It depends which kind of data you want to save and whether you will use it only internally or you have to exchange the data with an external service.
NSCoding is generally speaking a data serializer. A lot of built-in objects implements the NSCoder protocol that allows you to save them as a binary stream (file, in a BLOB of an sqlite, etc.) The NSKeyedArchiver gives you the plus of searching in such streams based on a string label, a bit like a dictionary but you can use only strings as keys. This approach is good if you occasionally have to persist some objects of different classes.
However if you have many objects of the same class, you'll better go for a database-approach, SQLite or CoreData. CoreData is practically a wrapper around SQLite that eases a lot designing your data model, and does the queries to the DB behind the curtains, without the need of writing SQL statements. In CoreData you define your classes, and each instance of the class can be persisted i.e. you can get back the values of the members of the object without having them always in the memory. This is a very convenient way to store a lot of structured data. For example if you'd write a web browser, you could store the bookmarks of the user with the name, URL, and maybe last visited time.
For XML and JSON there're no particular advantage if you use the data only locally to the device. If you have to communicate with some external service, you might consider to cache/save the XML / JSON objects as they are for later use. Other approach would be to regenerate this data from your internal data structures (see above) every time you need it.
If you design your data model yourself, I see even less point to use plists, but maybe somebody will correct me.
EDIT: I add here a short link reference for tutorials on how to use NSCoding, Core Data, and as a bonus, SQLite.
UPDATE 12.01.2016: If you are looking for persistence solutions I suggest you to also check out Realm.
与恶龙缠斗过久,自身亦成为恶龙；凝视深渊过久,深渊将回以凝视…
