short theoretical+practical reflection on the study of biases after reading the very interesting new PNAS paper https://www.pnas.org/content/early/2020/09/18/2005058117">https://www.pnas.org/content/e... by  @atomsrivet @summerfieldlab et al. we often study biases very broadly or very narrowly & try to provide explanatory theories at those levels

eg. cog biases arise bc brain simplifies information for efficiency vs confirmation bias occurs due to cost asymmetry of error detection etc. this approach limits our capacity to look into possible order/structure within domains of biases & inhibits our theory building capacity

great-but how to tackle this practically? in their new paper  @atomsrivet & colleagues look at 3 similar but distinct decoy effects and provide a rich dataset sampling stimuli meticulously over bidimensional multiattribute space. this allowed them to *fully* map decoy influence

the comprehensive range of decoy locations meant the authors could examine different factor(s) that comprise the decoy map. this is an opportunity to sing the gospel of SVD, with a fantastic thread by @daniela_witten https://twitter.com/WomenInStat/status/1285610321747611653">https://twitter.com/WomenInSt...

through an extensive modeling exercise, accounted here: https://twitter.com/atomsrivet/status/1308363205643849728">https://twitter.com/atomsrive... authors find support for an adaptive gain framework. interestingly, the distinct decoy effects can be examined as the manifestation of a singular phenomenon, where inputs are contextually compressed

this connects individual-level explanations for attraction, similarity,& compromise with higher-level explanation for neural coding efficiency. to sum up the journey: theory-> rich data-> extensive model comparison-> multi-level considerations-> more holistic understanding

so much for my daily session of procrastinating actual PhD work by reading exciting papers in different fields #phdchat #academictwitter