The real Doug Kaplan, the man behind Gravis Marketing

Editor’s Note: This arti­cle was sub­mit­ted by dawolf, one of our reg­u­lar read­ers and com­menters. We always wel­come your contributions.

Another Editor’s Note: I have pub­lished an update on this.

Much has been writ­ten of late regard­ing whether polls need to be “unskewed” due to incor­rect Democratic/​Republican reg­is­tra­tion. But what if there are results being made up sim­ply out of whole cloth?

One poll­ster that arrived on the scene recently is Gravis Mar­ket­ing who has shown a strong Repub­li­can house bias this cycle. A recent inves­ti­ga­tion of Gravis by Demo­c­ra­tic Under­ground made it clear that there were ques­tions sur­round­ing the company.

This is not the first time that a poll­ster has been accused of pub­lish­ing “poll results” that do not cor­re­spond to actual polled indi­vid­u­als. Nate Sil­ver at FiveThir­tyEight made such an accu­sa­tion of the now-​​defunct Research 2000 on June 29, 2010.

How­ever, ques­tions such as those raised by Demo­c­ra­tic Under­ground are not evi­dence of guilt. So is there any real evi­dence that the work they are com­mit­ting is fraudulent?

Here is a recent poll they claim to have con­ducted is cur­rently avail­able, with crosstabs. The exec­u­tive sum­mary states

On Sep­tem­ber 29th – 30th, 2012, Gravis Mar­ket­ing, a non-​​partisan research firm, con­ducted a sur­vey of 914 likely vot­ers in the state of Florida regard­ing their likely vote for a given pres­i­den­tial can­di­date, sen­ate can­di­date, and other social and eco­nomic issues impor­tant to voters.

Nine hun­dred four­teen likely vot­ers. This num­ber is key. It means one likely voter is 0.1094% of the total. If a likely voter didn’t answer one par­tic­u­lar ques­tion, this per­cent­age goes up, not down, so for any ques­tion the small­est pos­si­ble result, to three dec­i­mal places, is 0.109%.

That leads us to be able to rec­og­nize many impos­si­ble results, such as the one cir­cled in the table to the right.

Since 0.07 is less than 0.109, it would require just under two-​​thirds of a per­son to be unsure of whom to support.

Here is a sum­mary of their impos­si­ble results.

Do you need more proof? Lets go back to that first impos­si­ble exam­ple given above. We can go back to the raw num­bers by mul­ti­ply­ing out from the per­cent­age, to indi­cate how many peo­ple must have given a par­tic­u­lar answer in order to achieve the sup­plied crosstab results. This gives the table to the right:

Not a sin­gle answer has an inte­ger num­ber of peo­ple. Most are too far away from an inte­ger to have pos­si­bly been accounted for as a round­ing error.

Clearly, Gravis Mar­ket­ing should not be trusted, and their polls should not be used in any man­ner. They are cur­rently being included in the poll aggre­ga­tion being con­ducted by Nate Sil­ver at FiveThir­tyEight, and by Real­Clear­Pol­i­tics. Hope­fully this will be rec­ti­fied soon.