Assuming the data has been generated off of real world location data a couple of thoughts came to me that I'd thought I'd sense check with the group:
\n\nCellphone location most probably is reported as CER ( at least Android which, assuming this data is again modelled on the real world, should be a significant subsection of the devices used to checkin). However, the figures do look like meters - especially if it is modelled on US cellphone location having to have a ceiling of 300 meters accuracy (FCC 94–102).
The accuracy jumps could be related to software updates (GPS upgrade?) Or mobile base station upgrades (most cell towers use software defined radios, so I could imagine a network wide upgrade or a rolling upgrade program over time - that could fit well with some of the observed improvements in accuracy in time.) The other realworld issue with cellular networks (well GSM at least) is that in a 3G network the physical cells grow and shrink dependent on density of subscribers - so coverage will change dependent on how many people are in a cell.
Given these sorts of hypothesis, I'd expect to see a strongish correlation (and hopefully causation) between location and accuracy, as the network coverage (either wifi or Cellular) shouldn't differ massively between device (again assuming the time column maps to a minute time unit, and thus unlikely to see a massive update / refresh cycle of devices in play). Indeed, it might be possible to derive / build some sort of coverage map based on accuracy - i.e. heat map of accuracy by location.
\n\nIf I get time before the end of the contest, I'll have a go at that.
", "messages": [ { "id": 124657, "postDate": "2016-06-20T23:01:19.050Z", "content": "Assuming the data has been generated off of real world location data a couple of thoughts came to me that I'd thought I'd sense check with the group:
\n\nCellphone location most probably is reported as CER ( at least Android which, assuming this data is again modelled on the real world, should be a significant subsection of the devices used to checkin). However, the figures do look like meters - especially if it is modelled on US cellphone location having to have a ceiling of 300 meters accuracy (FCC 94–102).
The accuracy jumps could be related to software updates (GPS upgrade?) Or mobile base station upgrades (most cell towers use software defined radios, so I could imagine a network wide upgrade or a rolling upgrade program over time - that could fit well with some of the observed improvements in accuracy in time.) The other realworld issue with cellular networks (well GSM at least) is that in a 3G network the physical cells grow and shrink dependent on density of subscribers - so coverage will change dependent on how many people are in a cell.
Given these sorts of hypothesis, I'd expect to see a strongish correlation (and hopefully causation) between location and accuracy, as the network coverage (either wifi or Cellular) shouldn't differ massively between device (again assuming the time column maps to a minute time unit, and thus unlikely to see a massive update / refresh cycle of devices in play). Indeed, it might be possible to derive / build some sort of coverage map based on accuracy - i.e. heat map of accuracy by location.
\n\nIf I get time before the end of the contest, I'll have a go at that.
", "rawMarkdown": "Assuming the data has been generated off of real world location data a couple of thoughts came to me that I'd thought I'd sense check with the group:\r\n\r\n1. Cellphone location most probably is reported as [CER][1] ( at least [Android][2] which, assuming this data is again modelled on the real world, should be a significant subsection of the devices used to checkin). However, the figures do look like meters - especially if it is modelled on US cellphone location having to have a ceiling of 300 meters accuracy (FCC 94–102).\r\n\r\n2. The accuracy jumps could be related to software updates ([GPS upgrade][3]?) Or mobile base station upgrades (most cell towers use software defined radios, so I could imagine a network wide upgrade or a rolling upgrade program over time - that could fit well with some of the observed improvements in accuracy in time.) The other realworld issue with cellular networks (well GSM at least) is that in a 3G network the physical cells grow and shrink dependent on density of subscribers - so coverage will change dependent on how many people are in a cell. \r\n\r\n\r\nGiven these sorts of hypothesis, I'd expect to see a strongish correlation (and hopefully causation) between location and accuracy, as the network coverage (either wifi or Cellular) shouldn't differ massively between device (again assuming the time column maps to a minute time unit, and thus unlikely to see a massive update / refresh cycle of devices in play). Indeed, it might be possible to derive / build some sort of coverage map based on accuracy - i.e. heat map of accuracy by location. \r\n\r\nIf I get time before the end of the contest, I'll have a go at that.\r\n\r\n\r\n [1]: https://en.wikipedia.org/wiki/Circular_error_probable \"CER\"\r\n [2]: http://gis.stackexchange.com/questions/25115/probability-of-ios-android-device-location-accuracy\r\n [3]: http://www.gps.gov/systems/gps/modernization/", "votes": 1 } ], "comments": [], "raw_markdown_by_id": { "124657": "Assuming the data has been generated off of real world location data a couple of thoughts came to me that I'd thought I'd sense check with the group:\r\n\r\n1. Cellphone location most probably is reported as [CER][1] ( at least [Android][2] which, assuming this data is again modelled on the real world, should be a significant subsection of the devices used to checkin). However, the figures do look like meters - especially if it is modelled on US cellphone location having to have a ceiling of 300 meters accuracy (FCC 94–102).\r\n\r\n2. The accuracy jumps could be related to software updates ([GPS upgrade][3]?) Or mobile base station upgrades (most cell towers use software defined radios, so I could imagine a network wide upgrade or a rolling upgrade program over time - that could fit well with some of the observed improvements in accuracy in time.) The other realworld issue with cellular networks (well GSM at least) is that in a 3G network the physical cells grow and shrink dependent on density of subscribers - so coverage will change dependent on how many people are in a cell. \r\n\r\n\r\nGiven these sorts of hypothesis, I'd expect to see a strongish correlation (and hopefully causation) between location and accuracy, as the network coverage (either wifi or Cellular) shouldn't differ massively between device (again assuming the time column maps to a minute time unit, and thus unlikely to see a massive update / refresh cycle of devices in play). Indeed, it might be possible to derive / build some sort of coverage map based on accuracy - i.e. heat map of accuracy by location. \r\n\r\nIf I get time before the end of the contest, I'll have a go at that.\r\n\r\n\r\n [1]: https://en.wikipedia.org/wiki/Circular_error_probable \"CER\"\r\n [2]: http://gis.stackexchange.com/questions/25115/probability-of-ios-android-device-location-accuracy\r\n [3]: http://www.gps.gov/systems/gps/modernization/" } }