Getting two different results from simulation and rel IBMQ device. Which is more accurate?
If you can run the simulation in the first place, then the simulation is probably more accurate.
Current quantum computers are quite noisy. Simulations are effectively perfect, unless you intentionally put noise into them. For any small circuit a simulator will be faster, cheaper, and more accurate. Assuming your goal is to know what the circuit is supposed to return, as opposed to your goal being to benchmark hardware (e.g. show an error mitigation technique works).
The main issue with simulations is that you can't simulate large circuits. In a few situations, allowing the simulation to have a noisy result allows it to simulate a slightly larger circuit. Sampling random circuits is one example. When simulating large enough random circuits, making the simulation possible can require making the simulation noisier than the actual hardware.