Mickey Kaus kindly linked to my post last Friday on the relatively high ratings of Hillary Clinton by rank-and-file Democrats, but added: "I’d still like to see Edmund Muskie’s approval ratings in 1970, just for comparison’s sake."
Fair enough. Here you go: In a Gallup poll of 3,489 adults conducted in January 1970, 63% of US adults said they "liked" Muskie, 15% said they "disliked" Muskie and 22% did not answer. I do not have a cross-tab of those results by party identification, but what we have supports Kaus’ rhetorical point: High early polling numbers – either in favorable ratings or vote preference — do not guarantee anyone a party nomination. A little over a year after this survey, George McGovern mobilized anti-war sentiment to defeat Muskie in the Democratic primaries. That’s a caveat worthy of emphasis.
However, MP should probably put his "explainer" hat on for a moment and warn readers against comparing Gallup’s 35-year-old measure too closely with the more recent ratings of Clinton and others. Putting aside all the differences in the popular and political culture, the 1970 Gallup poll was conducted in person (not on the telephone), it asked whether respondents "liked" or "disliked" public figures (rather than whether their impression was "favorable" or "unfavorable") and asked respondents to answer with numbers rather than words. The 1970 Gallup interviewers showed respondents a card with a scale that had labels on the endpoints. The numbers ranged from +5 (like very much) to -5 (dislike very much). The results I wrote about last week were based on questions that asked whether their impression is "very favorable," "mostly favorable," "mostly unfavorable" or "very unfavorable" (that is the Pew Research Center language; the Diageo/Hotline poll uses the word "somewhat" rather than "mostly").
Experimental academic research shows that on otherwise identically worded questions, rating scales using both positive and negative numbers (-5 to +5) get significantly more positive responses than ratings scales that run from 0 to 10. In one study published in Public Opinion Quarterly (Schwarz et. al. 1991), the average positive rating rose from 63% with a 0 to 10 scale to 85% using a -5 to +5 scale. How results from a -5 to +5 scale would compare to a question using word labels only, I cannot say. But these results warn us to be careful about making face value comparisons of such ratings.
But as long as we have the "way back" machine open, here is another question from that same 1970 Gallup survey whose language is far more comparable to one asked more recently:
In view of the developments since we entered the fighting in Vietnam, do you think the U.S. made a mistake sending troops to fight in Vietnam?
52% yes
29% no
19% no opinion/no answer
Gallup now tracks an almost identically worded question regarding Iraq (the results are from their most recent survey, conducted December 16-18, 2005):
In view of the developments since we first sent our troops to Iraq, do you think the United States made a mistake in sending troops to Iraq, or not?
52% yes
46% no
2% unsure
And of course, less than two years after the 1970 survey, President Richard Nixon was reelected, defeating George McGovern by a 61% to 38% margin.
UPDATE: Here are links to the Gallup data from January 1970 which will only work for those with a Gallup.com subscription: the questions on Muskie, Vietnam and the full poll.
Gallup asked the like-dislike question about Edmund Muskie again several times before the 1972 primaries and Muskie grew even more popular. The results were 73%-16% ( like dislike) in May 1970, 75%-15% in October 1970 and 77%-15% in February 1972.
“Experimental academic research shows that on otherwise identically worded questions, rating scales using both positive and negative numbers (-5 to +5) get significantly more positive responses than ratings scales that run from 0 to 10.”
This is interesting–and it’s reason #4872 (or so) why I’m glad the subject of my science is inanimate. Is there a quick answer as to why there’s this difference?
My hunch (and this isn’t an area where my hunches tend to carry much weight) would be that many people aren’t used to using negative numbers. Afraid to make a mistake, they end up sticking to what’s familiar–tacitly treating the scale as zero to five. Everybody knows 5 is better than 1, but I can see how folks might be confused about whether negative 5 is better than negative 1.
A couple of weeks ago a friend who has been active in NH politics for 25 years and I were trying to identify the line below which a front-runner is vulnerable in very early polling in NH.
So we actually took out The Making of the President, 1972 and read about the ’72 NH primary. Muskie polled over 60% in NH well into February.
We both are veterans of Hart in ’84 (my friend has a laminated front page of the Boston Globe from the morning after) and well remember Mondale didn’t look vulnerable until Nightline released a poll the night before the primary showing Hart and Mondale tied at 30.
I am actually researching this, but here is what we concluded with respect to early NH polling and front runners:
As long as a front runner is over 40, he is safe. In part this is simple mathmatics – if one cadidate gets 40%, then to be beaten another has get over 2/3 of the remaining votes (which is difficult in a multi-candidate race).
Exceptions to this rule:
’72 Muskie (though he was in a two man race in the NH primary)
’84 Mondale was over 40 most of the way
If the front runner polls between 30 and 40, he is in trouble, but still may win. Examples:
Dole in ’96 (the prior year he was usually around 35)
Dukakus in ’88 (got 38%, but the field splintered)
Reagan in ’80
BTW There is a rumor that the pollster at UNH has Hillary in the mid-20’s.
I imagine that the more significant difference between polling techniques in 1972 and today is not the scale for responses but the way the sample is composed and handled. Note that by today’s standards, 3000 interviews would be a huge sample — and I’m guessing that it was a “test bank” supposed to approximate the demographics of the electorate rather than a random sample. For reasons I do not understand, today’s random samples of roughly 800 are thought to be more predictive than a composed sample of several thousand….
Also, because polling has become a much more significant part of campaign strategy, campaigns track a lot more information than just favorability — and adjust accordingly.
So that unlike Muskie or Mondale, Sen Clinton is unlikely to miss a major shift in public opinon within the Democratic party electorate; Muskie and Mondale both thougth they would win the NH primary because they didn’t have a good sense of the Democratic primary electorate.
The point being that if HRC is polling at 40% in NH a year from now (which I do not expect), barring scandal, she’ld more or less have the state sewn up and her challengers would start looking at other early states to try to slow her down.