4 ms·

It's terribly inaccurate, based on my experience. For the Feb 2011 (most recent available stat they have), they were wrong by -314%. They should make the fact t

by jggube 16y ago

It's terribly inaccurate, based on my experience. For the Feb 2011 (most recent available stat they have), they were wrong by -314%. They should make the fact that their service is inaccurate obvious; otherwise, they're misleading many people. Their data points are from 2 million people. In the U.S. alone, there are 230 million+ interet users: http://www.google.com/publicdata?ds=wb-wdi&met=it_net_user&idim=country:USA&dl=en&hl=en&q=internet+users http://www.google.com/publicdata?ds=wb-wdi&met=it_net_us... I only took basic stats courses in college, but I know enough to know that's nowhere close to a big enough sample size to be statistically significant.

gwern 16y ago
> I only took basic stats courses in college, but I know enough to know that's nowhere close to a big enough sample size to be statistically significant. Just like you can't get a statistically significant result about the 300 million Americans by phoning a few hundred people.
edanm 16y ago
"I only took basic stats courses in college, but I know enough to know that's nowhere close to a big enough sample size to be statistically significant." I am not a statistician, but I'm pretty sure you're wrong about that. The real problem isn't the small sample size (2 million is really large), it's probably that they're not representative.
earl 16y ago
If that were a random sample you'd be completely wrong -- that's more than enough to get extremely accurate numbers. What's more likely is that compete is based on a panel [1] [2], and the members of that panel tend not to include the type of people that read tech sites. [1] http://techcrunch.com/2010/02/25/compete-data/ http://techcrunch.com/2010/02/25/compete-data/ [2] http://www.quora.com/Where-does-Compete-get-its-data-from http://www.quora.com/Where-does-Compete-get-its-data-from

by jggube 16y ago

gwern 16y ago

> I only took basic stats courses in college, but I know enough to know that's nowhere close to a big enough sample size to be statistically significant. Just like you can't get a statistically significant result about the 300 million Americans by phoning a few hundred people.

edanm 16y ago

"I only took basic stats courses in college, but I know enough to know that's nowhere close to a big enough sample size to be statistically significant." I am not a statistician, but I'm pretty sure you're wrong about that. The real problem isn't the small sample size (2 million is really large), it's probably that they're not representative.

earl 16y ago

If that were a random sample you'd be completely wrong -- that's more than enough to get extremely accurate numbers. What's more likely is that compete is based on a panel [1] [2], and the members of that panel tend not to include the type of people that read tech sites. [1] http://techcrunch.com/2010/02/25/compete-data/ http://techcrunch.com/2010/02/25/compete-data/ [2] http://www.quora.com/Where-does-Compete-get-its-data-from http://www.quora.com/Where-does-Compete-get-its-data-from