The Rasmussen Individual State Robo Polls start to get really kinky....

[The Vorlon]

These Robo-Polls seem to be going a tad off the rails...

Might need an oil change or a tune up or something... reboot that hard drive.... Upgrade to XP Pro or something...?

They're are not really "polls" per se, but simply adding up the State by State data already collected from May 1st, to May 31st, as part of his National Tracking poll.

For what it is worth....

Bush is up by 2% in Virginia (47% to 45%)
This is Bush doing 7% worse than in 2000

http://www.rasmussenreports.com/Virginia%20Jun%204.htm

Bush is up by 1% in Pennsylvania 45% to 44%
This is Bush doing 5% Better than he did in 2000

http://www.rasmussenreports.com/Pennsylvania%20Jun%204.htm

Even if we assume there is no methodological errors of any kind in Rasmussens polls (unlikely...) based on the very small sample size of just 400, we can expect that only 45% of his results will be within 3% of the quoted candidate lead, and about a 1 in 3 chance that any individual poll is out by more than 5%..

Some of these state results are starting to look a bit strange...

[Reds4]

It seems that any state which a candidate (either Bush or Gore) won by about 5 to 10 points last time, is showing closer this time. For instance, Virginia down to 2 points for Bush (which is almost undoubtedly wrong), NC down to 4 points, Bush up 1 in PA when losing by 5 last time,  Bush up 1 in MO when winning by 3.5 before. It seems Rasmussen polls and his weightings are set where almost every state that is within 10 or 11 points or so shows as a dead heat in his polling. I'm counting on the mason-dixon state polls much more and I know Vorlon is doing the same.

[muon2]

Taking a statistical study that was designed for one task, and then using its innards to draw other conclusions is a tough task. The problem is not just one of statistical errors. Systematic bias in the larger sample, even if controlled in that sample, is magnified when subsets are examined.

It is much better to take a number of restricted samples, then weight them into a large sample. In fact I think you demonstrated that the worked in the case of the individual state polls predicting a national poll.

[The Vorlon]

Taking a statistical study that was designed for one task, and then using its innards to draw other conclusions is a tough task. The problem is not just one of statistical errors. Systematic bias in the larger sample, even if controlled in that sample, is magnified when subsets are examined.

It is much better to take a number of restricted samples, then weight them into a large sample. In fact I think you demonstrated that the worked in the case of the individual state polls predicting a national poll.

Which roughly translated means this......

When you do a national sample, the sample as a whole may be right, but individual samples within states are not representative.

If you do a national poll, within in a state like Idaho you might call 5 or 6 people, which - needless to say - is not a representative sample of Idaho.

You do this 30 times and add up your 30 un representative samples of Idaho and you get one much bigger, but still unrepresentative sample of Idaho.

---------------------------

I expect that for his National polls Rasmussen is using something called a "cluster sample" bought from a firm that specializes in producing telephone calling lists (very creatively named "Survey Sample International") to have his robots make his calls from.

Cluster samples are designed to give you representative sample for the entire population, but not within small sub sets.

For example, the sample he is buying is balanced to indeed give you 51% Women and 49% men nationally, but it is not guaranteed to give you 51%/49% in say Ohio or New Jersey..

In the big states (California, texas, maybe down to say New Jersey or so) I might work ok, but after that it gets pretty dicey IMHO.

If I were to take a guess.... (and labled as such) based on what I know of how Cluster samples usually fill out when you actually try to call them..

Bush will do worse than expected in the EST time zone, about right to a bit better than expected in the central time zone, and also a bit  better than expected in the mountain and pacific time zones...

Bush +2 in Washington State anybody?

Again this is a semi-educated guess... , but I'd be willing to bet a reasonably fresh bag of donuts..

These errors are over and above the BIG sampling random error from only 400 sample size, which will be in additional to any survey design error...

Maybe Rasmussen is just doing this to save money?

These "polls" cost him nothing - he has already collected all the data, he just has to add them up and he gets "free" polls to use... ?

[Smash255]

I wouldn't be that suprised if Virginia is as close as this poll suggest.  Northern Virginia in the D.C burbs especially Fairfax County (a county Bush won in 2000) has shifted to the left & is a county Kerry will probably win comfortably (6-10%) this year.  The shift in the D.C burbs from Right wing leaning to left wing leaning will have an impact & its shown by this poll.  Now before we can say its going to be battleground or not we need to see more polls, but the heavily populated Southern Suburbs of D.C have trended to the left so this poll could be accurate

[JNB]

 If the polls are accurate, I think it shows there have been a few small shifts. My gut feelings is that Bush is doing worse among white males than he did in 2000, on the other hand, he seems to be doing better among females and Hispanics than he did in 2000, and I think this may explain some of the odd polling results.

[Rococo4]

I wouldn't be that suprised if Virginia is as close as this poll suggest.  Northern Virginia in the D.C burbs especially Fairfax County (a county Bush won in 2000) has shifted to the left & is a county Kerry will probably win comfortably (6-10%) this year.  The shift in the D.C burbs from Right wing leaning to left wing leaning will have an impact & its shown by this poll.  Now before we can say its going to be battleground or not we need to see more polls, but the heavily populated Southern Suburbs of D.C have trended to the left so this poll could be accurate

I would be surprised if it was that close in Virigina....  it may be closer than 00, but Bush is still going to win the state.

[Fmr. Gov. NickG]

There may be a trend to this effect, but I'm beginning to suspect that all of Rasmussen's polls are being weighted to appear artificially close.  

I've just updates pollbooth BTW, but I didn't include any of these month-long robo survey.  
Strange lack of national polling this week, too.

[agcatter]

I kind of go with Gov. NickG on this one regarding Rasmussen's state polls.  They seem to indeed show states much closer than they should be time and again.  I mean, I don't anymore think Va. is that close than I think Bush is close in NJ or tied or leading in Oregon or Penn.

Vorlon may have explained why.  I almost prefer the one night state polls that he did in 2000 rather than the 30 day polls where he just adds up the data for a month.

[The Vorlon]

There may be a trend to this effect, but I'm beginning to suspect that all of Rasmussen's polls are being weighted to appear artificially close.  

I've just updates pollbooth BTW, but I didn't include any of these month-long robo survey.  
Strange lack of national polling this week, too.

I am with you on this.  Even if we put aside the GRAVE doubts I have about this from a statistical validity point of view, that fact that the samples are so small (400) and the time frame so long (30 days) I think makes it more likely we should toss them than use them...

Unreleated but some good news on the polling front...

A bunch of the big newspapers in Florida just got together and signed a deal with Mason-Dixon to poll the state for them, rather than use their own individual crews of Ragtag pollsters they typically use.

This is great because we'll be getting some good sized samples, and frankly because Mason-Dixon is pretty much acknoledged by everybody in Florida as being the only folks who can actually get the right numbers in that state. (Florida is just a nightmare to poll BTW - I hope M/D is charging them top $$$ because they will earn every penny of it in Florida)

[muon2]

Taking a statistical study that was designed for one task, and then using its innards to draw other conclusions is a tough task. The problem is not just one of statistical errors. Systematic bias in the larger sample, even if controlled in that sample, is magnified when subsets are examined.

It is much better to take a number of restricted samples, then weight them into a large sample. In fact I think you demonstrated that the worked in the case of the individual state polls predicting a national poll.

Which roughly translated means this......

When you do a national sample, the sample as a whole may be right, but individual samples within states are not representative.

If you do a national poll, within in a state like Idaho you might call 5 or 6 people, which - needless to say - is not a representative sample of Idaho.

You do this 30 times and add up your 30 un representative samples of Idaho and you get one much bigger, but still unrepresentative sample of Idaho.

Thanks for the English language translation of my post.

I do get terse at times when I'm rushing a message midday.

[NHPolitico]

These Robo-Polls seem to be going a tad off the rails...

Might need an oil change or a tune up or something... reboot that hard drive.... Upgrade to XP Pro or something...?

They're are not really "polls" per se, but simply adding up the State by State data already collected from May 1st, to May 31st, as part of his National Tracking poll.

For what it is worth....

Bush is up by 2% in Virginia (47% to 45%)
This is Bush doing 7% worse than in 2000

http://www.rasmussenreports.com/Virginia%20Jun%204.htm

Bush is up by 1% in Pennsylvania 45% to 44%
This is Bush doing 5% Better than he did in 2000

http://www.rasmussenreports.com/Pennsylvania%20Jun%204.htm

Even if we assume there is no methodological errors of any kind in Rasmussens polls (unlikely...) based on the very small sample size of just 400, we can expect that only 45% of his results will be within 3% of the quoted candidate lead, and about a 1 in 3 chance that any individual poll is out by more than 5%..

Some of these state results are starting to look a bit strange...

Bush's lead in the South was based on cultural issues-- guns, porn, church and state, abortion, etc.  Those issues aren't resonating this year. That means these voters are choosing based on other issues-- the economy and Iraq. It also means that they will vote more the way other states do. So, you have Bush polling the same in almost every state. So, you have Bush either winning big or losing big as voters break one way or the other in every state.

[NHPolitico]

 If the polls are accurate, I think it shows there have been a few small shifts. My gut feelings is that Bush is doing worse among white males than he did in 2000, on the other hand, he seems to be doing better among females and Hispanics than he did in 2000, and I think this may explain some of the odd polling results.

I saw a poll that showed just that-- a shrinking gender gap. Voters of all backgrounds and sex organs are thinking about the race in the same way. Bush has to get them to think about it his way or he'll lose big.

[Smash255]

Rasmussen Jersey polL Kerry up 51-39

[JNB]

 I know it costs extra $$$$, but I would like to see more polls broken down by rage, gender and Conservative, moderate, liberal view.

  In 2000, Bush cleaned Gores clock among white males by winning 62-36, and he had almost 90% of the conservative vote.  My gut feeling is he is down to the mid to high 50s among white males and around 75% among Conservatives. His stands on immigration and Trade has not helped him with many among this group.

[Filuwaúrdjan]

If the polls are accurate, I think it shows there have been a few small shifts. My gut feelings is that Bush is doing worse among white males than he did in 2000, on the other hand, he seems to be doing better among females and Hispanics than he did in 2000, and I think this may explain some of the odd polling results.

I've been thinking the same thing...

[TeePee4Prez]

 I know it costs extra $$$$, but I would like to see more polls broken down by rage, gender and Conservative, moderate, liberal view.

  In 2000, Bush cleaned Gores clock among white males by winning 62-36, and he had almost 90% of the conservative vote.  My gut feeling is he is down to the mid to high 50s among white males and around 75% among Conservatives. His stands on immigration and Trade has not helped him with many among this group.

I think what helped him win Ohio and come close in Pennsylvania last time was the fact that he came off as a compassionate moderate.  He garnered a lot of union support which is unheard of from Republican candidates.  He hit on his moral agendas really hard and practically assured labor he would not bite.  I don't know about Ohio, but I do not see him winning PA even though Rassmussen suggests it's possible.  The Philly papers and readers' comments are grilling him like a charcoal hamburger especially since NIck Berg was killed.