* Finding incorrectly placed locations in map

[AdrianBruce]

... BTW: Are you only getting Variance reports and none of the other four types?

No, I've Variance, Deviation, Plot Pair and Singletons. Variance and Singletons were simply the ones I started thinking through in detail first.

So far I'm interested to find a number of issues that others might consider pedantry but I can't unsee until fixed - e.g. "Isle of Man" is my usual form but it threw up one "Isle Of Man" as either a place (or part of one?)... Tut tut.

So, rather interesting!

[BillH]

Mike,

I am using a technique very close to your suggestions.

I did leave standard deviation at 2.0 because I actually found quite a few places that were mapped incorrectly. It looks like most of these were mapped a long time ago using the FH geocoding rather than Map Life Facts which is what I usually use. I was actually surprised at how many plots were wrong... in the wrong county and even in the wrong state. So far I have probably corrected at least 30 or 40 places and I have quite a ways to go.

I was wondering. I am using 3 right most place fields instead of 2. Would it be possible for the plugin to remember the last value used rather than defaulting to 2? If not, no big deal.

Bill

[tatewise]

In this prototype plugin, I chose not to remember the settings, because I was not sure how useful it would be.
Once the Places have been corrected the plugin will be of limited future use, unlike some plugins that get used over & over.
I was also interested in what range of settings and perhaps other settings that might be needed.

To set the Rightmost Place Parts to 3 you could Edit the plugin script and change line 18 to local intPmax = 3

[BillH]

Mike,

I guess I was thinking that I would probably run this plugin every so often to ensure that I have not introduced new errors in plotting.

I will go ahead and update the plugin as you explained.

Bill

[tatewise]

If the settings were remembered, would you want them saved separately per Project or the same for all Projects?

[AdrianBruce]

Just to reinforce what Bill said above - I would envisage rerunning this every so often to check the geocoding in particular. I have a number of oddities (so this plug-in has proved seriously useful, thanks) but without understanding quite how I got these oddities in the first place, I currently feel that an audit run every so often would help reassure me.

I've checked my Singletons and Plot Pairs - all are OK, so I've set Low Usage Place Parts to zero, rather than Exception them - my logic there is that if geocoding errors do creep in again, then I don't want to have Exceptioned some error places.

That leaves Deviation - plus some Repetition.....

Repetition is my name for a completely different check that I've done manually - I sorted the place-names tab on Latitude / Longitude, scrolled down and found a massive number of places with the same Latitude / Longitude - all in the middle of Loch Tay! It turns out that my geocoding for (just) Scotland gives or gave co-ordinates in the middle of Loch Tay. Some obscure places - e.g. Wester Caputh, Perthshire, Scotland - were geocoded as if they read just Scotland. Curious - not even Perthsire, Scotland. Even stranger, Montrose, Forfarshire, Scotland was also in the middle of the Loch but I just regeocoded it and it worked fine, so why it went wrong in the first place (or which geocoding routine I used), no idea.

Back to Mike's Plug-In...
I'm slowly working through my Deviations (ooh, err, missus ) where I found some amusing errors - some obscure places in Scotland had been mapped to Snodland in Kent! And some obscure places in England had ended up in England, Arkansas, USA!

Trying to get the right Standardized value (to match the Google Geocoder used in the Map Life Facts plug-in) can be an art. I couldn't get it to pick up Rattray in Perthshire - I got one in Aberdeenshire. It mapped OK if I standardized to Blairgowrie and Rattray (they're twin settlements) - at which point I saw my Rattray was on the map so nudged the icon to the right point and removed the standardization.

As for West Derby - well, that had previously mapped as the west part of Derby, Derbyshire. Of course. West Derby, Liverpool, Lancashire, England doesn't geocode well - it ends up near Ormskirk. Instead, West Derby, Liverpool, England geocodes fine... Less is more.

So now I'm just working through the data - I may be some time but many, many thanks Mike.

[tatewise]

Adrian, the plugin seems to be more useful than I had expected

Did you see my question about whether you want settings saved separately per Project or the same for all Projects?

[ColeValleyGirl]

Mike, FWIW, I'd save the settings per project. When I use it (when I ever get time) I'd want to run it on the same project at intervals for the same reason as Adrian and Bill. And each project might have different place structures (my Staines in Ohio have a different structure from my main project -- one is US only, and one is UK/US/Canada/Australia/New Zealand, for example.)

[AdrianBruce]

I think that I'd also like by project saving of settings.

[BillH]

If the settings were remembered, would you want them saved separately per Project or the same for all Projects?

Mike, I have always had only one project so I would be ok with having them saved for all projects. I guess I really don't have a preference.

Like Adrian, I have found this plugin very useful. I have found a lot of mis-plots, spelling errors, etc. that I've been able to correct. In researching how to correct the errors reported, I have also found other related errors. Thanks again for a very nice plugin.

Bill

[LornaCraig]

...It turns out that my geocoding for (just) Scotland gives or gave co-ordinates in the middle of Loch Tay...

Loch Tay is indeed right in the middle of Scotland. I don’t know exactly what algorithm FH uses, given that countries are of course irregular shapes, but it definitely looks about half way up and half way across!

For this reason when I have only the name of the country I block the geocoding unless it's a very small country, because at best it's pointless and it can be misleading.

- some obscure places in Scotland had been mapped to Snodland in Kent!

Yes, I've seen that sometimes. That one really baffles me!

[tatewise]

Try attachment Place Exceptions plugin Version 0.4 Date 19 Jun 2021. [ Deleted as now in Plugin Store. ]

This adds a Variance Usage Threshold to set the maximum number of Places to qualify for a Variance report.
If set to 0 it inhibits all Variance reports. If set too high there may be many false positives reported.

This version of the plugin remembers the settings separately for each Project.

The published version will probably need a Help & Advice button to display a Plugin Store Help page with similar advice as I provided with the posting of Version 0.3 earlier.

[AdrianBruce]

Thanks Mike - having the parameters save is good, and having the Variance Usage Threshold makes me feel (rightly or wrongly) happier with the content of my Place Exceptions list.

I've now checked my Variance, Plot Pairs and Singletons and either dealt with or "accepted" them. So I can now zeroise the relevant parameter numbers, rather than rely on the PE List. That means that if Tilston and / or Tilstone ever get bad geocodes in the future, they won't be Excepted from reporting on that, since I'm not putting them in the PE List any more.

Edit:
I do use the Place Exceptions List for Places of the form "?Nantwich, Cheshire, England" which is a place-holder for "Probably Nantwich but I can't read it / not quite sure about it..." and also for my Constituency place-names which I'm not sure whether I should be mapping or not... This may or may not make sense!

[AdrianBruce]

... some obscure places in Scotland had been mapped to Snodland in Kent!

Yes, I've seen that sometimes. That one really baffles me!

Not particularly important but I wonder if there is some text matching that thinks "Snodland" similar to "Scotland"?

[BillH]

This adds a Variance Usage Threshold to set the maximum number of Places to qualify for a Variance report.
If set to 0 it inhibits all Variance reports. If set too high there may be many false positives reported.

Mike,

I guess I really don't quite understand what this means. What would I set it to in order to get the same results I got with version 0.3 (in other words all variances over 2.0)? The value can only be 0 to 4 so what is the difference between 2 and 4?

Bill

[AdrianBruce]

Mike,
We mentioned Hawaii earlier on. I have 2 such place names:

• Honolulu, , Hawaii Territory, USA, standardized to Honolulu, , Hawaii, USA;
• O'ahu, Hawaii, USA (the island for the above but without the Territory designation);

Both have reported a Deviation of 3.6 from USA ~ W-NW. Both are correctly Geocoded according to the FH maps.

I am presuming that my USA ~ W-NW stuff is a number of places in California plus some in Washington State. I got the impression from the earlier posts that the two Hawaii places would end up looking for their distances within Hawaii, USA and therefore not report as having a Deviation from the W & NW of the USA, so I was not expecting to see them reported as Deviation from there. Am I wrong? (I wasn't sure how many places there have to be in those more detailed places)

[BillH]

Mike,

I have added an exception to the named list for Seattle, King County, Washington, USA? (note the question mark at the end). I have the question mark to indicate I'm unsure of this place. It still shows up on the report.

image1.jpg (33.63 KiB) Viewed 4563 times

image2.jpg (43.97 KiB) Viewed 4563 times

Bill

[BillH]

Adrian,

I have two locations in Alaska and they have deviations of 10.2 and 10.3 for NW-N. I figured they were just being compared to Washington State places that I have.

Bill

[AdrianBruce]

Bill - doesn't the named list have to be named exactly "Place Exceptions"? Mine is and it seems to work OK to exclude what I want excluded...

[tatewise]

Yes, the Named List name must exactly match the plugin name of Place Exceptions.
Bill, Place Exception Plugin Exception List is some other list entirely!

Bill, I think you are confusing Variance with Deviation reports.
The Variance Usage Threshold specifies how few Place parts must have the 'deviant spelling' to be reported.
The previous plugin Version 0.3 had that threshold set to 2 internally.
If more Place parts than that threshold have the same spelling then it is assumed to be correct and not deviant.
( The Standard Deviation setting for Deviation reports is the same as for plugin Version 0.3. )

To perform any rational plot assessments there must be a cluster of at least 3 plots in any particular Area.
So only 1 or 2 plots in the Alaska, USA Area or the Hawaii, USA Area are not considered significant enough.
Unfortunately, that means they get reported as Deviation outliers.
If the plugin did not report those then it would not report genuine outliers.
In this case, you say they are being reported in the USA ~ NW-N sector and USA ~ W-NW sector respectively.
I would expect Alaska to be in the USA sector between bearing NW & N so that is OK.
But I'd expect Hawaii to be in the USA sector between bearing S & SW or SW & W depending on the USA central plot.

[BillH]

Not one of my better days.

I didn't even think. I should have known the plugin would look for a named list with the same name. How else could it possibly know which named list to use. I read right over the name in the text box in the plugin and still didn't get it right. Oh well...

You're right... I did confuse the two. Maybe I should just go to bed now and call it a day...

Thanks for the explanation on Hawaii and Alaska.

Bill

[tatewise]

A workaround for the Alaska/Hawaii case is to create & geocode a dummy 3rd Place record for Alaska, USA or Hawaii, USA.
Then the cluster of 3 plots will be enough to avoid them being reported as Deviations.
That is perhaps better than adding the 2 Deviation Place records to the Place Exceptions Named List.

[AdrianBruce]

A workaround for the Alaska/Hawaii case is to create & geocode a dummy 3rd Place record for Alaska, USA or Hawaii, USA.
Then the cluster of 3 plots will be enough to avoid them being reported as Deviations ...

That's actually what I've just done, before I read your response, and it works successfully. I actually now have places for:

• Hawaii Territory, USA;
• O'ahu, Hawaii Territory, USA - one of the islands in the archipelago;
• Honolulu, O'ahu, Hawaii Territory, USA - which sits on said island.

This actually allows me to put notes against Hawaii itself, such as when it was a Territory, and that I intend to record Hawaiian places with an intermediate level of the island, thus not following my usual US practice of the county (or equivalent). (It happens to be more exact and therefore more useful, to record the island. Not that I have more than 2 or 3 events there!)

So, that makes me happy as far as Hawaii goes, thanks.

[AdrianBruce]

I suspect that, by now, I'm getting into diminishing returns using the Place Exceptions plug-in. For instance, since most of my Cheshire Places are in south-east Cheshire and mid-Cheshire, then Birkenhead (which was in Cheshire) is, and always will be, a number of deviations from the bulk of Cheshire places. I probably need to make it an Exception.

But what the Place Exceptions plug-in can't do, is highlight places that are incorrectly coded but are still within the correct county / whatever. For the sake of recording it, my technique is to bring up a map of my places in (say) Lancashire and just run a visual check - it's surprising how many cases I can usefully ask: "Did any of my relatives really live there?"

To do this, go to the Places tab, filter it to "Lancashire", select all the Places that are in "Lancashire, England", and use Map icon / Map Locations of Selected Place Records.

This will take you to the FH "Native" Geocoder - zoom the map right out and check you haven't picked up places outside the UK, zoom back in and do a visual check, clicking each in turn until you've checked them all or got bored. I found that my instincts were strangely true and that some markers were in places they shouldn't be. I hadn't realised that there are two Pendletons in Lancashire, for instance. Naturally, mine were plotted over the wrong one.

[tatewise]

Yes, inevitably the Plugin has its limits because it is statistically driven.

Once you have populated the Place Exceptions named list with appropriate exceptions, it might be interesting to increase the number of Rightmost Place Parts to form smaller Area clusters and reduce the number of Standard Deviations to allow less deviant plots to be reported. That might find 'abnormal' plots like your example for Lancashire.

It should already be identifying places for Lancashire, England that are plotted outside the UK, unless there is a significant cluster of them, but increasing the number of Rightmost Place Parts might help.

Of course, it cannot detect all abnormalities such as plotting the wrong Pendleton!