Preference surveys report individual choices regarding air alternative choice set. This type of data is paramount to forecast the product/service demand as it makes possible to determine the importance the consumers/users pay to the different product/service characteristics. The data modelling is usually performed through Legit. As a probabilistic model, Legit is based on assumptions about the consumer's behaviour, which might not be real (for instance, the model assumes there is no taste variation among individual). Besides, researchers have been using a linear utility function in order to avoid model complications, even though it is known that the individual' s behaviour in a choice process is not linear. This paper reports literature main research findings on the subject, as well as, the influence of designs in the calibration results. Experiments are conducted using simulation tools. Results showed that, for data holding Legit assumptions, the bigger the distance between the attribute vectors, the better the calibration results from Legit modelling as it reduces the possibility of the optimisation algorithm to get stuck in a fiat region. On the other hand, it is likely that the data could break the homoscedasticity Logit assumption on the error term.