-efficient actions to improve it. The European Customer Satisfaction Index (ECSI) system is based on a structural model for such monitoring, a model that is estimated by partial least squares (PLS). The robustness of PLS applied on an ESCI-like measurement model is studied in the present paper. Monte Carlo simulations show that PLS is reasonably robust against multicollinearity, skew response distributions and various types of model misspecifications. This is true, in particular, when estimating the customer satisfaction index.

A control chart based on ranks

A control chart technique that considers exponentially weighted moving averages (EWMA) of the ranks of the observations is proposed. This nonparametric technique is outlier-resistant and performs well if one is concerned about sudden shifts in the proce..

Customer satisfaction in the Austrian food retail market

This paper discusses a new customer satisfaction index which is based on the American Customer Satisfaction Index (ACSI) model but takes into account company- or sector-specific satisfaction drivers. The concept is developed for the Austrian food retail market. Suitable satisfaction drivers have been identified from the literature and talks with experts. Application of factor analysis shows that these variables are well represented by three factors, viz. the emotional perception, the assessment of the customer orientation and the price-performance ratio. An empirical investigation based on 1000 households shows that the emotional factor and the price-performance ratio but not the assessment of the customer orientation contribute significantly to the explanation of the data. These empirical results show that the extension of the model by satisfaction drivers is a necessary feature for the ACSI model to be adequate, at least in the case of the food retail market. From the point of view of the user, an advantage of the new customer satisfaction index is that the values of the coefficients of the satisfaction drivers give clear hints on potential improvements.

On structural equation modelling for customer satisfaction measurement

(2000). On structural equation modelling for customer satisfaction measurement. Total Quality Management: Vol. 11, No. 4-6, pp. 820-825.

Statistical Analysis of "Structural Change": An Annotated Bibliography

The typical “structural change” situation is — from the point of view of a statistician — as follows: To cope with a particular economic phenomenon a model is specified, and it is suspected that for different periods of time, or for different spatial regions, different sets of parameter values are needed in order to describe the reality adequately; the “change point” which separates these periods, or regions, is unknown. Questions that arise in this context include: Is it necessary to assume that the parameters are changing? When, or where, does a change occur or — if it takes place over a certain period of time — what is its onset and duration? How much do parameters before and after the change differ? What type of model is appropriate in a particular situation (e.g., two-phase regression, stochastic parameter models)?

Demand for international telecommunication time-varying price elasticity

Abstract In this empirical study, the price elasticity of the demand for telecommunication between Sweden and three destination countries, Germany (FRG), United Kingdom, and USA, was investigated. Within the constant parameter framework the structural relations identified to explain the telecommunication demand are similar for the three countries. The short- and long-term price elasticities of the demand derived from the models are in the range that is reported in other empirical studies. However, recursive estimation and the analysis of recursive residuals support theoretical assumptions of time variability of price elasticity. Kalman filtering in various versions, differing in the way of determining the hyperparameters, as well as moving local regression analysis are applied to estimate the trajectories of the price elasticity.

On price elasticities of international telecommunication demand

Abstract In this empirical study, the price elasticity of the demand for telecommunication between Sweden and six destination countries, Germany (FRG), United Kingdom, USA, and the three Scandinavian countries Denmark, Finland, and Norway, is investigated. A linear regression model in logarithmic transforms of all variables is specified. To allow for time-varying regression coefficients the moving local regression technique is used to estimate the trajectory of the coefficients. The results of our study can be summarized as follows. For all countries except the USA, the (absolute) price elasticity increases over the observation period 1976–1990. The US price elasticity is close to zero, whereas the demand is relatively price elastic for the European countries. Our analysis shows that a model with constant price elasticities would be misleading when observations over a longer time period are analyzed. An appropriate method of analysis is suggested.

Multiple testing of pairs of one-sided hypotheses

SummaryTwo-sided test procedures fork real parameters should point out in the case of rejection whether the left or the right alternative can be assumed. This sets up a multiple testing problem fork pairs of one-sided hypotheses. Holms (1979, Scandinavian Journal of Statistics 6:65–70) sequentially rejective test provides a solution the critical levels of which are slightly improved. Considerable improvement is obtained when the hypotheses are redefined to be disjoint in pairs.

Structural analysis and measurement of customer perceptions, assuming measurement and specifications errors

In previous research (e.g. Cassel et al., 1999,Journal of Applied Statistics26, pp. 435-446; Cassel et al., 2000, Total Quality Management, 11, pp. 897-907) the bias and precision effects, for various specification problems, on the structural analysis and measurement of customer perceptions have been studied. In this paper the combined effects of some of the most significant problems, misspecifications and skew response distributions, and certain types of measurement errors, are studied. The measurement errors considered reflect the bias due to the inertia of respondents in adapting to changes. In the situation of repeated measurements, the later response tends to be biased towards the former response. This effect can be modelled by implementing certain change filtering assumptions. A simulation study using the European Performance Satisfaction Index (EPSI) structure demonstrates that the EPSI measurement methodology using partial least squares is very robust with respect to the problems introduced in the simulations. This is true, in particular, when estimating the inner structure of the model and when estimating the outer measurement relations of the model. The estimated values of the latent variables (such as the customer satisfaction index) are affected in an expected way.