The definition of "unique visitors" is quite up to debate I think - would you cound the same person twice if they come in the morning and evening, for example? What if the same person accesses a site from home and from work? Does it count as two different unique visitors? For a given month's unique visitors, would you count the same person twice if the person comes in week 1 and week 3? They are some of the conceptual issues.
There are a ton of operational issues. We have clues such as IP address, cookie, and the information about the PC/device (screen, OS, etc.) None of these will correspond to a person for sure. Nielsen has another information for their panel, and is able to tell if two visits are by the same person even when IP addresses, cookies, and PC info do not show they are. In most of the case, web publishers cannot. For this part, panel data based on some account based identification works really better.
But then, I am not sure if panel data is safe from errors resulting from shared PCs and devices. I imagine that is not completely out of control, but a tough issue to address. 1) perhaps you have to ask the panel to log on each time with a password - preventing others from being tracked. But then it will lose track of data if the panel simply does not log on for laziness or forgotten password; or 2) perhaps they can ask the tracking software to be installed only on a person's un-shared PC, un-shared Windows XP account, or simply limit the panel to those people who do not share a PC. In any case, that will compromise the representativeness of the data, especially from families.
Another weekness of a panel data is that it is incomplete in the simplest sense - just a small sample of large and diverse population. Unless you have a very good sampling methodology based on good knowledge of how diverse users behave online differently, small-size panel data will overrepresent some groups and underrepresent other groups, producing skewed estimates for some or many web sites (skewed upward or downward).
I think that these are somewhat inevitable, and has always been the problem with the TV rating. Indeed, TV station people and advertisers did not believe in rating numbers, but they used them anyway because there were no alternative. In case of web stats, however, there are competing sources, and there are debates, and that might, in the end lead to a little better methods.
Looked from another angle, this is because the data collection is less costly in case of web statistics.
by
infosoc
2008-02-21 01:50
web traffic stats
·
statistics
·
Nielsen
·
comScore
http://www.news.com/Doing-a-number-on-Web-site-traffic/2100-1024_3-6231334.html?tag=st.num
-
cached
-
mail it
-
history
simpy
