**Summary**

My estimate of the median ADSL2+ broadband download speed *across* Australia, and amongst *all* ISPs, is between 8 and 9 Mbps.

—

**Some background noise**

Back in January I queried Internode’s and iiNet’s claim that the *median* ADSL2+ download speed in Australia was 12Mbps. Their estimate was based on a “heatmap” using data collected from subscribers connected to inner metropolitan Sydney exchanges. When you’ve been working with data as long as I have you get a feel for the stuff. So iiNet’s/Internode’s conclusion that *half* of all subscribers were achieving speeds of 12Mbps *or better* just didn’t sit right with me. My gut said that it should be significantly less than that. In fact a crude “ring model” based on simple concentric circles indicated that only a quarter of subscribers should be getting those kinds of speeds.

These days I have the mind of rusted iron *thingamajig*, so I went on to forget all about the subject. That is, until a few days ago, when the topic generated some very useful feedback in the comments section of the post – courtesy of Matthew Moyle-Croft, Network Engineer at Internode. Matthew explained that my ring model failed to consider different gauges/attenuation of copper cables as distance from the exchange increased.

—

**Flawed models**

Thanks to an earlier comment from “AustCC” I had already realised that the model also failed to take into account overlap between exchange boundaries. The impact of overlapping boundaries and distances between exchanges on speeds is crucial. I was quite foolish in not considering it. Indeed, in a March 2008 presentation to the *National Consumer Congress*, iiNet themselves illustrated how overlap and exchange distance can impact on median download speeds (the relevant slides are number 26 and 27). In inner suburbs, exchanges are typically only 2km apart with a median download speed of ~15Mbps. Outer suburbs exchanges are spaced further apart with a median download speed of ~9Mbps. Looking at the Internode/iiNet heatmap, you see the area of Sydney that has been selected appears to be more of the “inner suburb” variety. This could potentially lead to a biased result. Now before anybody accuses me of being “rude” again, I mean biased in the statistical sense… as in non-representative. I didn’t and don’t think iiNet and Internode are lying by fabricating or selectively choosing results.

—

**Other ISPs**

What about other ISPs? I read the fine print of several ISP websites to see what these companies were claiming about *their* expected ADSL2+ download rates. TPG, for example, state that “around 70% of TPG ADSL2+ customers can achieve speeds greater than 10Mbps”. Exetel, as another example, report an “average speed obtained from 100 test users was 8Mbps down.”

So between a flawed model and various competing claims (from various competing companies), where does that leave my hope to estimate broadband download speeds in Australia?

—

**Survey results**

Of course the most authoritative source of customers’ download speeds comes from *actual users* of the technology. In particular, the thousands of respondents to the Australian Broadband Survey, administered every year by Whirlpool. Around 10,000 *very tech savvy* subscribers reported their own ADSL2+ download rates in late 2008. This, in essence, represents the “wisdom of the crowd” (which I happen to hold in very high regard). The results have been reproduced below:

**“What line/sync speed are you getting from your ADSL service? (this question was asked of ADSL 2+ subscribers)”**

speed | no. respondents | % respondents |
---|---|---|

<2 Mbps | 464 | 4.6% |

2-5 Mbps | 1,889 | 18.6% |

5-10 Mbps | 3,453 | 34.0% |

10-15 Mbps | 1,598 | 15.7% |

15-20 Mbps | 1,430 | 14.1% |

20+ Mbps | 550 | 5.4% |

don’t know | 770 | 7.6% |

Source: whirlpool.net.au/survey/2008

—

** ADSL2+ download speed… percentile estimates**

With a bit of statistical chicanery, it’s possible to generate estimated percentiles from grouped data. For example (excluding the “don’t knows”) 9384/2=4692 people were below the median speed, which falls in the 5-10Mbps cohort. 464+1889=2353 achieve speeds of <5Mbps, leaving 4692-2353=2339 people. Assuming an even distribution throughout the [5,10) Mbps range the estimated median is:

M = 5+5*(2339/3453)=8.4Mbps.

Applying the same logic to the remaining estimated percentiles yields the following results:

percentile | ADSL2+ download speed
(minimum, estimated) |
---|---|

10th | 18.6 Mbps |

20th | 15.4 Mbps |

30th | 12.4 Mbps |

40th | 9.8 Mbps |

50th (median) |
8.4 Mbps |

60th | 7.0 Mbps |

70th | 5.7 Mbps |

80th | 4.2 Mbps |

90th | 2.8 Mbps |

—

**Conclusion**

I continue to hold Internode and iiNet in the highest regard. However, the results in the above table match my statistical “intuition” much more closely than *any* claim from *any* company. **I’m confident that the median ADSL2+ broadband download speed across Australia, and amongst all ISPs is, in reality, somewhere between 8 and 9 Mbps.**

Filed under: ISP statistics, research, statistics | Tagged: ADSL2+ speeds, broadband speed, statistics | 8 Comments »