Online interactions have nowadays a huge impact on everyday's life. People use social networks daily both for personal and work relationships. These interactions are inevitably influenced by the few elements present on online profiles. The adagio 'an image speaks a thousand words' is true also in this context: online profile portraits can have a great influence on interactions. Recently multimedia quality assessment opened the field to broader aspects as image aesthetics, interestingness and memorability. Higher level quality dimensions have been indicated as important and proposed in literature under the concept of 'image psychology'. This is particularly true for portraiture images. How to find and measure such factors is still an ongoing research subject. In this paper we refine some results on our previous research work, where we focused on face digital portraits context perception. Those results underlined the importance of some high level content features, e.g. the dress of the portrayed person and scene setting, in categorizing image. Here we consider different statistical approaches to factors of influence analysis, underlining positive and negative points of each one. Our reference model is the one of our previous work: logistic regression was adopted to model category fit based on images features.