Page 1 of 1

Test towards reference - how to understand?

Posted: 15 Jan 2009, 18:31
by Lars-Ake
Today I got a question on the Test Towards Reference button for collections.
The question goes:

"When I “test toward reference” a collection, the first column of report seems to have some random lacks. I send you a copy as an example.
...some reports are missing several lines, others none...
If it’s a bug, I don’t see the pattern related to it.
With another collections of 51 members, it happens the same. With other two collections (10 and 13 members), this problem doesn’t happen."

Image

My answer:

Without too much analysis, I think this is NOT due to an error in CDendro.
The first big column is supposed to be dated as
"FIRST COLUMN GROUP IS BASED ON MEMBER OFFSET RELATIVE TO BEST MATCH FOR COLLECTION"
Edit by me: (The alternative would be to have it based on the dating of the reference, but that is currently undated, see a comment by Torbjörn (tAxelson))

Then the various members, one on each row, can be placed on a certain place on the time scale depending on their offsets within the collection.
In cases when the REFERENCE has no data for that place of the time scale, then there cannot be any correlation values displayed from a "match" between the sample of that row and the reference.
Then you get these blank areas.

Regards/Lars-Ake

Re: Test towards reference - how to understand?

Posted: 16 Jan 2009, 12:44
by taxelson
I do agree with Lars-Åke. Just to clarify:
The best match of the short "reference" (the undated sample here selected as reference) towards the mean of the whole collection is at 1918, and than CDendo assumes that as the right date for the "reference". The blank lines appears for those members in the collection which have no data for the years 1882-1918. Although, according to the table there is no reason to think that 1918 is right date for the "reference". If you have an idea about what may be the right date, you can set the "references" date to that year and try again. You will than find the correlation values etc according to that date towards each sample in the collection in the first column. If you have even another possible date, you may set the "reference" also to that and try another time. In this way it is possible to study the pattern of correlations at different possible dates and from this maybe judge the right one. Unfortunately a sample of only 36 years lengths is undatable "on its own merits" anyway.