Battleground State Poll Analysis: Part One - Strategic Vision and American Research Group

Part One of my battleground state poll analysis begins with the final polls conducted by Strategic Vision (SV) and American Research Group (ARG).

SV was previously a Republican polling firm but refrained from partisan polling during this election. I have been informed that they are attempting to secure bi-partisan or non-partisan work and that their polls were conducted and released in an attempt to establish a more balanced reputation.

ARG is a nominally non-partisan polling firm whose results have a reputation of leaning in favor of the Democratic candidate. Thus, a comparison of SV and ARG may be instructive.

The battleground states selected for this study are Iowa, Florida, Michigan, Minnesota, New Hampshire, New Mexico, Ohio, Pennsylvania and Wisconsin. All results are listed with the support of President George W. Bush first and the support of Senator John F. Kerry second.

Florida:
Results: 52-47
SV: 50-46
ARG: 48-50

Iowa
Results: 50-49
SV: 49-46
ARG: 48-47

Michigan
Results: 48-51
SV: 45-44
ARG: n/a

Minnesota
Results: 48-51
SV: 48-48
ARG: n/a

New Hampshire
Results: 49-50
SV: n/a
ARG: 49-48

New Mexico
Results: 50-49
SV: n/a
ARG: 47-48

Ohio
Results: 51-49
SV: 48-46
ARG: 47-49

Pennsylvania
Results: 49-51
SV: 46-47
ARG: 47-50

Wisconsin
Results: 49-50
SV: 48-46
ARG: 47-48

Predicted Winner
SV had a record of 4-2-1 and ARG had a record of 3-4-0. SV is given a tie for its prediction of a tie in MN. Advantage SV.

Average Error In Spread
SV had an average error of 2.3 points. ARG had an average error of 2.3 points. Although ARG called more races incorrectly, its spread was correct in IA and WI and reduced its average error. No advantage.

Bias Towards Each Candidate
SV underestimated Bush's support by an average of 1.9 points and underestimated Kerry's support by an average of 3.6 points. ARG underestimated Bush's support by an average of 2.4 points and underestimated Kerry's support by an average of 0.7 points. Thus, SV had a pro-Bush bias of 1.7 points and ARG had a pro-Kerry bias of 1.7 points. No advantage.

Conclusion

Although this analysis contains a fair amount of rounding (and rounding of rounded numbers) and is based on state vote totals that are subject to change as absentee ballots are counted and other revisions are made, it appears that each polling firm did demonstrate a bias consistent with each's reputation.

SV had a bettor record in its predictions than did ARG. Nonetheless, each firm's average spread error and degree of bias was equal. For that reason, no conclusion can be made as to the comparative accuracy of these two firms. I suspect that their performance will be only fair to middling when compared to other firms in later installments of this feature.