Attention will provide a more accurate view of a customers engagement time on page. For the average webpage or piece of content, about 48 seconds of attention is a good benchmark.
Time on page measures the difference between when opens a webpage and then clicks another link. Or opens a new tab and visits the same domain. A common fallacy is that time on page represents all users, when it only captures those that complete those actions. So it is not reflective of an overall audience. It tends to skew up.
The core metric of time on page is flawed in collection which can make it hard to get consistent data. So it is better to use Attention.