Empirical question: How to create proper bounding boxes

I’m following along with the fast.ai course and am hoping to train a model that detects tabular data. I want to clarify bounding box best practices. Would differences in bounding box tightness in the following two images affect the model’s mAP?

Somewhat tight bounding boxes:

Tight bounding boxes:

Not sure whether there’s a right answer here. All else equal, (tighter boxes = more time labeling images = smaller dataset) while (less tight boxes = less time labeling images = larger dataset). Does anyone have a sense of the tradeoff?