Note on Statistical Limits

The systematic errors in a method, which are often difficult to determine or eliminate, govern what we prefer to term the accuracy of a determination. The accuracy of the analyses described here is, to the best of our knowledge, complete and no avoidable systematic errors should occur, although it would be unwise to dogmatize on this point. 

However, each method will certainly give rise to more or less random errors which govern what we prefer to term its precision. If these random errors obey a Gaussian distribution they are characterized by a standard deviation, r, which can be estimated by performing a sufficiently large number of replicate determinations. The range on each side of any one experimental result in which the true result may lie will depend upon the v value for the determination concerned and the degree of probability with which we are satisfied. In most industrial measurements the 95% confidence limit, which is approximately 2a, is considered adequate and 2 r (not r) has been used to express the precision of the methods described in this manual. 

The precision, P, quoted for each method is such that, if n determinations are made, the true result will lie in the range  

The values for P quoted here have been determined on homogeneous samples of water and thus apply to the method as such, and do not include random errors such as from over-side sampling, about which it is next to impossible to make generalizations. 

The values for P have also been determined under as realistic conditions as possible (on board ship if feasible) by workers familiar with the methods. As P often varies with the magnitude of the property being measured it is quoted for two levels of concentration or for the concentration level most likely to be encountered. 

---

Note The range and precision depends very much upon the amount of nitrate in the water. The above applies to near-surface waters with less than about 10 g-at/liter. In deeper waters the precision is much poorer. In deep Pacific waters the value for P (see note on statistical limits) is little better than 0.75 fig-at N/liter. 

The limit of detection will depend upon the volume of water filtered and the sensitivity of the fluorometer. IMth a 2-liter sample about 0.01 mg chlorophyll detectable with surety. Hie predsion is very much dependent on the amount of pigment being determined but P see NOTE ON statistical limits) is bettor than 8% of any value of chlorophyll a exceeding 0.5 mg/nP. 

---