The first report was the most recent quarterly report from Akamai. They rated the US as 16th globally with average landline broadband download speeds at 12.6 Mbps. This number is up 10% from the previous year’s average speed of 11.5 Mbps. Akamai derives their speeds from looking at millions of actual downloads around the world.
At about the same time the FCC released its latest report on broadband and they said that the average download speed in the US for September 2014 is now 31 Mbps, almost triple the 10 Mbps reported in March of 2011. That’s a huge disparity with the FCC’s numbers at almost 2.5 times higher than what was measured by Akamai. How can the numbers be that far apart?
Akamai is measuring the actual download speed of millions of transactions on the Internet. This is how fast things are actually downloading and, because of the huge number of transactions that they look at, the number should be very accurate.
The FCC on the other hand is using a very odd measuring technique. First, they rely on speeds that come from 5,000 homes in the US that have volunteered to have their speeds measured. It’s pretty obvious in looking at the FCC results that these homes are not a random sample. I would suspect they aren’t measuring too many homes with dial-up, for example. If their sample is not random, then the results of the FCC tests cannot be multiplied out to accurately represent the whole country. That is one of the basic precepts of statistical sampling.
Another reason for the difference is that the FCC is basically looking at speed tests and not at actual downloads. Speed tests measure the highest speed that can be achieved to a specific test site created for that purpose. The FCC further says that their sample shows customers’ actual speeds are not much different than subscribed speeds. But my recent experience is that this disparity still widely exists. For example, in helping a western city look at broadband issues, we recently asked citizens there to take the Ookla speed test and report to us the results of the speed tests along with the speed they are paying for. The results were eye-opening with about 80% of customers seeing speed test results much slower than what they were subscribing to. In some cases the results were drastically different with households that subscribe to 100 Mbps services seeing actual speeds under 20 Mbps.
Another difference between the two sets of numbers is the little-understood fact that you can’t download anything faster than what the Internet is sending you. Somebody with a gigabit fiber connection from Google is going to see the same download speed from watching a Netflix HD movie as somebody on a 20 Mbps cable modem connection. Netflix doesn’t care what connection you have and sends the same stream to everybody, as does a lot of the Internet. Akamai is measuring this real speed and not the theoretical speed from speed tests.
The problem with the FCC’s method from a mathematical perspective the FCC’s method will skew the average towards the faster speeds in the sample. In a very simple example, if you sampled one gigabit customer and nine dial-up customers you would say that the average data speed was a little over 100 Mbps. While that would be true for those ten people, that is obviously not a number that has any practical relevance.
What is troubling is that the FCC makes policy decisions based upon the results of these tests. For instance, they routinely publish the percentage of homes that now meet their new 25 Mbps download standard to be considered as broadband. On the surface there is nothing wrong with what the FCC is doing, and for those 5,000 customers the reported speeds really is the average for that group. The problem is that those results show a different Internet than is really out there. For example, the FCC just also declared that 39% of the homes in the country can’t get real broadband at all. If the FCC is going to report an average speed for the whole country, those homes ought to be represented somehow in the the average.
There might be help on the way since the FCC is finally starting to require large ISPs to tell their customers the truth about their networks, including actual download speeds. There are new reporting formats circulating now at the FCC that the ISPs are going to have to provide to every customer.
There is one promising take away from the FCC numbers which is that the large ISPs are increasing customer speeds. 2016 promises to be a year when many households will see a boost in speeds. I know Comcast just recently upgraded my speed, with no announcement, and my 50 Mbps cable modem now seems to regularly be getting 88 Mbps download. Almost every big cable companies has announced initiatives to increase customer speeds this year, across the board.
But this difference in speed reports does remind us that you always have to dig a little deeper when you see a statistic like average download speeds. There are different ways of measuring speeds and the FCC way of measuring speeds is suspect in several ways. What they published is not an untrue number for the way they measured it – but the number is just not relevant when talking about the US broadband experience.