On Evaluating the Integration of Reasoning and Action in LLM Agents with Database Question Answering