I'm sure people in the field know this, but I'm not in the field. Under what conditions (be they on the manifold or the map) is a diffeomorphism from a differentiable manifold $M$ to itself homotopic to the identity? Is this "usually" the case (by some definition of "usually")? And if it is homotopic to the identity, can we choose it to be a homotopy of diffeomorphisms?

Your latter question the answer is generally no. A diffeomorphism $f : S^n \to S^n$ has homotopy class given by its degree $\pm 1$. But the homotopythroughdiffeomorphisms (usually called isotopy) classes are the group of exotic smooth structures on $S^{n+1}$ provided $n \geq 5$. There are large classes of manifolds for which the homotopyclasses of diffeomorphisms are reasonable. Hyperbolic $n$manifolds for $n\geq 3$ have the property that homotopyequivalences are homotopic to isometries. This is "Mostow rigidity". So homotopyclasses of diffeomorphisms are the same things as homotopyclasses of homotopyequivalences in this case, which is $Out(\pi_1 M)$, since hyperbolic manifolds are $K(\pi,1)$spaces. If you generate 3manifolds via Heegaard splittings there is a sense in which most 3manifolds are hyperbolic, so the above gives you an answer in one instance of your question. But in general there's not much known about the forgetful map $$\pi_0 Diff(M) \to \pi_0 HomEq(M)$$ Perhaps the largest obstruction to understanding this map is that we know so little about $\pi_0 Diff(M)$. In high dimensions surgery theory gives you some tools. 


In differential topology, there is the notion of "isotopy". A isotopy of M is a selfdiffeomorphism which is linked to the identity by a 1parameter family of selfdiffeomorphisms. The set of isotopies is denoted Diff0(M), since it is the neutral connected component in the topological group Diff(M). The quotient group Diff(M)/Diff0(M) is called the Mapping class group (MPG) and has been much studied for surfaces. MPG(any surface) is generated by the Dehn twists. For the 2torus, MPG(T2)=SL2(Z). Indeed any matrix A in SL(2,Z) is a diffeomorphism of R^2 preserving Z^2, and thus a diffeomorphism of T^2=R^2/Z^2. The reason why it is not isotopic to the identity for A \neq id is that it is not even homotopic to the identity, since its action on \pi_1(T^2)=Z^2 is also A itself! MPG(the disk minus k points) is the braid group on k braids. For surfaces it is true that a diffeomorphism which is homotopic to id is an isotopy (see Gabai and many others); but in the genaral case it needs not be. In general, a selfdiffeomorphism f of M has to act by the identity on the homotopy groups and homology groups of M to be homotopic to the identity, and you will easily make many counterexamples. 

