StatLLM: A Dataset for Evaluating the Performance of Large Language Models in Statistical Analysis