Stones Cry Out Annex

2005-02-07T20:01:00.001-08:00

2005-02-07T20:01:00.000-08:00

2005-02-07T20:00:00.001-08:00

2005-02-07T20:00:00.000-08:00

2005-01-19T22:14:00.000-08:00

2005-01-19T22:11:00.000-08:00

2005-01-19T22:00:00.003-08:00

2005-01-19T22:00:00.002-08:00

2005-01-19T22:00:00.001-08:00

2005-01-19T22:00:00.000-08:00

2005-01-19T12:14:00.000-08:00

2005-01-17T11:28:00.000-08:00

2005-01-17T01:14:00.000-08:00

2005-01-16T21:16:00.000-08:00

2005-01-16T20:00:00.001-08:00

2005-01-16T20:00:00.000-08:00

2005-01-16T18:15:00.000-08:00

2005-01-16T16:23:00.000-08:00

2005-01-16T07:47:00.000-08:00

2005-01-14T15:26:00.000-08:00

2005-01-06T11:10:00.000-08:00

Rick Brady

Rick Brady

2005-01-03T01:34:00.000-08:00

2005-01-02T23:58:00.000-08:00

Recall that the authors wrote: "...assuming proper poll methodology, no discriminatory voter supression, and an accurate and honest vote count...the probability that Kerry would have received his reported popular vote total of 48.1%, or less is one in 959,000..." (p. 10).

2005-01-02T23:20:00.000-08:00

There appears to be a typo in the Simon/Baiman paper. In the footnote on page 10, they show P(0.481)=1-NORMDIST(0.481, 0.508, 0.005686, True)=0.0000010424. However, when I run the exact parameters listed in the paper I get: P(0.481)=.0000010259, which yields odds of 1:974,772, not 1:959,000. Another conservative error.

2005-01-02T21:33:00.000-08:00

2005-01-02T19:12:00.001-08:00

2005-01-02T19:12:00.000-08:00

2005-01-02T19:11:00.002-08:00

2005-01-02T19:11:00.001-08:00

2005-01-02T19:11:00.000-08:00

2005-01-02T11:30:00.000-08:00

2005-01-02T11:12:00.002-08:00

2005-01-02T11:12:00.001-08:00

2005-01-02T11:12:00.000-08:00

Stock Market Option Support Cross Cuts Age and Part Affiliation

2004-12-22T13:24:00.000-08:00

According to the poll, 71 percent of Republicans and 41 percent of Democrats support a Stock Market Option for private investment of Social Security Funds.

Support for the Stock Market Option gradually declines with the age of the respondent, but 67% of 18-29 year olds, 60% of 30-39 year olds, 58% of 40-49 year olds, 46 percent of 50-64 year olds, and 30 percent of 65+ year olds would support an option. If the older populations were held harmless with the plan, I suspect the support among those held harmless would likely increase.

Expectation of Full Benefits Low Among Those under 50

2004-12-22T13:23:00.000-08:00

According to the Washington Post/ABC News Poll, 93% 18-29 year olds, 85% of 30-39 year olds, and 72% of 40-49 year olds expect to receive full social security benefits.

ABC News Distortion of Push Poll

2004-12-22T13:21:00.000-08:00

Poll: Few See Crisis in Social Security

And just a quarter of Americans agree with President Bush's portrayal of the Social Security system as in crisis — making it more difficult to raise support for fundamental change…

Well, that is one way to look at the data. But contrast that with this statement by the Washington Post about the same data:

74 percent think the system faces either major problems or is in crisis -- as Bush has asserted.

Talk about a distorted headline!

Washington Post Reporting of their Push Poll

2004-12-22T13:18:00.000-08:00

Details Cloud Support for Social Security Plan

The president also has at least general support from 53 percent of the public for the concept of letting people control some of their contributions to invest in the market. It is on the specifics that Bush faces problems. Support dropped to an even split when people were told that the cost of the transition to a new program could reach $2 trillion over time, as some forecasts project. And 62 percent said they would not participate in such a program if it meant their retirement income would go up or down depending on the performance of their stock picks -- which is the essence of Bush's plan.

The question did not refer to the cost of the “plan”, but instead asked if respondents would “Support a stock-market option funded by $2 trillion in borrowing.” Big difference. Forget also that the details of the Bush plan aren’t known and therefore the $2 trillion figure is probably has a margin of error of +/- $1.5 trillion. Who knows? Bush hasn’t released his plan.

2004-12-20T06:43:00.000-08:00

Go Collin!

2004-12-16T14:25:00.000-08:00

2004-12-12T16:12:00.000-08:00

Please send e-mail to rick at alohalee dot com if you would like the spreadsheet for these tables.

2004-12-08T06:32:00.000-08:00

Note the formula I used for comparing a poll result to an established standard. The denominator of this formula is the calculation for standard error assuming a simple random sample. All I need to do is calculate the standard error for cluster sample and insert it into the denominator to yield a comparative Z-score and determine whether the variance between the exit polls and the election result is statistically significant in each state.

The formula to calculate standard error for any sample where you do not know the Design Effect as described by Nick above, is simply the confidence interval of a poll divided by the confidence level of that poll. If you know these two pieces of information, you can calculate the standard error of any poll that would yield a normal distribution after drawing 100 samples.

Therefore, I posed the following question to Jenn Agiesta of Edison Media Research,

How can I find out the margin of error associated with the exit polls within each of the 11 battleground states? Your FAQ site says it is typically around 4%, but that isn't very precise. (Note: I assumed 95% confidence level, but should have asked)

Jennifer replied,

If you have the sample size for each survey (or if you send me a list of your 11 battlegrounds which I can use to get you the sample size), you can use this chart to determine the margin of error for each question you're looking at. (Note: The chart says the confidence interval is associated with a 95% confidence level)

Because Dr. Freeman's data were extrapolated from the gender split as reported by the CNN web-site, the 50% row most closely approximates the gender split of samples taken in each state, which served as the basis of determining the aggregated Bush/Kerry/Other proportions used by Dr. Freeman and myself. Admittedly, using 50% is not precise because the exit poll samples included a larger proportion of women than men and varied by state. I've requested more detailed confidence intervals associated with each state based on the exit poll gender split per state, but have yet to receive a reply.

Yet the 50/50 row is very close. The gender splits in each state ranged from 42/58 to 50/50, with most fairly close to 50/50.

2004-12-08T06:07:00.000-08:00

Last week Mark Blumenthal pointed me to this post by MyDD with a warning from Nick Panagakis of the National Council on Public Polls.

Mr. Panagakis begins by fisking Dr. Freeman's approach to the analysis,<>

The conclusion that "exit polls differed significantly from recorded tallies" in the three states is incorrect.

However, Freeman's page 6 footnote is correct: "This analysis assumes a simple random sample. If on the other hand, states were broken into clusters (e.g., precincts) and then the clusters (precincts) were randomly selected (sampling individuals within those selected precincts), the variances would increase."

By necessity, exit poll samples are cluster samples. The number of precincts in states typically number in the thousands. Wisconsin, for example, has 3,700 precincts. Illinois, a larger state, has 10,000.

Standard error assuming a simple random sample is calculated, but only as a first step. A confidence level of at least 99% is assumed - higher than the customary 95% - probably because of the higher standard of precision for exit polls and the number of races involved, about 100 across the states including the race for president and races for senate and governor on November 2.

A measure called the Design Effect must then be calculated to adjust the standard error for the cluster sampling effect. The magnitude of the Design Effect depends on the average number of interviews per precinct in each a state sample. The smaller the number of average interviews per precinct in a state, the smaller the design effect. Design Effect also differs by characteristic and can be much larger for characteristics highly clustered by precincts such as race. Design Effect is a variance measure so the square root is used to multiply the standard errors.

Without knowing the number of precincts sampled, you can't calculate the Design Effect. But Design Effect square roots are said to have typically ranged from 1.5 to 1.8 in the November exit poll. I used 1.6 as a "best estimate".

Conclusion. All of the state estimates above are well within their error calculations below.

Ohio, n = 2020. Sqrt (.5 X .5) / Sqrt 2020 X 2.6 X 1.6 = +/- 4.6%.

Pennsylvania, n = 2107. Sqrt (.5 X .5) / Sqrt 2107 X 2.6 X 1.6 = +/- 4.5%.

Florida, n = 2862. Sqrt (.5 X .5) / Sqrt 2862 X 2.6 X 1.6 = +/- 3.8%.

2004-12-08T06:03:00.000-08:00

Recall that I wrote the following:

...There are some inherent limitations of this model, which are explained in greater detail here. (link)

...Although, Mystery Pollster does discuss the issue of random sampling, vs. cluster sampling, which is admittedly a weakness of both Dr. Freeman's and my analysis. I tried my best to deal with this issue here as linked to in my post Exit Polling 2004: Significantly Variant from Election Result?. While the differences between a random sample and a cluster sample may increase the margin of sampling error associated with the exit poll, and therefore the Z-scores and levels of confidence presented in Exhibit 2 above are likely inflated, I'm not sure that this would change the results much. Even Mystery Pollster admits this additional unaccounted for error is likely not fatal to Dr. Freeman's results. To refine my analysis, I will look at this a bit more and see if there is a way to alter the equation to incorporate the higher standard error associated with the cluster sample. This may take some time... (link)

...Dr. Parker of SDSU does not know of a way to adjust the equation to incorporate a higher level of error associated with the cluster samples, without more information on how the cluster sample was established and what standard error is associated with that sample. Like Mystery Pollster though, he doesn't think the error introduced by a cluster sample as opposed to the simple random sample, if done properly, would void the results. (link)

...I've been thinking more about my exit poll posts and how much the differences between a simple random sample and a cluster sample can effect my conclusions. Frankly, I think the affect could be greater than either Mark Blumenthal or my professor let on. I've been in contact with folks at Edison Media Research trying to nail down the exact margin of error for each of the exit polls so I can refine my calculations. (link)

2004-12-03T19:57:00.000-08:00

2004-11-30T07:49:00.000-08:00

2004-11-29T07:51:00.000-08:00

2004-11-27T13:26:00.000-08:00

One of my classmates got a 179 on her LSAT. She was accepted to Harvard Law, Stanford Law, Yale Law, and PhD programs at Harvard and University of Chicago. She chose to do a dual JD/PhD at Harvard and wanted to finish both in 5 years. I think she was smoking something because from everything I saw, she was a major ditz. Often, she showed up to class unprepared and had a habit of turning in work late and incomplete. She said she was so "burned out." That was her excuse for everything and I didn't want to work with her. She might have been a brilliant test taker, but she didn't know how to apply herself. I wonder how she is doing right now.

2004-11-25T23:23:00.000-08:00

Hugh Hewitt traffic. Notice drastic spike on election day.

2004-11-25T23:22:00.000-08:00

DailyKOS traffic. Notice drastic dip on election day.

2004-11-24T14:06:00.000-08:00

2004-11-24T13:58:00.000-08:00

I wrote this in haste and realize this statement isn't really accurate. My statement assumes as simple random sample. Exit polls are typically based on clustered samples. Therefore the standard error of the poll sample could be higher than assumed by my statement, which would lower the confidence level and increase the cofidence interval.

2004-11-24T13:15:00.000-08:00

I've noticed a problem with this table. The results are fine, but the Kerry Predicted column is way off. This was an error in trascription and not with the spreadsheet from which this table is based.

Unfortunately the software HELLO service that I use to upload images to blogger is down at the moment so I cannot correct the error. If you would like the spreadsheet that I used for this table, please send an e-mail: rick at alohalee dot com

2004-11-24T07:29:00.000-08:00

In his working paper, Dr. Freeman wrote,

A random sample of a population can be modeled as a normal distribution curve. Exit polls, however, are not random samples. To avoid prohibitive expense, exit poll samples are clustered, which means that precincts, rather than individuals, are randomly selected. This increases variance and thus the margin of error because of the possibility that precinct voters share similar characteristics that differentiate them from the rest of the state in ways that past voting behavior would not predict.

For the purposes of my analysis, the exit poll sample is assumed to attempt to achieve a normal distribution. The exit polling company is trying to achieve a survey result which most closely approximates the mean of samples in a normal distribution. In other words, if the organization were to take 100 samples and then take an average of the results of those samples, this average (mean of samples) should approximate the election outcome. The odds of approximating the election outcome with a single sample (as opposed to a mean of 100 samples) are higher the larger the sample size.

2004-11-23T21:43:00.000-08:00

This was hinted at by Dick Morris in his article.

The exit pollsters plead that they oversampled women and that this led to their mistakes. But the very first thing a pollster does is weight or quota for gender. Once the female vote reaches 52 percent of the sample, one either refuses additional female respondents or weights down the ones one subsequently counted.

This is, dear Watson, elementary.

2004-11-23T21:15:00.000-08:00

Dr. Freeman wrote in his paper,

It's likewise possible that the data were already partially calibrated to tallies, but given the news story and the abrupt change at 1:30 (EST), that too seems unlikely. If, in fact, the data already had been partially calibrated, however, it would mean that the pure exit poll numbers favored Kerry to an even greater extent" (pg 6).

I think he nails it here. If the data circulated by to clients had been partially weighted to election returns available at the time, then the error should be reduced and therefore his findings (and mine) are actually understating the variance!

2004-11-23T19:17:00.000-08:00

2004-11-23T17:31:00.000-08:00

2004-11-23T17:30:00.000-08:00

2004-11-20T13:22:00.000-08:00

Z-test used to compare a proportion derived from a sample to an established standard proportion.

Source: Rea and Parker. 1993. Designing and Conducting Survey Research: A Comprehensive Guide. 2nd Editon. San Francisco, CA: Jossey-Bass.

2004-11-20T12:21:00.000-08:00

Simplified table to use to determine whether the Z score is critical or within the realm of sampling error.

2004-11-16T09:24:00.000-08:00

The caucus could decide that Orin Hatch needs the position for two more years because he faces reelection in 2006 and would not have the benefit of a Chair position to help his campaign.

There could be any other number of "smokescreen" reasons to deny or delay a Specter chairmanship.

2004-11-16T07:33:00.000-08:00

"I wished he hadn't said it," Sessions said. "But if you read what he said in its entirety, it wasn't a direct threat to the president, although it would have been better left unsaid. It may simply be that there's a difference between being the chairman and making those statements and being a single member of the committee."

2004-11-13T11:36:00.000-08:00

My wife and I truly appreciate your generosity and
we'll be sure to be good stewards over your gift.

God's blessings.

The Brady's

2004-11-12T13:29:00.000-08:00

Part 1: Blogroll HTML

2004-11-12T13:26:00.000-08:00

Part 2: Blogroll HTML

Politics Defined

2004-11-12T11:41:00.000-08:00

"Politics is an ever-present social activity. It occurs whenever individuals seek to reach agreement on a course of common action despite their disagreement over the goals. Politics arises whenever people who have different, even oposing, inerests accepts that they must cooperate if their interest is to be served. Politics, so defined, occurs within the smallest, most intimate associations - as within families and among friends - as well as at city hall or in Washington, D.C. Successful politics almost always depends on bargaining and compromise as each side tries to get as much of what it wants from the agreement, whether friends are trying to decide on a restaurant or Congress and the president are wrangling over the budget. The essays in this book attest to the bargaining and compromise that occur in Washington, at least whenever politicans succeed in creating national policy."

---Samuel Kernell and Steven S. Smith, editors, in their introduction to "Principles and Practice of American Politics, Classic and Contemporary Readings."

2004-11-11T14:00:00.000-08:00

The formula you could use to determine whether a particular poll proportion
is significantly variant from another poll proportion.

Source: Rea and Parker. 1993. Designing and Conducting Survey Research: A Comprehensive Guide. 2nd Editon. San Francisco, CA: Jossey-Bass.

2004-11-11T13:59:00.000-08:00

Once you calculate the "Z" score using this formula: 1) calculate the degrees of freedom (still need to get that formula); and 2) look up your value on this table to determine if the variance is "significant" or if the variance is due to "sampling error."

Source: Rea and Parker. 1993. Designing and Conducting Survey Research: A Comprehensive Guide. 2nd Editon. San Francisco, CA: Jossey-Bass.

2004-11-11T13:57:00.000-08:00

This is just an example of Z scores.
This table is actually several pages long.

Source: Rea and Parker. 1993. Designing and Conducting Survey Research: A Comprehensive Guide. 2nd Editon. San Francisco, CA: Jossey-Bass.