Working hard to get ahead [Makers and Takers]

[Teflon93]

:doh2: got ya…and I’ve made a dufus of myself

Not at all—I should have explained why I chose per capita. Another way would have been to take the total income divided by total workers, but I didn’t feel like doing all that work when per capita income figures were handy.

Mathematically, a flat 20% tax would replace much of the revenue currently generated by income taxes, but there would be enormous political problems, given how many people do not pay income tax at all. Moreover, big Catholic families would have a much larger tax bill as children would become liabilities and not deductions under this scheme, if implemented in pure form.

The trick is not only to raise the required revenue, but to do so in a manner which will not get you thrown out of office or the legislation rejected altogether. Once we’ve exempted so many people from the tax rolls, it may be an impossible sell to get them to vote to go back on.

 

[cynic]

An interesting and easy to comprehend (very easy) analysis on the data used in this book:

Brandon K. Schultz:

"I’ve re-analyzed the surveys that Schweizer reports (which are readily available online) using SPSS 15.0. Based on my replication, there are several interesting methodological “choices” that the author makes to draw such grandiose conclusions.

Even though the surveys measure political views using ‘continuous’ items (e.g., a response format that ranges from 1 “Extremely Liberal”] to 7 “Extremely Conservative”]), the author compares only the highest extreme (7) and the lowest extreme (1) throughout the book. From a statistical standpoint, this is problematic because it ignores trends in the middle and looks only at the relatively few people who place themselves at either political extreme (on the General Social Survey, this equals 4.7% of the sample, or 2,394 of 42,096 respondents; this number drops even lower when comparisons are made due to missing data in the comparison variables).

Here is a representative sample of the problems this causes: On page 20 Schweizer analyzes the General Social Survey and claims that 23% of liberals and only 14% of conservatives feel that Jews are especially violent. I re-ran this analysis (using Schweizer’s exact methodology) and here are the results when you examine the whole political spectrum, going from 1 (Extremely Liberal) to 7 (Extremely Conservative):

1=22.7%; 2=12.2%; 3=9.1%; 4=10.8%; 5=14.6%; 6=11.8%; 7=14.1%

See any anomalies? Hmmm… That “Extremely Liberal” group looks funny, doesn’t it? And it’s nothing like groups 2 or 3–the folks who called themselves “Liberal” or “Slightly Liberal,” respectively. The problem is that very few people identified themselves as Extremely Liberal. In this instance, that 22.7% is 17 of 75 respondents. So, on the basis of what 17 of 75 people said, Schweizer wants us to believe that Liberal ideas tend to make people anti-Semitic. You just have to ignore the fact that only 34 of 279 people (12.2%) who called themselves “Liberal” (group 2) agreed, which, by the way, is a lower percentage than that of people who called themselves “Extremely Conservative.” But that’s not worth mentioning, is it? Schweizer sure doesn’t think so.

Are you getting a sense of the goofiness yet? Well, here’s another example in case you still have blind faith in Mr. Schweizer:

On page 142 Schweizer uses the same survey (GSS) to show that Liberals use drugs and alcohol to cope with anger at “five times the rate” of Conservatives. Here we go again (Remember, 1 = Extremely Liberal and 7 = Extremely Conservative):

1=30.4%; 2=5.9%; 3=6.4%; 4=6.3%; 5=2.3%; 6=6.3%; 7=5.3%

Wow, that Extremely Liberal group is doped up, isn’t it??? Well, that’s because 7 of 23 “Extremely Liberal” people agreed that they drank or used drugs to cope with anger. That’s right, SEVEN of TWENTY-THREE people. What about “Liberals”? Not so much. Only 8 of 135 agreed. Get the picture? Well, just in case, let’s do one more:

On page 142 Schweizer uses the GSS to conclude that Liberals experience extreme rage more often than do conservatives (22% to 15%), based on their responses to a scale of rage that went from 1 (little rage) to 10 (extreme rage). Looking for those folks who rated their rage a 10 (the same way Schweizer does), here is the full political spectrum:

1=21.7%; 2=13.3%; 3=12.1%; 4=13.1%; 5=13.1%; 6=13.6%; 7=15.8%

Man, those “Extremely Liberal” folks are sociopaths, huh? Yup, all FIVE of them. That’s right, 5 of 23 – that’s how Schweizer got 21.7%. And yes, Schweizer did wrongly round 15.8% down to 15% for the Extremely Conservative group for some reason. Perhaps 16% sounded too angry. (And again, let’s just conveniently ignore the fact that smaller proportions of Liberals reported the same level of rage than did Conservatives or Extreme Conservatives.)

Schweizer says that the GSS is an authoritative survey and he’s right. But when you chop up data any way you please, you can “prove” anything, even with the best survey data in the world. Trust me, using Schweizer’s same goofy methodology and the same surveys, the “Extremely Conservative” folks look pretty bad on items related to attitudes toward Blacks, interracial marriages, segregated schools, whether whites have a “right” to live in all-white neighborhoods, and whether wives should have reproductive choices. But that’s junk statistics, and it’s not worth reporting, even here on Amazon for free. Unless, of course, the Hoover Institute wants to fund my new book project: “Acres of Fakers.”

If you want to see if “liberalism” is related to how people behave socially, you should start by looking at how these various survey items CORRELATE with one another. At the very least, correlations would make use of all the data, and you can ask the question: Are people MORE angry the MORE liberal they claim to be? This seems to be the kind of question that Schweizer would love to answer, but apparently he didn’t like the results. I’ve looked at the correlations and they are microscopic. For example, the correlation between political views and opinion of Jews as violent is .03.In other words, there are no meaningful relationships between how we rate ourselves along the political spectrum and the behavioral phenomena that Schweizer tackles in his book (at least not that I’ve discovered yet).

So in short, Schweizer’s research is wrought with undeniable statistical errors and shortcuts. As such, this book is valuable in two ways: 1) it underscores the reason why the academic peer-review process is vital in true scholarly work, and 2) it also demonstrates the quality of “research” funded by the Hoover Institute at Stanford. "

There has been a long drawn out debate about a book which largely mis-represents the views of it’s target.

And what a waste of time. Perhaps the OP should get hold of a copy and *read *it along with the surveys before posting, instead of simply the sound of the title… the sub-title… the abstract…

 

[Teflon93]

The statistical “debunking” seems to have missed a couple of salient points. One of these is that error due to the stratification of samples can be assessed only when one has some notion of the true proportion between sample size n and population size N. Now, it is true that the shape of the curve representing sampling error is dependent early on with the sample size n (by this we mean that regardless of population size, increasing the size of a sample by n=n+1 for very small samples will have a greater impact on reducing sampling error than reducing the population size in the frame of reference). So the question is, how many samples were in the “Extremely Liberal” bucket?

If n < 10, the critic would have a point. If n > 30, I don’t think they have one.

The way to figure this out is to calculate the sampling error for each strata.

I don’t see anything in the excerpt to indicate that this was done.

It is common in such surveys to correct for differences in sample sizes of responses by weighting. This is largely due to the fact that conservatives and liberals may not respond to such a survey in equal proportion—response rates may be different. Left unweighted, this results in overrepresentation of one group or another and skews results (you see this all the time in political polling done by media outlets rather than professional pollsters).

Another issue with the debunking is it presumes that people are equally likely to identify themselves as liberal as they are conservative. In fact, if I recall correctly, the current trend is for conservatives to be about 3 times more likely to self-identify as conservative than liberals will as liberals. This is because liberals do not believe the “liberal” label plays well in the political marketplace. This is why Barrack Obama isn’t running as a “proud liberal” despite his sky-high ADA rating as the most liberal U.S. senator. The problem for the survey should be obvious—it will skew all responses from the liberal side toward the center, regardless of what respondents truly believe. One way to get around this is to ask demographic questions which determine one’s affiliation rather than relying on self-identification; I do not know if this was done in the study.

The net-net is people identifying themselves as “extremely liberal” would be far smaller than the population identifying themselves as “extremely conservative”. This could be dealt with either by acknowledging a greater margin of error for extremely liberal results or by weighting or by filtering using demographic questions. There certainly are other valid techniques; I don’t see much evidence of discussion of this in the “debunking”. It possibly was done but has been dumbed down for the layman to the point where it is invisible to the practitioner.

 

[cynic]

.
Another issue with the debunking is it presumes that people are equally likely to identify themselves as liberal as they are conservative. In fact, if I recall correctly, the current trend is for conservatives to be about 3 times more likely to self-identify as conservative than liberals will as liberals. This is because liberals do not believe the “liberal” label plays well in the political marketplace. This is why Barrack Obama isn’t running as a “proud liberal” despite his sky-high ADA rating as the most liberal U.S. senator. The problem for the survey should be obvious—it will skew all responses from the liberal side toward the center, regardless of what respondents truly believe.

Which is main point isn’t it. There is such a small number of respondants identifying as “extremely liberal”. Whatever way you look at it this is flawed.

 

[Teflon93]

Which is main point isn’t it. There is such a small number of respondants identifying as “extremely liberal”. Whatever way you look at it this is flawed.

Not at all.

When you say “such a small number” you’re presuming that it is absolutely so (without boring everyone with the formula, let’s just say that “such a small number” in this context is <10) or that the confidence interval for that strata is unacceptable. The first case is not true and the second is unknowable without the data; I’d have to at least have the sample standard deviation to plug the numbers and guess.

When I take the absolute worst case—discrete data—I get a confidence interval of 24%. That’s not good, but it isn’t awful either, and how much it impacts the data depends on the mean response.

I’m very skeptical when people claim they “replicated” any dataset. You can’t really do that without throwing away lots of information you need for statistical analysis, especially information about variation.

I’d be more impressed with a transparent analysis from the study’s raw data. I do not know whether the author has made it available (perhaps that’s why “replication” was necessary). In my opinion, he should make it available, because transparency’s good.

But be wary of folks with axes to grind about the conclusions of the study coming back and asking rather silly questions based on the surface of the data. One good check of this is whether they reference variation at all or just talk means. The latter is an indication of some gigantic gaps in understanding or a really dumbed-down analysis for the sake of the audience.

And for anyone who doesn’t do this type of analysis for a living, please keep in mind that what they say is partially true—anybody can torture data to confess anything they want to. It is also true that anybody can catch them in the act provided they have the raw data.

This “debunking” simply doesn’t seem conclusive or serious to me, and stems I would imagine from someone who would have been an “extreme liberal” in the study wanting to gut it. This doesn’t mean the original study was valid; it does make you wonder about the “debunking”. But perhaps there was more to it than posted, too.