I give an argument using the correct definition of isotriviality (relating it to the weak definition), and the reader who is familiar with Deligne-Mumford stacks can see that the method is actually applicable quite generally to separated Deligne-Mumford stacks locally of finite presentation over an algebraically closed field (these hypotheses ensure the existence of a coarse moduli space; perhaps it can all be done more generally).

Let $k$ be an algebraically closed field and $E \rightarrow S$ an elliptic curve over a *reduced* $k$-scheme $S$ such that $j:S \rightarrow \mathbb{A}^1_k$ is a constant map to some $j_0 \in k$. Let $E_0$ be an elliptic curve over $k$ with $j$-invariant $j_0$. We claim there is an etale cover $S' \rightarrow S$ such that $E_{S'} \simeq (E_0)_{S'} := (E_0) \times_{{\rm{Spec}}(k)} S'$. In other words, the weak notion of isotriviality implies the right notion when $S$ is reduced. This will proceed by "standard" EGA-style reduction steps (and the proof will show by conceptual reasons that the reducedness hypothesis on $S$ cannot be dropped when $j_0 = 0, 1728$).

Consider the functor on $S$-schemes ${\rm{Isom}}(E,(E_0)_S)$ whose $T$-points are ${\rm{Isom}}_T(E_T, (E_0)_T)$ for an $S$-scheme $T$. This is represented by a finite type unramified $S$-scheme, and we just need to check that it is etale surjective over $S$ (so then we can take it to be $S'$, over which there is a "universal isomorphism"). We can assume $S$ is affine, then by limit arguments we can assume it is finite type over $k$, and then local (and essentially finite type over $k$) with residue field $k$ provided we prove the etale property for the Isom-scheme over the closed point. We can pass to the completion, so now the problem is rather concrete: if $R$ is a *reduced* complete local noetherian $k$-algebra with residue field $k$ and if $E$ is an elliptic curve over $R$ with special fiber $E_0$ such that $j(E) \in R$ is equal to $j_0 := j(E_0)$ then we claim that $E \simeq (E_0)_R$ (so the Isom-scheme of interest is the base change to $R$ of the Aut-scheme of $E_0$, so it is etale over $R$ as desired).

Let $A$ denote the universal deformation ring of $E_0$, so $E$ corresponds to a local $k$-algebra map $A \rightarrow R$ which we want to show kills the maximal ideal of $A$. Since $R$ is reduced, it suffices to show that some power of the maximal ideal of $A$ dies in $R$. Let $G_0 = {\rm{Aut}}_k(E_0)$ be the finite automorphism group of $E_0$, so the completed local ring $O$ at $j_0$ on the coarse moduli space $\mathbb{A}^1_k$ is the ring $A^{G_0}$. (I am omitting the rigorous proof of the "well-known" fact that $\mathbb{A}^1_k$ is the coarse moduli space for the moduli stack of elliptic curves in a *good* sense which includes this description of the completed local ring at geometric points. If such a proof is unclear then someone can always post an MO query about it. This link with deformation rings is the heart of the argument, as will soon become clear.) The map $A^{G_0} \rightarrow A$ is *finite* and local (since $G_0$ is a finite group), so the maximal ideal of $A^{G_0}$ generates a max-primary ideal of $A$. By hypothesis $A^{G_0} \rightarrow R$ kills the maximal ideal, whence $A \rightarrow R$ kills a power of the maximal ideal. We have seen that this suffices for our needs (due to the reducedness of $R$). QED

definingisotriviality (in case your problem is caused by using a definition which is too restrictive). – Boyarsky Jul 1 '10 at 19:22geometrically connectedetale double cover $X' \rightarrow X$, the corresponding quadratic twist $\mathcal{E}$ of a constant elliptic curve $E \times X$ is a new elliptic curve over $X$ which is not isotrivial in the sense of the OP, but it is isotrivial in theright sense: becomes constant over anetale coverof the base (namely $X'$!). – Boyarsky Jul 1 '10 at 20:21