it depends on the protocol used. It could use the information that owns the protocol for example, the parameter that counts the number of retransmissions
If you use queuing theory, you can count the number of packets delivered. We can start as follows,
We can find the probability that a node successfully transmits on the i-th retransmission by considering the packet collision probability and retries (R) as (1-p)p^i, for i=0,1,2 ... R. p heere is the constant conditional collision probability which can be calculated as p= 1-(1-t)^(N-1). Here t is steady state probability that tagged node transmits in a random time slot given that it has packet to transmit. and N is the number of nodes.
Then we need to calculate #of retransmission per packet for the successfully transmitted packets as so on.
You should first define the concept of "deliverability" that you are using. There are at least two different kind of delivery, a local delivery (hop-by-hop), or a target delivery (end-to-end). On the first one, you are determining how many of the sent message could be arriving effectively, and could be used on the next node, a neighbor, or a node inside the RF range of the original sender. On the former one, you have to estimate how many of the generated message, or detected situations are being effectively received and can be used in their target node, that could be a normal sensor node, or more commonly could be the Sink node.
There could be different strategies that could be used on simulation models, or in real implementation, to measure or estimate this metric.
I have been using a version of a packet delivery rate, as a metric to compare different protocols options in some papers, which reflects the fraction of data packets that was sent by any sensor node, in the whole network, and that was effectively received at the sink node. If you are interested I am sending you links to the papers:
Gonzalez, R.; Acosta, M., "Evaluating the impact of acknowledgment strategies on message delivery rate in wireless sensor networks," Communications (LATINCOM), 2010 IEEE Latin-American Conference on , 15-17 Sept. 2010
Gonzalez, Ricardo; Huerta, Monica. "An evaluation of strategies to improve the message delivery reliability on WSN without the use of acknowledgement messages". Workshop of the 4th IEEE Latin-American Conference on Communications 2012 . Cuenca, Ecuador. November 2012
Additional information about message delivery can be found in:
González, R. H., Loureiro, A. A., & Mini, R. A. (2010). QoS: Requirements, Design Features, and Challenges on Wireless Sensor Networks. In H. Jin, & W. Jiang (Eds.)Handbook of Research on Developments and Trends in Wireless Sensor Networks: From Principle to Practice (pp. 56-78). Hershey, PA: Information Science Reference. doi:10.4018/978-1-61520-701-5.ch004
I hope that this information could be useful for you.
Ricardo Gonzalez
Universidad Simon Bolivar
Caracas -Venezuela
Conference Paper An evaluation of strategies to improve the message delivery ...
If I have not interpreted it wrong it is quite similar to(or can be derived from) DATA THROUGHPUT.
If you are using a single hop point to point network(star) the data received at the PAN Coordinator can be captured and parsed to compute the throughput.
Instead if you are using a multi-hop(mesh or tree) network then ensure that you identify the packet that is sent from each device uniquely. Again the data has to captured through the PAN Coordinator.
Assuming that you know the packet transmission details i.e. How much data is transmitted and at what interval.I think retries etc. won't matter if packet is delivered(in 1attempt or multiple attempts)