Hello! Me again!
Have you heard of Media Metrix?
They claim to "deliver the most accurate and comprehensive reporting
of Internet behavior in the workplace using a sample of more than
100,000 panelists" although, 100,000 sounds to me to be a irrelevantly
small sample?
And again among the US college and university student population of 9
million they claim to have a "40,000-member sample" which again seems
too small to be significant?
Similarly, this claim:
"Complete Global Coverage Through comScore's 500,000-member non-U.S.
sample, Media Metrix XPC provides total worldwide data based on users
in more than 200 countries."
Sounds too small to provide what they claim.
Then they claim 2 million participants, at this URL
http://www.comscore.com/method/method.asp
But I am not a statistician, nor do we have one on our team, so I'd be
interested in your opinions.
Matt
Full Version: Media Metrix?
Matt Probert wrote:
| QUOTE |
They claim to "deliver the most accurate and comprehensive reporting of Internet behavior in the workplace using a sample of more than 100,000 panelists" although, 100,000 sounds to me to be a irrelevantly small sample? And again among the US college and university student population of 9 million they claim to have a "40,000-member sample" which again seems too small to be significant? |
The vast majority of statistical surveys involve only about 1000 or so
participants. I'd say, from a statistical point of view, that the sample
sizes you present are quite significant.
| QUOTE |
| Then they claim 2 million participants, at this URL |
They could have 2 million participants, but only sample 100,000 at a time...
Just remember - when you hear polls reporting people's opinions or political
positions, that's typically only about 1000 or so people being polled. And
the all-important "Nielsen" ratings uses only 5,100 families to determine
the television viewing habits of the entire country (see
http://www.nielsenmedia.com/)
--
Tony Garcia
Web Right! Development
Riverside, CA
www.WebRightDevelopment.com
"Tony" <[Email Removed]> wrote in message
news:[Email Removed]...
news:[Email Removed]...
| QUOTE |
| Matt Probert wrote: |
| QUOTE |
| The vast majority of statistical surveys involve only about 1000 or so participants. I'd say, from a statistical point of view, that the sample sizes you present are quite significant. |
Provided they are demographically weighted in a manner that reflects the
population they are purported to represent.
Martin Harran wrote:
| QUOTE |
| "Tony" <[Email Removed]> wrote in message news:[Email Removed]... Matt Probert wrote: The vast majority of statistical surveys involve only about 1000 or so participants. I'd say, from a statistical point of view, that the sample sizes you present are quite significant. Provided they are demographically weighted in a manner that reflects the population they are purported to represent. |
My point being that a sampling of 100,000 is certainly more significant than
the typical size sample, which was the question being asked.
--
Tony Garcia
Web Right! Development
Riverside, CA
www.WebRightDevelopment.com
On Tue, 12 Jul 2005 16:57:57 -0700, "Tony"
<[Email Removed]> wrote:
<[Email Removed]> wrote:
| QUOTE |
| The vast majority of statistical surveys involve only about 1000 or so participants. I'd say, from a statistical point of view, that the sample sizes you present are quite significant. |
Much larger than the usual size and plenty large enough to give good
data assuming that they have a somewhat representative sample. Size
is much less important than sampling method. You want to be sure that
they are looking at the right population (i.e., their sample isn't
mostly people very different from the group that you are interested
in), that they aren't relying on a self-selected group (i.e., a web
survey that only a select group of people will ever find), etc.
An example:
In the famous (at least on this side of the pond) polling debacle,
where the post-election headline read "Dewey defeats Truman", the main
reason for the erroneous prediction was that they were using telephone
surveys at a time where many less affluent people didn't have phones,
and therefore weren't sampling an important segment of the population.
Increasing the sample size wouldn't have helped at all, because that
sampling bias would still have been there.
--
MGW
"MGW" <[Email Removed]> wrote in message
news:[Email Removed]...
news:[Email Removed]...
| QUOTE |
| On Tue, 12 Jul 2005 16:57:57 -0700, "Tony" <[Email Removed]> wrote: The vast majority of statistical surveys involve only about 1000 or so participants. I'd say, from a statistical point of view, that the sample sizes you present are quite significant. Much larger than the usual size and plenty large enough to give good data assuming that they have a somewhat representative sample. Size is much less important than sampling method. You want to be sure that they are looking at the right population (i.e., their sample isn't mostly people very different from the group that you are interested in), that they aren't relying on a self-selected group (i.e., a web survey that only a select group of people will ever find), etc. An example: In the famous (at least on this side of the pond) polling debacle, where the post-election headline read "Dewey defeats Truman", the main reason for the erroneous prediction was that they were using telephone surveys at a time where many less affluent people didn't have phones, and therefore weren't sampling an important segment of the population. Increasing the sample size wouldn't have helped at all, because that sampling bias would still have been there. |
Nicely put!
I'd just like to add that there was nothing wrong with the telephone survey,
per se.
The error was in the way the data was handled and reported.
"Dewey defeats Truman"
When all they really knew was variables for:
" Xpercent of telephone owners Awilling to complete our survey, from this
Ylocale, Breport they voted for Zcandidate"
Variables are A,B,X,Y,Z
(Assuming the owners were checked to be the actual owners, otherwise that
would be another variable.)
--
"Tennis is a funny game; unbelievable highs and the lows are just as low..."
(JOHN MCENROE) BBC1
http://www.private-eye.co.uk
MGW wrote:
| QUOTE |
| On Tue, 12 Jul 2005 16:57:57 -0700, "Tony" <[Email Removed]> wrote: The vast majority of statistical surveys involve only about 1000 or so participants. I'd say, from a statistical point of view, that the sample sizes you present are quite significant. Much larger than the usual size and plenty large enough to give good data assuming that they have a somewhat representative sample. Size is much less important than sampling method. You want to be sure that they are looking at the right population (i.e., their sample isn't mostly people very different from the group that you are interested in), that they aren't relying on a self-selected group (i.e., a web survey that only a select group of people will ever find), etc. |
There are lies, Damn lies, and Statistics.
--
Tony Garcia
Web Right! Development
Riverside, CA
www.WebRightDevelopment.com
On Wed, 13 Jul 2005 02:30:24 +0100, "Mak" <[Email Removed]>
wrote:
wrote:
| QUOTE |
| Nicely put! |
Thanks. I admit to having a masters degree in statistics, though I
have *never* worked in market research. ;-P
| QUOTE |
| I'd just like to add that there was nothing wrong with the telephone survey, per se. The error was in the way the data was handled and reported. "Dewey defeats Truman" When all they really knew was variables for: " Xpercent of telephone owners Awilling to complete our survey, from this Ylocale, Breport they voted for Zcandidate" Variables are A,B,X,Y,Z (Assuming the owners were checked to be the actual owners, otherwise that would be another variable.) |
LOL
--
MGW
Writing in news:alt.www.webmaster
From the safety of the cafeteria
MGW <[Email Removed]> said:
From the safety of the cafeteria
MGW <[Email Removed]> said:
| QUOTE |
| On Wed, 13 Jul 2005 02:30:24 +0100, "Mak" <[Email Removed] wrote: Nicely put! Thanks. I admit to having a masters degree in statistics, though I have *never* worked in market research. ;-P |
Don't believe ya - everyone works market research, it's a key element of
the [m|d]ating game.
--
William Tasso
** Business as usual
Tony wrote:
| QUOTE |
| There are lies, Damn lies, and Statistics. "The fact is that, despite its mathematical base, statistics is as much |
an art as it is a science. A great many manipulations and even
distortions are possible within the bounds of propriety."
Darrell Huff, 1954, 'How to Lie with Statistics'
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.