The impact of the pupil size artifact on pupil-based eye-tracking data in reading tasks: Assessment and compensation

Eye tracking in reading research requires high spatial accuracy due to small, densely arranged areas of interest. However, despite widespread use of pupil-based eye trackers in reading studies, a major source of systematic inaccuracy remains largely unaddressed: the pupil size artifact (PSA) - apparent gaze shift in fact caused by pupil dilation even when the eye remains stationary. Using pupillary light reflex manipulations and reading tasks under controlled luminance, we quantified gaze inaccuracy due to the PSA and compared correction methods. We observed systematic apparent gaze shift of up to 2 <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mmultiscripts><mml:mrow></mml:mrow> <mml:mrow></mml:mrow> <mml:mo>∘</mml:mo></mml:mmultiscripts> </mml:math> as pupil sizes varied from 2 to 6 mm. Horizontal PSA showed contralateral patterns (median slopes: 0.38 <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mmultiscripts><mml:mrow></mml:mrow> <mml:mrow></mml:mrow> <mml:mo>∘</mml:mo></mml:mmultiscripts> </mml:math> /mm), while vertical PSA increased with pupil size (up to 0.86 <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mmultiscripts><mml:mrow></mml:mrow> <mml:mrow></mml:mrow> <mml:mo>∘</mml:mo></mml:mmultiscripts> </mml:math> /mm for larger pupils). Even under constant luminance, pupil size varied substantially during reading (median 95% ranges 0.78-1.38 mm). We compared two correction approaches: line assignment (standard in reading research) and PSA recalibration (modeling pupil-size-induced apparent gaze shift across the screen). Both methods effectively corrected vertical apparent gaze shift, showing similar average correction offsets and sensitivity of correction to pupil size, suggesting that line assignment implicitly compensates for vertical PSA effects. However, only PSA recalibration addressed horizontal apparent gaze shift, reducing overall gaze shift by over 50% and improving the performance of line assignment algorithms. Our findings underscore the importance of accounting for PSA in reading research. We offer practical recommendations for improving gaze accuracy in eye-tracking reading studies.