Psychophysics with children: Evaluating the use of maximum likelihood estimators in children aged 4–15 years (QUEST+)

Abstract

Maximum Likelihood (ML) estimators such as QUESTþ allow complex psychophysical measurements to be made more quickly and precisely than traditional staircase techniques. They could therefore be useful for quantifying sensory function in populations with limited attention spans, such as children. To test this, the present study empirically evaluated the performance of an ML estimator (QUESTþ) versus a traditional Up-Down Weighted Staircase in children and adults. Seventy-one children (4.7-14.7 years) and 43 adults (18.1-29.6 years) completed a typical psychophysical procedure: Contrast Sensitivity Function (CSF) determination. Some participants were tested twice with the same method (QUESTþ or Staircase), allowing test-retest repeatability to be quantified. Others were tested once each with either method (QUESTþand Staircase), allowing accuracy to be quantified. The results showed that QUESTþ was more efficient: In both children and adults, approximately half the number of ML trials were required to attain comparable levels of accuracy and reliability as a traditional Staircase paradigm, and plausible CSF estimates could be made in even the youngest children. The ML procedure was also as robust as the Staircase to lapses in concentration, and its performance did not depend on prespecifying correct model priors. The results show that ML estimators could greatly improve our ability to study sensory processes and detect impairments in children, although important practical considerations for-and-against their use are discussed.

Development of the spatial contrast sensitivity function (CSF) during childhood: Analysis of previous findings and new psychophysical data

“…The full distribution of ages is shown graphically in the Supplementary Methods . Elements of these data have been reported previously when considering the performance of different psychophysical algorithms in children ( Farahbakhsh, Dekker, & Jones, 2019 ). However, the developmental effects reported in the present paper have not been previously analyzed or reported.…”

“…All participants attempted to complete two CSF assessments, either: 2 × staircase ( N = 21 children, 15 adults), 2 × ML ( N = 16 children, 15 adults), or one of each ( N = 34 children, 13 adults). However, as detailed previously ( Farahbakhsh, Dekker, & Jones, 2019 ), some participants ultimately contributed data for only one assessment, either because they were too young to complete two assessments ( N = 12 children), or due to technical errors in the initial implementation of the ML procedure leading to invalid/unusable data ( N = 6 children). In practice, the data from the two psychophysical methods were highly correlated (see Supplementary Results ), so all CSFs were averaged within-subjects, to yield one CSF estimate per observer (114 CSFs total).…”

Can Psychophysics Be Fun? Exploring the Feasibility of a Gamified Contrast Sensitivity Function Measure in Amblyopic Children Aged 4–9 Years

“…The core psychophysical algorithm (the "back end") consisted of a QUEST+ (13, 14) (Maximum Likelihood) procedure, similar to the qCSF (7)(8)(9)(10)(11)(12). It was the same algorithm that we have described in detail previously (6). However, in previous works it received input from a conventional four-alternative forced choice (4AFC) psychophysical task, whereas here the "front end" input was provided from the unconstrained, gamified procedure described above.…”

“…One key difficulty is that conventional CSF assessments are too slow to be performed routinely in young children. Thus, while letter charts [e.g., Pelli Robson charts (5)] can provide a rapid summary measure of overall contrast sensitivity, to measure contrast detection thresholds precisely, and to do so across multiple, specific spatial frequencies, typically requires a protracted psychophysical procedure composed of several hundreds of trials (10 min) (6).…”

“…These "maximum likelihood" (ML) algorithms evaluate all previous trials, along with all possible outcomes to any subsequent stimulus, in order to determine the most informative stimulus to present next. This makes ML assessments faster than conventional psychophysical procedures [e.g., adaptive staircases (15,16)], allowing the whole CSF to be measured in around 30-100 trials (3-10 mins) (6,8,10). Furthermore, in situations such as home monitoring, where the same individual undergoes repeated testing, data from any previous assessments can be entered as "prior information, " further reducing any subsequent test durations.…”