Are they sufficiently different? I read some of them and they are very similiar:
Having conducted a number of surveys, I can tell you first hand that the wording of the item can GREATLY impact the results. We’ve sometimes spent an hour or more quibbling over a single word. So what you describe as "very similar" may in fact
be very similar but lead to very dissimilar results nonetheless.
That being said, I’m not as concerned about differences in the wording of that portion of each item - (I’d be willing to bet that order effects would play a more significant role) – a much bigger problem is that he’s combining data that was collected using different scales. Here a just a few illustrative examples:
ABC: Favorable, Unfavorable, or Unsure?
CBS: Favorable, Not favorable, Undecided, Haven’t heard enough or Refuse to answer?
NBC: Very positive, Somewhat positive, Neutral, Somewhat negative, Very negative, or Don’t know?
USA: Favorable, Unfavorable, Never heard of it, or Unsure?
AP: Favorable, Unfavorable, Neither, Unsure?
This is a huge problem for lots of reasons!
To illustrate a few by comparing the first two polls:
1) Respondents for ABC have almost a forced choice comparison between Favorable and Unfavorable – "Unsure" is included as a catch-all, but people can interpret that in a variety of ways which (I assume) is why a lot of the other polls break it out.
2) Respondents for CBS are asked to indicate their opinion on a *much* different scale. Whereas the ABC poll tries to push people into a simple dichotomy, the CBS poll allows people ample room for that gray area with “undecided” and “haven’t heard enough” options. That latter option is especially noteworthy in that it brings certainty into the equation, prompting a “favorable/not favorable” response only when you’re reasonably sure you won’t change your mind in the future.
They’re really two different questions. You would expect that someone with limited exposure (maybe they caught a few negative headlines) – would classify their opinion as “unfavorable” on the ABC poll, and maybe “haven’t heard enough” on the CBS poll.
Predictably, people are less willing to commit to a favorable/unfavorable opinion when they are given choices that more closely reflect their opinion. People consistently report "unfavorable" opinions at roughly twice the rate as those in the CBS poll:
ABC
36% Favorable
48% Unfavorable
16% Unsure
CBS
18% Favorable
25% Unfavorable
19% Undecided
36% Haven't heard enough to decide
2% Refused to answer
Going back to the trendline, hopefully the above is sufficient to show that his trendline doesn't have a consistent meaning - what it means to "unfavorable" changes depending on which polls are used to determine a given point on the curve. Further, it's not difficult to imagine that spurious trends could appear simply because of how the CBS poll and other "low predictors" are temporally distributed in relation to the ABC poll and other "high predictors."