Friday, November 16, 2007

Occasional errors

I'm making a final blast through 20th century Birth, Death and Marriage records to see if I can identify close living relatives.

After identifying Ellis Brain's birth and death details, further slogging showed that his mother (my great-grand aunt) died in 1945, and her husband in 1940. I've ordered Maria's death certificate to see if further children or other next of kin are listed.

The next project has been to identify the death or marriage of my great-grand-uncle Ellis Richard Williams. I know he was alive up to about 1935 from his telephone book listings. He's aged about 58 at that time, so there is plenty of scope for a search. Neither his mother, nor his siblings Maria and Griffith made it to 70, but his father lived to 76.

On the Genes Reunited site, I've been working through from 1934 (in case his telephone listing survived him by a year) up to 1961 without a good match. The big problem is that the GR index on these records is completely abysmal.

A typical set of search results from Genes Reunited. In this case, the JUN pages are indexed incorrectly.

Each year you have to look at a page from each of the four reporting quarters. Like a phone-book, the page can begin almost anywhere, but if I search on Ellis Williams, then I would expect to find him somewhere on the page, or at least a pair of names like Elizabeth and Emily which bookend a place where Ellis might appear. In actual fact, quite a lot of the pages are one page off, and some are much more. Because I'm paying for each page view, these errors mount up in cost.

It took 3 attempts to report this to Genes Reunited. They kept coming back with suggestions about Birth and Census records (which have different issues) despite the fact I kept saying it was the Death records. Finally, they admitted this happened "occasionally" and gave me some more credits to do searches on nearby names that might elicit the page I desired. That's why you see "Drummond Williams" in the screen shot above: I was trying to get it to look for names a little earlier in alphabetical order than "Ellis Williams". You can't simply navigate back or forward one page in the records: you have to run a new search on a different name and hope that triggers the correct page.

At times a sequence of "Ellis Williams" (or any other name you might search on) will cross a page boundary, but the poor indexing and search results don't illuminate this issue.

1934-1940: my annotated list of actual page sequences, with markup to show those that were incorrectly returned from a search on Ellis Williams.

So, with the extra credits given, I went back to review the files I had downloaded. So far I've looked at all the quarters from 1934 to 1951. In only one year did all the pages match my search. In all others there was at least one wrong, and in several, three out of four. This means a 25-75% error rate - quite a bit more than "occasional".

I'm now waiting for Genes Reunited's response, given
  1. the obvious problems in the basic service
  2. the costs to the subscriber
  3. the additional time taken to re-run searches and download new pages (each of which comes in a PDF, so takes some time to load), not to mention a day's turn around to get a reponse from their support staff
  4. the fact that the known problems with the search are not being actively communicated to subscribers.
Other agencies with bibliographic data will usually provide a link that allows you to instantly report a page that is illegible or which doesn't correspond to the search you've done. GR does not provide this, despite being "officially the number 1 family website in the UK by visits"

No comments:

Post a Comment

Flickr slideshow