I think Shepherd's positive view of causation can be captured by thinking of it as an account, for any given thing, of the structural or constitutive causes of its changing the way it does or being the way it is. It is, of course, silly to assume (although sometimes it does seem to be assumed) that everything we call 'causation' will necessarily count as causation in the same way; even if Shepherd's view doesn't cover all causation (and I don't think it does), it nonetheless captures an important region of our causal reasoning to a high degree of fidelity. Here's a way to see the big picture of what she is doing.
1. Let's suppose there is a great big glob of bread dough on the counter. This glob of bread dough has what Shepherd calls 'qualities', but which I will call 'features' (because 'qualities' is potentially confusing). The whole set of features, and their relations to each other, in some sense is the bread dough.
2. Now suppose you punch your fist into the dough and leave an impression. Your fist has features, too; and the whole set of the features of your fist, given how they are related to each other, is, in some sense, your fist.
3. Now suppose you want a causal explanation of why the bread dough has the impression it does. There are a number of things you could be wanting in this case, but one thing that might satisfy you is an account of all the relevant features of the dough, all the relevant features of the fist, and how these features shift and change relative to each other. So, in this case, facts about the nature and consistency of dough (to whatever degree of specificity you want), facts about the nature, shape, and firmness of your hand (to whatever degree of specificity you want), and facts about how these features interact as your fist is plunging into the dough to result in the impression (to whatever degree of specificity you want).
I think it is clear that this is, in fact, the sort of thing we would want to know, at least in some cases, if we asked for a causal explanation of the impression in the bread dough. This is a case of change; let's take a static case.
1. Let's suppose you have a cup on a table. The cup has features (and the whole collection of those features is the cup); the table has features (and the whole collection of those features is the table).
2. Let's suppose you want an explanation for why the system is the way it is (e.g., why the cup is on the table rather than falling through it). Again, there are a number of things that you could be wanting here, but one of the things could be this: an account of the features of the table, and the features of the cup, and how they are related to each other, such that the cup has to be the way it is (on the table rather than falling through it).
NB: In this case we can either be taking the system diachronically (why the cup-table system is the way it is through time) or synchronically (why the cup-table system is the way it is at a given time).
Again, I think it is clear that this is the sort of thing we sometimes really are asking for when we ask for a causal explanation. The dough case is about how all the features involved cause the change to be the way it is; the cup case is about how all the features involved cause the system to be the way it is. Hence we are identifying the structural or constitutive cause of the system through time or at a time.
And note that Shepherd is exactly right that this allows us to regard the form of reasoning in these cases as analytic and necessary. By 'form of reasoning' I wish to highlight that not all causal explanations of this kind are necessary, but that the general logical form of the reasoning can be put in necessary terms. In a sense what we have is an equation:
(LMS) S = (a #1 b #2 c ...)
Where S is the system changing or remaining the same, and a, b, c, &c. are the features of the system, while #'s are the various interrelations of the features. Both sides of the equation are necessarily the same, but our knowledge of the features and interrelations on the right-hand side can be more or less extensive. For instance, we can tell the system is changing (e.g., if the cup is falling through the table), but there might be features of the system we have not yet identified.
This sort of causal reasoning, therefore, works just like the application of mathematics or the balancing of chemical equations (Shepherd uses analogies to both mathematics and chemistry). Given that (LMS) is necessarily true, we can conclude, from any change that there must be a (constitutive) cause of the change. And this is also necessarily true. If we change one side of the equation, we must necessarily change the other side to preserve the equality. If we know the system is changing (through sense-experience, for instance), we can (and do) immediately conclude that there is something changing in the features of the system. (LMS) does not tell us what all those features are; it is a description of the form of constitutive-causal reasoning. The precise details of what the system is, and what the features are, have to be filled in, e.g., by scientific investigation. This is why (LMS) does a great job capturing much of what is going on in certain types of laws involving 'ceteris paribus' clauses. Given (LMS) we can also divide the right-hand side of the equation into (known features & interrelations) + (unknown features & interrelations). And this allows us to have add all sorts of various qualifications to our causal inferences.