Baffle step is actually directionality shift. Directionality increases, so on-axis SPL does too. It is frequency sensitive, in that directionality only increases at frequencies where the baffle is acoustically large.
When a travelling wave encounters a baffle edge it is suddenly unloaded which which creates a new sound source originating at the baffle edge with the OPPOSITE polarity radiating back toward the driver and normal polarity radiating around the corner.
At low frequency the baffle edge sources will not be signifigantly delayed in time from the driver and since they are polarity inverted relative to the driver the two cancel effectively cutting by half the acoustic load at the driver.
As frequency increases the energy from the baffle edge sources becomes smaller and eventually will not interfere with the output from the driver.
The power response doesn't change, but the on-axis response does. It occurs whether a speaker operates in freespace or half space.
The baffle transition increases total radiated power by 3db.
The axial level is increased by 6db and the dispersion pattern goes from full-space to half-space.
The radiated area is cut in half but the axial level increases by 6db which adds up to a total radiated power increase of 3db.
When you perform an outdoor ground plane measurement, what you're doing is to measure the speaker in a pure anechoic half-space, with no reflectors. The idea is to place the speakers within 1/4 wavelength of the ground, which prevents it from causing reflections that would interfere with the source. It's a launch point boundary, simply constraining the radiating angle without causing anomalies from reflections.
Speakers being within 1/4 wavelength to the ground is not a requirement for groundplane measurements, it's not even practical since in some cases a driver will be more than a foot from a baffle edge and this would make your highest measurement frequency 280hz.
Even with small speakers you would be limited to under 1khz.
If the cabinet is all the way on the ground then the diffraction from the baffle edge on the ground will be postponed until it is reflected by the ground to the opposite baffle edge effectively moving that edge three times further from the driver.
If you lay a tall, thin array speaker on its side, then two important things are accounted for. One is the distance between sound sources and the ground are small. This prevents the ground from causing destructive interference, and makes it act as a pure half-space boundary. The other is the baffle still defines the radiating pattern, so baffle step is still measured. At frequencies high enough that the baffle is acoustically large, the radiation pattern is reduced to quarter-space. At lower frequencies, the radiation reverts to half-space.
If you lay a tall thin speaker on it's side you effectively have a baffle twice as wide with two drivers (or two rows of them in the case of a line array) side by side which still will have a baffle transition but one lower in frequency.
Whether the speaker is lying on its side or standing upright, the baffle halves the radiating angle at frequencies where it is acoustically large. That's key here. On its side, the baffle step transition is between halfspace and quarter space. Upright, the baffle step transition is between freespace and halfspace. But in each case, the baffle step transition halves the radiating angle. This means you will get a reliable measurement that includes baffle step influence either way. What you won't get, when measuring on its side, is all the non-minimum phase interactions from all the reflections and the changing directivity from moving between half-space and free-space due to the distance to the ground.
This is where the confusion is comming from.
First of all don't think in terms of the ground causing the speaker to radiate half-space/quarter-space rather than full-space/half-space.
Instead think of the ground as an acoustic mirror.
You place your mic on the ground causing reflections from the ground to be coincident with the direct arrival.
Now if you imagine a speaker in space (not on the ground) it's sound will arrive at the mic at the same time as the sound from the speakers 'image' in the mirror which is the ground.
In effect you now have two of those speakers their sound arriving at the mic at the same time the level being 6db higher than what it would be for a single speaker.
Subtract 6db and you have a groundplane measurement.
Placing the speaker on the ground will cause the baffles of the speaker and the other speaker which is the speakers image in the acoustic mirror to join and you will then have a speaker with one dimension doubled along with the number of drivers.
Two separated speakers their sound arriving at the mic at the same time produce the same result as a single speaker except 6db louder but when you get the speaker too close to it's image the proximity interferes with the baffle effects.
A tall skinny speaker has the advantage that the baffle effects are mostly due to the baffle edges to the sides.
Lying the speaker on it's side for a groundplane measurement would double the horizontal dimension of the cabinet which accounts for most of it's baffle effects.
The top and bottom baffle edges are responsible for the minority of baffle effects in tall skinny speakers which are vertically symmetrical, MTM's, drivers in the middle or line arrays.
For tall skinny tower type speakers with the drivers at the top, the top baffle edge will also be important and the bottom baffle edge will be unimportant.
Standing a tall sikinny speaker on the ground for a groundplane measurement will cause the baffles to join doubling the height dimension but since the bottom edge of the baffle contributes relatively little to baffle effects you still can get good results with the speaker on the ground.
You still get the sound from the speaker and the it's image at the mic and the level is 6db higher than a single speaker.
You will want to angle the speaker down to get the mic on-axis.
Also you will want to be shure not to angle the speaker down too far or the speaker and it's image will reflect off of each other and cause interference (keep the angle of the speaker/image baffles extreme to each other to direct reflections away from the mic).