To Poll or Not to Poll?

Skimming through the news this morning, I noticed a couple of articles of interest. First:

Gallup poll shows double-digit lead for Bush, with momentum

A new Gallup poll puts President George W Bush 13 points ahead of challenger John Kerry in the US presidential race.

The spread -- 55 to 42 per cent -- is among likely voters, Gallup said, and the poll was conducted between 13-15 September.

Among registered voters, the spread is 8 points -- 52 to 44 per cent.

And second:

What If the Polls Are Wrong?

Election Surveys That Screen Out
'Unlikely' Voters Might Be Outdated
September 17, 2004

Presidential elections are poll-driven. The candidate ahead in the surveys usually gets better coverage, and the results energize supporters. The one behind often comes across as doing little right, and campaigns and constituencies lose confidence.

But what if the polls are wrong, and we aren't surveying the real likely electorate?

This might be more than an academic issue. A number of polls this presidential race show a gap in the preferences of registered voters vs. likely voters. In these models, the president usually does better with likely voters, the figure most news organizations emphasize. To get to likely voters, all polling organizations use what is called a "screen," asking questions to determine who is likely to actually turn out on election day.

These screens differ greatly, as there is no consensus among experts on what works best. "This is an art, not a science," says Peter Hart, the prominent Democratic polltaker who has helped conduct The Wall Street Journal/NBC News survey for 15 years.

This controversy will be fueled by today's just-released Gallup poll that shows George Bush with a 13-point lead over John Kerry. That is at variance with other surveys this week, which suggest a tight race with a much smaller Bush tilt. But the likely voters margin also is considerably larger than the eight-point advantage in Gallup's registered voters in this survey. The likely voters match-up invariably gets more attention.

Really, the reason this all bothers me is not just that the results are leaning towards Bush. The question on my mind is "Does public opinion lead the polls, or do the polls lead public opinion?" (The WSJ article seems to be dancing around this issue, although they're probably more worried that Bush might not actually win.)

The question gets more problematic when push-polling is factored into the equation. Push pollsters actively try to manipulate the results of their polls by using slanted questions. For example, "Would you vote for such and such a candidate who has recently been assocated with such and such horrible scandal or crime." (Although, in point of fact, push pollers are usually more sophisticated than this crude example would indicate.) The expected answer is "No!"

Push polling is obviously a subversion of the democratic process, but to what extent do more legitimate polls influence public opinion, rather than simply reporting it? Even if a pollster asks a perfectly unbiased question, such as, "If the election were held today, which candidate would you vote for," the poll will likely show that one candidate has a slight lead, and this will in turn impact the public perception of the campaign, and even the campaign itself. (It also helps further the media's portrayal of the election as a horse race, but that is a whole other issue.) It could be argued that, since an unbiased poll really does reflect public opinion at a certain point in time, that it therefore cannot have a negative impact on the campaign. But this fails to acknowledge the concern that public opinion often resembles the movements of a school of fish: when part of the school starts moving in a different direction, other parts follow. In this analogy, the following parts of the school are what are commonly referred to as "undecided voters". The poll makes them aware of which direction the other fish are swimming, thereby allowing them to follow along, rather than deciding for themselves which way they are going to swim.

Still, it might not be obvious at first why this is a problem. But consider this: If conducting an unbiased poll can tend to push voters towards whichever candidate has an edge, this puts pollsters in the position of being able to favor the leading candidate, without actually conduting a biased poll. All they have to do is wait until their favored candidate has a slight edge, conduct their poll at that time, and then loudly proclaim the results. And how do these pollsters figure out when their favored candidate is ahead? By following the small number of genuinely legitimate polls. So a genuinely unbiased poll can bias the campaign, even though it doesn't intend to.

It's also rather interesting to speculate what an American Presidential campaign would be like without polls, but I've already spent too much time writing this. :)