STATISTICA & APPLICAZIONI - 2005 - Special issue

Questo numero speciale della Rivista Statistica & Applicazioni riporta una selezione di 16 articoli presentati al Convegno “Valutazione e Customer Satisfaction per la Qualità dei Servizi”, svoltosi presso l’Università di Roma “La Sapienza” nel 2005.

A vast literature has recently concerned the measurement of quality dimensions such as access, effectiveness, performance and outcome of health services supplied by national health care providers. The main concern is to achieve a classification of administrative areas with respect to observed quality indicators. We describe a simple and effective procedure to achieve this goal which allows powerful testing of the hypothesized cluster structure. We describe the performance of this method on a dataset on preventable hospitalizations (PPH) in Italy during 1998, in order to highlight clusters of regions with homogeneous relative risk.

Aim of this paper is to show a statistical methodology which takes into account external information (linear constraints on the variables coefficients) available during a customer satisfaction analysis. This methodology which is called Restricted Co-Inertia Analysis (Amenta and Ciavolino, 2005a) incorporates the external information by rewriting the objective function of the Co-Inertia Analysis (Chessel and Mercier, 1993) according to the Restricted Eigenvalue Problem (Rao, 1973), in order to integrate the interpretation of the analysis results.

In some recent articles, emphasis has been given to the partition of the Goodman-Kruskal’s tau index using orthogonal polynomials for the study of the non symmetrical relations in three-way contingency tables. New graphical techniques that consider such a partition and allow for the analysis of asymmetric relationships have been proposed, including three-way ordinal non symmetrical correspondence analysis (Simonetti, 2003). Such a procedure takes into account the presence of an ordinal predictor and response variables. In this paper we demonstrate the applicability of such a technique for the patient satisfaction evaluation.

The application of a dynamic version of the Customer Satisfaction analysis requires panel data, but this kind of data is not easily available for service enterprises. It’s possible to construct cohorts of individuals, using socio-demographic variables invariant over time, if Independent Repeated Surveys are available: following these cohorts over time is the same than using panel data. The aim of this paper is to verify the possibility of implementing Latent Variables Dynamic Models on pseudo-panel data in order to evaluate the features of the dynamic satisfaction of urban transport service’s users.

This paper deals with the problem of testing for dropping out (by censoring all graduate students) in analogy with survival analysis. A conditional test based on the probability of dropout and on the distribution of observed time to dropout is given. The application of the proposed solution to all students who began one of the three new laurea programs in the Civil Area of the Faculty of Engineering at the University of Padua is presented. A simulation study shows that the proposed solution presents a very good overall performance.

The presence of heterogeneity among the 19 boroughs (Municipio) of Rome gives rise to a group effect on the evaluations expressed by citizens about perceived quality of local public transport. By fitting an ordered logit multilevel model with random intercept it is useful to explain the heterogeneity in terms of individual and contextual explanatory variables. The results show that citizens’ opinion on local public transport quality depends not only on individual characteristics, but also on contextual indicators concerning demographic and environmental features of borough in which any citizen lives.

We define the activities of the sanitary services as “the activities performed by technological and not technological components”. In this paper, we study the relation between the technological components quality and the service quality. By using a probabilistic approach, we propose a stochastic index to relate the technological components reliability to the sanitary services quality. Finally, the index is evaluated by means of a Bayesian method shown on a Monte Carlo application.

In the paper we propose some simple dynamic indicators for monitoring the performance end efficiency of the university courses. For each student enrolled in the same year we consider the correspondent trajectory given by the number of the obtained CFU with respect to the time; the data are analyzed in the perspective of the functional data analysis methodology.

This paper deals with the problem of the treatment of ordinal data arising from the evaluation of statistical units (projects) expressed by judges with respect a set of predefined common criteria. The methodology proposed to create objective interval scaled measures is the Rasch Multifacet modes. The application regards a set of project, that was selected for financing by the Lombardia regional Government. We found that the scale used to evaluate the projects contains to many modalities: as we aggregate this modalities to get a simpler scale most of misfitting observations disappeared and the criteria used to evaluate the projects look pretty good in term of fit and reliability. Some judge show very lenient and other were, on the contrary, very restrictive in their judgement: the model adjusted adequately for these differences, producing ranking based on the estimated measures that were quite different from that obtained from the observed scores.

In this paper we propose an explorative study for the evaluation of Patient Satisfaction in hospital through linear and non linear Discriminant Partial Least Squares. This technique permits to study the dependence relationships between the ordinal variable “satisfaction” in function of variables of different nature, highly correlated, which reveal patient judgements on service quality. To decide on the optimal model dimension we compute the Generalized Cross Validation (GCV), as alternative to the PRESS criterion, well known in literature to check model validity and stability (Cross Validation).

Pisa 2000 cognitive data on reading and mathematical literacy of the Italian 15 year-old students have been analysed applying multilevel statistical methods to estimate the Italian school effectiveness. The Italian student sample was examined utilising both reading and mathematical assessment data as response variable, and Italian macroareas; student socio-economic and cultural factors; school climate variables as predictors. As both in reading and in mathematics achievement the differences between Northern and Southern Italy are significant, a set of multilevel random intercept models was fitted to each macro-area both for reading and mathematics data. In each macro-area, school effect is more important for reading than for mathematics, and socio-economic status variable has everywhere a strong impact on student achievement.

The paper deals with non linear dimension reduction methods in presence of ordinal data. More precisely, our attention focuses on isomaps, a multidimensional scaling method based on the so called geodesic distance, which connects a pair of units along a path which goes trough its k nearest neighbours. We propose a criterion for the choice of k, discuss some issues posed by the use of isomaps for ordinal data and present an application to the study of health service user satisfaction.

In spite of its paradoxical behaviour, “Kappa” statistic has become a popular tool for measuring interobserver agreement.

The aim of this paper is twofold: firstly to point out the inadequacy of the “Kappa” statistic in the context of University Student Satisfaction; secondly to propose a procedure for assessing and testing agreement among multiple raters which is based on a statistic not affected by “Kappa” paradoxes.

Another advantage of the proposed statistic is that it has a well-known limit distribution when either the number of subjects or the number of raters is large.

Customer Satisfaction Analysis is mostly based on the study of the deviations between customer expectations and customer perceptions on the quality of the product/service. According to this definition, the paper considers the CS in its own nature of interval value and proposes to apply fuzzy regression models for the CS estimation. The main assumption of the model is that the CS depends essentially on the perceived and expected satisfaction levels. The model capabilities are tested on a real dataset demonstrating how the fuzzy approach is promising in the CS analysis.

Frequently, in social and political context, an important problem is the assessment of public sector activities (typically education, health, social services) in order to compare institutions or operators. One way to evaluate service quality is to consider ''appreciation'', i.e., the quality of the service as perceived by users. This is usually performed by collecting users’ responses to several items of a questionnaire. The Rasch model, introduced originally in psychometrics, can be used to analyse users’ responses and allows the association of a measure of “quality” to each item and a measure of “satisfaction” to each user. The aim of this paper is to introduce users’ measures obtained by the Rasch model into a multilevel analysis, to study relations between users’ satisfaction and other variables included in a hierarchical structure. An application to student satisfaction in university courses is then proposed.