Quantifying biases in test-negative studies of vaccine effectiveness