vol. 13 núm. 2 (2020)

Estimados lectores, desde el equipo editorial de la revista Comunicaciones en Estadística queremos enviarles un cordial saludo y agradecer el apoyo de todos nuestros autores, revisores y lectores. Fiel a su línea editorial, la revista Comunicaciones en Estadística continúa presentando trabajos de investigación con aplicaciones en diferentes áreas del conocimiento

High volume of data is generated daily, especially on social networks. The usage of this data as a source in the study of the agent’s behavior in the stock market have been gaining interest, specifically in the machine learning field. Hence, in this article; a study about the predictive power of this kind of data over the future price variation direction of a stock is made, using the texts published in the StockTwits social network and machine learning techniques.

The orthodox economical models propose that the human being is rational, selfish and maximizing to make their consumption choices. The evidence from behavioral economics challenges these assumptions by proposing new models to study human choice. Studying the process of choosing crop care products in tomato growers in Colombia, this research sought to compare statistically and conceptually the RUM (Random Utility Maximization) and RRM (Random Regret Minimization) models built via modeling of discrete choice concluding that RRM models achieve better goodness of fit to describe choice behavior and purchases of nematicides in samples of Colombian tomato producers, so they constitute a viable model for designing new products, estimating their potential market share and fix prices using psychological and economical principles.

When modeling count data, the poisson model is typically used, in which the equidispersion (ED) assumption is assumed, where the mean and variance are equal. When this condition is not easy to justify, different alternatives have been proposed, some more flexible than others in terms of accounting for both overdispersion (OD) and underdispersion (UN). One of them is the COM-Poisson model which was recently proposed and has been evaluated in inferential terms. The investigation presented here aims to compare the COM-Poisson model predictive quality with respect to the Poisson model and establish the loss in efficiency that occurs when the inadequate model is fitted when the property of equidispersion is not satisfactory. A simulation study determined that adjusting the inappropriate model either over or underdispersion does not represent in most cases, a gain or loss of the predictive quality. Two case studies illustrate ours findings obtained here.

TIMSS is an internationally applied instrument for measuring trends in Mathematics and Science achievement in the fourth grade of primary and eighth grade of secondary school. Given the design of these tests, it is possible to obtain traceable and comparable results, which allow governments of different nations to generate, maintain or eliminate public policies with the objective of improving their education system. Naturally, these tests have acquired great relevance at the global level. Considering the above, the objective of this academic work is to propose an estimator using Small Area Estimation that allows for improving the quality of the estimates currently obtained with the sampling design used in TIMSS 2015. In addition, this methodology allows us to estimate the result that would have been obtained by a country that did not present this test in 2015, but has auxiliary information related to academic achievement. To build the estimator of academic achievement, the Fay-Herriot model is used using auxiliary variables such as demographic and socioeconomic information and scores obtained in previous applications of the test. Finally, this paper will show that the use of Small Area Estimation in international tests improves the efficiency of the results obtained and, therefore, allows for better decision-making in education policy.