Con­duct­ing pref­er­en­tial look­ing stud­ies with infants online: Unique chal­lenges and solutions

Lori­jn Zaad­no­ordijk, Trin­i­ty Col­lege Dublin
@LorijnSZ

By load­ing the video, you agree to YouTube’s pri­va­cy pol­i­cy.
Learn more

Load video

Always unblock YouTube

Infant research fre­quent­ly relies on look­ing behav­ior. In many cog­ni­tive domains, look­ing time par­a­digms are used, for exam­ple, to assess what infants antic­i­pate, whether they can dis­crim­i­nate between stim­uli, or whether they have learned some­thing dur­ing the exper­i­ment. In addi­tion, recent­ly there has been renewed atten­tion to how infants’ look­ing behav­ior in their envi­ron­ment shapes what and how they learn. Infants are not pas­sive­ly receiv­ing infor­ma­tion, they active­ly seek it out by attend­ing to cer­tain types of organ­isms or objects over oth­ers at dif­fer­ent stages in development.

As such, ide­al­ly, devel­op­men­tal sci­en­tists could run large-scale, cross-sec­tion­al look­ing behav­ior and learn­ing stud­ies. Such stud­ies, unfor­tu­nate­ly, are time con­sum­ing and cost­ly. There has been increas­ing atten­tion to reduc­ing the costs and improv­ing the sam­ple sizes of exper­i­men­tal infant stud­ies, which are typ­i­cal­ly low in infant stud­ies (Bergmann, et al., 2018). The logis­tic and prac­ti­cal chal­lenges relat­ed to com­ing to the lab as well as the small age ranges can be iden­ti­fied as par­tial­ly respon­si­ble. Online data col­lec­tion offers the pos­si­bil­i­ty to acquire larg­er sam­ples as par­tic­i­pants can stay at home and par­tic­i­pate at a con­ve­nient time. This makes large-scale stud­ies more fea­si­ble in terms of both recruit­ment time as well as reach­ing a more diverse pop­u­la­tion. Being able to con­duct look­ing behav­ior stud­ies online would there­fore open excit­ing nov­el research possibilities.

How­ev­er, unlike in the lab, where state-of-the-art eye-track­ers may be avail­able, the look­ing data that is acquired online is based on a web­cam, which con­tains more noise and does not pro­vide the gaze direc­tion or a ref­er­ence axis to remove the effects of head motion. Man­u­al cod­ing of these data is labor-inten­sive and prone to day-to-day and inter­rater vari­abil­i­ty. An auto­mat­ed approach would thus be desir­able and improve repro­ducibil­i­ty. How­ev­er, although algo­rithms such as Ama­zon Recog­ni­tion can rel­a­tive­ly reli­ably indi­cate whether the infant is look­ing at the screen, detect­ing whether the infant is fix­at­ing on the left or right side of the screen is still prob­lem­at­ic (Chouinard, Scott & Cusack, 2019).

In addi­tion to test­ing infants online, we have been explor­ing meth­ods to improve the accu­ra­cy of gaze esti­ma­tion algo­rithms in online infant exper­i­ments. In this pre­sen­ta­tion, I will describe the unique chal­lenges that devel­op­men­tal sci­en­tists face when tran­si­tion­ing to online test­ing. I will touch upon a range of top­ics (from ethics to analy­ses) while describ­ing our expe­ri­ences with online look­ing behav­ior stud­ies and solu­tions we and oth­ers have explored. Final­ly, I will present var­i­ous state-of-the-art resources and ini­tia­tives that play a key role in online test­ing in the devel­op­men­tal sci­ence community.