Precision and accuracy

(5) Wild readings. All data should be recorded. None of it should

be rejected at the time of the reading.

However, when a wild reading is

encountered, the meteorologist has a choice of three alternatives:

(a) Rejection. You can assume it to be a gross error and reject

it.

If desired, another reading can be made to take its place.

Rules

should be established for rejection.

For example, when a value has a

probability of occurrence of less than 1/20, it is to be rejected. In doing

this he is assuming (with a 20 to 1 gamble) that it is a gross error rather

than a valid random-dispersion error.

(b) Weighting.

You can assume it to be valid but with a low

probability of occurrence.

The proper treatment here is weighting, not

giving it the strength of the other readings.

(c) Routine procedure. You can ignore that it is a wild value and

treat it the same as the other values. Of course, you are taking a chance.

If it were a gross error (mistake), it would damage the value of the

average. If it were a valid random error, who knows but what it is correct

and the other values are random extremes.

g. Precision and accuracy.

(1) The dictionary defines these words each in terms of the other,

indicating that they mean about the same thing. In metrology, we need to

make finer distinctions. Precision has meaning only when applied to a group

of measurements, never to a single measurement.

It refers only to the

reproducibility of the results of measurement. The smaller the spread, the

greater the precision.

Accuracy refers to the degree to which the

measurement indicates the truth. It can pertain to a single measurement or

the average of a group of measurements. Figure 4 represents the results of

firing 10 shots each at four different targets. If a good quality gun were

clamped in a vise but little effort were made to aim it, the score would be

as in A, figure 4. If it were a poor gun but well aimed, we might get a

score such as B, figure 4. It may be noted that precision depends upon the

absence of random errors within the gun itself, whereas accuracy is a

function of systematic errors outside of the gun (poor aim, wind, bullet

drop, etc).

A person firing the gun from his shoulder could introduce

errors of both types.

152