In The Symbolic Uses of Politics, Murray Edelman had this to say about public opinion polling:
A related kind of ambiguity pervades political process as well: uncertainty about how much public support or opposition for programs exists or can be created. Because opinion is constructed and volatile, all indicators of it are problematic. Poll reports are therefore another device for the reduction of ambiguity to clarity.
Polled individuals are abstracted both from their everyday lives and from political discussion and action shared with others. Their opinions are therefore also abstract — not necessarily related to any course they would pursue when involved in political activities different from answering an interviewer (or, perhaps, voting). In this artificial situation expressed opinion depends upon verbal cues together with changing memories of past situations and anticipations of future ones. Polls and surveys generate numbers that have the dramaturgical look of hard data and the epistemological look of shifting fantasies.
While it may be foolish for political actors (including underdog challengers) to ignore popular sentiment, it is perhaps even more foolish to ‘chase after the wind’ of public opinion; to treat it as if it were something that existed concretely, as an unbending thing, rather than as an ever-constructed construction, whose articulation a political actor has to constantly contribute to and contest — if that actor hopes to be a contender, and not merely a commentator on the sidelines.
(Thomas Gilbert and Andrew Loveridge made interesting related points last year in the Berkeley Journal of Sociology.)