Modeling Dynamic Control in Normal Aging

Modeling Dynamic Control in Normal Aging Brian D. Glass (b.glass@qmul.ac.uk) Biological and Experimental Psychology Centre, School of Biological and Chemical Sciences, Queen Mary College, University of London, Mile End, London E1 4NS, UK. Magda Osman (m.osman@qmul.ac.uk) Biological and Experimental Psychology Centre, School of Biological and Chemical Sciences, Queen Mary College, University of London, Mile End, London E1 4NS, UK. behavioral characteristics and strategies in the dynamic control task. Abstract Complex and dynamic decision making environments are common throughout life, but little is known about how normal aging influences performance on these types of scenarios. To determine performance differences associated with normal aging, we test older and younger adults in a dynamic control task. The task involves the control of a single output variable via multiple and uncertain input controls. A computational model is developed to determine the behavioral characteristics associated with normal aging in a dynamic control task. Older adults exhibit a positivity effect, congruent with previous research. Model based analysis demonstrates a unique performance signature profile associated with normal aging. Method Procedure In the present dynamic control task, the participant attempts to control a single outcome value towards a goal. To do so, on each trial the participant chooses values for three separate cues. These cue values are then combined via the dynamic control equation (Equation 1) then summed with the outcome value plus some normally distributed random noise (standard deviation = 8). In this way, the participant's cue selections guide the outcome value. The outcome value is initialized at 178 with a goal value of 62 and a “safe range” ( ±10 around the goal value) Keywords: dynamic decision making; learning; normal aging; computational modeling Introduction Normal human aging is associated with cognitive changes that lead to differences in the way older adults approach and perform in decision making tasks. Specifically, older adults appear to suffer from executive control deficits (Braver, et al., 2001; Kray, Li, & Lindenberger, 2002; Ortega, et al., 2012). However, emerging evidence suggests that older adults can utilize compensatory strategies to return performance to or beyond baseline levels (Glass, et al., 2012; Huang, et al., 2012; Worthy, et al., 2011). Equation 1. where y(t) is the outcome on trial t, x 1 is the positive cue, x 2 is the negative cue, and e is an error term randomly sampled from a normal distribution with a mean of 0 and SD of 8. The dynamic control equation was designed such that one cue has a positive impact on the outcome value, one cue has a negative impact, and a third cue has no impact. The impact of the cue is not labeled or available to the participant, thus the participant must learn to control the outcome value based solely on the resulting movement of the outcome value on each trial. After each trial, the cue input values are reset to 0. The participant can then freely select input values for each of the three cues before confirming the choices. While previous research has focused on classic paradigms such as category learning, task switching, and single- response choice procedures, little is known about normal aging in dynamic control tasks for which the participant controls multiple input variables in an integrative and uncertain task environment. Such complex dynamic environments are analogous to many real-life situations. For example, we make several distinct health choices on a daily basis which influence our overall health and wellbeing in uncertain ways. These types of environments are often noisy and the specific influence of the various choices is often unclear or unspecified. A critical feature of this control task is that the outcome value can swing below the target, meaning the participant must dynamically adapt in order to maximize performance. After an initial learning phase, the participants completed 2 Test blocks of 20 trials each. The first Test block was a Congruent Test in which the starting value and goal criterion were equivalent to the learning phase. The second Test block was a Transfer Test with a different starting value and goal value than the earlier phases. At the beginning of each block, the control task was reset to the initial state. The present research contrasts older adult and younger adult performance in a dynamic control task designed to simulate such real-life dynamic decision making environments (Osman & Speekenbrink, 2011). A novel computational modeling technique is developed to assess individual