NWO Vidi Grant for Andreas Alfons

Econometrics Department Director Professor Patrick Groenen is very pleased with the performance of Andreas Alfons: ‘In his VIDI project, Andreas focusses on rating scales as often used in surveys and online product reviews. Andreas intends to develop novel statistical methods for such data that are robust against outliers. I believe that his research is timely and will be of practical importance to many researchers.’

Adhemare de Rijk, Funding Manager Research reacts: ‘Andreas is able to explain complex statistical methodology not only in an understandable way, but also with great enthusiasm. This recognition by the NWO is very good news and I am very proud of him.’

Short summary of Andreas Alfons’ NWO Vidi research project

Empirical research in the social sciences relies heavily on the statistical analysis of data measured on rating scales. Results of such research have a tremendous impact on society: for example, policy makers monitor opinions in society via rating-scale data collected in surveys, psychologists collect rating-scale data in experiments to gain new insights into human behavior, and companies let test groups rate various aspects of a new product before launching it. 

Due to modern technology and the internet, data collection is now easier than ever. As a result, data sets are growing ever larger – consider, for instance, product ratings on web stores such as Amazon – but at the same time reliability of the data is decreasing with online data collection. Even though rating scales by definition have a limited range and thus do not exhibit extreme values, observations can still be outliers if they go against the correlation structure of the majority of the data. Such correlation outliers are likely to be present, especially in big data, yet currently outliers are typically ignored in the social sciences. If they are addressed, the common practice is to first delete outliers and then to apply standard statistical methods for further analysis. From a statistical point of view, outlier deletion is incorrect and may lead to wrong conclusions due to inflated significance. However, to date there are no statistical methods for rating-scale data that can cope with outliers.

With his research Andreas Alfons will try and solve this problem by developing a collection of outlier-resistant methods for the statistical analysis of rating-scale data. He will develop an outlier-resistant method for extracting the relevant information from big (incomplete) rating-scale data.

Erasmus School of Economics wishes to thank the Research Office (especially Adhemare de Rijk and Pim Jansen) and all faculty members at the mock interviews for their invaluable support during the application process.