Con il patrocinio di Toulon-Verona Conference Università degli studi di Bari “Aldo Moro” Università degli studi di Foggia Politecnico di Bari

Conference on “Choice and preference analysis for quality improvement” and seminar on experimentation

University of Bari University Hall and Student Center (former “Palazzo delle Poste”) 9-10 July 2015

BOOK OF ABSTRACTS

2015 Stampato: luglio 2015

ISBN 978-88-88793-87-0

© Copyright 2015 by Università degli Studi di Bari Aldo Moro

Tutti i diritti di traduzione, riproduzione e adattamento, totale o parziale, con qualsiasi mezzo (comprese le copie fotostatiche e i microfilm) sono riservati 3

SUMMARY

ZAVARRONE E., GRASSIA G. , VITAGLIANO M. Photo fooding: Some remarks on measurement processes ...... p. 5

SANTARCANGELO V., BUONDONNO A., ODDO G., STAFFIERI F.P., SANTARCANGELO N., MARAGNO M., TRENTINELLA F. Web misinformation: A text-mining approach for legal accepted facts...... 6

TOMA E., D’UGGENTO A.M., RICCI V. Factorial analysis of end-to-end performance evaluation of Italian higher education institutions ...... 7

PISCITELLI A., D’AMBROSIO A. Sensory evaluation of seven white wines to define consumers preference by key intrinsic attributes to wine choice ...... 8

BOLZAN M., MAROZZI M. Issues in designing an index of trust for public institutions ...... 9

MARIANI P., ZAVANELLA B. Business enterprises and economic system: subjective expectations ...... 10

D’ALESSANDRO M.T, GAUDIANO G., COLUCCI M., SANTARCANGELO V., ROMANO A , MINERVA T. Process mining: Review and a case study...... 11

FABBRIS L., SCIONI M. A fractional design to elicit graduates' preferences for job characteristics...... 12

FABBRIS L., RAPISARDA L., SCIONI M. An experiment to define the optimum salary for fresh graduates...... 13

MANDARANO A., TRETTEL A. Misurare il comportamento istintivo durante la fruizione di messaggi pubblicitari, logo aziendale e navigazione internet con la rilevazione dell’attività cerebrale, emozionale e di puntamento dello sguardo: il caso di Telecom Italia/TIM ...... 14

DEGAN L., GUIDETTI C., MORICHI M. Analysis and implementation of a NPS (Net Promoter Score) Survey: The Ariston Thermo case ...... 15

GIORDANO G., LAURO C.N., SCEPI G. Embedding covariates in conjoint analysis models ...... 16

GALASSO R., GRASSIA F., GRASSIA M.G., ZAVARRONE E. PLS path modelling for the evaluation of patients' satisfaction of a department ...... 17 4

MARIANI, P., RANCATI, E., GORDINI, N. Determinants of student satisfaction in higher education: The case of Bicocca University of ...... 18

ZANELLA A., CASCINI E. Quality control of perishable goods sale points ...... 19

D’ANGELLA A., LUPO N., CONTINI D. Mapping di impianti elettrici e tecnologici: misura della qualità per servizi di installazione e manutenzione ...... 20

PALAZZO L., SALVATI D.B., RAGOZINI G. Comparing operative units in a large company through archetypes in a benchmarking perspective...... 21

SCOLORATO C., SANTELLI F., RAGOZINI G. Understanding customer satisfaction determinants through models for ordinal data ...... 22

LIBERATI C., MARIANI P. Visualization and monitoring of dynamic customer satisfaction ...... 23

MASSARI A., PERCHINUNNO P., GIRONE F. Statistical models for categorical variables for measuring student satisfaction at the ...... 24

D’UGGENTO A.M., NISIO A. Introducing performance management in universities. The case of the University ‘Aldo Moro’ of Bari ...... 25

CASCINI E., FACCINI P. Statistica e ricerca industriale: ottimizzazione simultanea delle caratteristiche di un catalizzatore Ziegler-Natta...... 26

CASCINI E. Ordinamento degli score derivanti dall’analisi delle componenti principali mediante un indicatore di qualità complessivo...... 27

SANTARCANGELO V., ODDO G., PILATO M., VALENTI F., FORNARO C. An opinion mining application on OSINT for a reputation analysis of public administrations ...... 28 5

Photo fooding: Some remarks on measurement processes

ZAVARRONE E.*, GRASSIA G. **, VITAGLIANO M.* *IULM , **University ‘Federico II’ of Naples [email protected], [email protected]

Some apps, developed for monitoring the healthy food habit, are based on systems that measure calories and nutrition in meals. Typically the apps user takes a photo (through i.e. smartphone) of the plate and could have an estimate of consumption of calorie and nutritional facts components (Pouladzadeh et al., 2013) in real time. These apps come from recent studies on food habits based on the use of digitalization as alternative to the classic PAPI questionnaire. The food digitalization presents advantages as low level impact and the speed of data acquisition but some limits too. For the digitalization measurement, any formal guidelines or protocol has not been proposed and the proprieties as validity and reliability risk being neglected. Scholars (Gemming et al., 2015, Martin et al. 2014) recognize this limit and they have highlighted some cautions in the capturing photo phase (distance from smartphone to the plate, lightness, type of resolution of the device used): not optimal condition to exposure can determine biased estimate of the calories and the nutrient components. It is evident these systems can be efficient if an accurate digitalization measurement process can be applied. This paper proposes an approach for evaluating the quality data in digitalization process through identification and testing of the minimum condition request for the correct functioning of these systems. The analysis has been implemented on photos of ice-cream under several measurement conditions. The conversion in bitmap and after in image matrices . These matrices can be evaluated in a heatmap approach. The ideal plate, that satisfies the measurement aspects, comes from the application of severe rules of clustering methods (Zhao, 2014). 6

Web misinformation: A text-mining approach for legal accepted facts

SANTARCANGELO V.*, BUONDONNO A.**, ODDO G.**, STAFFIERI F.P.***, SANTARCANGELO N.****, MARAGNO M.*****, TRENTINELLA F.****** *Centro Studi S.r.l., **iInformatica Srls, ***Studio Legale Staffieri, ****BNG SpA *****Ordine degli Ingegneri della Provincia di Matera, ******Università Telematica E-campus [email protected]

More and more information appear in the web every day, in our world interconnected by the web. However, the growing of misinformation (unintentional inaccurate information) and disinformation (intentional inaccurate information) of the web contents introduces a lot of noise in the analysis and results of “Big Data”. In this work we show a review about the information web spoofing and introduce an innovative system as solution about the investigation of the notoriety of web data. Our system is based on a web crawler that performs a “text mining” activity on web data about an input text data considered by the user. The system considers the recurrence of the web data (input), weighing every entries found thanks to a knowledge base that associates a notoriety weight (from -3.0 for worst notoriety to +3.0 for best notoriety) to the website (e.g. the website www.ansa.it has notoriety +3.0, the website www.wikipedia.it has notoriety +2.0, the website nonciclopedia.wikia.com has notoriety -3.0). The knowledge base, developed by us, considers over 1000 websites and it can be shared and improved for research tasks. Our tool permits to provide a prototypal “filtering” system for the misinformation that is a ‘Big data’ problem for the web contents. It can be useful for search engines and also as tool for web browsers (it is able to suggest the notoriety of a displayed web content). As case study, in this paper, we consider the usability of a web data content as “accepted fact” in legal context, that represents an actual open topic. The importance of this topic can be related to the no update of the regulations and from the growing of fake information on the web. In this work we show the legal context about ‘accepted fact’ in Italian legislation and propose our ‘filtering system’ as possible tool for legal field that aims to provide a statistical estimate about the legal acceptance of a web data. In fact, the results obtained by the system permit to return a weighted score about the notoriety of the data that become an useful feedback to consider the web-data as an “accepted fact” in legal context. 7

Factorial analysis of end-to-end performance evaluation of Italian higher education institutions

TOMA E.*, D’UGGENTO A.M.*, RICCI V. ** *Department of Economics and Mathematics, University ‘Aldo Moro’ of Bari; **Statistics, Research and Programming Services, University ‘Aldo Moro’ of Bari [email protected], [email protected], [email protected]

We carry out an end-to-end bibliometric performance analysis of Italian higher education institutions (HEIs) using data from the latest (2014) release of the Scimago Institutions Rankings (SIR). We track six variables through the following chain: input- output-excellence-outcome-productivity. Factorial analysis (FA) then allows us to ascertain that the primary indicators are orthogonal and represent a quantity and a quality/productivity dimension respectively. Productivity of research is computed either in term of output or outcome. The quantity dimension is size-dependent while the quality and productivity dimension is size- independent. We also carry out an analysis of performance according to the geographical area where Italian HEIs are located. 8

Sensory evaluation of seven white wines to define consumers preference by key intrinsic attributes to wine choice

PISCITELLI A.*, D’AMBROSIO A.** *Department of Political Science, University ‘Federico II’ of Naples; **Department of Industrial Engineering, University ‘Federico II’ of Naples; [email protected]; [email protected];

The complexity of the wine market has constrained researchers to try different methods to define how consumers choose wines. Price, brand, region of origin, grapes and gained awards are the key extrinsic attributes used by different consumer groups when choosing wine. This paper addresses the problem of measuring the key intrinsic attributes that characterise the wine, based on specific sensory categories and the impact on consumers in terms of preferences. To this end it has been conducted a sensory evaluation experiment on seven white wines by 11 judges; for each wine, the judges had to express a preferential judgment by stating a value in a ten-point scale in which one means the worst, ten means the best. The experiment was conducted by following a double-blind control procedure. We analysed the wine attributes with a multidimensional unfolding methodology using PREFSCAL algorithm (Busing et al., 2005). The sensory evaluation of wines is used in order to define a procedure to draw a profile of consumers, their discriminatory capability and to understand how the key intrinsic attributes influence the consumer’s preferences. Keywords: Sensory analysis, Unfolding, Consumer’s preferences, Key intrinsic attributes. 9

Issues in designing an index of trust for public institutions

BOLZAN M.*, MAROZZI M.** *, ** [email protected], [email protected]

A large and growing body of research shows the low and declining trust in public institutions, in particular in political parties and representatives. Governments are very concerned about this downtrend in trust because it can lead to falling democratic participation and falling engagement in civic activities, tax evasion and decline in law- abiding behavior, which in turn can lead to problems of governmental legitimacy and even threaten the survive of a democracy. Mistrust decreases social capital which has an important role in promoting economic development, wellbeing and safety, in reducing poverty and crime. Recently, and in particular from the 2008 financial crisis, public institutions have faced reduced funding as well as even stricter requirements for transparency and efficiency. Reforms have been introduced, like ethics regulation to ensure that public officers display trustworthy standards of conduct. Many government administrations launched projects to measure their citizens' trust with public institutions. Customer satisfaction with specific public services or trust with specific public institutions have been studied widely, but less attention has been paid to measure citizens' general trust in public institutions. Unfortunately, it is quite difficult to measure trust because of its elusive meaning. The problem of measuring the general level of trust in public institutions is addressed using composite indicators. In particular, the most important design issues in designing an index of trust in public institutions are discussed, among them: how to summarize, standardize, weight and pool together the data. Uncertainty analysis is shown to be very useful in assessing the robustness of the index. A practical application to data from the last round of the European Social Survey is presented. The data refer to the trust in seven public institutions: country's parliament, legal system, police, politicians, political parties, European parliament, United Nations. 29 European countries are considered. The results show that Scandinavian countries are the most trustful countries whereas former communist countries are generally ranked among the least trustful ones. The Iberian and/or Mediterranean countries of Cyprus, Spain and Portugal are the only non former communist countries ranked among the least trustful ones. 10

Business enterprises and economic system: subjective expectations

MARIANI P., ZAVANELLA B. Department of Economics, Management and Statistics, Bicocca University of Milan [email protected], [email protected]

Forecasts are a major problem for planning activities in the firm framework. The statistical and econometric models used for this purpose require knowledge of official data, which are communicated with a certain time lag. Models formulation and application also require time. From all this it follows that models forecasts are made available with a sure delay. Furthermore, statistical models are founded on the historical realization of the analyzed variable; if a sudden change occurs in the structure of the variable, this break can take some time to be captured by the models. An important source of information that could help to answer these questions are surveys on subjective qualitative expectation addressed to firms. Actually, entrepreneurs are often able to pick up some signals from the economic system, on the contrary, statistical models cannot perceive unforeseen event in advance. There are a lot of qualitative survey, one of these is carried out by Federmeccanica and investigate the sentiment of the producer of metallurgic sector of Italian industry. Among the questions contained in this survey two are of particular interest: the production in the last quarter is increased, decreased or is unchanged? ; what is the expectation for the next quarter? In this application the analysis is conducted as follows. At a first step, balances of forecast are compared with the final data of the same quarter, to establish coherence between the two answers. Then the balances series is graphically compared with official data of industrial production index numbers. Finally are constructed simple univariate models for short term forecasts and the obtained results are compared with the ones supplied by Federmeccanica, to test the forecast capabilities of this survey as already discussed. This procedure is of particular interest because in the last years an unforeseen great crisis occurred and subjective expectation can be very useful in these cases. The original contribution is the use of subjective expectations in a model based on Istat data. Keywords: Measurement of expectations; strategic decisions 11

Process mining: Review and a case study

D’ALESSANDRO M.T*, GAUDIANO G.*, COLUCCI M.*, SANTARCANGELO V.***, ROMANO A**** , MINERVA T.***** *Sudelettra S.p.A., **L’Antincendio S.r.l., ***Centro Studi S.r.l., ****iInformatica S.r.l.s. *****Università degli Studi di Modena e Reggio Emilia [email protected]

The process mining is defined as a research ‘to discover, monitor and improve real process (i.e., not assumed processes) by extracting knowledge from event logs readily available in today’s system’ [P. van der Aalst]. Discovering a process models by event logs and by actual process monitoring can be seen as a research field belonging to Big Data Analysis oriented to process development. In fact, process mining aims to design a Process Model (Discovery Mining), by identifying the differences between a real process and a model (Conformance Checking), to extend and improve the devised process model (Enhancement). This work shows the state of the art of Process Mining considering the potentiality of data retrieval on Enterprise Resource Planning (ERP) Systems (e.g. Microsoft Dynamics NAV) as well as how the Process Mining performances can be enhanced by using Semantic Networks. In a case study on ERP Systems, we show an application of Process Mining related to a Maintenance Extinguishers process. The considered process involves different entities: customers, maintainers, administrative employees, fleet and park of devices for fire protection. The interactions among these entities determine the process dynamics, that can be identified by process mining analysis of the ERP systems. Due to the analysis of logs and records (e.g. worktops, invoices) on business management systems, it has been possible analyzing the process of maintainers and define a theoretical model describing the real process of Maintenance Extinguishers (Discovery Mining). The parallel analysis of some questionnaires compiled by maintainers about their tasks and the fieldwork analysis of their activities have permitted the real process representation. The comparison between the process model and the real process (Conformance Checking) have allowed identifying some process discrepancies, inconsistencies and inefficiencies and defining some improvement actions, aimed at quality development of the business process 12

A fractional design to elicit graduates' preferences for job characteristics

FABBRIS L., SCIONI M. Statistics Department, University of Padua [email protected], [email protected]

A relevant feature of the surveys held through electronic questionnaires is the widespread use of statistical experiments, that are typically embedded into survey designs. There are experiments, called 'fractional', in which the alternatives obtainable combining the response options of all the experimental variables are so numerous that it is necessary to sample and administrate them in a random fashion in order to reach the knowledge targets. We applied a fractional experiment to a survey to collect data about the criteria that early graduates adopt as guidance for choosing a job among the offered ones. The survey was conducted by interviewing, with a self-administered questionnaire, a sample of students of Padua University who graduated in 2014. The data collection was assisted by a CAWI (Computer Assisted Web-based Interviewing) system. The questionnaire was designed to highlight the mental process of early graduates when they are offered a job in scenarios that portray the current Venetian local markets. The experimental procedure consisted of a set of questions to describe graduates’ choice mechanism based on the simulation of their choice of the job which, among those offered, best fitted their expectations. Jobs were characterised by a set of interrelated (‘conjoint’) features (e.g. work–contract duration, career perspectives, distance from home, flexibility of work time, level of autonomy). The experimental attributes were as follows: (a) width of the choice set (varying between 2 and 6) administered to graduates, (b) number of variables that described the job offer (this number was inversely proportional to the choice set width so that the product between the two numbers be always 12), and (c) the way graduates identified the preferred attributes of the chosen job offer. Any job characteristic was defined by two possible alternatives. A number of 3,600 questionnaires was collected. The paper presents the experimental design and the basic indicators for measuring the effectiveness of the design. 13

An experiment to define the optimum salary for fresh graduates

FABBRIS L., RAPISARDA L., SCIONI M. Statistics Department, University of Padua [email protected], [email protected]

The economic downturn has had a significant effect on the composition and characteristics of the labour market. So, even though employment conditions of graduates are better than those of secondary school leaving certificate holders, the labour market has become difficult even for graduates. Fewer jobs are available and low-paying contracts or internships are often offered to fresh graduates at their first work experience. The salary, although it is not the only thing that concerns graduates, and sometimes not even the most important, is a relevant aspect when they evaluate whether to refuse or accept a job. What kinds of salaries do graduates expect to earn upon graduation? The salary expectations of students match the reality of the job market and the willingness to pay by entrepreneurs? To answer these questions, we applied a completely randomized design with a single factorial experiment. Graduates were not asked directly the minimum salary they would accept upon graduation, but their willingness to accept a job was elicited experimenting the admissibility of 4 different salary levels: 600, 800, 1000 and 1200 euros. Moreover, in order to define the value system of graduates, we investigated the importance attached to a set of ‘perks’ singly posing questions about the tradeoffs between giving up 100 € salary and getting other benefits. The data come from a survey conducted by interviewing, with a self-administered questionnaire, a sample of students of Padua University who graduated in 2014. The data were collected through a CAWI (Computer Assisted Web-based Interviewing) system. A number of 3,600 questionnaires was collected. This paper presents the main results of the analysis of data processed through multivariate analyses. We highlighted the differences in the graduates’ willingness to accept a job, according to the level of the achieved degree, the subject of the study programme, the final mark and other social and curricular variables. 14

Misurare il comportamento istintivo durante la fruizione di messaggi pubblicitari, logo aziendale e navigazione internet con la rilevazione dell’attività cerebrale, emozionale e di puntamento dello sguardo: il caso di Telecom Italia/TIM.

MANDARANO A., TRETTEL A. Telecom Italia, BrainSigns [email protected], [email protected]

Il principale tema di ordine metodologico che l’intervento intende affrontare riguarda la necessità e la possibilità concreta di poter misurare l’emozione nelle persone durante la fruizione di messaggi pubblicitari. E’ infatti noto dalle neuroscienze come i sistemi cerebrali collegati all’emozione giochino un ruolo fondamentale, e spesso sottratto al controllo cognitivo cosciente, durante le decisioni che generiamo ogni giorno. Si propone una presentazione delle applicazioni in TelecomItalia/TIM di tecniche di neuromarketing per misurare la reazione istintiva di emozione e di interesse dei consumatori durante la fruizione di materiali di comunicazione pubblicitaria. Infatti, il settore di ricerca sulla comunicazione di TelecomItalia/TIM ha attivato nel corso degli ultimi cinque anni una collaborazione stabile con l’Università Sapienza di Roma per la messa a punto di un sistema scientificamente appropriato di rilevazioni dell’impatto cerebrale e emozionale della propria comunicazione pubblicitaria sui vari canali di diffusione disponibili. Verrà illustrato come si possano misurare con tecnologie avanzate e rigore scientifico l’impatto emozionale e di interesse generato dalla i) comunicazione pubblicitaria relativamente a spot e annunci stampa, ii) percezione di logo in studi di re-branding e iii) ottimizzazione della navigazione sul sito web. Si tratterà di come l’emozione sia necessaria alla formazione di memoria di brand e di come sia un driver fondamentale di scelta o di acquisto con meccanismi che sono poi alla base di molta parte del nostro decision-making quotidiano. Verrà mostrato come le informazioni acquisibili mediante la rilevazione neurometrica consentano di indagare su una sfera importante dei comportamenti di preferenza e scelta dei consumatori che completano ed integrano quelle ottenibili con dichiarazioni esplicite verbali in interviste o focus group. Molto spesso le emozioni percepite durante le stimolazioni sensoriali a cui siamo sottoposti nella vita di tutti i giorni non si appalesano al nostro controllo cosciente, pur guidando invece il nostro comportamento percepito come “spontaneo” o naturale. Spesso infatti non siamo in grado di giustificare razionalmente i nostri comportamenti di acquisto, che sovente differiscono di molto dal nostro intento iniziale a favore di un acquisto “deciso sul momento” in base a motivazioni puramente istintive. La novità tecnologica e metodologica di questo intervento è che tali emozioni (ed i relativi processi cerebrali che le accompagnano) possono essere misurati durante la fruizione di messaggi pubblicitari in condizioni sempre più “ecologiche”, vicine cioè a quello che accade nella vita di tutti i giorni. Questo è possibile grazie ad avanzamenti metodologici e tecnologici dei dispositivi di misura dell’attività cerebrale ed emozionale dell’uomo che oggi possono tranquillamente essere impiegati al di fuori dei laboratori scientifici delle università o dei centri di ricerca. Saranno per questo riferiti esempi concreti dell’esperienza diretta di TelecomItalia/TIM negli ultimi anni in cui questi metodi di ricerca neurometrica sono stati inclusi positivamente nel processo di brand research, a integrazione dei metodi più tradizionali, con strumenti atti ad essere impiegati anche fuori dai laboratori universitari. Saranno presentate evidenze concrete di come i comportamenti di preferenza e scelta dei consumatori indagati con tali tecniche innovative abbiano prodotto suggerimenti utili in azienda per il miglioramen–to/ottimizzazione della comunicazione, nelle modalità di presentazione dei servizi offerti e nella fruibilità di contenuti digitali e non, con riferimento anche agli orientamenti di target user e non user. 15

Analysis and implementation of a NPS (Net Promoter Score) Survey: The Ariston Thermo case

DEGAN L.*, GUIDETTI C.**, MORICHI M.** *Galgano & Associati Consulting, **Ariston Thermo SpA [email protected]

Ariston Thermo Group decided that Customers must be their Best Promoters, so to monitor Customer Satisfaction and Loyalty trend by a periodic NPS (Net Promoter Score) Survey, focused on evaluation of Service, in the different market areas. The origin of this study lies in the presence not always clear and coherent of documentation currently available concerning the statistical bases, which NPS index is based on. NPS is a management tool that can be used to gauge the loyalty of a firm's customer relationships, as an alternative to traditional customer satisfaction research and claims to be correlated with revenue growth. The NPS, is based on the fundamental perspective that every company’s customers can be divided into three categories: Promoters, Passives, and Detractors. By asking one simple question ‘How likely are you to recommend ARISTON THERMO GROUP to your friends/colleagues?’ you can track these groups and get a clear measure of your company’s performance through your customers’ eyes. NPS is calculated by subtracting the percentage of customers who are Detractors (score 0-6) from the percentage of customers who are Promoters (score 9-10): the Promoters are considered generators of positive contacts, Detractors negative contacts, whereas Passives (7-8) are considered not having any influence on promoting the Firm’s Products &Services. To model the NPS, promoters are labeled +1, passives 0, detractors −1, in the given proportions p+1 of promoters, p0 of passives, and p−1 of detractors, with p+1 +p0 +p−1 =1. NPS Index corresponds to the mean of x-scores distribution. If everyone is a Promoter, the Net Promoter Score is 1 (or 100%); if everyone is a Detractor, it is -1 (or -100%). The study characterizes the statistical model of NPS, in order to design a NPS survey statistically significant and representative and applies appropriate estimation and testing criteria to periodic Ariston Thermo NPS Survey. In particular, for understanding the statistical proprieties of NPS index, the variance of x-scores distribution is studied: how it varies with respect to NPS values and the proportions of Detractors and Promoters. The study provides some suggestions concerning suitable sample sizing (at least 450 responses per quarter), based on proportions estimated by a recent pilot Survey, and shows result of pilot test. During 2015 Ariston Thermo Group NPS project shall be implemented in , Germany, China, Turkey, Russia, UK and Spain. In total the project will involve about 20,000 respondents/year. Questionnaires will be submitted by phone interviews (CATI). For detractors, an internal manager closes the loop by recalling them and whenever possible help them to solve their issues with the aim of improving their satisfaction level. First results in Italy for 1st Quarter 2015 are 49% for Product NPS and 54% for Service NPS. While the former is stable, the latter shows a higher variability among months. As detractor level is stable, results from service NPS questionnaire will supply information about how to move consumers from passives to neutral cluster. 16

Embedding covariates in conjoint analysis models

GIORDANO G.*, LAURO C.N.**, SCEPI G.** *Dept. of Economics and Statistics, ; ** Dept. of Economics and Statistics, University ‘Federico II’ of Naples [email protected], [email protected], ; [email protected]

The aim of this paper is to define an integrate scheme of analysis where different kinds of information are retrieved in suitable data matrices according to: i) the role of the variables (active/illustrative; response/predictor), ii) the kind of data (quantitative/ qualitative) and iii) the purpose of the analysis (exploratory/confirmatory). We start from the formulation of the traditional Multivariate Linear Regression model that considers a set of multiple responses linearly related to a set of predictors. According to the quantitative or categorical nature of data, different specification of the model can be derived (e.g. MANOVA, MANCOVA, Seemingly Unrelated Regression, see Timm, 2002). The interest in this formulation arises from the specification of the metric approach to Conjoint Analysis, (Green and Srinivasan, 1990), which is one of the most famous techniques to analyze preference models at individual level. The data structure of the Conjoint Model is fully consistent with a multiple regression analysis with multiple preference responses revealed by a set of judges. In this paper it is proposed a unified approach to investigate the typical data structure of Conjoint Analysis models in the framework of Exploratory Data Analysis. Firstly, the Factorial Approach to Conjoint Analysis is recalled, (Lauro et Al., 1998) and then, we introduce an L-shaped data structure where several variables are available as external information (socio-economic variables observed on the judges). By taking into account the kind and the role of information retrieved in each data matrix (either metric or dummy variables), we aim at showing how to mine preference data from this peculiar data structure. We derive a unique formulation to investigate both the relationships among the different sets of predictors. Namely, we express the preference response variables as a function of two sets of predictors: inner and outer arrays in the language of Design of Experiments (Giordano and Scepi, 1999). In the analysis of preference models, this allows to take into account the effects of the stimulus attributes and the Judges characteristics on the elicited preference. Finally, an inter-relationships coefficient matrix will be derived. The geometric interpretation of these methods can help to understand their common traits and allows the graphical displays and the interpretation of the different models. Indeed, the powerful graphical tools of exploratory methods have proved to be valuable in many applicative fields (Giordano and Scepi, 2012; Giordano et Al., 2011; Lauro et Al., 2008). References Giordano G., Scepi G. (2012). Network Data as Contiguity Constraints in Modeling Preference Data. In: AA.VV. Challenges at the Interfaces of Data Analysis, Computer Science and Optimization. Springer, Berlin Heidelberg: pp.233-241. Giordano G., Scepi G. (1999), Different informative structures for quality design, Journal of the Italian Statistical Society, 8, 2-3, pp.139-149. Giordano G., Lauro C.N., Scepi G. (2011). Factorial Conjoint Analysis Based Methodologies. In: AA.VV.. Classification and Multivariate Analysis for Complex Data Structures. Springer, Berlin Heidelberg: pp.17- 28, Vol. XIX. Green P. E., Srinivasan V. (1990), Conjoint Analysis in Marketing: New Developments with Implications for Research and Practice, The Journal of Marketing, Vol. 54, 4, pp. 3-19. Lauro C.N, Scepi G., Giordano G. (2008). Conjoint Analysis based methodologies for the ex-ante evaluation of regulatory impact. SA-IJAS, Vol. 3-4, pp.233- 249. Lauro C.N., Giordano G., Verde R. (1998), A Multidimensional Approach to Conjoint Analysis, Applied Stochastic Model and Data Analysis, 14, Wiley, pp.265-274. Timm N. H. (2002) Applied Multivariate Analysis, Springer, New York. 17

PLS path modelling for the evaluation of patients' satisfaction of a department

GALASSO R.*, GRASSIA F.**, GRASSIA M.G.*, ZAVARRONE E.*** *Dept. of Social Science, University of Naples Federico II; **General directorate, ISTAT; ***Dept. of Marketing, Communication and Consumer Behavior, IULM University Milan [email protected], [email protected], [email protected], [email protected]

Patient satisfaction surveys are gaining in importance and should be at the heart of any healthcare facility. It's valuable to get a holistic view of what patients really think about the care and treatment they receive. Patient satisfaction can be measured by methods based on latent variables, i.e., variables that are not directly observed but deducted from mathematical analysis. Those methods include the partial least squares (PLS) path modelling aimed at defining optimal linear relations among latent variables in order to assemble the best set of predictions. Aim of this paper is show an application of patient satisfaction survey made on the Department of Day Surgery of the Caserta' Hospital. One hundred patients of this Department ware interviewed from April and to May 2011, a month after their surgery, with a CATI technique. The Questionnaire, consisting of 36 items has investigated on seven categories of performance: quality of the facilities, quality and clarity of provided Information, quality of relationship with surgeons and nurses, quality of the received care, quality of life, overall patient satisfaction. We decided to consider the quality of life after the surgery as latent variable that affects satisfaction. The use of this latent variable was an innovation for a patient satisfaction model. For the analysis of data, collected through the questionnaire, was used the PLS path modelling. The model was constructed by correlating three exogenous latent variables (Facilities, Information and Service, Relationship), with the latent endogenous variable Perceived Quality. This variable and the variable Quality of Life were correlated to the endogenous variable Satisfaction. At the end, the variable Satisfaction was correlated to the endogenous latent variable Loyalty. The results were very interesting: overall Perceived Quality was the variable with greatest (54%) positive impact on patient satisfaction, but Quality of Life had a good impact to (36%). As previously mentioned, the use of this latent variable is an innovation, and therefore his good impact means a good result. For the manifest variables, professionalism of surgeons and nurses positively affected the level of satisfaction, conversely waiting times, quality of post surgery care adversely affected the level of satisfaction. In conclusion, PLS path modelling may represent a valuable tool to measure quality in the setting of managed health care since it allows for the identification of areas where the service can be improved. References: Grogan S., Conner M., Willits D., Norman P. Development of a questionnaire to measure patients' satisfaction with general practitioners' services. British Journal of General Practice, Ottobre 1955 Pagano A., Rossi C. La valutazione dei servizi sanitari. in Gori E., Vittadini G. Eds. Qualità e valutazione nei servizi di pubblica utilità, Serie gestione d'impresa-direzione, Etas, Milano, 1999 Tenenhaus M. La régression PLS: Théorie et pratique. Editions Technip, Paris, 1995 18

Determinants of student satisfaction in higher education: The case of Bicocca University of Milan

MARIANI, P., RANCATI, E., GORDINI, N. Bicocca University of Milan [email protected]; [email protected]; [email protected]

In order to face the global competitive scenario, Italian universities are trying to adopt a marketing approach to better serve their students and increase service quality. Service quality has been examined to measure customer satisfaction (Lassar, Manolis & Winsor, 2000; Abu Hasan, 2008; Khodayari & Khodayari, 2011). At the same time, marketing as made a shift from product-oriented marketing to a more service-oriented marketing approach, even product companies. Since competition has increased amongst service-minded companies, the quality of the service is crucial for companies in order to satisfy and maintain customers. Nowadays, it is possible to argue that the higher educational system (such as universities) is a type of service company/service provider. Services offered by a university include, but are not limited to, teaching. It is of great importance that the teaching quality is significantly high, since competition to attract, maintain and foster students amongst universities are fierce today (Abu Hasan, 2008). Many researchers have, in fact, argued that it is possible to consider a student as a customer (Shank et al., 1995; Joseph et al., 2005). In addition, it is only the meeting between teacher and student that constitutes the service quality evaluation; it is the complete and full offering from the university. In other words, it is the student’s evaluation of everything the university offers that is of importance. Besides service quality, customer satisfaction has also been under the scope the last three decades (Devasagayam et al., 2013). Many researchers have found strong links between service quality and customer satisfaction (Lee et al., 2000; Sureshchandar et al., 2002; Ooi et al., 2001). Previous research has stated that customer satisfaction may be seen as a process, and not only as a final evaluation in retrospect (Oliver, 1997). In other words, it is possible to measure customer satisfaction during the process of consumption of the service. According to previous researches (Harvey, 2001, 1997; Lee et al., 2000; Donald & Denison, 1996; Morrison, 1999; Marsh, 1991; Rich et al., 1988; Guolla, 1982), it is assumed that there are four major groups of factors, which seem to affect student satisfaction: 1. Institutional factors; 2. Extracurricular factors; 3. Student expectations; 4. Students demographics. Some of these factors are similar to Harvey’s 2001 study. According with Harvey most universities around the world conduct satisfaction surveys among the students regarding the services they provide. These services include: learning and teaching; learning supports facilities; support facilities; external aspects of being a student; the learning environment. In this study, services one and two are classified under the heading “teaching factors”; services three, four and five are classified under “organizational aspects”. In addition to these, institutional and demographic factors are also included in order to come up with a more comprehensive framework. The aim of this study is to determine the level and the factors of higher education students’ satisfaction with the institutions they are attending. Firstly, the concept of satisfaction will be defined. Secondly, a conceptual framework to demonstrate the relationship between the factors that lie behind university student satisfaction will be presented. Thirdly, the theoretical background will be discussed through an empirical analysis on the Bicocca students. Through the results of the study, University of Milano-Bicocca could better understand: a) its strengths and weaknesses; b) reasons why students decide to enroll; c) what corrective actions implemented to increase the student’s satisfaction, the number of student, and to increase the corporate image and awareness of University of Milan-Bicocca. This study is useful to any university in the world, in order to measure and analyse the level of satisfaction over the time. 19

Quality control of perishable goods sale points

ZANELLA A.*, CASCINI E.** *Catholic University of the Sacred Heart of Milan, **AISS-Italian Academy for Six Sigma [email protected], [email protected]

The paper concerns the distribution process of perishable goods, at which we hinted in a former paper. The novelty is related to having recourse, in a more accurate way, to an information system which also in this context we assume to be sufficient and complete, that is so that it includes all responses of interest and is apt to suggest suitable corrections to the controllable variables in order to remove possibly trends of responses over time. We consider s, s=1,2,…., 5 points of sale and define real quality as Qo / Qs with Qo the same amount of a specific product (order) made available to each point of sale, Qs the product quantity sold by the point of sale sth . Further we introduce virtual quality as Qo co/ Qs cs where co is the unitary cost of the specific product, assumed to be a constant irrespective of sale point s, cs is the unitary proceeds average from selling, depending on s. Economic grounds suggest that it must be co/cs < 1. Likewise we have Qoco∙ / Qs cs < 1, since costs can’ t be larger than proceeds . If we consider real quality too small, that is the sold quantity Qstoo small with respect to the available quantity Qo , we may refer to the virtual quality and try to increase sales by a price reduction, that is a reduction of cs . We shall investigate by means of suitable simulations the relationship between sales increase and price reduction. 20

Mapping di impianti elettrici e tecnologici: misura della qualità per servizi di installazione e manutenzione

D’ANGELLA A., LUPO N., CONTINI D. Impianti elettrici Contini Domenico, Matera [email protected], [email protected], [email protected]

La misura della qualità negli impianti cablati, a seguito di una corretta installazione, è necessaria per garantire il giusto funzionamento degli stessi ed è indispensabile per verificare l’efficienza dei dispositivi di sicurezza installati atti a prevenire l’incolumità dell’utilizzatore. La presenza di differenti tipologie di cablaggio (telecomunicazioni, networking, distribuzione elettrica) in un edificio così come in un qualsiasi sistema articolato, fa sì che la manutenzione dei suddetti impianti diventi un processo sempre più complesso e delicato, dove le scelte d’intervento devono essere fatte anche considerando le varie interazioni tra gli stessi. Ulteriori problematiche si possono palesare, ad esempio, per impianti già esistenti o a seguito di passaggio di proprietà dell’immobile ospitante l’impianto. In quest’ultimo caso, si possono perdere molte informazioni sugli impianti esistenti nell’edificio, di conseguenza un mapping preventivo o durante le fasi di ristrutturazione consentirebbe una più agevole conoscenza della struttura impiantistica e una più semplice modalità di intervento in caso di manutenzione ordinaria o straordinaria. (L’articolo 86 del DLgs 81/08 ha formalizzato per tutte le attività di manutenzione, l’obbligo di redigere un registro dei controlli sull’impianto elettrico). La procedura proposta consente in maniera semplice ed affidabile la tracciabilità degli elementi impiantistici dell’edificio o di qualsiasi altro sistema complesso, mediante l’utilizzo di tecnologie innovative. Tali tecnologie (RFID) sono basate sull’individuazione o memorizzazione automatica di un certo tipo di informazioni inerenti il sistema cablato considerato. Queste informazione, univoche sul dispositivo, vengono associate, in un database, ad una serie di dati inerenti l’impianto in questione. Queste indicazioni saranno utili all’operatore che si occuperà della manutenzione o allo stesso proprietario dell’impianto in quanto si potrà conoscere, in tempo reale, tutte le specifiche dello stesso e dunque prevenire un guasto, per esempio dovuto a decadimento naturale dei materiali, o intervenire tempestivamente e con grande precisone a guasto avvenuto. 21

Comparing operative units in a large company through archetypes in a benchmarking perspective

PALAZZO L., SALVATI D.B., RAGOZINI G. Department of Political Science, University ‘Federico II’ of Naples [email protected], [email protected], [email protected]

Archetypal analysis is a statistical method that synthetizes a set of multivariate observations through a sort of pure individual types, not necessarily observed points, which lie on the boundary of the data scatter. Archetypal analysis was first introduced by Cutler and Breiman (1994), and has found applications in many fields, from physics to market research. In this work we propose to use archetypes to design a data driven three{steps benchmarking procedure (Porzio et al., 2008). This procedure is well suited for internal benchmarking, and it can be also exploited for external benchmarking, if appropriate ' data are provided. Formally, the archetypes a j are a convex combination of the ' ' ' ' observed data X: a j   j X , with  ij  0 and  j 1  1; i; j. On the other hand all the data points can be expressed in terms of archetypes: ' ' ' xi   i A , where A is the archetype matrix and the  i are the weights of the archetypes, ' ' with  ij  0 and  j 1  1. The archetypes can be used as benchmarks by using the following steps: (i) identification of extreme performers (archetypes); (ii) description and selection of benchmarks; (iii) comparison of observed performers with benchmarks. In this work we applied this method to analyze a Telecom Italia business{segment, Open Access, namely. The company continuously measures the performances of local operative units in order to monitor the results, and to establish a ranking among units aiming at providing productivity bonus. However, the company does not exploit its information for internal benchmarking purposes. Furthermore, as the performance indicators are measured over time, in order to preserve such variability, we would intend to use archetypal analysis for interval data (D'Esposito et al., 2012). In such a case each unit will be represented as interval on each performance indicator. The intervals will be given by the minima and the maxima over time for each indicator and for each unit. 22

Understanding customer satisfaction determinants through models for ordinal data

SCOLORATO C., SANTELLI F., RAGOZINI G. Department of Political Science, University Federico II of Naples [email protected], [email protected], [email protected]

In this paper we present the main findings of a case study related to customer satisfaction survey by Telecom Italia, the biggest Italian company of telecommunication. In particular, we analyze the degree of satisfaction about the performances of the customer support service and its determinants. Data had been collected during three years and consider several aspects of the service. The available dataset covers the whole Italian national territory and it is structured according to territorial units. The aim of the paper is to identify the main determinants of the global Customer Satisfaction degree in order to understand the reasons of different levels of satisfaction in time and space, in order to improve the offered service. As Customer Satisfaction is measured on an ordinal scale, we decide to use the Categorical Regression Models (Scott Long, 1997) and the Cub Models for ordinal data which have been introduced for interpreting and fitting ordinal responses (Piccolo, 2003). In this way, we can compare and integrate the different results gaining insights in our analysis. First analyses show significant effects of different customer segments on the global Customer Satisfaction degree. In fact, we observe some differences between the satisfaction degree concerning Consumers and Business segments along with some territorial differences. As for the determinants of the global satisfaction degree, it turns to be affected by the clearness of the received information and by the elapsed time between the moment of fault reporting o and its resolution, while it results less important the perceived courtesy and competence of the technicians. Furthermore, as by-product of the analyses we found that the measurement scales adopted in the survey is not adequate. We propose then different solutions to such an issue. 23

Visualization and monitoring of dynamic customer satisfaction

LIBERATI C., MARIANI P. Department of Economics, Management and Statistics, University of Milano-Bicocca; [email protected], [email protected]

Customer Satisfaction for banking services is, arguably, a construct that develops and changes over time for a number of different endogenous and exogenous factors (modification of products, transparency of banking transactions and financial services, customer relationships, changes of market conditions an so on). Measuring change requires a longitudinal perspective, and most of the times such perspective is missing in the market research. This paper aims to analyze the customer evaluation evolution of the banking services, in order to catch differences among the clusters and time lags through a dynamic factorial model (STATIS). Moreover, while existing techniques compare temporal trajectories using dissimilarity measures, an additional innovative aspect of this work lies in the proposal of an new index can summarize some characteristics of the trajectories such as distance covered, the shape and direction. A real case of study on customers of an Italian bank is presented. Results based on a sample of 27000 instances per 3 waves, obtained via a questionnaire framed according to SERVQUAL model, reveal the effectiveness of such an approach. 24

Statistical models for categorical variables for measuring student satisfaction at the University of Bari

MASSARI A., PERCHINUNNO P., GIRONE F. Department of Management and Private Law, University ‘Aldo Moro’ of Bari [email protected], [email protected], [email protected]

Student satisfaction is an important issue for universities to assess the quality of their courses; it can be measured by several factors such as the quality of academic staff, teaching equipment, services and logistical support. From this point of view, students are considered as customers of a public service, whose perception is useful for decision makers to improve future planning of training activities, as stated by the National Agency for Evaluation of University and Research in its QA process guidelines. Every year, information about Italian university students’ satisfaction is collected by the Evaluation Committee through a survey named “Rilevazione dell’opinione degli studenti sulla didattica”. The student fills in one questionnaire for each subject and, therefore, also for one corresponding professor. The questionnaire is composed of 35 questions and is divided into 7 sections: the first contains information about the student (age, gender, course, average mark, academic year, in course or not, …); the second is about programs and texts; the third concerns the teacher and some features of his lessons (regularity, clarity, interest, teaching equipment, …); the fourth contains some questions to measure the benefits deriving from lesson attendance; in the fifth section there are some questions about the examination; the sixth deals with classrooms, laboratories and equipment) and the last one measures the satisfaction level and interest in the subject. In the last academic year, the survey has involved both students and professors, in order to assess the congruence between the views expressed by the protagonists of educational processes. We use the latest data from this survey, collected from the students of the University of Bari, in order to analyze their satisfaction level, considered as a multidimensional concept laid out by various components and also connected with the student’s profile. According to the goals of the research, some statistical models for categorical variables (multiple correspondence categorical analysis and multidimensional scaling) have been used. Since the questions involve qualitative responses, along with the evaluation of the expectations and perception of quality and of student satisfaction, multiple correspondence analysis has the advantage of allowing a graphic representation of the phenomenon able to focus attention on the structural characteristics of the students’ satisfaction. The results of the analysis could be useful to the decision makers of the University of Bari to take further actions to improve the whole satisfaction of their students. 25

Introducing performance management in universities. The case of the University ‘Aldo Moro’ of Bari

D’UGGENTO A.M.*, NISIO A.** *Dipartimento Scienze economiche e metodi matematici, Università di Bari ‘Aldo Moro’; **Dipartimento studi aziendali e giusprivatistici, Università di Bari ‘Aldo Moro’ [email protected]; [email protected]

The aim of the paper is to highlight the contribute of statistical techniques in the implementation of the performance management system in a Public administration and, in particular, the analysis is carried out on the route followed to adopt the performance cycle in the University of Bari ‘Aldo Moro’. From 2009, in fact, Italian universities have been engaged in the adoption of a performance management system and the National Agency for Evaluation of University and Research (ANVUR) has been recently entrusted to manage it. Since the 1980s, most of the OECD countries have been concerned by a broad process of managerial reform inspired by the principles of the New Public Management (De Bruijn, 2002) which highlighted, among others, the need to focus on measuring, managing, and assessing performance. In public administrations, and also in universities, performance management is considered to be the tool to improve the efficiency, effectiveness, quality and fairness of policies, programs, projects and services; moreover, it draws and focuses the attention of decision makers on the results and their causes and, at the same time, makes employees aware of what is expected from them and informs the stakeholders of the purposes, objectives, results of the organisation and the methods through which they were achieved (Bracci et al., 2014). The adoption and effective use of performance management in Italian universities, more than in public administrations, is arduous and often encounter issues, mainly due to the lack of a “quantitative approach” in measuring the productivity, the efficiency and efficacy of activities but also consisting in a partial development of the system, a formal approach in the design, an incomplete measurement of the performance dimensions, a failed use of the information produced. The adoption of a performance management system takes place according to the following steps: design (of the performance management system), implementation (introducing the system in the organisation), use (of the information the system produces), evaluation or assessment (continuous and systematic of the achievement of the system’s purposes); impacts (recording/analysis of the effects of the performance information produced by the system concerning the issues on the agenda, decision-making processes, individual behaviour and the same organisation). Starting from the experience of the University of Bari, in the paper we show that, in each step, statistical techniques play an important role: for the identification of variables and dimensions, in the data collection and organization, in the mapping of the University’s databases; in the general mapping of the working processes and rethinking of the functioning logics of the organization (as in a Balanced Scorecard), in the definition of elementary indicators, performance measurements, weights, and so on. Finally, they could allow the analysts to achieve one synthetic indicator that expresses the score of a unit.

Keywords: performance management system, university performance, statistical techniques. 26

Statistica e ricerca industriale: ottimizzazione simultanea delle caratteristiche di un catalizzatore Ziegler-Natta

CASCINI E.*, FACCINI P.** *AICQ – ASA, **Lyondellbasell [email protected]; [email protected]

La metodologia statistica applicabile con profitto all’industria, ha raggiunto, oggi, un notevole livello di sistematizzazione. L’attendibilità dei metodi di misura, la valutazione dello stato di controllo statistico dei processi, la risoluzione di problemi complessi, lo svi– luppo e l’ottimizzazione delle caratteristiche di nuovi prodotti, sono i settori nei quali ha raggiunto un livello di perfezione tale, da costituire una base affidabile da cui partire con sicurezza per affrontare i casi reali. In questo lavoro, sviluppato per ottimizzare simultanea– mente le 3 caratteristiche fondamentali di un catalizzatore stereospecifico Ziegler – Natta, per la fabbricazione di polipropilene, viene mostrato, in concreto, quali e come alcune di queste metodologie possano essere modificate con profitto. Come è noto, infatti, un qualsiasi lavoro di ricerca industriale, se può beneficiare di teorie statistiche consolidate, può contribuire, nello stesso tempo, ad un loro ulteriore sviluppo. Il lavoro è organizzato in questo modo: 1) viene introdotto brevemente il significato di catalizzatore stereospecifico; 2) viene mostrata la insufficienza, dal punto di vista statistico, dei dati di produzione storici, ai fini dello sviluppo e della ottimizzazione del prodotto, fornendo un esempio concreto del concetto espresso nel testo di Box, Hunter, Hunter (John Wiley, 1978) a pagina 487: Statistics for Experimenters, dove, per affermare la indispensa– bilità della sperimentazione, viene mostrato, con un certo umorismo, un normale comporta– mento aziendale del top Management nei riguardi di un ingegnere di processo, che chiede la disponibilità delle attrezzature, per realizzare un piano sperimentale … Il rischio è special– mente grande, se uno ha la temerarietà di suggerire la necessità di un esperimento programmato per risolvere un problema particolare … A meno che uno non abbia un carattere molto forte, è molto probabile che si trovi messo alla porta, con il consiglio di recuperare i dati tra quelli registrati negli ultimi 10 anni …; 3) viene indicato il piano sperimentale che è consistito nella progettazione e nella realizzazione di una serie di esperimenti fattoriali completi e frazionari; 4) vengono descritte le analisi statistiche utilizzate per la ottimizzazione complessiva delle tre caratteristiche fondamentali del prodotto; in questa fase è proposto un metodo alternativo di indagine, rispetto a quello classico degli esperimenti centrali composti. Un’idea dei risultati complessivi ottenuti può essere tratta dal confronto delle due equazioni che seguono, riferite ad una delle tre caratteristiche considerate; la prima è il modello vigente prima del piano sperimentale e la seconda il modello derivato dal piano sperimentale. y1 = 14,5 + 10,4 x1+0,271 x4 (s = 9,12) 2 y1 = 65 – 12,6 x1 + 1,87 x1 – 2 x4 + 1,875 x2 x4 +2,25 x3 x5 (s = 2,60) Le conclusioni del lavoro sono: a) la ricetta di fabbricazione è stata cambiata in modo drastico, con notevoli vantaggi pratici b) è stata mostrata l’efficienza della programmazione statistica degli esperimenti per la ottimizzazione dei prodotti c) è stato proposto un metodo per la determinazione del termine quadratico del modello, nel caso di risorse sperimentali limitate.

Cascini, E. (2002) Applicazioni industriali recenti, Atti del Convegno AICQ: Approccio statistico nel controllo dei sistemi per la qualità e proposte formative, Milano Faccini, P. (2005) Tesi di Laurea in Tecnologie Fisiche Innovative, Università di Ferrara. Relatori: E. Cascini, Titolare del corso: Tecniche statistiche di qualità - Università di Ferrara, A. Fait: Ricerca e Sviluppo - Basell 27

Ordinamento degli score derivanti dall’analisi delle componenti principali mediante un indicatore di qualità complessivo

CASCINI E. AICQ - ASA [email protected]

Uno dei risultati derivanti dall’analisi delle componenti principali è la possibilità di assegnare una serie di score a ciascun item. Ad esempio, dall’analisi dei dati del seguente riquadro: Caratteristiche 1 2 3 4 5 6 7 Item 1 0,40 0,68 0,92 0,32 0,66 0,12 0,22 Item 2 0,30 0,72 0,90 0,28 0,77 0,14 0,25 Item 3 0,45 0,58 0,88 0,40 0,78 0,18 0,99 Item 4 0,48 0,66 0,79 0,30 0,65 0,26 0,30 Item 5 0,46 0,65 0,12 0,38 0,66 0,30 0,42 Item 6 0,90 0,72 0,70 0,30 0,77 0,14 0,25 Item 7 0,55 0,70 0,85 0,26 0,66 0,14 0,90 Item 8 0,30 0,72 0,90 0,28 0,77 0,14 0,88 Item 9 0,48 0,20 0,79 0,30 0,65 0,26 0,30 Item 10 0,48 0,66 0,10 0,42 0,60 0,32 0,40 come si può verificare, eseguendo i calcoli, ed indicando con sik, il k.mo score dell’item i, (i = 1, 2, …, 10; k = 1, 2, …, 7) risulta, per il primo item, ad esempio, arrotondando al secondo decimale: s11 = 0,83 s12 = 0,31 s13 = 0,77 s14 = -0,24 s15 = 0,48 s16 = -0,46 s17 = 0,43 (1) Definito l’indicatore: Iqi = (γ12vi1 + γ22vi2+ ... + γ72vi7) /( γ12 + γ22 + ... + γ72) per l’item i, con γk peso di importanza della caratteristica k, variabile tra 0 e 1, con vik il valore della prestazione k dell’item i, si ottiene, come si può verificare, eseguendo i calcoli, ponendo tutti i γk = 1, e assumendo che i valori di vik siano quelli indicati nella tabella riportata sopra: Item Iq 1 0,474286 Come si può verificare, considerando come variabili indipendenti gli elementi 2 0,480000 della matrice sik, della quale la (1) costituisce la prima riga, e Iq variabile 3 0,608571 dipendente da essi, si ottiene: 4 0,491429 5 0,427143 Iq = 0,164 x s.1 – 0,056 x s.2 + 0,188 x s.3 + 0,01 x s.4 + 0,107 x s.5 – 0,189 x s.6 + 0,174 x s.7 6 0,540000 7 0,580000 La relazione lineare tra Iq e s.k è deterministica (coefficiente di correlazione 8 0,570000 uguale a 1), ed è valida per qualsiasi insieme di pesi. Questa circostanza, 9 0,425714 dimostrabile analiticamente, può essere utilizzata per ordinare i valori di score 10 0,425714 derivati dall’analisi delle componenti principali.-. Si farà vedere come gli item possono essere ordinati esattamente mediante questo indicatore. Nella figura sottostante, a titolo illustrativo, sono indicati gli item in funzione dei primi 3 score. 28

An opinion mining application on OSINT for a reputation analysis of public administrations

SANTARCANGELO V.*, ODDO G.*, PILATO M.*, VALENTI F.*, FORNARO C.** *iInformatica S.r.l.s., **Università Telematica Internazionale "UniNettuno" [email protected]

The United States Department of Defense defines the Open Source Intelligence (OSINT) as “The intelligence discipline that pertains to intelligence produced from publicly available information that is collected, exploited, and disseminated in a timely manner to an appropriate audience for the purpose of addressing a specific intelligence and information requirement”. Then, OSINT representing a large amount of the public accessible data on the web (from websites, blogs, social networks) can be considered a Public “Big Data” example. OSINT is also strictly related to Opinion Mining (also known as Sentiment Analysis), a discipline aiming at retrieving the opinion of a subject from web contents, reaching also the reputation analysis scope. As known, Opinion Mining Process can be summarized into 4 steps : Target Definition (1), OSINT Extraction (2), Sentiment Analysis (3), Score Return (4). After choosing a keyword/phrase, a crawler extracts contents related to the user input from OSINT data (2). Sentiment Analysis (3) examines the polarity of the filtered data extracted thanks to the use of a Sentimental Thesaurus, associating a polarity (negative, neutral, positive) for each term of the extracted excerpts, determining the Score (4). Starting from the state of the art, we feel that this work can propose an interesting innovative solution for Reputation analysis on OSINT data of public administrations. The devised system is based on an own Italian thesaurus for opinion mining, then it is particularly tuned for Italian context analysis. To achieve better performances our system considers also the use of “Semantic Network” as support to the crawler and the filtering phase. We think that this application could improve the service quality level of public administrations, as it referred to people opinions of the web. It can be also useful to obtain the reputation score about a person, an office, a project/subject/theme of Public Administration realizing an indirect customer satisfaction system and integrating the data of reporting application (as the App “Comunicamelo.it” of “Comune di Erice”). In fact, the integration with public opinions of websites, blogs and social networks gives relevancy, accuracy and quality to this kind of tools.

ISBN 978-88-88793-87-0