Tryon’s Rat Experiment

Tryon’s Rat Experiment is a multi-decade selective breeding animal experiment begin in the 1930s which rapidly bred enormous differences in a complex psychological trait, maze-running, demonstrating core principles of behavior genetics.
biology, psychology, statistics, genetics, Wikipedia
2020-01-132020-01-14 notes certainty: highly likely importance: 7

Try­on’s Rat Ex­per­i­ment is a mul­ti­-decade se­lec­tive breed­ing an­i­mal ex­per­i­ment in the 1920s–1940s which em­ployed au­to­mated maze-run­ning ma­chin­ery to min­i­mize mea­sure­ment er­ror and, us­ing trun­ca­tion se­lec­tion, bred two differ­ent strains of rats: “maze-bright” and “maze-dull” rats, se­lected for high & low maze-run­ning per­for­mance.

Within a few gen­er­a­tions, the rats showed in­creas­ing differ­ences in maze-run­ning per­for­mance, and the two strains even­tu­ally had non-over­lap­ping dis­tri­b­u­tions. Try­on’s Rat Ex­per­i­ment rapidly bred enor­mous differ­ences in a com­plex psy­cho­log­i­cal trait, demon­strat­ing core prin­ci­ples of be­hav­ior ge­net­ics: the her­i­tabil­ity of even psy­cho­log­i­cal traits far re­moved from stan­dard ex­am­ples of ge­net­ics like coat col­or, and the abil­ity of se­lec­tion pro­duce large pop­u­la­tion-wide changes in a short time for even highly poly­genic traits like maze-run­ning.

The eti­ol­ogy of the changes in per­for­mance were sub­se­quently in­ves­ti­gat­ed: the per­for­mance changes were not on the g-fac­tor of in­tel­li­gence, but were more maze-run­ning-speci­fic, and have neu­ro­log­i­cal cor­re­lates.

The ex­per­i­ment was wide­ly-cited in psy­chol­ogy and early be­hav­ior ge­net­ics, and par­al­leled var­i­ous later ex­per­i­ments in se­lec­tion on com­plex be­hav­ioral traits.


Fig­ure 4 from Tryon 1940 show­ing near-com­plete di­ver­gence after 7 gen­er­a­tions.


Rosen­thal & Fode 1963 tested the abil­ity of ex­pectancy effects to cre­ate ‘maze-bright’-like differ­ences in rats; un­der­grad­u­ate stu­dents were told ran­dom rats were maze-bright or dull, and there were sub­se­quent small differ­ences which were just barely sta­tis­ti­cal­ly-sig­nifi­cant.

While they do not ex­plic­itly cite Tol­man or Try­on, they use maze-run­ning as the task and the term ‘maze-bright’, and ap­prov­ingly quote Pavlov about how sup­posed ge­netic effects may in fact be solely en­vi­ron­men­tal. Given how well-known Tryon was, the spe­cific term, and Rosen­thal’s broader re­search par­a­digm which is purely en­vi­ron­men­tal and his claims to make enor­mous changes in highly her­i­ta­ble traits by sim­ple ex­pectancy effects like the (de­bunked) Pyg­malion effect, Rosen­thal & Fode 1963 has been in­ter­preted as im­ply­ing that Try­on’s rats were not ge­net­i­cally mod­i­fied and pos­si­bly the ob­served differ­ences were merely Tryon et al’s ex­pectancy effect­s—­maze-bright­ness be­comes a self­-ful­fill­ing prophe­cy.1

This is false for many rea­sons:

  1. p-Hack­ing: Rosen­thal & Fode 1963 is highly du­bi­ous to be­gin with: the p-val­ues are all one-tailed (de­spite Rosen­thal’s other ex­per­i­ments sup­pos­edly show­ing that ex­pectancy effects are com­plex & their di­rec­tions diffi­cult to pre­dict, mak­ing one-tailed un­jus­ti­fied ex­cept to make the p-val­ues much smaller2), and de­spite that, still gen­er­ally only just p < 0.05 (a clas­sic in­di­ca­tor of a re­sult that will fall prey to the ). Other re­sults of Rosen­thal like the Pyg­malion effect have sig­nally failed to repli­cate, and there do not ap­pear to be any repli­ca­tions of Rosen­thal & Fode 1963.
  2. Too Small Effect: The ex­pectancy effect is far smaller than the ge­netic effect ob­served within a few gen­er­a­tions, much less the fi­nal gen­er­a­tion. The ge­netic effects of se­lec­tion can eas­ily ac­cu­mu­late; the sup­posed en­vi­ron­men­tal effect would not.
  3. Ir­rel­e­vant to Au­to­mated Ex­per­i­ments: The mech­a­nism of bias as pro­posed is im­pos­si­ble—­Tol­man & Tryon in­vested great efforts into au­to­mated maze-run­ning ma­chin­ery pre­cisely to make the mea­sure­ments as ac­cu­rate & un­bi­ased as pos­si­ble. Un­der­grad­u­ate stu­dents could not have bi­ased the maze-run­ning mea­sure­ments by han­dling the rats dur­ing test­ing be­cause there were no hu­mans in­volved dur­ing test­ing! (Rosen­thal does not men­tion this in ei­ther ver­sion.)

  1. Iron­i­cal­ly, Rosen­thal would later in­voke his an­i­mal re­sults as in­creas­ing the prior plau­si­bil­ity of the Pyg­malion effect: “If an­i­mals be­come ‘brighter’ when ex­pected to by their ex­per­i­menters, then it seemed rea­son­able to think that chil­dren might be­come brighter when ex­pected to by their teach­ers.” (Rosen­thal & Ja­cob­son 1968) But ↩︎

  2. This ca­sual use of one-tailed tests when con­ve­nient shows up in Pyg­malion effect re­search as well, de­spite un­ex­pected re­sults.↩︎