The absolute error is the maximum deviation between the ideal straight line and the actual transfer function, including the quantization steps. The minimum absolute error is therefore ½ LSB, due to quantization.
Absolute error or absolute accuracy is the total uncompensated error and includes quantization error, offset error, gain error, and non-linearity. Offset, gain, and non-linearity are described later in this document.
Absolute error can be measured using a ramp input voltage. In this case all output values are compared against the input voltage. The maximum deviation gives the absolute error.
Be aware that the absolute error represent a reduction in the ADC range, and one should therefore consider the margins to the minimum and maximum input values to avoiding clipping keeping the absolute error in mind.