Prompting Strategies for Enabling Large Language Models to Infer Causation from Correlation