fmgo12 Which measure of accuracy is used by Location functions?

[jbante]

The Location and LocationValues functions both report an accuracy for the returned location in meters, but I can't find any documentation reporting what measure of accuracy is being returned. Is it CEP? CEP95? 2DRMS? None of FileMaker's documentation says, and neither does Apple's documentation on the CLLocation class.

[paul_howard]

Hey, Jeremy. Me again. Don't me to seem stalker-ish, but I thought I'd take another shot at answering your question, without the 140 character limit. :-)

 

Since I have a little bit of a personal interest in this question myself, I thought I'd do some more research, which turned up three stackoverflow threads that I thought particularly relevant:

 

 - http://stackoverflow.com/questions/11945455/gps-reported-accuracy-error-function

 - http://stackoverflow.com/questions/3411629/decoding-the-cllocationaccuracy-consts

 - http://stackoverflow.com/questions/10634734/what-does-horizontalaccuracy-exactly-mean

 

Note that the last one (I think) exactly re-states your own question, and that the Accepted Answer is, "Call Apple in Cupertino and ask for one of their GPS engineers." One of the other threads correctly notes that another option would be to study CLLocation.m, but that's probably as likely to happen as getting the answer directly from one of Apple's engineers.

 

That said, if I were an Apple engineer, I would feel uncomfortable telling developers that, for instance, kCLLocationAccuracyNearestTenMeters will give you a result that is accurate to within ten meters if I was using anything less than a 95% probability range to calculate that. I don't think it would be unreasonable to assume that using "kCLLocationAccuracyNearestTenMeters" is roughly equivalent to saying, "Give me a location with a R95 value less than or equal to 10m." 

 

But I think the only way you'll ever find out for sure is to do extensive testing, along the lines of the previously linked http://blog.oplopanax.ca/2012/11/measuring-smartphone-gps-accuracy/, but also testing with various constants/values for horizonalAccuracy and, in the case of FM Go, timeout.

 

Hope that helps.

[jbante]

I put together a file to help us try to figure this out empirically. I need everybody's help, though. Here's how it works:

 

1. Download the file to whatever iOS devices you have with FileMaker Go (12) on them.
2. Find a spot, and get comfortable. It will be helpful if you temporarily turn auto-lock off.
3. Start the file recording location measurements. It will keep repeating until you tell it to stop or the battery dies.
4. Leave your device alone for a long time. The longer the better. A few hours would be great, if you can manage. 30 minutes probably isn't enough.
5. Repeat 2-4 in as many different spots as you can.
6. Share your sample data with me. The file can export a CSV and email it to me.

 

Many folks are concerned about their privacy and reluctant to share data on their whereabouts, but there's no need to worry. The file will randomly offset all your data so that I have no practical clue where you've been — even if you send me the FileMaker file instead of the exported CSV. (But it does this in a way that won't upset any of the analysis it's doing to answer the question at hand.)

 

... Apparently, the forum software doesn't let me upload an uncompressed fmp12 file, so you'll have to unpack the zip before transferring it to your iOS devices.

 

... Update: I tweaked the variance calculations. If you notice they're behaving weird, try downloading this newer version of the file.

AccuracyTesting.fmp12.zip

[David Wikström]

I'm running this on an iPhone and an iPad mini right now. It would have been kind of nice to have an option to show actual coordinates/elevation so one can compare the two devices, not just samples within the same device?

[David Jondreau]

How about hosting it?

[jbante]

David W., I agree that would have been nice, but I wanted to be very careful with folks' privacy. I know a lot of people are touchy about this sort of thing.

 

David J., hosting it presumes that folks will be able to connect to it from all locations, but I'm also interested in seeing what the Location functions do in spots too remote for any data access. To keep from investing too much effort in this, I decided a local file would be the expedient one-size-fits-all solution.

[David Wikström]

I changed the script to not encode the location - I'll run some tests that I'll keep to myself ;-)

[jbante]

I'm doing a blog post series on using the Location functions. Today's post is about what I've figured out so far from the location sample data I've received:

 

http://scarpettagroup.com/location-whats-accuracy/