Linear regression with or without forced origin

[Steve Reimer]

One thing that we haven't been able to effectively address is a quick and easy method for determining error at the low calibration levels. If you could report your result at your low standard level plus or minus a percentage then all this becomes more evident. A result of 0.05 ug/L + or - 75% ( + or - 0.04 ug/L) might be a little more obvious to someone looking at the data. A value of -0.01 (+ or - 0.04) ug/L might not be so disturbing.
The MDL calculation in 40CFR136 appendix B was for devised wastewater. EPA has lost the lawsuit on that one twice and a replacement has been in the works for over ten years. I will be long retired before that one is dead. The drinking water program is moving away from requiring it and solid waste is deleting all mention of it except for method development.

[Peter Apps]

Steve

Has anyone looked at doing an uncertainty budget for low level calibrations ? - admittedly they can quickly get very complex and cumbersome for even moderately complex analytical methods, but they do reveal what the main sources of uncertainty are.

Alternatively, if the MDL is defined as the lowest level that the method can deliver RSDs of less than 30% (or pick a number) in replicate analyses (rather than extrapolating from clean standards) then the estimate should be representative of actual practise, robust and fairly straightforward to do.

Peter