for every passive device the NF equals its own attenuation, so splitters have 3dB, and at combiners are not guaranteed the input signals are the same, thus they should have 0dB in that case I guess.
If you combine two identical signal, perfectly in phase they amplitude will double while the noise (uncorrelated) will sum in power. Theoretically you can gain 3 dB NF. In practice you will have a lower gain since the two waveform will not be perfectly identical in phase and amplitude and also the combiner will have imperfections and losses.