Why doesn't LayerNorm have a normalization dimension parameter?

How does LayerNorm determine the dimension(s) for normalization using only x, scale, and bias?
If normalization is to be performed along the nth dimension, how should scale and bias be set?