Greg Sterling posted this http://bit.ly/4xk3xm and I reposted a different compete.com comparison switcing out a few sites. http://bit.ly/6X28ao ]]>

comScore is basing numbers off of a sampling of a subset of total internet users. When they cite “Americans conducted 14.7 billion searches…” I think they’re possibly taking their sample set of searches, then multiplying by what they estimate the ratio of their sample set of users compared with overall number of internet users to come up with an estimate of total American search numbers.

I could be wrong – it may be the total number of searches from their sample set, in which case the actual numbers would be larger.

Either way, comScore’s numbers aren’t the actual, total numbers of Google searches/searchers. Google’s numbers would be based on actuals that they see in their data center. Their ratio of 1:13 is likely also imprecise by some amount — it’s likely roughly rounded to whole numbers 1:13.

The amount of error in comScore’s estimate, multiplied by the difference in Google’s rounding will result in a high likelihood of really large error, considering the large quantities involved when you multiply these numbers. And, you’re multiplying by yet another estimate — the number of users who don’t go past page one of results.

So, if you’re going to quote the multiplied number in a blog post, just acknowledge in passing that there’s likely some significant error involved, since comScore’s #s are based on an estimate projected from a representative panel combined with other sources, plus the ratio Google mentioned was likely rough, plus the %age users sticking on page one is also roughly estimated. (No need to detail all the sources of error in the computation.)

I’m not snubbing your estimate — I think your estimate is worth blogging about, because it shows the amount of usage going on in Maps and it will be one of the only estimates anywhere that gives some idea of how many “pageviews”/”impressions” may be happening in that section of Google!

I don’t know statistics well enough to compute the +- error estimate with your final number. I can just see that there’s significant uncertainty when multiplying these multiple variables, and I think it’s best to acknowledge that when publishing the number you’ve arrived at — it will take the wind out of anyone that would be inclined to point out the fuzziness before they make any comments on the article.

The only estimate that I am confident of in this comparison? **A lot**!

Until we can get compare or hitwise to study it from a consistent set of data we won’t really know.

]]>Good logic..but I think you need to filter between subsequent searches…and subsequent searches that involve a map.

]]>from Google’s post:

Proportion of Google users in the United States making more than one query per day: 7 out of 10

so approx. 70% are repeat. how does that play into the comparison?

27, 19, and 10 million unique visitors for the 3 referenced sites in December versus about 1 billion local searches in Google.

that’s comparing the number of searches to the total unique visitors of the other 3 sites. we need to compare the unique numbers against uniques not total searches.

if 70% (of the 970 million) are making multiple searches, then is it fair to say the unqiues for local searches is 270 million not 970?

nonetheless, they still dwarf everyone, but i am trying to make it an apples to apples comparison.

am I missing anything?

]]>