We are making roll measurements (atan2(ay,az)) using the LSM9DS1's accelerometer. We see extremely little error, just fractions of a degree. Since this is within our required accuracy, we are thinking that we don't need to do any calibration to get our roll measurement. Is this true? We are confused by the datasheet which seems to suggest that errors as high as 90 mg are possible. If a 90 mg error occurred in the Y axis, this alone would generate a 5 degree error in our roll measurement, yet we are seeing almost no error. Is this calibrated in the factory? Should we do our own calibration to ensure that we have errors less than 1 degree? Thanks!