chrisishardcore.com

Comments: Crack for the Weak?

Sorry, Chris. Your daily numbers are bogus. A note on the Rasmussen website says:

"Today's decline for Kerry is primarily the result of dropping Sunday's data from the three-day rolling average."

This indicates two things: 1) That the 4/4 numbers (what you call the 4/5 numbers on your site) were strongly Kerry, that is, greater than the current 3 point spread. But your guess for this date is 45-47.
2) That the latest daily was approximately in line with the previous two. But your guess of 44-44 is not in line with the previous two days 40-48 and 42-47.

In general, it's impossible to determine the daily numbers from a rolling-average poll without at least one known daily number to start from. If you don't know one real daily number, and guess instead, your guess will affect every single number from that point forward. And every daily guess will have no more value than your original guess.

Given today's clue, we can make a better guess at the dailies. I'd guess the following. Note that what you label as 4/8 is really the 4/7 sample.

4/7: 43-46 (known to be approx in line with previous)
4/6: 43-46
4/5: 43-47
4/4: 40-50 (known to be a big Kerry day)

Your guesses for these dates were: 44-44, 40-48, 42-47, and 45-47.

The new guesses are undoubtably wrong, but much closer to reality than your current guesses. Hopefully, Rasmussen will slip up and tell us one real daily number. But even then we would slowly lose accuracy due to rounding.

Posted by Tom McDonald at April 8, 2004 01:30 PM

Unfortunately, Tom is right. I used to be an analyst for a polling firm in NVA. Unless you have a true daily from which to start, your numbers are undoubtedly incorrect. Even if you had a true daily, as time goes on, like Tom accurately states, your numbers would eventually be off track due to rounding. However, Tom forgot to applaud you for making the effort, which I am sure was simply a mistake, so I will thank you for him. Keep up the good work.

Posted by David A. Curtis at April 8, 2004 03:59 PM

David, you're right, should have given kudos. Chris is awesome, combining the best qualities of blog compulsion and political obsession. Rock on.

Posted by Tom McDonald at April 8, 2004 04:29 PM

Was going to post what Tom did, but then saw his post. Also, you've micalculated your moe.

Posted by VoR at April 11, 2004 11:15 PM

Why do you need to "guess" your margin of (sampling) error? It can be estimated quite readily by calculating the reciprocal of the square root of the sample size. To wit:

1/sqrt(1500) = 0.0258 = 2.6%, to 2 significant digits
1/sqrt(500) = 0.0447 = 4.5%, to 2 significant digits

This estimate is accurate as long as the population size is much larger than the sample size, as is the case here. Also, it only estimates random (sampling) error, not systematic error (for which there is no ready estimate).

For more info, consult Bevington:

http://www.amazon.com/exec/obidos/tg/detail/-/0072472278/103-9489233-1379857?v=glance

Posted by Lara Inis at April 21, 2004 01:01 AM

Tom McDonald writes: "In general, it's impossible to determine the daily numbers from a rolling-average poll without at least one known daily number to start from."

In fact, for a series of 3-day rolling averages (as in the case of the Rasmussen polls), you would, in principle, need 2 additional constraints, e.g., 2 daily numbers or 2 weekly averages. Otherwise you would be trying to solve for n unknowns given only n-2 linear equations. [In general, for a series of p-day rolling averages, you would need p-1 additional constraints.]

Since Rasmussen now appears to be releasing weekly averages as well, you're in business. All you need are the numbers for a two-week period (12 3-day rolling averages and 2 weekly averages) giving you 14 linear equations in 14 unknowns. The rest is textbook algebra.

Posted by Lara Inis at April 21, 2004 03:08 AM

Post a comment