Some Site Statistics

[whole diary]

[grahame]

I keep an eye on forum statistics ... and yesterday morning, I added Google Analytics into the Coffee Shop in order to get an idea of the geographic spread of visitors.  Here are the top six locations from which this site was visited:

Code:

1.   London   119   6.49 
2.   Bristol   37   10.46  
3.   Cardiff   21   8.29  
4.   Plymouth   20   6.80 
5.   Oxford   18   6.67
6.   Bath   17   20.12

Reading the data ... an example row:
Plymouth was 4th most popular, 20 different visits averaging just under 7 pages viewed per visit.

Of a total of 481 different visits, 457 were from within the UK^▸ (United Kingdom).

It's good to see hundreds of visitors in what was probably just a typical day, and people sticking around looking at multiple pages too.   I suspect there's quite a few commercial sites who would like real visitors coming through at these rates, and it does confirm that what's posted here gets seen / read / answered.

[devon_metro]

How accurate is Google analytics w.r.t location of users? I know my Bristol address thinks i'm in Ealing London which might skew the data!

[grahame]

How accurate is Google analytics w.r.t location of users? I know my Bristol address thinks i'm in Ealing London which might skew the data!

It gives a pretty good general idea - 4 out of 5 are likely to be within 20 miles, and 19 out of 20 (or even better) correct by country;   those are figures I've seen mentioned and look sensible with the data involved.  I also track our own company high volume web site and see very different patterns indeed. All of which confirms a general statement that "the overwhelming majority of people who visit our site are doing so from the area served by trains operated by First Great Western".

Here, as an interesting comparison, is our own company website where we added analytics for some pages starting about 3 weeks ago, and have built up since - you'll see a very different pattern within the UK^▸ (United Kingdom):

Code:

1.	London	2,213
2.	Melksham	645
3.	Bristol	379
4.	Cambridge	207
5.	Birmingham	172
6.	Manchester	172

And overseas too, for example:

Code:

1.	San Francisco	449
2.	San Jose	268
3.	Sunnyvale	260
4.	San Diego	249
5.	Los Angeles	201

[LiskeardRich]

The location stats will be skewed at the moment, as all bt broadband users in devon and cornwall currently report as bristol due to technical issues. This problem is unique to BT users only

[Btline]

Any interesting locations in the UK^▸ (United Kingdom) for high usage? i.e. out of the FGW^▸ (First Great Western) network

[EBrown]

Any interesting locations in the UK^▸ (United Kingdom) for high usage? i.e. out of the FGW^▸ (First Great Western) network

Well I'm generally logging on from Kent, Canterbury. Although KentMAN, Janet's (Joint Academic NETwork) RNO (Regional Network Office) I believe is Ashford, Maidstone or Medway not Canterbury, so that may be the routing point. (Techno-babble for fast academic network - although we should be moved onto SuperJanet5 soon; which is 1Gbps not 100mbps as a user speed.)

I also log on from a server in San Diego occasionally!

That provides a little data skew.

[bobm]

I think BT Broadband is notorious for skewing locations. Although my broadband is in Swindon it shows as Milton Keynes. Sky on the other hand is very accurate and if I am in Bracknell it says I am. It's very annoying because I have been trying to get some location data from my own websites and it is hard to be confident in the results. (Not that I have anything like the programming skills of grahame!) 

[Brucey]

TalkTalk shows me as being from Brighton on Google Analytics, when we are just north of Portsmouth.

I have a lot of visitors from Milton Keynes, who are on Orange (landline, not mobile).  I could never work out whether the location was skewed or because I have occassionally promoted my website to people who live near Milton Keynes.

[devon_metro]

Another interesting one perhaps, 3G connections? Do they provide the location of the mast providing the signal or whereever the Mobile operators servers are based?

I would try but I don't get 3G coverage in deepest darkest Zummerset

[grahame]

A slightly different database says 2/3 are within 25 miles:

http://www.maxmind.com/app/city_accuracy

We can't rely on accurate placement to a town or city for an individual hit, but the statistics give us a general picture and become clearer when compared to the same diagrams / stats for other sites (I'm now running Analytics on 4 different domains). During next week when I have a full 7 days of data, I'll come back with a report of that period and some elements of inconsistency will be smoothed out.   Mind you, with Easter in the data period, I won't be claiming a "typical week" as I think Dr Beeching once did with the week before or after Easter, and look at the legacy he left us!

[grahame]

During next week when I have a full 7 days of data, I'll come back with a report of that period and some elements of inconsistency will be smoothed out.   Mind you, with Easter in the data period, I won't be claiming a "typical week" ...

9th (Easter Monday) to 15th April inclusive:

1.   London   1,408   5.51
2.   Bristol   372   13.65
3.   Oxford   247   7.04
4.   Reading   218   4.67
5.   Birmingham   185   5.12
6.   Bath   179   23.38   
7.   Plymouth   162   7.83
8.   Cardiff   129   6.36
9.   Southampton   112   8.66
10.   Worcester   92   7.10

[TonyK]

TalkTalk shows me as being from Brighton on Google Analytics, when we are just north of Portsmouth.

If I use IE8, I am in Cardiff. If I use Opera, I am in Poland. If I look out of the window, I and in Bristol.

Dziekuje! (Or should I say Diolch yn fawr?)

[devon_metro]

I'm apparently in Crawley now, my home address in Bristol has recently been showing Doncaster!!

[thetrout]

Some members are right in saying that Data can be skewed.

It all depends on which DSLAM (The system that your Home/Work Router terminates to when using ADSL/VDSL etc) and also who your provider is. For example, If you're in the South West and on PlusNet for example. Your connection almost all the time will show up as in London (Most likely the Global Switch Data Centre - where a few ISP's have their terminating kit).

Virgin Media is reasonably accurate, as they have street cabinets, that all route to a major node. So if you're for example the Frome, Westbury or Warminster areas, you'll terminate to the Trowbridge Node. So you'll normally show up in Trowbridge or as near as.

BT is a nightmare of an Internet Network. As they are known as the "Principle Operator" of the PTSN (Public Telephone Switching Network) Exchanges, they can terminate your connection anywhere that there is capacity at the time. So you may be in Exeter for example, but your connection may show up somewhere really obscure and miles away from where you actually are.

Also the majority of ISP's give out Dynamic IP Addresses. So if you unplug your router and plug it in again, the chances are you'll get a completely new IP Address, which could show up as somewhere completely different.

o2 Broadband has a habit for skewing geographic data too. My backup connection at my office in Frome normally shows up in Bracknell or Slough - of which I am nowhere near either.

Interestingly enough though. It depends on the operator and if they have an LLU agreement with BT in the area. What this is, is where the ISP have installed their own equipment and backbone connections into a BT Exchange for their customers. TalkTalk do this for example. If you're in a TalkTalk LLU area and you want to switch just your broadband to another ISP. It doesn't work, as TalkTalk use their own backbone lines and servers for the telephone lines and in actuality, all they pay BT Openreach (Company who own the telephone network; a completely different company from BT Retail and BT Business Retail) for is the rental for the copper wire from the BT Exchange to the customer premises (There's an acronym for that too, think it's CPE but it's 1:58AM and my mind is fried from staring at Exchange Server Logs Files all day ) They also pay a service fee for the line in case a problem develops on the line between the CPE and Exchange and they are required to actually fix it.

3G is a different kettle of fish altogether. My understanding (Although this may have changed since) was that you don't have a "Public IP" on a 3G network. What actually happens is the provider is running a massive NAT network. There is always 1 PoP (Point of Presence) depending on the location. I did a test on this on Hutchinson 3G. Travel from Taunton to London and the last time i did this my terminating location was Slough - which didn't change. This would make sense as there is a huge amount of Data Centre's and Internet Pipes in that area. Rackspace (Mods please note that this isn't an advertisement as I am not connected with them in any way) have one of their 9 Cloud Data Centres in Slough. You may have seen recently the "Bang Goes The Theory" regarding IT and Mobile Security, where they did an interview with a Cloud Hosting Provider. Although the program said they couldn't declare the providers name... Anyone in the Internet Connectivity area of IT knows it was Rackspace!

Incidentally, last year Thames Water dug up the Fibre Cables in Slough. This caused everyone on traditional BT 20cn and 21cn connections i.e. connections that use the BT Network to effectively piggyback to the ISP network (As is the case with an Internet Connection on any rural exchange or exchange with no LLU providers) to completely crash. It took out the entire southwest. However, users on LLU services such as o2 and TalkTalk were mostly unaffected as they use their own equipment.


Sorry, I really went on there didn't I! But I run a business which focuses on Internet Collaboration hence why I pick up this useless information... Which one day may actually prove useful to somebody!

*edit* removed a typo which was a typo, but said typo manifested itself as a rather rude word which has happened twice this morning!

[grahame]

A fascinating and detailed insite there into the various issues involved - and to add that looking at raw log files on a server, it's sometimes possible to work out the connection type, and sometimes not.  All of which means we can't trust individual entries, and the who setup gets biased towards certain places. Where it is useful is in comparison between sites, or between two parts of the same site - looking not at the actual numbers but at the relative proportions.  Often, that will confirm what you hope (such as "most of our users seem to be in GW^▸ (Great Western) land, compared to a nationwide site") but occasionally it will point out an apparent oddity which can be worth investigating.