HPLC Products

There are four major sources of random error in HPLC analysis. These are:

Errors are additive, and because these errors are in the main random errors they may sum to zero. Problem is, they may not!

So representing the error by the symbol ? the change in the result R is:

 ? R          =              ? M           +                ? V            +                ? I              +                ? A

   R                             M                                  V                                I                                    A

Where M = Mass, V = Volume, I = Injection volume, and A = Integration Area

So what does this mean in practice? If the sample mass is slightly too low, the dilution is made slightly over the line, the loop is not quite filled, and the integration tick marks are slightly too far up the side of the peak, then the final result will be out by the sum of each of these errors. In the above case the result will be lower, because each error gave rise to a lower result, but the opposite could equally well occur, or some other combination. So if each error was only 1%, the result could be out by +/- 4%.

How could this possibly happen?! A tiny amount of the sample could fall on the balance pan and be included in the weighing but not in the sample solution. Or we could weigh out 99mg instead of 100mg. With dilutions, especially with smaller volumes, it is not difficult to go slightly over the line. Or if the solvent was volatile, to leave the top of the master solution whilst making dilutions.

If using an autosampler, there will always be a certain (albeit small) level of error, especially with a variable volume injection model. But sample carryover, the inclusion of some wash solution etc could introduce errors here. With integration, it is not easy to get the start and end tick marks in the right place for every peak. Things to watch especially are that the Peak Threshold has not been set too high, and that the peak width is appropriate for the peaks of interest. Either of these can cause late or early peak starts and ends.

Calculating Standard Deviation. Using the example given earlier of a +/-4% error, if the actual error was calculated for a series of independent analyses, it would be found that the mean random error would be lower because the individual errors can compensate for each other (for example if the Mass error was positive, and the Volume error was negative). Calculating the standard deviation (SD) for this example gives a value of +/-2%. The advantage with calculating Standard Deviation is that for Random errors, they follow the Statistical Normal Distribution model so we can show that 68.3% of results occur within +/-  1 x SD, 95.4% occur within +/- 2 x SD, and 99.7% occur within +/- 3 x SD.

So what to do now? It is good practice to quantify the possible errors that can exist in a given method, so that we can specify the accuracy limits of the results. If these turn out to be unacceptable, we need to look at individual errors and try to determine which has the greatest impact on the total error. It would be a shame to throw away a perfectly good pipettor, if it is the integration settings which are screwing up the results!