It is well known that model selection induces bias (see e.g. https://t.co/tDXngqVLqV). One can 1) suggest an alternative and messy approach to improve model selection bias (i.e. model-averaging), or 2) just keep from inducing that bias in the first place.
@bmwiernik @IsabellaGhement @bolkerb @AdanZBecerra1 @BatesDmbates Here is a good one: "Model selection bias and Freedman’s paradox" https://t.co/NZQ5bwicaQ Notice in this table there is no selection. Now, some might select, but the problems apply there a
@epi_punk @ken_rothman The intro, results, and discussion from this is relatively free from maths: https://t.co/XI5XJjiAbm