"I think I understand how this works in circumstances where there are obvious negative real-world consequences, but I'm stuck on examples where the consequences are less obvious, or too obvious: 'When you hit me, I feel pain, because broken bones hurt.' 'When you call me stupid, I feel sad, because my self-esteem is damaged.' "
I should say first of all that the three-part message technique usually isn't suitable for kids. When a child is being bullied by another child, it probably won't work (other than because it might startle the bully into silence momentarily to hear such an utterance coming at him or her) and it's not likely to get a response much more articulate than "Yeah, stupid head!" or some such thing. When the hostile language aimed at a child is coming from an adult the child is always outranked, and a three-part message is likely to be interpreted as insolence or showing off, which makes it unsafe. There will be exceptions, certainly -- and I'd very much like to see adults within a household modeling the three-part message for the children and encouraging its use in that language environment -- but as a general principle the three-part message is intended for use by adults.
I also don't feel that the three-part message is useful when the abuse that's going on is physical, as in "When you hit me...." Hilleviw is correct that the physical abuser is already well aware that hitting causes pain. That needs no explanation, and a three-part message isn't likely to bring about any change in the attacker's behavior. Dealing with physical abuse is an entirely separate topic. But there are useful things to say about hilleviw's second example: "When you call me stupid, I feel sad, because my self-esteem is damaged."
The problem, as hilleviw notes, is with part three -- the statement of consequences. "My self-esteem is damaged" can't be verified in the real world, and is therefore something it's all too easy to argue about. When there really is no real-world consequence, the complaint isn't suitable for formulation as a three-part message and the interaction will have to be handled in some other way. Trying to construct the message and discovering that it can't be done is in itself useful information; when you've done that, you've learned something worth knowing. Often it means that you really don't know why the behavior bothers you, which may mean that you're not yet ready to make that complaint.
I was once called in as troubleshooter in a hospital where the hostile language that two doctors (both male) were using with the nurses (all of them female) was creating serious stress and conflict; the doctors had established a pattern of stopping by the nursing station and throwing a lot of verbal abuse at any nurse(s) who happened to be there. Administrators had tried counseling both of the doctors and had gotten nowhere; the nurses had tried to persuade the doctors to speak to them courteously and had gotten nowhere. When I suggested the three-part message technique to the nurses, all they could come up with was messages like "When you call me stupid, I feel angry, because your words are disrespectful and insulting." That statement was certainly true, and the nurses were entitled to say it if they liked, but it wasn't a well-formed three-part message -- the third part couldn't be verified and was subject to argument.
I had the nurses set up a very simple checklist, just one page of paper. For each day, the nurses made a check mark for two items: when one or the other of the doctors came by the nursing station and used abusive language; when a nurse made any sort of error, no matter how trivial, during a shift. They did that for six weeks, and then they counted up the totals and did an equally simple statistical analysis. The nurses were then able to construct the following three-part message: "When you throw hostile language at the nurses at the nursing station, they feel angry, because the average number of nursing errors during that shift increases by seven percent." And they had the data to prove that the third part of the message was verifiable.
This worked. The two doctors agreed to modify their behavior, and they followed through on the agreement. Of course this was not a scientific experiment; of course correlation does not prove causation. Of course the doctors should have been willing to be courteous for no reason other than that it is the right thing to do. But the goal of a three-part message isn't to raise the other person's consciousness, or educate that person, or establish scientific evidence, or anything of that kind. The goal is to request a change in behavior in a fashion that doesn't cause the other person to lose face, so that there's a reasonable chance that the other person will actually make that change. The nurses' message served that purpose, and it solved the problem. Two doctors who weren't willing to be courteous on the basis of a claim that their rudeness caused hurt feelings, was disrespectful, was morally unacceptable (and so on), were willing to be courteous on the basis of a claim that their rudeness was associated with an increased rate of nursing errors. They didn't turn into nicer people, they just made a requested change in their language behavior.
It didn't matter to the nurses that the doctors might be thinking of them as airheads who fell apart and made errors every time somebody snapped at them. It was already abundantly clear to them that the doctors had no respect for them as nurses or as people. Improving the doctors' character wasn't the nurses' obligation and they weren't interested in taking on that job. What mattered to the nurses -- their communication goal -- was that the doctors should change their behavior, thereby decreasing the stress level on that floor of the hospital.