>
> It ∗is∗ flawed because the sample isn't random - they only use data from
> alexa users,
The term you want to use is "representative", implying that some subset of
users reasonably indicates what the population of users does. In this
case, as you pointed out, that population is vague, Alexa's 'method' of
assigning 'popularity' definitely is not considered to be any form of
valid statistical sampling design (in statistical jargon Alexa's method is
termed 'self-selection' [which is why you should not trust conclusions
drawn from 'surveys' based on questionnaires printed in magazines,
open access web questionnaires, 1-800-xxxxxxx radio/tv station 'polls'
etc].
In statistical sampling, "random" connotes that there is some
well-defined a priori probability mechanism used to select units into a
sample or that the mechanism of empirical entries of units into a sample
is well-defined. As you pointed out, neither probably exist for the
tallies at this site. In fact, having MOTM be at the 'favored' top of the
list (which I have no problems with) gives them a higher chance (whatever
that may be) of being accessed. It is even an issue if these tallies are
representative of Alexa users (say I click on Paul's site 100 times just
to boost his site's tallies, just to "mess up" the set of tallies).
> who probably have certain tendencies because of the
> characteristics of the demographic.
Exactly on target, Paul.
for example, go to their "100 busiest
> sites" and you'll see alexa right near the top (#5 or 6 or something). of
> COURSE this is waaay out of wack, because alexa users ALL go to the alexa
> site, therefore messing with the data.
>
Not really "messing with the data". Data are data - it is the process
(statistical and otherwise) of collecting the data that can be at issue.
>
Sorry for the novella, and not targeting Paul W. at all. Just felt
compelled to comment on an issue that I know something about (maybe given
that I know so little about lots of other things :-) ).
Jeff
=^..^=
@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@
Jeffrey S. Pontius
Associate Professor
Department of Statistics
Kansas State University
Manhattan KS 66506-0802
pontius@...@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@∗@