The question
A regional stream-sediment survey from the 1960s and 1970s covers the district around a client's project: about 3,500 samples. All were assayed at the time, mostly by comparing colours by eye. About a third were later re-assayed with a modern instrument. Can the old values stand in for modern ones at the two thirds of samples that were never re-measured?
The data available
| Old measurements | Modern re-assays | |
|---|---|---|
| Samples | All of them | About one third |
| Precision | Steps of 5 ppm | Fine |
| What it captures | Part of the metal in the sample | Nearly all of it |
The survey was run in 27 batches, each with its own habits. Doubly-measured samples from eight neighbouring map sheets were added to steady the comparison.
What was done
- Every count in the data description was reproduced before any analysis, and sample positions were corrected.
- A translation rule from old value to modern value was built for each batch, each with its own margin of error.
- Old and modern values were combined into one map per metal, with a translated old value counting for less, according to its margin of error.
- The result was tested by hiding known answers: the modern assay of some samples was withheld, each method had to predict it, and the predictions were compared with the truth. This was done for scattered samples, for whole 10 km blocks and for whole batches.
The result
| Metal | How old and modern values relate | Decision |
|---|---|---|
| Copper | A steady offset in every batch. Prediction error 13 % smaller with the old values than without. | Old values used |
| Lead | No relation. | Old values not used |
| Zinc | Depends on the batch; no gain away from modern samples. | Old values not used |
The obvious shortcut would have done harm. Rescaling the old values with one rule for the whole survey gave errors 30 % to 125 % larger than ignoring them altogether, for all three metals. The copper map gained a quarter more ground; the lead and zinc maps were kept to modern data, and the client knew why.
What it cannot say
The maps are regional guides, not measurements. A new sample would fall within a factor of about 1.7 of the mapped copper value nine times out of ten, a margin that held in the tests. They show which zones several kilometres wide are richer or poorer; they do not fix the position of a contour to better than about a kilometre, and they say nothing about ground no stream drains.