Some journals (specifically on biology or chemistry) accept only MS Word files as manuscripts, while many journals on mathematics or physics prefer LaTeX. Even Science prefers MS Word.
Why do they insist on proprietary MS Word, instead of accepting PDF + plain text? I don't think they typeset papers on MS Word.
I cannot answer for all journals but can provide insights into a few. One issue concerns the authors rather than the journals themselves. If a community has little need for equations then the need for LaTeX is also perceived as small. Hence a journal would not implement something that they think has no use.
In my own field of Earth Science, the usage of LaTeX varies between sub-disciplines. In Glaciology, being a quantitative subfield, a majority use LaTeX while in other more descriptive subfields no-one uses LaTeX. It is easy to see that this is also reflected in what formats the journals accept. Since the editors of the journals usually come from the sub-disciplines their journals cover, the editors are used to using the same authoring tools as others which further cements the existing structure.
On top of that the individual publisher may add capacity to journals. Elsevier, for example, has a LaTeX-class that can be used for their journals, while Wiley, for example, do not. This can help journals take the step to also include LaTeX contributions even though editors may not be users themselves.
The main divide that I notice in preferred input format is between journals that routinely reformat your paper and those that do not.
Journals wanting Word generally ask the author just just send a pile of text and figures, which they will stitch together to look precisely how they want it to look, with special fancy styling, etc.
Journals that want LaTeX generally ask the author to send a PDF file looking almost exactly as it will for publication, plus the LaTeX so they can recompile with the right page and issue information.
The first is essentially a leftover from the days before computer formatting, when a journal would be getting a bundle of paper, which would need to be typeset by hand in any case. These days it is most likely to be preserved in either high-end journals that can afford a significant paid staff or else in subfields that simply haven't made the cultural transition. Word is then a "lowest common denominator" that, unfortunately, the world has generally settled on for styled documents (though some places will also accept non-proprietary formats: for example, PLoS ONE also accepts rtf)
LaTeX lets a journal run much more leanly, since it places more burden on the authors. When something goes wrong with LaTeX, however, it's likely to go much more problematically wrong because it's possible for authors to include some awfully fancy programming in LaTeX (I know a person who wrote a piece of database management software entirely in TeX). To handle that, you need a much more programming-savvy journal staff, or else a large and well-maintained automated backend like IEEE's PDF eXpress to help you detect and manage the problems that come with freedom.
It is much easier to find copy-editors with Word skills than Latex skills. I am aware of two journals that used to accept submissions written in Latex that switched to Word-only because they could not find editors with the skills needed for Latex, and of another that is considering this switch.
There is a big problem with Latex as an authoring format, which is that it is a sophisticated and not terribly readable programming language, one that encourages people to hack up idiosyncratic macros for use in their articles. If you have 10 papers to be edited into an issue of a journal, life is much harder if each of the papers defines its own macros to do roughly similar things in incompatible ways, with possible and hard to deal with namespace collisions. This also means that the skills needed for editing in Latex are higher than those needed for editing articles written in Word.
That said, journals that can handle this problem gain freedom from vendors in their publishing operations. It is no accident that Latex is more popular with smaller publishers than with the giant publishers.
Similar to jakebeal, Peter Jansson, and this https://forums.adobe.com/thread/1046940 website. I would propose the reason for format is usually based on the end goal, which determines why someone prefers a format. Some journal or conference that uses professional graphic designers or wants the comfort/freedom of an easily formatted manuscript may want to use inDesign (one of the standards for making book layouts).
Even if inDesign is not the end point, it could just be easier to layout a microsoft template, or could just be easier because the typesetter for a journal is accustom to those options.
