Strain and size of the grain/particles of crystal will influence the Bragg peak width. However, the most common effect is actually the instrument parameters (Slit widths, beam width, sample, etc).
In general everything which violates the major requirement of an infinitely extended crystals and the exact periodicity of the crystal lattice. Some typical applications have been already mentioned by G. O. Lloyd. Presently I have in mind chemical segregation, topostratic effects (loss of periodicity in only one direction), or phase transformations (e.g. ZrO2). The reason however is always the same: the loss in infinite periodicity in 3, 4 or any dimension. Moreover, all factors which affects the measurement procedure, above described by some typical parameters like dimension of the source, collimation, divergence, defocusing, misalignment of the sample up to the sample "preparation".
Good basic question, but the answer has many layers of convolution depending on the geometry, optics and detector used.
In summary,
The first obvious contributor as mentioned is the incident beam shape via "instrument parameters (Slit widths, beam width, sample, etc)". Point or line source.
Crystallographic "strain state" and defect type/density/distribution.
In the typical case of standard conventional diffractograms from powder samples I must assume you folks mostly are using a conventional 0D point counter where all the XRD signal coming into the exposed detector face is counted as one relative intensity value. This excludes you from all the wonderful details of the sample material's Nano structure ubiquitously present in real time XRD signal.
Once you move past the veil of conventionality you will discover the possibility of deconvoluting the individual parameters mentioned earlier from the 3D XRD data ever present. With a 2D detector it is possible to "rock" the "powder" sample and determine coherent diffracting domain size and deconvolute them with relative ease in reciprocal space (Omega space). Here is an example illustrating this for a LaAlO-NdNiO sample, mono crystalline with large "sub grain" miss-orientations.
The task does get harder with Nano powders. But the relative advantage of 2D and speed supersedes any perceived advantage of higher dynamic range with 0D"spatially blind" detectors. Besides the 2D detector data can always be "dumbed down" to a 0D analysis any way, as many do even now!
Probably, the easiest way to explain this is by invoking the geometrical representation of Bragg's law i.e. when a reciprocal lattice vector lies on the Ewald sphere, diffraction occurs according to Bragg's law. If you leave aside the various instrumental contributions to peak broadening and other systematic factors, ideally, the peak shape should have been like a a Dirac delta function. But in REAL crystals, the reciprocal lattice points lying infinitesimally close to the Ewald sphere also contribute significantly to the diffracted intensity. Hence you observe the broadening.
Another way to explain this is according to the Heisenberg's uncertainty principle. When a detector measures the scattered intensity, a photon of particular energy is being detected and this is a purely quantum mechanical process, which MUST not violate the uncertainty principle. In this case, the uncertainty is in simultaneous precise measurement of energy (E) and time (t). Thus there must be a definite spread in of the energy of X-ray photon, unless the measurement is carried for infinite 't'.