During the early stages of scale development, many items within our item pool were found to have a strong ceiling effect with low score variability. In order to attempt create a meaningful measure with useful score variability, the decision was taken to reduce the size of the item pool and only include items with the most central mean scores and the largest standard deviations. Only these items were put forward into subsequent exploratory and confirmatory factor analysis. Two questions. One - does this seem sensible? Two - can you suggest examples from the literature where the same has been done or is recommended? With thanks.
This makes good sense. You create a scale to distinguish between different individuals. Some are low on some traits and high on others. If you have items that aren't discriminating between people (i.e., individuals tend to respond the same and thus the item has a low SD), then the item isn't useful in your scale.
A good reference for this would be any psychometric text book. In particular, in item response theory (IRT), items that have a low discrimination are not as desirable. This likely isn't the type of reference you're looking for, but it does support what you are doing.