In the last 40 years in much of the western world dropout rates in higher education have remained almost unchanged. Although student retention seems to be the most studied and discussed aspect, nearly every empirical study on the causes of dropout in higher education and even more the impact of retention actions carried out by universities, in most cases have achieved modest results. This paper argues that this fact finds its explanation, to a certain extent, in the nature of the methodological approaches and factual supports of the empirical studies that most of those actions were based on. In this regard, there are strong arguments and empirical evidence that reveal the deficient nature of the factual basis of the most accepted models, theorizations and measurements on dropout in higher education. Among them are those that underlie the models proposed in 2012 by Vincent Tinto and Adam Seidman, the two main current references on the subject. The most significant questions point to the low reliability of the inferences produced from the application of surveys, especially the national survey of student engagement, very recurrently applied throughout the western world in empirical studies on dropout in higher education.