Does Visual Information Play a Decisive Role in Vision-Language-Action Model Driving Behavior?