Use "Test towards reference" for dating


  • The Test Towards Reference command (button) can be used to check how well your dated collection members match towards a dated reference. This may be very suitable when you are building a new local collection and use a reference from a nearby area. You may even turn on the block checking mechanism to better see how well each member matches.

  • The Test Towards undated Reference command can be used to check how well a number of reference curves from different regions best match towards an unknown mean value sample which is then selected as being the reference. This is useful either when you have no reference for the area from where your sample comes, or when your sample is of unknown origin, e.g. from a ship.
    Right-click the button for searching over a limited area of the time scale!

    In this case we have a collection with all available mean value curves from various regions. The sample to be dated is then selected as the reference. By running the button-command "Test towards undated reference" (...undated ref.) we get a list of how each member of the collection best matches towards that sample. Such a table may be difficult to read when there are many members of the collection. Therefore you may want to check the option "With condensed report" found just below that command button. That way you get an extra sorted list with frequently recurring year numbers like this:

     
    Year numbers occurring more than once: 
     1792   11 
     1769    3 
     1824    2 
     1921    2 
     2010    2 
    

    You may want to uncheck those members giving low quality matches towards your reference. This can be done by hand (click) or through the button "Select best matching members". Then  you can run the button-command "Test towards undated reference" again and get a new report with better overall results.

  • Note: When the "Test towards reference" command is run towards an undated reference, it behaves exactly as the "Test towards undated reference" command.

    The rest of this section describes with examples how these two command buttons can be used together with the "With condensed report" option.

For this type of crossdating operation, you need a collection with all your usable references.

The collection above contains reference curves (=members) from several sites, some of them marked with red dots in the map below.
Each reference curve (=member) has been created from its own collection with the button command "Create mean value sample".
Then each such mean-value-sample has been saved in a .d12 file.
These normalized files have then been crossdated to each other and added to the dated collection shown above.

Note: Saving a mean value sample as a .wid file is not as useful as a .d12 file, as it is handy to have the .d12 files easily available with all their data (mean of ring widths, normalized values and standard deviations).
- On the other hand, mean value .wid-files can be saved within a single .rwl file which is also very handy.

For this excercise we will compare a mean value curve from Aneby in Småland with the curves of the collection above.

Sites with some shorter or longer reference curves are shown on the map.
Aneby is marked with a white dot.
Now, I have (not shown here) opened my Aneby collection and created a mean value sample out of it and also selected that sample as the reference. In this case I already know that the Aneby collection should be dated to 2003, so it already has that dating information.

Before we click on the Test towards ref. button we see to check the checkbox "With condensed report" under that button! This will extend the report with a condensed version suitable for a dating report.

Some comments:

When instead using the command Test towards undated reference (the button "...undated ref.") the contents of the column groups in the report will be different.

To demonstrate this, I have created a 70 years long sample out of the Aneby mean-value-sample. (Use the "Create sample from block-button" on the Curve display tab of the sample.)

Then I have selected that small sample as the reference and then clicked Test towards undated reference (the button "...undated ref.") Though the reference is anyhow dated, its dating information is now ignored and the command searches for the best matching position for each collection member.

In this case the sum of all our references dates the "70 years long unknown reference-sample" to 2003.
Though members of the collection matches better at other years, mainly because of bad overlapping in time like the reference from Hurdal (Norway, only 22 years for overlap at correct match) and NMVK (Stockholm) which does not overlap at all at the correct position.
Now, for the sake of demonstration, let us think that 1916 was an interesting year.
How well do the other references actually match the 70 year Aneby data at the year 1916?
To check that, use the dating button (marked "Dated to 2003" or possibly "Set date of youngest ring") on the Workbench tab of the sample to make the sample "dated" to 1916! (The top of the dating button on the Workbench will then switch to the text "Dated to 1916".) Then turn back to the collection and click the button Test towards reference, i.e. "Test towards DATED reference".
The leftmost Corr-column of the uppermost table shows how all our reference curves "match" at the year 1916 - indeed with low values!

By first dating an unknown sample to a certain year, we can hereby get a table showing how well or bad several reference curves match the sample at that year!

Using "Select best matching members"
When your collection contains MANY members, some of them will have a LOW correlation to the "reference-sample" at the tested point in time.
Then you want to uncheck those members which are not of interest for your comparisons.
Click the Select best matching members button to open a fixed selection "window".

If you are only interested in matches setting the reference at 2003, then fill in that year and your minimum requirement for correlation coefficient, T-value and Overlap. (You may even leave one of these three last fields empty.) Then click "Uncheck now!".

If that resulted in too many unchecked members, click the "Check all" button on your form, adjust the values of the "Uncheck members..." window, and click "Uncheck now" again!

Then click the "Clear report" button and then the Test towards undated reference button again to get a report with less members involved.

Note: All reports are saved temporary in a .html file - Click the button "Open in NEW browser" to see the report.
Lars-Åke Larsson, Updated August 11 2013.


Copyright © 2013, Cybis Elektronik & Data AB, www.cybis.se