It seems more likely that it's actually the other way. Nate Silver has specifically said he built the model to have relatively high uncertainty, especially with the volatility of this year, so this seems more like the outcome of intentional decisions to not let the model be overly confident.
I think it's an error in the model structure. If their goal was to artificially increase uncertainty, there are more reasonable ways to do that than adding weird between-state correlations (like the uncertainty index which is part of the model). WA and MS definitely should not have a large negative correlation.
