Evaluation of Reasoning and Thinking Abilities of Large Language Models in Economic Problems (Appendix)

Experiment 1.1 consists of the Intuitive Alignment group and the Intuitive Contradiction group (200 questions each), testing whether the models rely on real-world knowledge bias. Experiment 1.2 is a control experiment with varied narrative order (200 questions each), examining the models’ sensitivity to text arrangement order.

The tested models cover mainstream series such as Llama, Gemma, and Mistral.