One of the news media's approaches to spicing up the particularly boring upcoming election is to report breathlessly on how new technology is changing democracy. News.com.au has an article today about the supposed usefulness of Google Trends to track what Australian voters are thinking. Excerpt:
Metering the internet search terms used by Australians, Google's special tool can compare voteres' searching habits. 'Liberal' is currently more searched for than 'Labor', except in the ACT.
Here's my explanation: 'Liberal' is not just the name of a political party, it's also used in lots of other contexts: liberal democracy, liberal arts, small 'l' liberals.
'Labor', on the other hand, is a word that, in Australia, refers to primarily to the political party. Anywhere but America, if you're not referring to the political party, you spell it differently: 'labour'. And 'labour', even if some Aussies do spell it 'labor', doesn't have as many popular meanings as 'liberal': there's childbirth, there's toil, and there's the labour force.
So, if the general public weren't searching much for either of the parties, you'd probably expect 'liberal' to be a more popular search term than 'labor' in Australia. And that's just what you get. The exception is the ACT, where so many people are in government, and might be researching their possible new bosses.
Conclusion: comparing search terms for 'liberal' and 'labor' tells us nothing new.
The article goes on to compare searches for 'health care' versus 'economy' or 'climate change'. If you go to Google Trends, you can see some of the problems with this approach.
The first problem is sample size. The article sez:
While the number of Australians using the search engine to find information on many key election issues has fluctuated throughout 2007, health care was the most consistently searched for term.
But people in the Northern Territory, the ACT and Tasmania are searching more for climate change than any other election issue.
Now, we don't know exactly what terms the journalist was tracking here, but we can do a very rough search of our own to get an idea of the order of magnitude we're talking about.
Say we compare "climate change", 'economy', and "health care" for Australia for the past year:
Link. Google Trends does not tell you how many people actually did a search; they give you a series of graphs with an unlabeled axis for the number of searches. But we can see from the shape of the graphs that we are probably not talking about tens or hundreds of thousands of searches.
It looks to me like it's more on the scale of dozens of searches. It's easier to get a sense of the numbers if you look at individual states. Climate change wins out in Northern Territory because nobody searched for the other two terms. In Tasmania, there's an exactly equal number searching for 'economy' and "health care", and that number is exactly three-quarters of the number that searched for "climate change". My bet is four people searched for "climate change" in Tasmania.
With the bigger states, it's harder to judge the numbers. But look at this plot for the term 'economy' in New South Wales:
link. The line jumps up and down erratically, and drops flat to zero for weeks at a time; my guess is that the rate of searches for 'economy' could be as low as between zero and thirty in a week.
This doesn't bode well for making comparisons between search terms; as a rule of thumb, if your sample size is as small as the readership of Brlogsbane, your statistics are useless.
Let's assume the numbers of searches are much bigger than I've estimated. There's still the problem of assuming that people's search terms give an indication of what's on their mind. Thing is, there are a million reasons to search for any particular term (are you interested in the issue of health care, or are you trying to find health care?).
Here's the searches I've made in the past few days, and why I made the searches:
- "inka essenhigh": I found an old post-it that I'd written "Inka Essenhigh" on and forgotten about
- "baked beans choice": I've been digging baked beans lately and wanted to see which brands Choice magazine recommended (SPC and Bi-Lo, if you're interested)
- "johnny luther htoo": I heard the incredible story of Johnny and Luther Htoo, the guerrilla leaders in Burma, on an archived episode of This American Life, and wanted to find out more
- "dirigisme": I read something that mentioned dirigisme and couldn't remember what it meant
- "black snake diamond role": Black Snake Diamond Role is an album I've wanted to buy for months, and was trying to find a good price for
Do these things say anything about what I'm concerned about? Hell, no.
You could probably gather from 'dirigisme' that I was reading about economics. If I searched for other words on that subject ('inflation', 'taxes', 'central bank'), I might be counted among the Aussies with the economy on their minds - which isn't the case.
If instead of "johnny luther htoo", I'd worded my search as "htoo burma", I'd be counted among the people following what's going on in Burma - which is true, but a coincidence.
Maybe more important is what I didn't search for: almost everything that's been on my mind lately.
If Google Trends was counting tens of thousands of searches on Labor or the economy, it might be useful as a blunt-instrument tool to get a sense of what internet-users were thinking about in politics. But it's not; it's even more useless than Newspoll!