How important is to get good data?
Will model training be affected by data quality?
What are some decent data collection practices?

Getting good data is crucial!
It sets the stage for everything your model will do. If the quality of the data is bad, the performance of the model will suffer as well.

Good data is like the ingredients of a recipe. If your ingredients are fresh and good quality, your dish is probably going to taste great. But if they are stale or lacking, the result will be poor.

Couple of the practices I would suggest:

1. Brainstorm first about the properties and parameters for your data-table.
2. Prioritize getting a variety of accurate data.