Credibility of Internet-based Polls


We've collected a very awe-inspiring data base of the candid opinions of those who've taken the time to complete this poll.  But how representative is it of the voting American public?  There are a number of ways you can judge for yourself if it's truly representative, something you can bet you'll never hear from Tom Brokaw.

The first thing you must know is that each isp address is dutifully recorded right next to the poll answers, along with any voluntary information regarding name, address, email address, phone number, fax number, sex, and race.  None of this information will ever be made public.  Only to prove that the information is valid, key individuals will be permitted to view it.  They will never receive a copy of it, though, so your confidential information will always be secured.

A drill down tool enables us to gather quite a bit of information about each isp address.  It also tells us that 10,000 people view this page each day, and that 95% of those who've taken the poll never visited the site before.  Visitors to the site find the poll through any number of search engines, with Google being the one with the most number of hits.  Following is an example of the the top 52 key words people searched for on Google, along with the number of times each word was searched, and its percentage of the total number of search words.

For example, 1,832 people searched for "rate", and the word "rate" represents 2.84% of the total words searched on Google for this page.  Are people who are searching for the word "rate" on the internet a particularly "racist" or "anti-semitic" sort?  Of course not.  They might be a bit more educated than the average surfer, but we must acknowledge that they would be a relatively unbiased cross section of the American public.  Ditto for the word "murder" which was searched 1,277 times and represents 1.98% of the searches.

It could be argued that those who search for the word "jews", which was searched 708 times, have a political or religious or social or some other statistical bias before they arrived at that page--but how do we know many of them weren't jews themselves, or were searching for information to SUPPORT jews.  And even IF they ALL arrived at that page with an agenda, they're only 1.1% of the searches on Google, which is only a small percentage of ALL the search engine hits, which means that any bias they MIGHT represent is not relevant.

What about those who searched for "blacks", which was 388 hits, or 0.6% of all the hits on Google?  Even IF they were a baised subset, they're far too small a percentage to have distorted the poll results by any significant amount.


   in   2,115  3.28 
   of   1,992  3.09 
   rate   1,832  2.84 
   the   1,817  2.82 
   murder   1,277  1.98 
   and   1,212  1.88 
   rates   1,178  1.83 
   divorce   881  1.37 
   world   875  1.36 
   iq   872  1.35 
   women   746  1.16 
   crime   740  1.15 
   jews   708  1.10 
   gre   676  1.05 
   scores   643  1.00 
   a   616  0.96 
   sat   583  0.90 
   jew   582  0.90 
   holocaust   547  0.85 
   statistics   541  0.84 
   adultery   539  0.84 
   us   538  0.84 
   population   500  0.78 
   to   495  0.77 
   rape   488  0.76 
   christian   441  0.68 
   by   440  0.68 
   jewish   435  0.68 
   porn   421  0.65 
   is   402  0.62 
   on   399  0.62 
   black   388  0.60 
   hoax   388  0.60 
   men   383  0.59 
   for   375  0.58 
   average   356  0.55 
   homicide   353  0.55 
   per   335  0.52 
   point   323  0.50 
   score   321  0.50 
   west   320  0.50 
   japan   295  0.46 
   laws   284  0.44 
   with   282  0.44 
   highest   274  0.43 
   gary   265  0.41 
   einstein   264  0.41 
   bible   260  0.40 
   paltalk   259  0.40 
   klebold   256  0.40 
   sex   255  0.40 
   worldwide   253  0.39


Now let's compare this to the assertion that a census of all 295 million Americans would be more accurate than this internet-based poll.  The first problem is that it would take more than a year to complete it, and the public opinion would have changed dramatically between the first answer and the 295 millionth answer.

The second problem is the hundreds or thousands of government employees who would be handling the data, any one of whom could make a simple spreadsheet error (like the simple, single spreadsheet error which cost Citibank $1 billion), and completely invalidate it.

The third problem is that collecting and collating millions of forms introduces numerous opportunities for fraud, key omissions, and (as with the Michigan affirmative action board which was ORDERED by the court to certify the state ballot on affirmative action, but refused), refusal to simply follow the law.

The fourth and killer problem is that if such a task we undertaken, and if the results showed that only 88% favored exiling blacks compared to 90% on this poll--not a single competent mathematician would dare claim  that the government poll was moreaccurate than this internet poll.  Why?  Because it's statistically improbable for so many human [read: government] hands to handle so much data without making numerous key errors.

But what would it prove?  Whether or not 88% or 90% is the actual figure--both are large enough to pass a constitutional amendment anyway.


Following is an example of the information provided by the drill down tool, with the isp masked to protect the identity of the respondent:


Referring Link
Host Name
IP Address --.--.29.110
Country United Kingdom
Region England
City London
ISP Onetel Dsl Pipe Allocations
Returning Visits 0
Visit Length 3 hours 39 mins 27 secs
Browser Firefox 1.0.7
Operating System Windows XP
Resolution 1280x1024
Javascript Enabled