In ``Steel, John R. On Vaught's conjecture. Cabal Seminar 76–77, pp. 193–208'' the following is proved:
Theorem. Let $\phi\in L_{\omega_1,\omega}.$ If every model of $\phi$ is a tree, then $\phi$ has either $\leq \aleph_0$ models or perfectly many countable models.
In page 206 of the paper the following is stated:
It would be natural to attempt to extend the above theorem from tree to arbitrary partial orders. However Arnold Miller has shown Vaught's conjecture for partial orders is equivalent to the full conjecture.
Question 1. How Miller's result can be proved?
Added remarks and new questions:
According to Enayat's answer, even the Vaught's conjecture for lattices, implies the full Vaught's conjecture. So it is natural to ask how much we can weaken our assumptions, for example
Question 2. Does Vaught's conjecture for Boolean algebras imply the full Vaugh's conjecture?
In general:
Question 3. For which structures it is known that the full Vaught's conjecture is implied by the Vaught's conjecture for those structures?