Common practice to is set levels so the meter reads  (average) - 8 when recording speech. This practice is partly due to the modulometer's characteristics as a quasi peak meter (quasi because it has the ballistics of a mechanical meter) and it also reflects the caution required in location dialogue recording where a lost take can represent many thousands of dollars.  Although not desirable by any means, a low level signal is better than one with distortion from peak clipping.