If we track all possibilities, then

first if condition gives us

T(n)=O(n/2)+T(n/2) equivalent to T(n)=O(n)+T(n/2)

second gives us

T(n)=2*O(n/2)+T(n/2) equivalent to T(n)=O(n)+T(n/2)

for the third one

You can easily see that all possibilities will be equivalent to T(n)=O(n)+T(n/4).

From these recursions you can deduce that T(n)=O(n) i.e. the time complexity is linear.

On your merge sort analogy: The array is being broken in a similar way but if you observe carefully we don't operate on each chunk unlike merge sort. Basically at each of logn levels in merge sort we are dealing with all n of them while here with n/(2^i) i.e. decay exponentially.