When collaborations go awry

Summary of the post (if you're in a hurry): pick your collaborators wisely. But if you don't have a choice, cross your fingers that they'll play nice.
Now I shouldn't complain too much - the collaboration that's gone awry hasn't catastrophically failed. But its frustrations have made me think a lot about making sure collaborations succeed. In this case, I had the offer of more, complimentary data to include in a manuscript for one of my thesis chapters. This collaborator is an expert in my small obscure field, encouraged and energized me through discussions and critiques, and urged me to publish as soon as possible. But, they are also retired, and I now know, on a different timescale than me. They assured me that their old data set perfectly supported my new findings, and we set about altering my thesis chapter to accommodate a new dataset. This was one year ago. Many months and miscommunications, I was ready to start getting tough when the collaborator informed that in fact, the data didn't support my findings and may in fact contradict them. After a morning of ARGH'ing to myself, I suggested splitting the paper into two and waiting on getting the old dataset published until we could adequately analyze the data. My collaborator agreed to consider that and promised me their notes on the new findings the next day. That was a week ago.
  This is why, if I could do it again, I would do things differently. It's early in my career, and I need those publications now, even at the expense of being linked with an expert in the literature. I would be firm about my deadlines (to be fair, something I would have to work on as well), not as polite in my emails asking about manuscript status, and in this case have advocated for publishing the ready-for-primetime paper right away. But it was because I'm an academic n00b that prevented me from taking a stand on the paper's status. Now I need to work on taking back 'ownership' of the paper to get it out the door. Here's hoping it doesn't take another year.