My research field was initially addressed to statistical analysis applied to multiple fields, with particular attention to the link between the causal model and the estimated one, noting that very often they do not exactly match. Therefore, I studied this problem, especially in the SEM approach, finding as solution a new estimation method which I applied both to simulated data and to real marketing cases. The SEM method is widely used in economics and psychology because it gives the possibility to study all the possible causal relationships among the variables and to derive the unobservable directly from the observable ones. The SEM model, initially created to analyse continuous variables, was recommended, under particular hypotheses, also for discrete variables. Indeed, the causal log-linear model, its variant for categorical variables, has problems in determining and calculating the intensity of some types of causal effects, such as the effect of one variable on another, mediated by one third. To make comparable the results obtained applying the SEM and the causal log-linear models to the same dataset, I developed a method and the relative package R which calculate the causal effects in the causal log-linear models. I also noticed that a limit of the SEM is given by the obligation to specify the functional form of the relationships among variables. To solve this further problem, some methods have been proposed in literature, but they are completely devoid of causal analysis and are applicable only to simple models. Then, I proposed an evolution of these methods and a causal theory suitable for them to be applied to complex models. A problem in the choice of the function to be used in the estimation process was also found in the regressions with the variable of interest included in the unit range or binary. The proposed link functions must be specified a priori and are almost exclusively monotonous. After having developed some non-monotone functions, I am building a nonmonotone function, which has as its special cases the main functions already present in the literature, so to leave to the data the determination of the relationship between the variable of interest and the explanatory ones. This function would allow the researcher to understand unexpected relationships among the variables and to evaluate their possible trends. In parallel, I am also developing the method of estimation of the parameters of this function so as to find one which best suits it. I am also validating the function with its relative estimation method using simulated data, so that I can know exactly the true relationship among the variables and thus be able to verify how close the new method is to the real model. From the first analyses the results are promising. My further research field is the study of big data and their application in the economic and pharmaceutical fields. When one works with this type of dataset, it is essential to be able to isolate the factors which have the greatest influence on the studied variable. Then, I created an algorithm which allows to determine these factors when they are binary and the variable of interest is continuous, demonstrating its validity with the use of simulated and pharmaceutical data. The selection of variables in a high-dimensional dataset is also useful for understanding market demands, for example when a company wants to introduce a new product. Subsequently, using a modified version of the bicluster analysis, I built a new algorithm which allows to understand the characteristics to be developed and those to be overlooked in the launch of new products, considering a specific target of customers. For the expansion of a company and its rise in the market it is essential to increase, stimulate and know its possible innovative factors. I worked on the development of an index to measure the state of innovation present in Australia and on the drafting of a report for Australian companies about the innovation. Innovation is mainly necessary in the health sector, where new techniques are always being tested. Since sleep disorders are closely linked to health problems, in recent years it has become essential to look for increasingly less invasive techniques which do not require hospitalization to control the various stages of sleep. Sensors have proven to be a good solution, but a specific algorithm is needed to interpret the data which they provide. Using some statistical techniques, including the factor analysis, which derives the unobservable variable from the observed one, and the time series change point detection methods, I developed an algorithm which determines the sleep phases and which allows to establish, based on an index, its quality.
Summary CV: I am a statistical consultant of the University of Sydney (research assistant). I was a teaching assistant at the Free University of Bolzano-Bozen. I was a post-doc researcher at the Ca’ Foscari University of Venezia and I collaborated with the Bocconi University. I presented my works at many international conferences and I am a reviewer of six scientific journals. I wrote a few papers and a package for R-project. I received my master’s degree in Economics in 2002 at the Ca’ Foscari University of Venezia, my master’s degree in Statistics in 2011 at the University of Padova and the PhD in Statistics in 2015 at the University of Padova with scholarship. I obtained my professional international master’s degree in Economics and Finance in 2004 at the Ca’ Foscari University of Venezia. I was a trainee-chartered accountant and a trainee auditor. Now I collaborate with M. Garbuio, a senior lecturer at the University of Sydney, and with R. Dussin, a chartered accountant.