Before answering your ultimate question, there are a couple of other points in your statement worth looking at.
It shows harsh boundaries where one CT has great access to transit and
right beside that CT is one that has extremely poor access
In your comment, you elaborated that these vector boundaries are Census Tracts. What I am taking from your statement is that you have symbolized your data in such a way that some census tracts have easy access to transit while adjacent ones do not. This creates what look like hard boundaries in your data. The boundaries, however, are Not determined by the
Access to Transit values, they are determined by the
Census Tract extents.
In short, smoothing them is simply going to cause a loss of resolution in your data, but not gain you anything other than it will look better. The only way in which this could work is if the Transit Access scores were aggregated to Census Tracts from a finer grained resolution. In that case, you could use the source data to create a vector layer that represents a finer classification of your data. This would create smaller polygons that may well cross the boundaries of the Census Tracts.
In addition, if you are planning on using these for future correlation analysis it is likely that other data is going to be aggregated at the Census Tract level as well. If you smooth your data, you will no longer be able to directly compare datasets because your boundaries will now be different.
It does not seem logical to be able to maintain your score value for future analysis. If you resample your data and smooth the values, then you are by definition, changing the scores. If you then convert that raster back to the same vector boundaries, the score is going to be different, based on the new values from the raster that you are aggregating.
My final thought would be there are two factors to take into account in making your decision.
- Does your underlying data support making a finer-grained dataset.
- For future analysis, if the other data you will be using for correlation is all going to be at the Census Tract level, then will you gain anything by moving this data away from a Census Tract aggregation?