US Media: Katrina and more

[cema]

1. What they said then, what they say now, and a few obvious comments: from Instapundit (and more references there, as usual).

2. Another study (UCLA) says US media is left-leaning. That's not new; but I wonder if there is not a deeper link between items 1 and 2. Maybe not. But some of the "what they said then but not now" things I mentioned above are closely related to favorite political issues: that the hurricane damaged blacks more than whites (not so), that more poor folks died than rich folks (ostensibly not so, which surprises me), etc.

Comments

[anhinga-anhinga.livejournal.com]

> 2

Something is wrong with the methodology, which makes the Drudge Report left leaning, and makes Joe Lieberman far to the left from the center...

(sorry about the login mix-up a moment earlier)

[cema.livejournal.com]

Methodology is a very weak point.

[anhinga-anhinga.livejournal.com]

Everything is weak about the study which finds the Drudge Report to be left leaning...

[cema.livejournal.com]

Drudge Report is obviously not left-leaning. However, the main bulk of its contents (when looking at the main page, www.drudgereport.com) consists of references to other media sources. Drugde selects which sources to quote and which articles to choose, but he works with what is available.

His own reports are not left-leaning, I am not sure if they are right-leaning or not, but definitely not left-leaning.

[anhinga-anhinga.livejournal.com]

Now, the mainstream media we are mostly exposed to (especially the newspapers) is obviously left-leaning.

Here is why it happens. The urban areas are left-leaning (as the well-known "skyscraper" voting map illustrates), and they are also associated with better education, higher IQs, and more interesting newspapers.

So the best and most notable newspapers are left-leaning, because their main markets are urban, and thus left-leaning. It's as simple as that...

[cema.livejournal.com]

I do not think it's that simple. I would expect other factors at work here.

[anhinga-anhinga.livejournal.com]

;-) Then these other factors probably cancel each other out ;-)

[cema.livejournal.com]

I am talking about factors that reinforce themselves, like reproduction of likeminded people in schools and universities.

[anhinga-anhinga.livejournal.com]

These people are always ready to "sell out" every time it is required of them. Those who are not usually don't keep their jobs for long.

I don't think that the newspapers are to the left of their primary markets, it is just that their primary markets are to the left of the average (because urban life is by definition more collectivistic). I think it's reasonably well described by "market considerations"...

Indeed, the journalists are to the left of their newspapers, but the business people running the newspapers are usually to the right of their newspapers (which also reinforces themselves, because of self-interest in tax policy and the like, and low tax rates on high incomes producing richer people). This is what cancels out: the journalists being to the left, and the executives and advertizers being to the right.

At the end of the day the long-established dominant newspapers are forced to be in tune with their primary markets, otherwise they would not be dominant. And the optimal balance between collectivistic aspects of life and individual independence moves somewhat to the left when people live closer together. Which is reflected in people's average political preferences, like what do they think about public transportation, private gun ownership, etc... "Market considerations" are an approximation of reality, but it's a reasonably good approximation...

[cema.livejournal.com]

You have a point, but, like I said, I expect other important factors to be at work. Cannot point them out now, no time to think, need to work...

[email-animal.livejournal.com]

After reading about 20 pages out of 62 of the paper itself, I can say the following. The authors of that study have invented their own definition of political bias - different from what is typically understood by it.

I am still working in my head the complete list of things I do not like about the study, but one of the problems is disingenuity. Their conclusion, which, I have no problem with accepting, reads roughly as follows:

Between 1995 and 1999, more news media outlets have exhibited patterns of citation of political think tanks similar with those exhibited by Democratic members of the Congress, than with those exhibited by Republican members of the Congress.

Now, what they do, and what I object to vehimently, is translate it into "newsmedia exhibits liberal bias."

I am commenting on this here because you and anhinga_anhinga are engaged in the discussion of what could possibly cause this "liberal" bias, potentially not understanding that in your head "liberal bias" means a different thing.

[syarzhuk.livejournal.com]

Это же очевидны - расистские луизианские власти отказываются заниматься дорогостоящей идентификацией чернокожих трупов. Поэтому среди *идентифицированных* белых непропорционально много.

[cema.livejournal.com]

Возможно.

[email-animal.livejournal.com]

The study does not outright belong to ljnauka, but the methodology, by which a congress with 55 Republican senators vs. 45 Democratic ones, and with about 25 more Republicans in the House gets a score of 50.1 on the (conservative=0/liberal=100) scale is flawed from the get-go.

Second, what the fuck is this: If a media outlet displayed a citation pattern similar to that of a lawmaker, then Groseclose and Milyo's method assigned both a similar ADA score.?

Lawmakers show a pattern of citing think tanks?

So, suppose I present a stinging crique of something supported by American Heritage Foundation, but in my critique, I only cite them, and no left-leaning think tank. Suddenly, I am exhibiting a right-wing citation pattern.

Collaborative filtering on objects of different types works poorly, and the validity of their scoring hinges on us believing that "if a senator had a speech in which he referenced X, Y, Z and a newspaper had an article in which it referenced X, Y, Z, then the newspaper article agrees with the senator politically"....


Finally, equating liberal/conservative bias with the frequency of mention of right-wing or left-wing outlets is also incorrect. Describing current situation in Iraq as either a raging success or a raging disaster (even in news coverage) can be done easily without quoting either liberal or conservative "think tanks".


Are quotes from government officials treated as neutral? Does it mean that an article that has Dick Cheney advocating torture and a counter-quote from the ACLU representative would be treated as evidence of liberal bias?

Does it mean that an article that cites "a highly ranked administration official who requested anonymity because his position in the administration would otherwise be jeopardized" selling administration's talking points with no counter from the left becomes classified
as "neutral"?

The latter cannot even be cross-listed with speeches in Congress, because congressmen do not refer in their speeches to "senior administration officials" or "attourneys with the knowledge of the case".

[cema.livejournal.com]

Methodology is a weak point here, certainly. I am not reading too much into it all. However, I am concerned about the quality of reporting (in general), and more so about the quality of editorializing. This is true about all sides, left or right or whatever (not everybody belongs to a camp). Applicable, in this case, both to the material of the study and to the study itself.

The study does not outright belong to ljnauka

Right. I am thinking about starting a group (did not find one in LJ) devoted to misusing statistics, both in journalistic studies and in science. Thinking about the name; how about stats_wtf? :-) Any better ideas?

[email-animal.livejournal.com]

I am happy to co-sponsor stats_wtf (-:

I have seen your complaints about media coverage over time. Yet, the crux of your complaints appears to be that the media is too liberal. Whereas the crux of my complaints is the media is simply not doing its job right.

It is so much easier to investigate a blowjob, apparently...

[cema.livejournal.com]

My complaints are that the media are not doing their job well. Thet they are too liberal is better rephrased as "they are too politically motivated"; a vocal minority are militantly right-wing, and that, of course, is not better at all.

A blowjob is not so much easier to investigate as the journalists have more first-hand experience (not necessarily in the literal sense) with it and are therefore better experts in what they are writing about. Better than many other issues. And it's a fun assignment too.

Speaking of stats_wtf, I have just created it and added you as a maintainer. If you have time, you can fill out the info. I would not set out any rules for posting just yet, more or less like I do in ljnauka. Time will show if we need to add any restrictions; usually this is not necessary.

I am not sure if we want to restrict ourselves to just statistics or talk about scientific methodology in general. I would say the latter belongs to ljnauka. There is going to be some overlap anyway, of course. However, not everything that belongs to stats_wtf is pseudoscience!

[syarzhuk.livejournal.com]

Joined stats! WTF?

[cema.livejournal.com]

:-)

Well, there is also code_wtf, and I am sure life will bring us more WTF kind.

[email-animal.livejournal.com]

I'll reply to your media comments later. Let's deal with administrivia first.

I am not listed as a maintainer for stats_wtf - maybe one first have to be a member (which I am now). Perhaps you can add me to the list of maintainers again.

Two immediate questions. One: English, Russian, or both? Two: well, I had to step out and forgot what it was. It will come back to me at some point.

As far as the scope, I think it should address the studies with wrong/poor methodology - be it in selecting what to measure or selecting how to measure it, or electing to interpret results incorrectly. A lot of this will have to do with statistics, but also with experiment/study design.

And indeed, some intersection with ljnauka may occur, but overall, this is different.

Administrativia

[cema.livejournal.com]

Maintainer status: added.

Language: both English and Russian, I think. This may potentially reduce the audience, which in fact may be a blessing. Or it might split it into two intersecting groups. We'll see. I would say, anything written in other languages certainly should be translated.

Almost all problems with applied statistics are with poor methodology. Desing of the experiment is often (at least loosely) related to statistics. I guess we may want to stretch it a little to accommodate borderline cases.

I would say, the main difference with ljnauka is that the latter deals with the cases where scientific methodology was not followed (therefore pseudoscience), whereas stats_wtf deals with either errors in scientific methodology or other scientific errors (therefore no pseudoscience, just bad science). Thin line.

Re: Administrativia

[email-animal.livejournal.com]

Maintainer status: added.

Thank you. I am already working on filling in the info section.

Agreed on the languaged. The key issue is that a lot of "food" for places like science_freaks comes from Mother Russia. I expect the majority of sources for stats_wtf to come from US/Europe, in a form of an English source.

To ensure we understand the use of terminology (statistics vs. experiment design), here are examples of what I consider to be purely statistical flaws vs. examples of what I consider to be experiment design issues.

Statistical flaw. (to be political). The following statement, The richest 1% of Americans pay 15% of all taxes (the numbers may be off, this is an example) may be true on the face. The conclusion: "They pay way too much in taxes" is wrong. The person making such argument purposefully ignores a much more important, from the point of view of tax-paying statistic: what percentage of total wealth of all Americans the top 1% has.
There is no problem here with experiment design: both measures are collected and readily available. The problem is in the analysis stage with delibarate misuse of statistics.

Experiment Design flaw. You saw that in the example discussed here. I will define "liberal" or "conservative" bias of an article in press based on the number of mentions of liberal/conservative think tanks and citations from their sources. I will then calibrate these numbers by comparing them to the patterns exhibited by known liberal and conservative members of Congress. The flaw is two-fold. First, the selection the metric is not validated, and in fact, there are specific reasons to believe that the metric is indeed invalid - for example, it assumes that any article that does not quote think tanks/their representatives is neutral - a provably wrong conclusion. Second, it assumes without validation that correlation in the pattern of citation between lawmakers and journalists is evidence of similar bias. Notice, that there may not be any flaws with statistics here - if we suspend belief, the actual statistics portion is probably correct.

Experiment flaw. (a bonus) Suppose I want to find out how perceptions about condom use change over time among female urban African-American teenagers. I target a school in an urban area, and conduct a survey of a group of 200 African-American girls. Six months later I come back to the same school and conduct another survey. Except, 170 of the girls who took the second survey did not take the first. Yet, I ignore this fact and report the change in responses to my questions as a real change of perceptions.
Here, the experiment design is fine (at assumes that all 200 girls are surveyed twice). The statistics is fine, the problem is in the fact that the experiment did not follow the design, and thus the reported conclusions are invalid. (PS. this is actually very close to a real life example I am familiar with.)

[email-animal.livejournal.com]

I have actually bothered to read the article (PDF, 62 pages), not completely, but relevant parts. I have to state the following. I still believe that they have flawed methodology, however, the argument moves into the realm of science. In particular, a couple of my assumptions (based on the press release) were wrong: (a) the study relates to the period of 1995 - 1999, when Clinton was the president (I thought they had studied 2000-2004 period), which also means Joe Lieberman could actually have been on a relatively liberal side of things - as opposed to today; (b) in their selection of citations, they were careful to use only positive citations and citations that reflected the policies of the think tanks. Critical citations and casual citations were omitted -- this renders some of my criticism moot.

I will post more thoughts on this to stats_wtf at some point. My first attempt resulted in an accidental erasure of about 2Kb of text.