Pollster.com

Articles and Analysis

 

ARG Etc.

Topics: 2008 , Divergent Polls , Sampling Error , The 2008 Race

The new trial heat match-ups posted yesterday by the American Research Group (ARG) for the national Democratic presidential primary generated some comments worth responding to, both here and on other sites.

First, consider the results, as presented on the ARG site. Results for each month are based on 600 "likely Democratic primary voters" - those who say "they will definitely vote in a primary or participate in a caucus in 2008:"

05-15%20ARG.png

One nit to pick in the interpretation by TPMCafe's Eric Kleefeld of an "expanding" Clinton lead in these results:

The same poll a month ago found that Hillary had 36% to Obama's 24% - meaning Hillary has expanded her lead by 5 points in ARG's sampling.

True, but remember sampling error 101: The "plus or minus 4 percentage point" margin of sampling error that ARG reports for each of these surveys applies to each percentage, not to the margin. So neither Clinton's three point increase since April (36% to 39%) nor Obama's two point decline (24% to 22%) is large enough to be statistically significant.

However, these results do show a statistically significant decline in Obama's support (from 31% to 22%) as measured by ARG's since it's first national sampling in March.

Having said that, we should keep in mind that most other polls released in March did not have the national race nearly as close as ARG. Our chart of all recent national horse race results (which as of this writing does not yet include now includes this latest ARG release) shows that most other polls had Clinton leading by the same (roughly) twelve point margin since March. Obama's standing has not changed since his surge of support in January and February.

ATopDems400051507.png

Finally, in addition to some pointed questions about their track record, commenter Alex Forshaw asked, "who is behind ARG?" As this question seems to come up regularly, I put it to ARG president Dick Bennett via email. Here is his reply:

We still rely on subscriptions to what used to be commonly known as omnibus surveys.** The difference, however, is that we package each state and national survey separately. Unlike a traditional omnibus, there is no set schedule or guarantee of a survey without repackaging. We do a lot of panel building and have been fortunate in the past 6 months because we have been tracking household telephone, cable, and Internet access around the country and that allows for the addition of the political questions. Campaigns cannot be subscribers (but I know that some campaigns receive the latest numbers shortly after our subscribers and before public release). We are always looking for new subscribers willing to allow the political questions as part of the surveys. Subscriptions are not inexpensive -- no $199 for the entire race deals -- and subscribers pay proportionately for each survey.

**An omnibus survey is one that allows multiple paying clients to buy questions.

 

MAP - US, AL, AK, AZ, AR, CA, CO, CT, DE, FL, GA, HI, ID, IL, IN, IA, KS, KY, LA, ME, MD, MA, MI, MN, MS, MO, MT, NE, NV, NH, NJ, NM, NY, NC, ND, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VT, VA, WA, WV, WI, WY, PR