I believe that 'types of coding' is a reference to shallow(structural), moderate(phonemic) and deep(semantic) processing. This is an encoding spectrum that places the three forms of encoding into levels, each requiring a different amount of effort.
the least effective type is shallow processing, which encodes information according to it's basic sensory properties (structure), such as size/shape/colour. For example: the size of a car.
Moderate processing involves the encoding of information According to it's acoustic/phonetic (basically auditory) attributes. For example: saying the word car aloud/silently.
Deep processing involves encoding information according to it's actual meaning (semantic=meaning), which links the information to its meaning that's already stored in LTM (yes elaboration).
For example: a car is a motorised vehicle designed to transport people from a to b.
While Deep proccessing requires the most effort, it is the most effective level because it encodes information in a semantic way, which happens to be the form that memories are stored in LTM(they are organised, intergrated and stored relative to their meaning). In terms of effort and effectiveness, Moderate processing is the half way point. Therefore shallow proccessing requires the least amount of effort, and is the least effective in the transfer and organisation of info into LTM.
Yes effortful encoding is important , with the more effort put in attending to info the more effective the higher the quality of encoding and transfer/organisation into LTM.
Since the atkinson/shiffrin model is a.k.a the Multistore model, I believe your teacher wants you to be able to explain how the info is encoded from STM, into LTM (as This ties in with the levels of encoding).
The semantic network theory stores information according to its meaning (can be in other forms but you don't need to know that) and since deep processing encodes info according to its meaning, the transferred *semantic* info from STM can be easily organised into LTM.