An external coder with no eye-tracking experience and naive as to expected outcomes was trained to code fixations from both groups.
On the other hand, current hand-coding methods (e.g., coding fixations looking at the videos frame by frame) can be extremely time consuming and, in some cases, imprecise if coding low-quality data sets.
Furthermore, the hand-coding graphical tool alone where the user simply needs to click on the screen to manipulate fixations is already much faster than previous hand-coding approaches (e.g., coding fixations analyzing videos frame by frame).
It can be argued, however, that having the results from the event detection algorithms as a basis to hand-code fixations can influence the coder's decisions for accepting or deleting fixations.
The automatic detection algorithms include a number of post hoc validation criteria aiming to obtain the cleanest results that the algorithms alone permit, in order to facilitate and speed the process of hand-coding fixations.
If an infant did not move his/her eyes from the central face during the trial, the response was coded as a fixation response on the face.
This does not necessarily mean that it is not possible to find short fixations in our data set: In case there are clean short fixations, they can also be manually coded after the automatic detection of fixations.
A fixation was coded when both the x- and y-coordinates were stable at one point, or in other words, when the 2-D representation of both x- and y-coordinates were displaying horizontal lines.
Additionally, fixations can be coded as smooth pursuit once they are created.
In the interest of improving the reproducibility of the experimental outcomes, we believe it is very important to include precise guidelines to define how the fixations are being coded.
