I understand the largest issue with PDPs is that it's assumed that the feature/s of the plot that are calculated are not correlated with other features.
Let's use something like height and weight to predict how fast a person walks. If i calculate the PD at height = 200cm, we'd average over the marginal distribution of weight. This means we'd include weights of 50kg - which is unrealistic/unlikely.
What would happen to a PDP if this assumption was violated? Would the PD be understated because of this?