Putting two complete varieties in a family over the projective line Let $X$ and $Y$ be two proper varieties of dimension $n$ over a field $k$. I'm looking for "reasonable" conditions, under which, there exists a proper and dominant morphism $f:V\to \mathbb{P}^1_k$, with $V$ a variety of dimension $n+1$, such that $X=f^{-1}(0)$ and $Y=f^{-1}(\infty)$.
 A: I will assume that by "variety" you mean (at least) "irreducible", otherwise as Daniel Litt points out the question is trivial.
Further, let's assume that $V$ is projective (and hence necessarily $X$ and $Y$), as ulrich points out, it will already be difficult enough.
Consider a common polarization of $X$ and $Y$ given by an $f$-ample line bundle. The first necessary condition is that the corresponding Hilbert polynomials on $X$ and $Y$ are equal. This follows from $f$ being flat.
So, if $X$ and $Y$ have the same Hilbert polynomial, then they are fibers of some huge family, which however may be disconnected and $X$ and $Y$ would have to be in the same irreducible component to satisfy your desired condition. This is pretty hard to force other than having a family over an irreducible base that admits both $X$ and $Y$ as fibers. 
Then, if $X$ and $Y$ are in the same irreducible component of the Hilbert scheme, then it is yet another pretty difficult question whether their Hilbert points can be connected by rational curves. It turns out that this is a rare condition. As ulrich mentions, for instance for a general curve of high enough genus, there does not exist a non-trivial deformation over $\mathbb P^1$. In other words, already for $n=1$ and for most $X$, regardless of the choice of $Y$ there is no such deformation. 
This suggests that having such an $X$ and $Y$ is very special. 
On the other hand, and perhaps this is more what you would like to see, for most curves (smooth projective) of small genus this is actually doable.
The key words you should be looking for is "rational connectivity" of either moduli spaces or Hilbert schemes. 
The above claims for curves follow from the fact that moduli spaces (stacks, really) for small genus are (uni)rational and hence any two generic point can be connected by a rational curve, while moduli spaces of high genus curves have non-negative Kodaira dimensions and hence they are not uniruled (so there are no rational curves through their general points).
As ulrich mentions the story starts with Harris-Mumford and Eisenbud-Harris. For current knowledge you should look at the work of Gavril Farkas. You can start here http://arxiv.org/abs/0810.0702 for example, and just look at his papers on arXiv.
Another paper to read is Verra's survey: http://arxiv.org/abs/1112.6095.
And of course, this only covers $n=1$ in your question. I believe very little is known in higher dimensions, other than that this is a very special property, even just for $X$ to be in a non-locally trivial family over $\mathbb P^1$.
