In news that I’m sure is shocking to absolutely no one, when last night’s election results are compared to the polls conducted by Rasmussen Reports, it turns out that Scott Rasmussen put a sweaty, slab-like, thumb on the scale in favour of Republican candidates at both the Presidential and US Senate levels. I decided to do a comparison between last night’s results and Rasmussen’s final polls so that I could quantify for all of our information just how much bullshit Rasmussen is shoveling.
Before I get into the meat of the subject, a few ground rules for this comparison. I looked at states for which Rasmussen Reports published polls with field dates falling within the last three weeks (for the vote for President) or the last four weeks (for the Senate races) before the election. If Rasmussen published more than one poll during that period, I looked only at the most recent one. I have calculated what I refer to as a “net miss”, which is stated as either R+X or O+X. That number is calculated by adding or subtracting the amount by which the Rasmussen poll number for each candidate differed from the final result for that candidate. If Rasmussen showed a given race with Romney five points over his final result and Obama three points under his final result, that would be noted as R+8.
First, let’s look at the race for President between Barack Obama and Mitt Romney. The headline number for the night, never mind the fact that it is functionally irrelevant, was the national popular vote. On November 6, Rasmussen released their final tracking poll, which showed Romney at 49% to Obama at 48%. The actual result for the national popular vote was Obama 50%, Romney 48%. So this yields a net miss for Rasmussen of R+3.
Turning to the individual states, Rasmussen conducted polling in nineteen states in the final three weeks of the race. They broke down as follows (in alphabetical order):
Rasmussen | Rasmussen | Actual | Actual | ||||
Romney | Obama | Romney | Obama | Net Miss | |||
Arizona | 52 | 44 | 55 | 43 | O+3 | ||
Colorado | 50 | 47 | 47 | 51 | R+7 | ||
Connecticut | 45 | 52 | 41 | 58 | R+10 | ||
Florida | 50 | 48 | 49 | 50 | R+3 | ||
Iowa | 49 | 48 | 46 | 52 | R+7 | ||
Indiana | 52 | 43 | 54 | 44 | O+3 | ||
Massachusetts | 47 | 52 | 37 | 61 | R+5 | ||
Michigan | 47 | 52 | 46 | 54 | R+3 | ||
Minnesota | 46 | 51 | 45 | 53 | R+3 | ||
Missouri | 54 | 43 | 54 | 44 | R+1 | ||
Montana | 53 | 43 | 55 | 42 | O+3 | ||
Nevada | 48 | 50 | 46 | 52 | R+4 | ||
New Hampshire | 48 | 50 | 47 | 52 | R+3 | ||
North Carolina | 52 | 46 | 51 | 48 | R+3 | ||
North Dakota | 54 | 40 | 59 | 39 | O+6 | ||
Ohio | 49 | 49 | 48 | 50 | R+2 | ||
Pennsylvania | 46 | 51 | 47 | 52 | – | ||
Virginia | 50 | 48 | 48 | 51 | R+5 | ||
Wisconsin | 49 | 49 | 46 | 53 | R+7 |
So what does this tell us about Rasmussen Reports polling of the presidential race? First, that it was bad. Really bad. Rasmussen didn’t manage to exactly nail a single race. Not a single one out of 20. Only in one race did they even get the margin between the candidates right. Second, that Rasmussen’s polls were tilted towards Romney. Of the twenty races polled, in fifteen of them Rasmussen had a better margin for Romney than actually occurred, whereas only in four races was this the case for Obama. Third, that the weight of Scott Rasmussen’s thumb on the scale for Romney averaged 2.5 percentage points. That is a non-trivial amount.
Bad as Rasmussen’s polling of the Presidential race was, as we are about to see, his polling of the Senate races was dramatically worse. As a note, for these races I note the candidates by party, rather than by name, and use R and D in the net miss notation. In the final month of the election campaign, Rasmussen polled eighteen states with Senatorial contests. Here’s how those races broke down:
Rasmussen | Rasmussen | Actual | Actual | ||||
Republican | Democrat | Republican | Democrat | Net Miss | |||
Arizona | 50 | 44 | 50 | 45 | R+1 | ||
Connecticut | 45 | 51 | 43 | 55 | R+6 | ||
Florida | 46 | 49 | 42 | 55 | R+10 | ||
Indiana | 42 | 45 | 44 | 50 | R+3 | ||
Massachusetts | 47 | 52 | 46 | 54 | R+3 | ||
Michigan | 39 | 51 | 38 | 58 | R+8 | ||
Minnesota | 33 | 56 | 31 | 65 | R+11 | ||
Missouri | 43 | 51 | 39 | 55 | R+8 | ||
Montana | 48 | 49 | 45 | 49 | R+3 | ||
Nevada | ?50 | 45 | 46 | 45 | R+4 | ||
New Mexico | 39 | 52 | 45 | 51 | O+7 | ||
North Dakota | 50 | 45 | 45 | 50 | R+5 | ||
Ohio | 48 | 50 | 45 | 50 | R+3 | ||
Pennsylvania | 45 | 46 | 45 | 53 | R+7 | ||
Virginia | 47 | 49 | 48 | 52 | R+2 | ||
Washington | 37 | 52 | 41 | 59 | R+3 | ||
Wisconsin | 48 | 48 | 46 | 51 | R+5 |
So what have we here? We have a complete and utter failure to call even one margin correctly, never mind a race as a whole. Out of the eighteen races polled, Rasmussen missed the margin high on the Republican side in fully sixteen of them. That’s 88.9%. The average net miss was also absolutely appalling. Rasumussen over-rated the margin by an average of R+4.4.
So here’s the overall situation:
Percentage of races (national presidential, state presidential and senatorial) called correctly: 0 (0/37)
Percentage of margins called correctly: 2.7 (1/37)
Average net miss: R+3.4
Let’s not beat around the bush. Rasmussen polling is garbage, and fundamentally untrustworthy. This isn’t a new state of affairs either. They had dramatically bad polling in the 2000 election and the 2008 election. Rasmussen places such a heavy thumb on the scale that the polling his company produces has no utility beyond its entertainment value. Anyone who considers themselves a serious political analyst should give serious consideration to simply avoiding Rasmussen polls in the future.
Image via Wikimedia Commons, data on polls sourced from Real Clear Politics, and information on vote percentages via the CNN Election Results Center.