Lately also JSON-LD came into play, but we don't really use it (guiltily?), so here we prefer to concentrate on the other two languages. 
Let's check advantages and disadvantages of both:

Microdata
Plus:
- it's very easy
- it is supported by Google, Yahoo and Bing in order to comply with schema.org
Minus:
- it is rudimentary and not very expressive, difficoult to implement

RDFa
Plus:
- it's supported by W3C, the highest authority for semantic web
- being built on the foundations of a complex language such as RDF, is potentially more expressive and implementable
- it allows you to connect web platform pages with a possible underlying ontology
Minus:
- it's a bit more complicated than Microdata
- it isn't explicitly supported by Google, Yahoo e Bing.

So what? It is not easy to take a stand, and yet, bearing in mind the advantages and disadvantages, we from Pensativa do it this way: for our personal platforms (those that we manage directly), we use RDFa which is more expressive, flexible, and allows us to relate the site with ontologies, while for the customers' platforms, if there are any underlying ontologies we use RDFa for the same reasons, otherwise we prefer to use Microdata which is surely read by the most important search engines.