One of my clients is Sysomos, which has done two “Inside Twitter” reports that involved crunching data from 11.4 million Twitter accounts. We’re talking about in-depth analytics.
So, it’s a bit of a head-scratcher to see Pear Analytics publish a report suggesting that 40.55% of all updates are “pointless babble” based on a sample of 2,000 updates taken over a two-week period.
I’m not suggesting you need to look at 11.4 million Twitter accounts for a report to be credible but 2,000 is just lame because it hardly represents a solid sample.
Here’s the conclusion from the Pear report, which includes a strange shout out to a filtering service called Philtro:
“With the new face of Twitter, it will be interesting to see if they take a heavier role in news, or continue to be a source for people to share their current activities that have little to do with everyone else. We will be conducting this same study every quarter to identify other trends in usage.
Since Twitter is still loaded with lots of babbling that not many of have time for, you should check out the Twitter filter, Philtro. These guys can not only help you filter the noise, but will also be allowing you to store the tweets you are most interested in real soon.”
3 Comments
I'm not sure the maths is with you on this one Mark. I'm not a statistical sampling geek, but if we were talking of an opinion poll being conducted with a genuinely random sample of 2,000 then we'd be 95% confident that the findings were correct to +/- 2.2% or thereabouts. That's a smaller margin of error than on nearly all opinion polls which are published.
It wouldn't justify quoting a result to two decimal places (as in 40.55%) but it would be reasonable to say 41%.
So if their sample was a random (or close enough) set of 2,000 tweets I think their conclusion is ok. Unless there's something I've missed about how the sampling error calculation should be figured out in this case?
Mark,
Thanks for the feedback. I'm not a sampling person either but 2,000 samples seems pretty pretty small to get an accurate picture of Twitter.
In case you've not seen it – there's quite an interesting discussion at http://stephendann.com/2009/08/15/pear-analytics-… including some responses from one of the people who did the research.