Results of impacts( ) for partitioned impact measures: An answer
After quite a few days of frustration I posed a question to the group about partitioned impact measures just some hours ago.
Apologies if that has grabbed anyone's attention already, but I have (finally!) located an answer.
Just a few words in case anyone else has a similar issue.
Based on things read (e.g., LeSage and Pace, 2009) and my basic understanding of what is going on, I had assumed that it was necessary for zeros to populate the direct W1 slot (immediate neighbours alone) and the indirect W0 slot (the local area) in partitioned impact measure estimates. My results did not match expectations, so my default position was that my results were flawed. Searching for similar output I found examples that both agreed and disagreed with that assumption.
I have just now located a reference that sheds some light on this:
C.D. Jensen and D.J. Lacombe, 2012. A note on partitioning effects estimates over space. Letters in Spatial and Resource Sciences, 5(1), 47-53.
On just a scan it is clear that it is not necessary that these slots contain zero entries (see page 52 in particular). This is associated with large spatially lagged variables in the Spatial Durbin model I had estimated. The paper is framed as a "correction", so I guess I was probably not alone in leaping to the conclusions I had.
Hope that helps. It has helped me. Apologies if anyone has jumped onto this already.