*5.7. Robustness Check*

During the guessing games, subjects needed to memorize a string of three or seven letters and recall the letters after they finished the guessing game. In this subsection, I present the results of this memorization task. Although the subjects were fully aware that if they failed to recall all the letters correctly, they would earn zero points for that round of the game, there were still some cases of wrong recalls due to reasons such as lack of attention or being too focused on the guessing game. I wanted to control the experimental results for such cases, as the subjects may have engaged in reasoning at higher levels when cognitive load did not fully apply. Table 14 shows the results of the memorization tasks. Most of the memorization tasks were perfectly performed. Not surprisingly, low cognitive load (three-letter memorization task) had more correct recalls, about 7% more than the high cognitive load task. The difference was significant at the 1% level.


**Table 14.** Results of the memorization task.

∗ indicates < 10% significance, ∗∗ indicates < 5% significance, and ∗∗∗ indicates < 1% significance.

To check whether poor performance of the memorization task affected the treatment results, I excluded the data with wrong recalls and performed the analysis again. The comparison pair was dropped from the sample if either game of the pair had incorrect recalls. This was performed to ensure that the cognitive load was fully in effect, so that high cognitive load added difficulties to thinking through the guessing games at higher levels, and the cost of reasoning was higher.

Table 15 presents the treatment results after the robustness check. Treatments that involved high cognitive load had more data points dropped. For example, the [HH-] to [HH+] comparison had 444 pairs of comparison in the original sample. After robustness check, about 100 pairs were dropped. However, the results did not change much compared to the results presented in results 1 to 3 (Sections 5.3–5.5). The changes were mostly within 1%. I can therefore safely conclude that the original results were robust. The quality of the memorization task (i.e., whether the letters were correctly recalled) was almost independent of the treatment effects. Even in the cases of wrong recalls, the effect of cognitive load still applied to the subjects.


**Table 15.** Summary of the robust results with incorrect recalls dropped.
