Yes, this does happen. It is called an information cascade. From Wikipedia:
An information (or informational) cascade occurs when people observe the actions of others and then make the same choice that the others have made, independently of their own private information signals.
There is lots of evidence that people will abandon their own opinions and conform to the opinion of the masses (see the Asch conformity experiments). Depending on the domain, there could be many explanations for this. A non-exhaustive list includes:
the decision-maker may be uncertain and believe that the masses, on average, will make the correct decision
the decision-maker may want to avoid being on the 'losing' end of things (social pressure)
the decision-maker may want to avoid responsibility
the decision-maker may not want to expend cognitive resources making a difficult decision
Do smart respondents resist this influence?
My brief research did not turn up any articles on the effect of intelligence on susceptibility to this effect. However it should be noted that there is no reason a "smart" voter should resist a cascade, and in fact it may be smarter to conform to the cascade. Consider the situation where your favored candidate is losing in the polls. Your second-favorite and least-favorite candidates are in a dead heat for first place. A rational voter might actually vote for his second-favorite candidate to prevent his least-favorite candidate from winning.
Does this happen in real-life major governmental elections?
While information cascades certainly exist, I'm not aware of any proof that an information cascade has affected a real-world, major governmental election. Which isn't to say this hasn't happened, only that the evidence is contentious. (It's generally hard to prove this kind of stuff outside a laboratory.)
Battaglini (2004) states:
Despite the fact that much attention in the theoretical literature has been focused on simultaneous voting mechanisms, many important decisions are actually taken in sequential
mechanisms. This is certainly the case of presidential primaries, and of roll-call voting
in legislatures; but it is often the case that even mechanisms that are supposed to be
simultaneous, are not in practice, because late voters generally have access to exit polls
and other information describing the choice of early voters.1
A famous example is the 1980
Presidential election in which the news that Ronald Reagan was winning by a landslide
changed expectation “suddenly and dramatically.”2
The “popular” intuition, therefore,
is that because of bandwagon eﬀects, timing matters. This view is also supported with
strong historical and experimental evidence (see Hung and Plott, 2000, and Morton and
Williams, 2000a and 2000b).
This literature is not my area of expertise, but this article seems relevant and provides several pertinent citations:
Battaglini, M. (2005). Sequential voting with abstention. Games and Economic Behavior, 51(2), 445-463. PDF