106 / 218
104
Chapter 5
respond as quickly and accurately as possible and had to respond correctly within the response
deadline to obtain the reward.
The main experiment consisted of 160 (or 240) trials and lasted ~35 minutes with a 30 second
break after every 32 (or 48) trials (table 1). In the break, the amount of money the participant
earned thus far was displayed on the screen and participants were informed in advance that
the total amount would be added to their financial compensation as a bonus.
Analysis
For each participant, we excluded the first trial of each block, trials with a response time
(RT) faster than 200ms, and trials on which participants failed to respond. For each trial-
type, [Reward (low, high) x Task switching (switch, repeat)], we calculated the proportion of
accurate responses. For the RTs, we first excluded the erroneous trials and then calculated the
mean RT for each condition.
Older participants usually respond more slowly compared with younger participants and
prefer accuracy over speed (Salthouse, 1996). Upon observing such a pattern in the current
data, i.e. opposite correlations between age and overall RTs, and between age and accuracy, we
assessed our effects in terms of changes in speed-over-accuracy strategy use. To this end, we
standardized the accuracy and RT measures into z-scores, inverted these scores for the RTs to
obtain a speed measure (i.e. higher z-scores reflect faster responding) and calculated a speed-
accuracy-tradeoff (SAT) score ((z-speed - z-accuracy)/2), whereby a higher score indicates
faster, but more inaccurate responses.
Data were analyzed using SPSS (IBM SPSS Statistics for Windows, Version 21.0. Armonk, NY:
IBM Corp.). A Shapiro-Wilk test of normality of the 4 trial-types revealed violation of the
normal distribution for one trial type (p = 0.049) in terms of the SAT, and all trial-types for
the RTs and accuracy were not normally distributed (all p < 0.001). We therefore performed
non-parametric tests, i.e. the related-samples Wilcoxon signed rank tests to assess the effects
of Reward (high vs. low), Task switching (switch vs. repeat) and the interaction between
Reward and Task switching. We report the standardized test statistic as
W.
From the mean scores on the 4 trial types [Reward (high/low) x Task switch (switch/repeat
trials)] we calculated (1) the reward effect (high - low reward), (2) switch effect (switch -
repeat), and (3) the difference between the reward effect on repeat trials and the reward effect
on switch trials, i.e. the degree to which an increase in reward decreases the switch effect. To
break down this effect, we also reported the results for the reward effect on switch and repeat
trials separately. Because age did not follow a normal distribution (Shapiro-Wilk p < 0.001),
we used a non-parametric Spearman’s ρ correlation – r (ρ) - to assess the relationship between
these measures and age.
The sample in this study consists of pooled data from several studies using the same paradigm
(
table 5.1
). Although the paradigm and instructions were essentially the same, the amount
of reward participants could earn on a high-reward trial (i.e. 10 or 15 cent) or across all trials