Efficient visual information sampling develops late in childhood 2016-2019

Peter , Jones and Linnea, Landin and Aisha, McLean and Mordechai, Juni and Larry, Maloney and Marko, Nardini and Tessa, Dekker (2020). Efficient visual information sampling develops late in childhood 2016-2019. [Data Collection]. Colchester, Essex: UK Data Service. 10.5255/UKDA-SN-854078

In playgrounds, traffic and around the house, we are at continuous risk of bodily injury. Children are particularly accident-prone, as reflected in the disproportionally high accident-rates of pedestrians aged 15 year and younger. In recent years many researchers, including myself, have made considerable strides in understanding how visuomotor abilities improve across development. Despite these advances in understanding, current approaches do not consider how children adjust for their changing abilities to avoid unnecessary bodily risk during everyday visuomotor decisions. For example, children are less efficient than adults at avoiding incoming traffic when crossing busy roads. Do they account for this correctly by waiting for larger gaps between cars before crossing?
Economic decision-making theories will be employed to model children's visuomotor choices. Economists identify the best financial investments by trading-off probabilities of positive and negative monetary outcomes. Likewise, such tactics can identify the best movement strategy (e.g., when to cross) that maximises safety and efficiency (e.g., avoids accidents but utilises safe gaps in traffic). This has proven very effective for modelling mature visuomotor behaviour, showing that adults often choose movements that optimise performance. Recently I pioneered this approach with children, showing that children aged 6 to 11 years make riskier visuomotor choices than adults during manual reaching. To understand and reduce the effects of risky action selection on childhood injury, we must characterise more broadly how visuomotor decision-making develops, and understand which neurocognitive processes drive this change. A combination of precise behavioural tests, mathematical modelling and neuroimaging will be used to address these fundamental questions.
This proposal consists of 3 main objectives that each form a necessary step towards understanding children's movements under real risk in real world situations; These are to (1) characterise children's risky visuomotor decisions in realistic circumstances, including whole-body movements and poor eye-sight, (2) identify which basic mental processes underlie children's immature visuomotor choices, and (3) investigate how these might be improved through training. By characterising changes in visuomotor decision-making in detail at the behavioural and neural level, these objectives will significantly advance our understanding of the developing visuomotor system in action and the mechanisms of visuomotor decision-making. Moreover, this project has great translational potential for improving childhood safety and well being in everyday life, by informing educational programs and generating new ideas for interventions to improve safety.

Data description (abstract)

The aim of the study was to investigate how the ability to trade off the benefits of visual information against the costs of this information develops during childhood adolescence and adulthood. We tested this by asking participants to sample dot location cues to find a fish hidden on a touchscreen for points. However, each location cue came at a cost, so the more cues you gather, the less points you can win for finding the fish. By measuring in a separate (fixed) condition how the likelihood of finding the fish increased with more dot-cues, we were able to predict the score-maximising strategy for each individual. We then tested if children, adolescents, and adults were able to identify this ideal strategy - we found that while adults did, children tended to sample fewer dots then they should have to maximise score, and also used more inconsistent sampling strategies.
We also used these data to analyse whether children use the same visual averaging strategies as adults do. As part of localising the target participants are asked to locate the middle of a cloud of dot location cues. To test whether children use the same strategy to compute the middle (in the current task the best strategy that will yield most points is the arithmetic mean) we compared their estimates of the mean with the output of various different mean computation strategies.

Data creators:

Creator Name	Affiliation	ORCID (as URL)
Peter Jones	University College London
Linnea Landin	University College London
Aisha McLean	University College London
Mordechai Juni	University of Santa Barbara
Larry Maloney	New York University
Marko Nardini	Durham University
Tessa Dekker	University College London

Sponsors:

Economic and Social Research Council

Grant reference:

ES/N000838/1

Topic classification:

Science and technology
Education
Psychology

Keywords:

visual sampling, decision making, risk-taking, child development, cost-benefit tradeoff

Project title:

Using economic theory to understand children's risky visuomotor decisions

Alternative title:

The development of perceptual averaging: learning what to do, not just how to do it

Grant holders:

Tessa Dekker

Project dates:

From	To
1 January 2016	31 August 2019

Date published:

07 May 2020 15:47

Last modified:

07 May 2020 15:47

Coverage and Methodology

Collection period:

Date from:	Date to:
1 January 2016	31 August 2019

Geographical area:

London

Country:

United Kingdom

Data collection method:

This was an experimental study, in which participants did a task on a computer that was custom-developed to address the main questions described above.
Design and procedure: Each participant completed a fixed condition and a free condition. We varied location cue reliability between subjects, by changing the width of the distribution from which the dot location cues were sampled from sigma=12mm to sigma=28mm (see below for details). Most adults did both cue reliability conditions, but we only analysed the first cue reliability condition they took part in. All subject completed the free condition, to measure sampling choices, and the fixed condition to model the ideal strategy that should be used to maximise score in the free condition (see below for details).
In the free conditions participants sampled probabilistic location cues (dots) drawn from a bivariate Gaussian distribution with sigma=12 or sigma=28, centred on a target (max Ndots was 20). They bought between 1 to 20 dots by pressing space bar, with the score won for a hit depreciating with 1 point for each dot (starting from 20 points). Because the middle of the dot-cues is more likely to overlap with the target/distribution mean as the number of dots increases (as variance around this location follows sigma/sqrt(Ndots)), sampling more dots increased the chance of locating the target. However, purchasing more dots also reduced the value of the target, so participants had to trade-off the benefits of more information against the cost. Once the participant felt they had bought a sufficient numbers of dots, they placed the cursor on a touchscreen in the estimated mean location of all the dots to find the target. They then were able to adjust their response until they confirmed their choice by pressing enter, upon which they got feedback about whether they had hit the target, and about their score. Each participant completed 100 trials of this condition.
In the fixed conditions, the design was identical except that rather than purchasing a number of dots of their choice, participants were presented with a fixed number of dots on each trial, and then located the mean of the dots to find the target following the same procedure as in the free condition. This condition was used to model the ideal strategy in the free condition; Specifically we used it to estimate for each individual, the hit probability for a given Ndots, and multiplied this by the points that could be won for this ndots to obtain the expected gain (hit probability x value) for that given Ndots. The ideal strategy is to sample the Ndots with the highest expected outcome. Each participant completed 25 trials with a fixed number of dots. Most children were presented with Ndots = [2 3 7 15]. Some adults were presented with all Ndots in this condition.

Observation unit:

Individual

Kind of data:

Numeric, Text

Type of data:

Experimental data

Resource language:

English

Access and Administration

Data sourcing, processing and preparation:

Sample: We include data from 5 age groups: 27 6 to 7-year-olds, 30 8 to 9-year-olds, 27 10 to 11-year-olds, 25 13 to 15-year-olds, and 29 adults. All participants had normal or corrected to normal vision and motor abilities, and had no known neurological abnormalities. Participants were recruited from local schools and databases so this was an opportunity sample. Measures: We recorded participants' aiming location on a touchscreen, as well as their hits (i.e., whether the touched location fell within a circular target area with a 12.761 pixel radius), and reaction times, and tested by now much their reaches deviated from the ideal location that would maximise score according to our predictions from the fixed condition. For further experimental details, please see publications listed under data description. Details on data collection and processing: The experimental paradigm was programmed and run in Matlab Psychtoolbox. Results from individual trials were exported to multiple .cvs files for each subject - creating a file per experimental block. The Config ID reported in the .csv file indicates whether the file contains trails in the fixed or the free experimental condition. These data read into and further analysed in matlab to compute model predictions and individual participant specific variables of interest. The results of these further data analyses are listed in SPSS .sav file.

Rights owners: