The basics of tracking polls - do they matter?

September 11th is not infamous anymore only for the obvious reason, it also marked the moment when there were four tracking polls published each day for the US presidential election (Research 2000/Daily Kos poll began). It's quite a number when compared to the last election or considering that it's still almost two months to the election day. However, a sheer quantity of polls doesn't necessarily quarantee reliable results. Therefore it's important to keep in mind a few basic facts about polls and especially about daily tracking polls.

There are four daily polls which are provided by Gallup, Rasmussen, Diageo/Hotline and the aforementioned Research 2000/Daily Kos. Gallup (yes, that Gallup) is a famous and old polling company which has been tracking this election daily since the primaries. Rasmussen Reports is another solid name in polling and it has also covered this election for some time now. Scott Rasmussen, the head of the company, is an evangelical Christian and the company has cooperated with Fox News but despite these facts he's widely considered to be a reliable and independent pollster. Diageo/Hotline started their daily polling on September 2nd and Reseach 2000/Daily Kos on September 11th. Daily Kos is a liberal blogging site but they have provided all the internals of Research 2000's polling so people can judge for themselves whether they are reliable or not. It's commendable to release as much of the polling internals and methodology as possible so Research 2000 is a welcomed addition.

There are certain things that has to be kept in mind when reading these polls. First, all four publish what are known as three-day rolling averages. This means that the average of polls from three past consecutive days will be published. On September 15th they publish the average of polls from 12th-14h of September. Each day the oldest day of polling will be dropped from the average and the previous day's poll is added to replace it. This has a few implications - the variance between the polls will not be huge because of the balancing effect of the previous days. Also, if one candidate one day takes a lead of 3% it is not clear whether this was because the previous day was excellent for that candidate or that the day which was dropped from the rolling average was a good day for him and therefore he lost that good number from his average.

Secondly, polls have always a margin of error and even though this is self-evident to many, it tends to be ignored in the heat of the race where passions run high. In the case of these four tracking polls, they report their 'MoE' to be between 2-3%. Any movement inside those numbers can be attributed to MoE and it's easy to read too much into a poll that has eg. Obama moving up one point when in truth it doesn't tell you any concrete, scientifical, statistical info at all.

Third thing to keep in mind is the methodology of different polls. This is very important and in essence the very core of polling. What questions do you ask is naturally a fundamental aspect but more importantly who do you ask them and in what proportion? All the pollsters try to estimate in which proportion the people will vote come election day. Few examples: women tend to vote more actively so many pollsters have the women-men ratio at approximately 52-48. Democrats have had the advantage of registered voters so all the pollsters ask more registered Democrats than registered Republicans. How big the difference is especially after the so-called 'Palin bounce' is really hard to estimate and so the pollsters have different estimates and this reflects directly into their poll results.

In conclusion we have four polling companies publishing three-day rolling averages on a daily basis. Margin of error is real and the methodologies of different pollsters matter a lot. Then how big an importance we should give to these tracking polls?

The answer is obvious to many but in practice many ignore it. What matter are trends. In any given day the margin of error may and will impact the numbers significantly because no pollster has resources to poll such a large number of people that MoE would be in reasonable levels. These three day rolling averages exist just because of that - that ables the pollsters to lower their MoE to 2-3%. However, the MoE is still there and 1-2% shifts up or down in these daily published polls alone do not tell you anything. Look at the past five daily polls and you start to get an idea. Compare one pollster's numbers with others - see if there are any common trends. Is McCain moving up in three of these four pollster's polls during the past five days? Is Obama consistently down 2-3% in many of these polls during many consecutive days?

This common sense attitude gets easily lost these days with the 'poll-mania' running wild. Political junkies, myself included, get easily trapped in the world of daily polls and start reading something into them that really isn't there. It's tempting and somewhat entertaining (depending on how your candidate is doing in the polls) but ultimately quite frutile. This syndrome does not limit itself into political junkies as news outlets are prone to make the same mistake sometimes. However, they usually put more weight into their own opinion polling (CNN, CBS etc.) and hype those polls out of proportion instead of the daily tracking polls. When it comes to the ordinary voter, they really don't follow the polls themselves but "surrender themselves" to the narrative that the media is touting at any one time. Tracking polls give the media more material to chew on but luckily the TV media for example hasn't started any "daily poll shows". Believe me, that day could come and it wouldn't be good news.

Tracking polls are mainly a concern of internet bloggers and enthusiasists at the moment. They are entertaining and when viewed through a larger scope they contain valuable information specifically in terms of trends. As temptating as it is to read into Gallup's daily poll that "Palin's interview gave McCain a one-point bump" it really isn't a very solid conclusion or even relevant in the big picture.

Remember, there are still 51 days left. That means we can enjoy at least 204 daily tracking polls. It's all perspective.