E.G.

To illustrate the distinction between relationship tokens and relationship types, consider: Huck's friendship with Jim is a different thing than Bert's friendship with Ernie. These two things are distinct tokens of the same type.

Relationship tokens are always grounded in further factors. For example, if X and Y are friends, this might be so in virtue of interactions between X and Y (say, X and Y regularly go bowling together); and/or feelings, desires, or beliefs of each about the other (say, X and Y know each other, care about each other, and respect each other); and/or other factors. In general, whenever a relationship exists, there are factors in virtue of which it exists.

According to 

the reductionist view: any given relationship token R just is the collection of factors in which R is grounded,

whereas according to

the anti-reductionist view: any given relationship token R is distinct from the collection of factors in which R is grounded.

I think we should prefer the anti-reductionist view.

***

The factors in which a token relationship is grounded can be irregularly distributed across a time interval over which the relationship continuously exists.

Over the course of, say, a year, Bert and Ernie might meet once a week on Fridays for bowling and might not otherwise interact. And their feelings, beliefs, and desires about each other might change during the course of that year. But this would not mean that their friendship comes and goes during that year. They're not only friends on Fridays, or only when they're thinking of each other, etc. They're friends all the time.

This observation fits easily within the anti-reductionist perspective. The anti-reductionist can say: They're friends because they meet on Fridays, and because they feel and believe certain things at certain times about each other, and so on—but their friendship is a distinct thing, a social reality that exists "over and above" those factors that give rise to it, so it's no surprise that their friendship exists continuously even though these factors don't.

It's like how a solar energy doohickey might (using battery storage) be able to provide a constant voltage at all times even as the sunlight that powers the doohickey comes and goes at different times. There's no mystery in that, because the sunlight is distinct from the voltage.

***

The reductionist might respond to these points by arguing that, once we correctly identify the factors in virtue of which a relationship exists, we'll see that they actually are continuous across time. But there is no good argument along those lines, as far as I can see.

For example, the reductionist might say: Bert and Ernie's friendship is grounded not in their intermittent interactions or in their intermittent attitudes, but is instead wholly grounded in their dispositions: e.g., their disposition to meet on Fridays given such-and-such conditions, their disposition to feel warmly about one another if suitably prompted, etc. These dispositions might not be intermittent even if the behaviors that express these dispositions are intermittent. If so, then it could be said that both the relationship and the factors in which it is grounded are continuous.

But this won't work because relationships are not generally grounded in dispositions. To see this, consider that Bert and Ernie could have the relevant dispositions without those dispositions ever being activated. For example, Bert and Ernie might be disposed to meet for bowling on Fridays if the bowling alley is open on Fridays; but it's never open on Fridays; so they never meet.

In the case where all of the relevant dispositions exist but are never activated, Bert and Ernie would not in fact be friends. In general: Mere inactivated dispositions do not a relationship make.

***

There may be a worry that anti-reductionism about relationship tokens is somehow spooky. But I think this matter is complicated.

I think it's fair to say that relationships are paradigmatic social phenomena. The factors in which we tend to think relationships are grounded will sometimes also be paradigmatic social phenomena. For instance, patterned interactions (e.g., regular bowling on Fridays) might be like this. Sometimes these factors will also, or instead, be psychological phenomena—e.g., warm feelings about one's friends.

If the grounding factors are social, then it's not clear how reductionism is any less spooky than anti-reductionism. Both pictures will be pictures of a world containing social phenomena. So, if one thinks, for instance, that

(a) anything that isn't physical is objectionably spooky (physicalist anti-spookyism),

and

(b) social realities, if any truly exist, are non-physical,

and

therefore, social realities are objectionably spooky,

then it's not entirely clear that a shift from anti-reductionism to reductionism about relationship tokens will result in an overall reduction in spookiness.

The physicalist anti-spookyism expressed in (a) seems to require either relationship eliminativism—the view that, in truth, there aren't any such things as relationship tokens—or some kind of physicalism about relationship tokens—the view that relationships exist but turn out to be physical phenomena.

It seems to me that that latter view (physicalism about relationship tokens) can probably be made to be consistent with some sort of anti-reductionism about relationship tokens. Here's what such an anti-reductionist would need to say: Although relationships aren't reducible to the further social and psychological factors in which they are grounded, relationships are reducible to physical phenomena, and so too are the further social and psychological factors in which relationships are grounded. All of that seems coherent to me. So, I think there is at least one sort of spookiness worry about the anti-reductionist view that can be defused.