A long time ago, I did an experiment where I modified the code of a RFM95 based node to send packets using error correction rate 4/8 (instead of 4/5). This did actually work, even though the LoRaWAN specification says to use 4/5. This is possible because the packet format used in LoRaWAN enables a specific LoRa header that tells the receiver what correction rate is used.
This was discussed in thread: Improve node performance by changing coding rate from 4/5 to 4/8?
Thomas @telkamp did a more advanced experiment, as described in the thread. The improvement was marginal.