This week I got a detailed response from Statistics Canada staff to my inquiry about their Canadian Internet Use Survey (CIUS): “Why does your report make no mention of how you define broadband?.”
(Slap forehead.) I learned a decade ago why you can’t define terms like “broadband” in a gen pop survey. At the time, I co-designed a series of national onliner surveys and discovered that if you ask respondents even slightly technical questions, they will be unable to answer. This constraint applies to goods such as personal computers and anything about their specs (clock speed, RAM, model, age, etc), as well as to service providers like ISPs (apparently quite a few Canadians used Loblaws as their ISP back in 2000).
Dumb question, interesting answer
Here’s how the Stats Can response begins:
The term ‘broadband’ is not defined in the Daily report you list since the survey does not use the term. Instead, the survey asks respondents about their connection type (telephone, cable, satellite, television, wireless or other). If the respondent indicates they use a type of connection at home other than cable or satellite, they are then asked a follow-up question: ‘Do you access the Internet at home using a high speed connection?’. (Cable and satellite connections are assumed to be high-speed). As such, the data on high-speed are self-reported.”
It made no sense to try to get respondents to group themselves around some speed threshold – 1.5 megs or any other – since the majority had no idea how to answer. This is pretty ironic given that what people buy from their ISP is a speed tier, but Stats Can tested the assumption:
Survey testing revealed that the majority of respondents were not able to identify precisely the speed of their connections, or whether it fell above or below a threshold such as 1.5Mbps. However, most expressed that high speed connections include connection types that are “not dial-up”.
Okay, but there’s a problem here: Stats Can conflates “broadband” and “high-speed.” The definitional dilemma this creates is that the CRTC still makes a distinction between these two terms: the former starts at 1.5 megs, whereas the latter is pegged at 128 Kbps.
Furthermore, the CIUS assumes cable and satellite connections are both high-speed. The decision to put them together may tidy things up methodologically. But the two platforms make strange bedfellows. Not only is DOCSIS 3.0 capable of downlink speeds of 50 to 100 Mbps, way faster than satellite; it also beats the pants off satellite in the uplink when a slow architecture like telco-return is deployed. Most of all, geostationary satellites have latencies that put them completely out of the running for important apps like telemedicine – three or four times higher than the 250 milliseconds it takes a signal to travel 22,236 miles into space and back, i.e. 900 to 1,000 milliseconds or more. Compare that to the 60 milliseconds or less required for emerging apps like consumer telepresence. (One indicator of how far out of touch we are is that in the CRTC’s 2010 Communications Monitoring Report, we find not a single mention of the term “latency.”)
Intensity of use – my kind of concept
Stats Can also pointed me in the direction of a working paper co-authored in March with Catherine Middleton of Ryerson University: Intensity of Internet Use in Canada: Understanding Different Types of Users.
The paper presents a series of formal statistical models for characterizing Internet usage in Canada. Like my idea of confining the onliner population to daily users, rather than occasional users, the paper starts with the idea that (daily) frequency of use can be combined with hours online (five-plus weekly) to create a profile of user “intensity.” Strange to say, this merge yields a high-intensity population of about the same size as I calculated in my previous post: 53% (my “core” online group), with the paper’s high-intensity group coming out at slightly less:
Of note here is the fact that fewer than 50% of Canadian Internet users were characterized as high intensity users in 2005 and 2007. Some would argue that the threshold for categorization as a high intensity user is quite low, but the data show that the majority of Canadian Internet users did not choose to use the Internet daily and for more than five hours per week from home. […] Based on the classification developed for this study, the majority of Canadians would not be categorized as high intensity Internet users in 2007 (p.8: my emphasis).
I’ll leave it to the reader to look at other details in the paper. I would like to comment on two points.
First, the reply email says the paper “discusses the definitions of high-speed and broadband used by the CRTC for example, and why these data should not be directly compared with those coming from the CIUS.” Now we know what’s brewing behind those closed doors in Ottawa: nothing. This is what the qualifying footnote (p.6) actually says:
The CRTC differentiates between high speed and broadband Internet connections. High speed connections are defined as those at or above speeds of 128 Kbps. Broadband connections are those at or above speeds of 1.5 Mbps. In 2008, 52% of Canadian households had a broadband connection, according to CRTC data. These subscriber statistics are based on residential data gathered from suppliers offering Internet services. The CIUS measures individuals and uses the term “high speed”. On the 2005 and 2007 CIUS, individuals were classified as having high speed access if they had a cable or satellite connection at home, or another type of connection (telephone, television, wireless, or other) that they identified as being a high speed connection. For these reasons, the two sources should not be directly compared.
And speaking of speeds… We know the CIUS makes no attempt to segment users by speed, because surveys can’t measure household line speed. That’s one reason why I suggest a modest change in terminology from “high speed” for anything that isn’t dialup… to always-on.
My suggestion is more than notional, because of what I read in the paper about using “speed” as a predictor of intensity of use. It can’t be so used, say the authors:
In 2007, about 88% of Canadian home Internet users accessed the Internet with a high speed connection, up from 80% in 2005 (CIUS). As such, most Canadian Internet users should now have the technical capacity to take advantage of the benefits that the Internet can offer. But as previous analysis of the Household Internet Use Survey (HIUS) data demonstrated (Middleton and Ellison 2008), simply having access to a high speed Internet connection does not mean that a household demonstrates high engagement with the Internet, as measured by intensity or scope of Internet usage. Consistent with this observation, the data above reveal that many Canadians are not high intensity Internet users, regardless of the speed of their Internet connections (p.9).
I disagree with the assumptions made here. It defies all common sense to claim that, controlling for other variables, a user on a high-speed connection running at 5 Mbps would be no more inclined to behave expansively online than a user on a high-speed connection running at 128 Kbps – 40 times slower. The authors make a point in their methodological notes of distinguishing continuous from dichotomous variables – and their framing is that respondents are either high-speed or dialup.
There are at least two different ways to attack this assumption. On one hand, we have well determined ways of knowing how much bandwidth certain applications and services need before they stop functioning. For example, high-resolution videoconferencing (1920×1080) needs symmetric bandwidth of at least 5 Mbps (i.e. down and up). True HD TV (1440×1080 interlaced) requires 15 Mbps in the downlink. We can quibble about how far down the foodchain you can take us with compression and still pretend it’s “HD,” but at some point below, say, 8 Mbps, you aren’t watching HD any more.
These technical thresholds make me wonder why the authors talk about what they term “the benefits that the Internet can offer.” You can’t benefit from real-time video games, or tele-surgery, without very low latencies. If you’re on a satellite connection, with 10 to 20 times more latency than is tolerable, those are two benefits you’ll never enjoy.
Moving to less solid ground, I’m equally convinced bandwidth will turn out to be highly correlated with intensity of use, in terms of both frequency and total time spent. It stands to reason you’d be more likely to try video communications with your friends and other emerging applications if you have the bandwidth, and the more bandwidth the better. Bandwidth has two important attributes: one being the speed with which files download, the other being the quality of streaming media, VoIP and other time-sensitive applications. These exert a strong pull on end-users (though as YouTube proved billions of times over, you don’t need “quality” video to be a hit on the Web).
UPDATE: Just re-reading the FCC OBI report referenced below (on actual broadband speeds), and it says the following: "...the amount of hours online is correlated with the speed of a residential connection - higher purchased actual speeds correspond to more hours online per month" (p.5).
We aren’t ever going to get the data we need from surveys, any kind of surveys, nor are the ISPs going to give us what we need. The only way to understand how bandwidth and behavior are related is to deploy tracking boxes in thousands of Canadian homes, so we can measure actual line speeds under many different conditions.
The FCC did that last month and discovered the average American broadband home is getting – in actual use – about 50% of the bandwidth advertised by ISPs. The chart below shows the FCC’s estimated distribution across the US of users by measured (not advertised) download speeds. (See FCC OBI Technical Paper No.4: Broadband Performance, p.15).
The OBI paper says they found the US advertised downlink mean to be 7-8 Mbps, with the actual mean being 50% of that, or less than 4 Mbps. In its 2010 Monitoring Report, the CRTC says the weighted average download speed for all residential plans as reported to them by Canadian ISPs is about 5.9 Mbps. This is not only an average; it’s also based on data submitted by the ISPs. Take with salt. Using the US numbers, therefore, we get a Canadian mean for actual line speeds of just under 3.5 Mbps.
Anybody care to prove this math is wrong?