MADISON, Wis. – The presidential race in Wisconsin hardly is in the bag for either side, according to recent polls, but the Democratc Harris-Walz team holds an edge. These are recent snapshots from polls commissioned by news media and conducted by reputable survey companies using statistical probability sampling with margins for error of three percentage point or less. Note: These percentages, as listed here, don’t include extraneous candidates like Robert F. Kennedy Jr.
> August 25 to August 28: Trump 47%, Harris 46%. Sample size: 1,082 likely voters. Conducted by Trafalgar Group.
> August 25 to August 28: Harris 48%, Trump 44%. 672 likely voters. By Redfield & Wilton Strategies. For London Telegraph newspaper.
> August 25 to August 28: Trump 50%, Harris 49%. 850 likely voters. By Emerson College. For The Hill newsletter.
> August 23 to August 26: Harris 53%, Harris 44%. 648 likely voters. By Morning Consult. For Bloomberg magazines.
> August 19 to August 21: Harris 49%, Trump 45%. 400 likely voters. By Fabrizio Ward. For Pinpoint Policy Institute.
> August 14 to August 20: Harris 46%, Trump 42%. 600 likely voters. By Spry Strategies. For American Principles Project, a super PAC.
Variables in polling
Polls are snapshoots at a given time and are not predictive. Also: The results are affected by numerous variables: When was a survey conducted? Who was surveyed, whether likely voters or registered voters? How large was the sample? Any sample of fewer than 384 people has what’s generally considered an unacceptable margin error. Surveys are expensive. Clients choose how much precision their budgets can handle. The larger the sample, the stronger the confidence in the findings but also the more expensive.
Foibles in measuring public opinion
Statistical probability sampling isn’t random. Questioning the first 500 people in a check-out Walmart will be accurate for what those 500 people think, but when it cems to elections: Who cares. Everyone in the group doesn’t need to be questioned but everyone needs to an equal opportunity of being questioned. For elections this means creating a list if every registered voter in, say, the state Minnesota. Or every likely voter. Or every citizen. This is an expensive and daunting process but an absolute necessity. Once a sample defined – say, registered voters — then an interval must be established to determine how many persons need to be sampled. It can be every 384th person, which is the minimum for 95% confidence that results will be within a margin of error of three points. An interval of every 500th will carry a higher confidence level. Absolute confidence will be questioning all 3,594,940 Minnesota registered voters but, of course, that’s not feasible. Remember too: The electorate is a moving target. People change their minds as issues change and as political camaigns progress.