The Pollsters
There were six organisations conducting polling of the presidential election:

• BVA
• CSA
• IPSOS
• IFOP
• LH2
• TNS-Sofres

Of which IPSOS was the most prolific during the last few weeks of the campaign, releasing six polls a week, whereas the other pollsters published their polls less frequently, for example once a week or every five days, or entirely irregularly. During the last week, some of the other pollsters released two or three polls, some even releasing two polls on the same day!


Lies, Damned Lies And Polls
My poll averages weren't without their flaws. The first major issue was that of absentions (that is, people who weren't planning on voting for any candidate) and undecided voters. Some pollsters mashed the numbers together, while some only reported the abstentions.

The other major issue was one of frequency. The pollsters released their polls at very different frequencies, meaning that every calculated average would not only be an effect of possible changes in opinion, but also of which pollster had happened to release a poll lately and what particular biases in the numbers their respective methodologies would cause. Beginning in March, IPSOS started releasing a new poll six times a week. This skewed the average towards IPSOS's findings, as any given average point from thereon out would contain two or three of IPSOS's latest polls.

And then there's of course various issues with the polls themselves (undecideds, redressement, selection of people to call etc).


The Data
Below you'll find two graphs: the first one contains the trend lines, starting January 8 and continuing until April 20 (the last day before the election first round polls were allowed to be published); the second one contains all the poll numbers in relation to the trend estimate.
I've also kept track and am keeping track of the polling for the second round; those graphs will be posted in a another diary after the second round on May 6.


If the graph was right, the election results would have been somewhere in the ballpark of:
Sarkozy: 28.3%
Royal: 24.3%
Bayrou: 17.9%
Le Pen: 13.9%

The actual election result was (98% counted):
Sarkozy: 31.1%
Royal: 25.8%
Bayrou: 18.6%
Le Pen: 10.5%

The last average before the election was thus reasonably close on Royal's and Bayrou's eventual numbers, while Le Pen was overestimated, and Sarkozy was underestimated.
The last polls before the election had this to say:

	BVA	CSA	IFOP	IPSOS	LH2	TNS-S.
Date	20.4	20.4	20.4	20.4	17.4	20.4
Royal	26.0%	25.5%	22.5%	23.5%	23.0%	24.0%
Sarkozy	29.0%	26.5%	28.0%	30.0%	27.0%	28.0%
Bayrou	17.0%	16.0%	20.0%	17.0%	19.0%	19.5%
Le Pen	12.5%	16.5%	13.0%	13.5%	14.0%	14.0%

No pollster had everyone pegged; the closest was BVA.


Events Affecting The Poll Numbers
Another important question is that of how much effect various events throughout the year had on the actual poll numbers. Two events in particular interested me.

Sarkozy was officially elected (though unopposed) as the UMP candidate for president on January 14, 2007. Shortly thereafter Sarkozy had his best month as far as poll numbers go, one pollster even giving him 36%. One can wonder how much of his rise can be attributed to the UMP congress (or, should I say, the media coverage of the congress), seeing as it had already been known for some time he would be candidate and the election was ultimately a formality.

Royal made a well received appearance on TF1's J'ai une question à vous poser on February 19, 2007. At that point she was already slightly rising in the polls, but it does look like she received a slight - and shortlived - bump in the polls.


Conclusions
As everyone underestimated Sarkozy and overestimated Le Pen, so did the poll average. My assumption was that all polls would be somewhat wrong, and that averaging would bring us closer to the truth. But if all the polls are wrong in the same way, averaging them won't remove that error. Ironically the poll average came the closest on Bayrou, who had been said to be rather difficult to poll.