Cognitive Load

The results of Experiment 1 suggest two interacting decision rules codeter- mine behavior in a complex, dynamic setting such as the Cournot Oligopoly. We found the response times asymmetry as predicted by the formal drift- diffusion model and conclude that imitation, in a dual-process view, is a more automatic, intuitive behavioral rule and myopic best reply a more controlled, deliberative rule. Dual-process theories distinguish automatic and controlled processes along the dimension of consuming cognitive resources. Automatic processes require only few cognitive resources while controlled processes are consciously reflected upon and consume relatively more cognitive resources. In this second experiment we want to further investigate the different nature of imitation and myopic best reply. We, therefore, set out to manipulate the cognitive resources available by implementing cognitive load and analyze changes in behavior.

Manipulating cognitive resources is based on the idea that if the working memory is exhausted, e.g. by an additional cognitive load task, the perfor- mance decreases. Working memory is a brain system which provides tem- porary storage and processing of information. Baddeley (1992) proposes a refined model that divides working memory into multiple components. The most prominent system is the central executive which is the attentional- controlling system that, among other features, is responsible for coordinat- ing information. If the central executive is taxed and working memory is

exhausted then processes which require cognitive resources cannot be ex- erted.

Exhausting the working memory is implemented by adding a secondary task surrounding the main decision task. Cognitive load tasks may consist of memorizing a number sequence (targeting the phonological loop) or visual pattern (targeting the visuo-spatial sketchpad) which has to be recalled after the main decision has been made. These tasks are frequently used in psy- chology and recently found its way into economics (Carpenter et al., 2013). Carpenter et al. (2013) manipulated the cognitive resources by asking their subjects to memorize a seven-digit number in the cognitive load treatment. Subjects in the load treatment subsequently performed worse in a number strategic games.

With respect to imitation and myopic best reply, the dual-process view leads to the following predictions. Taxing the central executive with cognitive load results in a shift to more automatic behavior and we will observe a higher frequency of imitation decisions in the cognitive load treatment.

H3. Under high cognitive load, more imitative decisions will be observed. In addition, the NoLoad treatment includes a replication of FullInfo treat- ment of Experiment 1 and the Load treatment itself includes a robustness check of the theoretical predictions tested in Experiment 1.

4.4.2 Experimental Design and Procedures

The cognitive load manipulation was implemented between subjects in two treatments, NoLoad and Load. The NoLoad treatment was a pure replica- tion of the FullInfo treatment of Experiment 1. The Load treatment had an additional cognitive load task, but was in any other aspect just as the NoLoad treatment (and FullInfo treatment of Experiment 1). Compared to Experiment 1, we increased the exchange rate to 20 Eurocents per 1000 points10

and the subjects were rematched within blocks of 12 participants after each part. The participants were rematched within a larger pool which

10

The exchange rate was increased because the average payoff in Experiment 1 was slightly below the wage rate demanded by experimental lab.

further decreased the probability of collusion.11

Besides those points, the most prominent change being the cognitive load task in the Load treatment, there were no other differences compared to the FullInfo treatment of Exper- iment 1 and the participants played 51 periods in three different tetrapolies, with 17 periods each.

The cognitive load task consisted of memorizing a seven-digit number (as in Carpenter et al., 2013) before each decision in the Cournot Oligopoly. The subjects had 10 s to memorize the number and had to recall the number after the decision in the Cournot Oligopoly. A correct recall of the number was rewarded with additional 750 points. We implemented the treatments in separate sessions each containing one treatment because of the additional task in the Load treatment.

The experiment was conducted at the Cologne Laboratory for Economic Research and programmed with z-Tree. We recruited 144 participants (57 females; median age 23 years) using ORSEE. During recruitment, we ex- cluded students majoring in economics, psychology, and business and stu- dents who had already participated in 20 or more experiments. Six ses- sions in short succession comprised each 24 participants. The sessions in the NoLoad treatment lasted around 1 hour and 25 minutes while the sessions in the Load treatment took longer and lasted around 1 hour and 45 minutes. Average earnings, including the show-up fee of 2.50 EUR, were 13.61 EUR and 20.12 EUR for the NoLoad and Load treatment, respectively. The par- ticipants in the Load treatment earned more than the participants in the NoLoad treatment due to the additional earnings of the cognitive load task. After excluding the earnings of the cognitive load task, the average earnings in the Load treatment was 14.06 EUR, including the show-up fee.12

11

We are aware that this also reduces the number of the most conservative independent block observations but our focus lies on the subject analysis based on the micro-foundations of the DPDM (Alós-Ferrer, 2018).

12

There were no significant differences in the earnings from the decision task between the two treatments (MWW test, N = 144,z = −1.489, p = 0.1365).