The short method of claiming that is that utilizing a mathematical mannequin knowledgeable by background info to set a previous distribution for logistic regression isn’t any extra “subjective” than deciding to run a logistic regression within the first place.
Right here’s an extended model:
Each as soon as in awhile you get individuals saying that Bayesian statistics is subjective bla bla bla, so each as soon as in awhile it’s value reminding individuals of my 2017 article with Christian Hennig, Beyond subjective and objective in statistics. Numerous good dialogue there too. Right here’s our summary:
Selections in statistical information evaluation are sometimes justified, criticized or prevented through the use of ideas of objectivity and subjectivity. We argue that the phrases ‘goal’ and ‘subjective’ in statistics discourse are utilized in a largely unhelpful method, and we suggest to interchange every of them with broader collections of attributes, with objectivity changed by transparency, consensus, impartiality and correspondence to observable actuality, and subjectivity changed by consciousness of a number of views and context dependence. Along with stability, these make up a set of virtues that we expect is useful in discussions of statistical foundations and observe.
The benefit of those reformulations is that the substitute phrases don’t oppose one another and that they offer extra particular steerage about what statistical science strives to realize. As an alternative of debating over whether or not a given statistical technique is subjective or goal (or normatively debating the relative deserves of subjectivity and objectivity in statistical observe), we are able to acknowledge fascinating attributes akin to transparency and acknowledgement of a number of views as complementary targets. We show the implications of our proposal with current utilized examples from pharmacology, election polling and socio-economic stratification. The purpose of the paper is to push customers and builders of statistical strategies in the direction of more practical use of various sources of data and extra open acknowledgement of assumptions and targets.