averaging is a good thing because what really matters is the trend assuming the polling companies keep using the same methods, we can debate all day which model is correct but we know the average over a period of time will with 100% tell us the direction the race is going even if we disagree with the baseline of the polling method. If the race more or less stays the same as it's done for the last 2 months we will these type of debates for the next 8 months.

A poll is a poll, but the point of averaging is to blend the different methodologies so you get a consensus. When you weight one methodology four times as strongly as other methodologies, you aren't producing a meaningful consensus.

I can see keeping each data point in the average for the standard period, but you should weight them to be proportional to the rest of the pollsters. You would essentially have a rolling average for the Rasmussen methodology at equal weight with any other methodology.

Does anyone know when ARG and others may start doing daily polls like they did for NH in 2004. When I worked on the Clark campaign in 2004 they were one of the best measures of where the race was going.

I did not argue that they should be dropped, only that their inclusion on a weekly basis skews the poll results some downwards, which is something to keep in mind.  

Here is what I wrote on the subject in your previous diary:

I am actually glad every weekly Rasmussen poll (outliers, since not confirmed by other polls) is included in the averaging of polls, as well as the averages usually showing polls WITH Gore included, even though in most polls (as just occured with the Cook poll) Clinton gains significantly when Gore is removed in a poll, and the polls without Gore would give the more accurate reading.  

Those two factors, if corrected, would push her average lead to 16%, or even more.  That is not the actual state of the race (the inflated poll theory, if true, may account for a good 3, 4%, perhaps even 5%.)   So, the actual race is closer to the 11.4% the RCP averages actually show, so in a way the constant weekly inclusion of an outlier poll and the constant inclusion of Gore in polls (even when the same poll has data without Gore available and shows strong Clinton gains) acts as an equalizer to not show too massive of a lead, one that is obviously not going to stay that high.  

This is a useful and thoughtful discussion.  Averaging polls does not eliminate systematic error from a particular pollster's numbers--it just spreads it around thinly enough so that errors in one direction might be compensated with errors in the other direction.  Since most publicly released polls have crossed at least one editorial decision to publish the poll, there can be some quality control.  Averaging can also deliver a second level of editorial judgement and quality control:  Whoever decides to create an average can make judgements about which polls to exclude on the basis that they do not contribute to overall accuracy.  The person creating the average can exclude results from pollsters who are systematic only in the erratic nature of their results (usually a reflection of the absence of any attempt at solid methodology).  Your excellent summary of results of final polls in 2006 Senate races suggests that an average of live interviewer state polls, an average of IVR polls, and results from YouGov America (formerly Polimetrix) each yield pretty good estimates of the vote.  Averaged together, they might offer an even better insight into the topline state of a race.  One important shortcoming to averaging is that it tends to focus all attention on the total line, at the expense of attention to the underlying dynamics, the pattern of results by subgroups.  The comparison of subgroup analysis from the RT Strategies polls (which I conduct for the Cook Political Report) and the YouGov America poll, helps us understand this second level of analysis, namely, the difference between the way subgroups play out in a live interviewer poll vs. a legitimate online poll.  Many more such comparisons will have to be made before we complete our test of the online methodology against the tried-and-true standard of telephone polling (the problem of cell-phone-only households notwithstanding).  Thanks for providing a forum for this experimentation.

Chris;

I like your analysis as it was very interesting and informative.

1. I would be curious to read the mystery pollsters reading of your theory.

2. I would like to see Pollimetrix's poll's on some of the early primary and caucus states to compare with our live polls.

3. I would like to see if Rasmussen is going to do any state polls on the primaries and caucus states?

It is very possible.  I saw polls that show her stated support more solid than Obama's or Edwards.  But, how many of those respondents are pushed into making a decision right there and then?

There is no question that at this point a large portion of all 3 candidate's support is still soft.  In Opinion Dynamics' latest national poll
 http://www.foxnews.com/projects/pdf/0517 07_release_web.pdf

this question was asked:

9. Do you feel like you have already pretty much made up your mind about which candidate you will vote for in the presidential election, or are you still deciding?

Democrats: 3%

A huge percentage of voters is still in the decision making process, although the poll results suggest that many are leaning towards Clinton right now.  Obama gets decent support, Edwards is further behind.   She can't make any big mistakes, but the same is of course also true for Obama and Edwards.  

While everything is still open and subject to massive realignment upon disastrous mistakes made, there are two issues that weigh heavily against Obama and Edwards, although both COULD be made up over time:  

1. Edwards gets very low support from those polled when it comes to first AND second choices.  His total support in terms of 1st and 2nd choices hovers typically around 23% to 27%.  That has got to be troubling, as it suggests that Edwards' support is low not just when it comes to THE choice, but also the cache of potential converts in his backpocket.  By contrast, Clinton usually gets around 61% to 65% while Obama is at about 45% to 49%.   It would be interesting to see some national polls go deeper down the list and show third and fourth choices to see how much convincing Edwards has to do to get those not preferring him today to change their minds.

2. Obama is fairly weak in state polling.  In the poll aggregates he is not ahead in a single state aside from his home state Illinois.  He needs to change that to have a chance.  If he moves up in Iowa and/or NH, he could possibly make waves.

My read is that Clinton is certainly the favorite to win. I would consider her a strong favorite at this point.  But, if she commits a major mistake, it will probably go over to Obama.  I don't really see a scenario that has Edwards winning it, unless BOTH Clinton and Obama commit huge blunders which get wide media play.  I am not sure how likely that double whammy is.  It certainly can't be counted on.    A suggestion that I have seen made repeatedly has him winning Iowa and using that win to catapult him to the top.  This seems like a very high-risk/low-yield strategy.  His better bet would be for him to have a string of outstanding debate performances and for his campaign to somehow catch fire, make those who initially discarded Edwards as one of their top 2 choices to take a second look again.

If Obama's Iowa and New hampshire team gets every "young" voter that attends his events or has shown interest in his campaign to show up he will win, I understand the skeptics but Lamont won a primary like this as so will Obama.

That was what the Dean campaign thought in 2004. It  never happened. Of the top 3 candidates Clinton and Edwards have the support of the groups who turn out the most on election day. Clinton has broad support across a number of demographic groups and Edwards has older voters. "Movement" type candidates have pinned their hopes on the youth vote in a lot of election cycles since Eugene McCarthy in 1968 but rarely has it ever come together in a way where it worked.

Maybe in the primary states (although Lamont is an exceptional case rather than the rule as the other candidates aren't remotely as resented as Lieberman was), but in the caucus states young voters tend to be too concentrated to be worth too many delegates.

I wrote early, the primaries will be decided by the 18-25 voters, period, with the independents thrown in for states that allow it.  Period.  Clinton needs to address indies and the young population, because she is/has lost ground there.

There is neither compelling data nor historical precedent to lend credence to the type of definitive voting strength that the writer envisions here.