>To rule out the spuriousness of the second kind, you need a longitudinal
>panel data for individuals to show what comes first, union membership or
>voting Democratic. Such data may not be available for individual elections,
>but they surely must exist for general party preference. This is not really
>my field so I do not know what data are available, but can somebody with
>expertise in that field shed more light on this subject?
A data geek could work with the National Election Studies <http://www.umich.edu/~nes/>. They're the best we've got, and they go back a long way.
Doug