Haven’t you been wondering that lately? Tridimas has a paper (link here scroll down) on that question, here is one empirical observation:
During the period 1957-2006, out of a total of 43 integration referenda, 23 were not constitutionally required but were called at the discretion of the incumbent government; 18 of these 23 resulted in a pro-integration vote as the incumbent government had sought. France has held three EU-related non-required referenda, all initiated by the President of the Republic. The UK approved EEC membership in 1975 in a non-required referendum, the only national UK referendum thus far. In 2003, seven of the nine referenda held by the new entrants to approve membership were not required. Again, none of the four referenda held in 2005 by Spain, France, the Netherlands, and Luxembourg to ratify the EU Constitutional Treaty were constitutionally mandated.
And the theory?:
In politics, some issues cause deep intra-party splits between the elected representatives of the same party rather than inter-party divisions among different parties. Constitutional issues, which concern questions of governance and national sovereignty of a state, are a prime example. It is then unlikely that the standard system of parliamentary politics will be able to resolve all those issues. On the contrary, it is more likely that the leader of the party in office will call a referendum to decide them. Ratifying changes to constitutional arrangements in a referendum confers legitimacy to their adoption (or rejection) by taking the decision away from parochial parliamentary majorities and putting it into the hands of the citizenry.
Ah, but there is risk! Still, it is not crazy to call the referendum, even though you might rationally prefer that the whole issue disappear.
I would add a further point to that model. Let’s say you think the core issue won’t go away of its own accord, arguably the case with both Brexit and the failed Colombian peace agreement. The choice is not “referendum vs. no referendum,” but rather “referendum today vs. giving my successors an option on future referenda.” And since a referendum can strengthen an incumbent, you realize that some future government might call one for self-interested reasons. In which case you might consider risking disaster now, responding to a kind of collective action problem through time. You may even fear that one of your successors will be irrational. And so some moment will feel like an optimal trigger point, though of course that will involve risk and probably more risk than is socially optimal. But is there not a preferred time to make your leap from the burning building? The pain of landing on your arm rather than your butt is not per se an argument for nixing the choice altogether.
Alternatively, consider the Italian referendum on reforming the Senate, due in December. The incumbent government may well lose the vote, but precisely because this is not a fundamental issue I predict they can simply continue in power, noting they didn’t have a strong mandate for change in the first place.