Hello,
Are you saying that when you lift the device the reading goes up? That's strange already.
When you lift the device the reading should go down. That's because the acceleration going up should subtract from that going down (gravity). You should have to move it pretty quickly to see the change though unless it is very sensitive.
However, once the device stops and remains still again the reading should again return to the previous reading before it was moved. In fact, once the device is kept moving with a constant speed the reading should go back to the original reading...as long as the speed does not change at all.
The device should also have an orientation associated with it. If it measures in all three directions then it should have a forward/backward, left/right, and up/down orientation, so that there should be some markings that indicate how it should be oriented for taking measurements.
This means that if the device was oriented correctly to begin with and you tilt the device, it will read a different value because it will think it is being accelerated on an angle now rather than just 'down'.
So if this was installed in say a model airplane and the airplane was sitting on the ground it should read acceleration in only one direction (down) but if the airplane nose was tilted upward it would read acceleration in two directions. So the up/down reading would go down while the forward/backward reading would go up (or down).
How many outputs does this device have? We first have to establish how it tells the outside world what it is actually measuring.
[ADDED]
I just took a look at the data sheet. What format is that video though?
I see that the chip either has one direction or two direction sensitivity. I'll assume you have the two direction unit.
If the chip is laying flat it should not read anything or at least not much, with maybe a small error due to the non perfect axis symmetry. So lets just say it reads very low with the chip flat. Now if one side of the chip is raised up (so the chip is at an angle relative to the floor) it will sense gravity as an acceleration in the direction which that side senses. So the two direction unit chip has four sides and a top and bottom. But lets call the two sides in the Y direction left and right, and the X direction front and back, and the top and bottom of the chip we dont have to worry about for now. If the chip is laying on the desk and the desk is horizontal, then if we lift the front of the chip such that the back is still touching the desk, we will (ideally) see the reading of the X output go more negative (it might go positive depending on the chip). That's because the acceleration is a vector and the chip will no longer be aligned by 90 degrees from that vector.
So how to deal with this...
The gravity vector would look like: <0,0,-9> but since the chip is reading <x,y,0> the chip reads nothing when flat <0,0,0>. When tilted with the x axis upward at an angle of 45 degrees, the reading would change to: <-6.4,0,0> because the x axis would sense part of that -9z vector component.
Now if the chip is ALSO moving forward and changing it's speed so that it experiences a front/back acceleration, that is going to add to the previous measurement so the reading will go up. If it was moving with the right change in speed, we might see <0,0,0> even though it was accelerating quite a bit.