Most favored narrative watch: Maddow’s expert says she’s wrong!

THURSDAY, MAY 3, 2012

But Maddow just blusters ahead: It’s a basic rule of pseudo-journalism:

You do not correct Rachel Maddow!

Last Sunday, on Meet the Press, Alex Castellanos dared to say that Maddow was wrong about a familiar factual claim. As it turns out, Castellanos was almost certainly right—more right in his comments than Maddow.

But correcting Maddow isn’t allowed—so Rachel swung into action. On Monday night, she opened her eponymous cable “news” program with a segment about this dispute. As she started, she kept repeating her basic factual claim:
MADDOW (4/30/12): The jobs we do in this country are still sort of surprisingly really segregated by gender. But whether you have an occupation that is male-dominated or you have an occupation that’s female dominated, there’s one thing that just about every single job in America has in common. Dudes get paid more for doing it.

If you are a driver, men get paid more. If you’re a manager, men get paid more. If you’re a janitor, men get paid for. If you’re a retail salesperson, men get paid more. If you’re a sales rep, if you are a cook, a chief executive, a security guard, a police officer, a customer service representative—in all those cases, men get paid more.

In 19 of the 20 jobs that are the most common occupations for men in this country, women lag behind what men get paid for doing that same work.

It’s also true in the most common jobs that women have in this country. If you are a secretary, men get paid more. If you’re a teacher, men get paid more. If you’re nurse, men get paid more. If you’re a cashier, men get paid more.

If you’re a receptionist, a financial manager, if you wait tables, men get paid more. Again, in 19 of the 20 jobs that are the most common occupations for women in this country, women lag behind what men get paid for the same work.

Overall, when you aggregate everybody working, women get paid 77 cents for every dollar that men get paid. For the same work, dudes get paid more.
True believers often think it: If they just keep repeating a claim, that will make the claim accurate. In this case, Maddow kept saying that men get paid more “for the same work.” And she used the statistic from Meet the Press, the statistic that launched this dispute:

“Women get paid 77 cents for every dollar that men get paid. For the same work, dudes get paid more.”

Those claims may still be technically accurate—but they’re grossly misleading. Consider what happened when Maddow ended her monologue and let an expert speak.

For the record: Before she introduced this expert, Maddow repeated her basic statistical claim five times. On tape and in person, viewers kept seeing her say it: Women get paid 77 cents where men get paid a dollar.

But uh-oh! Eventually, Maddow introduced Dr. Heidi Hartmann, founder of the Institute for Women’s Policy Research. This was the first Q-and-A:
MADDOW: I know that you are at the women’s, Institute for Women’s Policy Research. You have done some of the most important and most highly publicized work on this issue. Is there any way that the idea of a gender-based disparity is something that depends on how you look at it? Is this something other than a blunt truth about the American economy?

HARTMANN: Well, I mean, you obviously have by far the better part of the argument. You’ve got the Census Bureau and I might mention, the Bureau of Labor Statistics agreeing with you. Oh, also, I could mention, the U.S. General Accountability Office.

I think what the issue is for the Republicans is that they believe that that’s not—no matter how big the wage gap is, almost none of it is due to discrimination. And, of course, these numbers from BLS and Census Bureau are not really talking about discrimination. But the GAO study that I just mentioned did. They said that even when you put everything you can possibly think of in the regression equations, the statistical analyses to try to make that gap go away, you can’t explain at least 20 percent of it.

Now, most other studies place the part you can’t explain as a quarter to a half. So, a large part of the gap probably is due to discrimination. But that seems to be what the debate is.
Hartmann told Maddow she had the far better part of Sunday’s argument. Then, she quickly began to show that this claim isn’t accurate.

Duh! “Of course, these numbers from the...Census Bureau are not really talking about discrimination,” Hartmann said, referring to the “77 cents” figure which Maddow had now been reciting for two straight days. Having thrown that statistic under the bus, Hartmann cited a GAO study.

This study did attempt to measure discrimination, Hartmann said. And what did that GAO study find? According to Hartmann, the study said this:

“Even when you put everything you can possibly think of in the regression equations, the statistical analyses to try to make that gap go away, you can’t explain at least 20 percent of it.”

But twenty percent of “that gap” is only 4.6 cents. (That’s 23 cents divided by five.) According to Hartmann, the GAO study said that women are discriminated against to the tune of 4.6 cents on the dollar. Maddow had been saying the discrimination factor was 23 cents for the prior two days.

In fairness to Maddow, Hartmann cited some other studies (see above). She said these studies “place the part you can’t explain as a quarter to a half.” If Hartmann’s statement is accurate, this would mean that these studies say that women are discriminated against in their pay by 6 to 11.5 cents on the dollar.

Maddow had been saying 23 cents for two days. That claim is extremely familiar, but it’s almost certainly false; as Hartmann said, the Census Bureau doesn’t even intend for it to be seen as a measure of discrimination. It was that claim which brought the original objection from Castellanos.

One more point for today, this time involving the fact-check which was done on CNN. On Monday evening, Maddow played tape from that meandering fact-check, which was reported by Lisa Sylvester. But she didn’t play the piece of tape where Sylvester finally told Wolf Blitzer what her fact-check had found:
BLITZER (4/30/12): And so, the bottom line, though. When men and women have the exact same job, do women still only earn 77 cents on the dollar, if they're doing, working the same amount of hours, have the exact same job, in the exact same field?

SYLVESTER: They have—there is definitely a gap. It is, if you're looking at. But there are all kinds of other control factors, you know, what college somebody went to, what region of the country. If you're talking salaried workers versus part-time workers, the average for full-time workers, the difference is pay is 77 cents on the dollar.

Now, as you go along, as you control for other factors, even if you control for everything you could possibly imagine, all those things, the college, the hours worked— Men still make more than women, that gap narrows, it's about 5 cents of a difference. But it still is there, it's still real, and the truth is, men make more than women.
Gack! According to what Sylvester said, the actual discrimination factor is five cents on the dollar. That evening, Maddow forgot to play the tape of Sylvester making that statement. Tomorrow, we’ll show you what she played instead.

How much do women get cheated on pay? By all accounts, this is a very hard question to answer. But no expert even pretends that the figure from the Census Bureau is a measure of discrimination. People like Maddow recite that figure because—well, go ahead!

Go ahead! You explain! But when liberals recite that bogus figure, we're acting exactly like ditto-heads. Now that we've finally emerged from the woods, we don't look so great after all!

Tomorrow: What's the truth about this (complex) question? And why won’t Maddow tell you?

22 comments:

  1. Bob wrote: But twenty percent of “that gap” is only 5.6 cents. (That’s 23 cents divided by five.) According to Hartmann, the GAO study said that women are discriminated against to the tune of 5.6 cents on the dollar. Maddow had been saying the discrimination factor was 23 cents for the prior two days.

    Actually, the figure "5.6" should be "4.6" in this paragraph.

    Sylvester exaggerates when he writes: as you control for other factors, even if you control for everything you could possibly imagine.

    He didn't control for everything imaginable, because data isn't available for everything imaginable. E.g., for all we know, men may be better employees than women, on average. But, if that's the case, there's no way to factor that into any study.

    Or, take a measurable difference: IIRC women have more sick days than men. Even though this difference can be measured, there's no way to directly determine the value to the employer of an employee taking fewer sick days. Reliability should be worth more than just the difference in hourse worked, but I don't think there's any way to say how much more.

    ReplyDelete
    Replies
    1. It may be that, as you fantasize it, men are "more reliable."

      But you can't measure "reliability" only by "sick days."

      Men, we know, are far more likely to perform some act of violence that results in unexpected temporary or permanent work absences.

      It may well be that women are more reliable.

      Your anecdote gets us no where near pretending men are.

      Unless one is predisposed to believe they are, which would be a definition of sexism many would accept.

      YMMV, DavidinCA.

      Delete
    2. I totally agree, Anonymous. AFAIK there's no way to say whether men or women are better employees on average, and certainly no way to say how much better.

      My point was that there are conceivable differences that are not reflected in any study. So, a small adjusted wage difference of only 4.6% is too small to be significant. It's likely within the study's margin of error.

      I've lived through the revolution in women's employment. It occurred with dizzying rapidity. Today, there may or may not be a small amount of unfair sex discrimination remaining, but the battle is essentially won.

      Delete
    3. It's a common misconception that controlling for things can only decrease association. But controlling can strengthen association as well as weaken it, so there's really no good reason to say it "more like 4.6" than to say actually it could be "6.6". Moderate drinkers tend to smoke, control for smoking and the association of moderate drinking with health rises. Moderate drinkers also tend to be more social -- control for social connections and the association between drinking and health goes down. So you can't just say "Hey, it's 5.6 at most, and maybe less" -- door swings both ways.

      Margin of error on any salary study is also likely to be very very small for any common occupation. We have a ton of data on salary that we collect on a regular basis, and for a measure of central tendency like salary we'd need far less than the tens of thousands that instruments like BLS surveys ask monthly. Most studies using existing public data would find MoE to be an insignificant factor.

      Delete
    4. There's two kinds of error. I agree that the statistical MoE is apt to be small, because there's lots of salary data available.

      The potentially bigger source of error is model error. That's the risk that one's model isn't exactly right. Model error is potentially large. Furthermore, unlike statistical error, there's no way to measure the magnitude of model error.

      Probably the biggest unmeasurable factor is people's attitude. Asian-Americans considerably outearn other groups, even though the US has a horrible history of discrimination toward Asians, and some discrimination continues. But, I don't think a model could fully pick up the reasons for this difference.

      Delete
  2. "Tomorrow: What's the truth about this (complex) question? And why won’t Maddow tell you?"

    GACK!!

    I'll come back next week, I can't stand another day of this picayune discussion.

    ReplyDelete
  3. Ah! I get it! When you discriminate against women only 5 percent, it's not really discrimination.

    And when the actual pay gap is 23 cents on the dollar, not 5 cents, but all but 18 cents can be explained by "other factors", then that must not be a real problem.

    Which raises a question. Just how much are we allowed to discriminate if we can explain away most of it to "other causes," and say that only a fraction is the result of real, actual discrimination?

    Since we find now that an 80 percent/20 percent other causes/discrimination split is perfectly acceptable, can we go for 75/25? How about 50/50? Then we'd only be half-discriminating.

    ReplyDelete
    Replies
    1. "When you discriminate against women only 5 percent, it's not really discrimination."

      How do you know there is *any* discrimination?

      "must not be a real problem"

      If you think there is a problem, why don't you *define* it precisely along with proof it exists, so it can then be solved?

      Delete
    2. "How do you know there is *any* discrimination?"

      Uh, because all empirical data says that women earn less than men? Significantly less?

      It's just the height of denial and foolishness to say, "Yeah, but I can come up with a lot of reasons that say it isn't discrimination, so let's pretend we don't have a problem."

      Delete
    3. How do you know there is *any* discrimination?"

      Uh, because all empirical data says that women earn less than men? Significantly less?


      So, are you saying that there's major discrimination in favor of Jews and Asians? Give me a break. A raw comparison of earnngs without proper analysis tells us nothing.

      Delete
    4. "I can come up with a lot of reasons that say it isn't discrimination"

      I had never said there wasn't any discrimination.

      The correct approach, if *you* think there may be discrimination, is to show the evidence.

      If we are to believe that there remains a 5 cent difference that is unaccounted for for reasons that are not discrimination, you can't just that the 5 cents must be from discrimination when maybe you just can't think of any more alternatives. Again, you have to *demonstrate* discrimination.

      Delete
  4. I know Bob doesn't stoop to reading the comments on his own damn blog, but given how much he loves to cite NAEP findings about relative levels of educational performance broken down by race, is he similarly willing to say that only SOME of the disparity is evidence of discrimination, so anyone who cites the precious NAEP as evidence of racial discrimination is as dimwitted and crowd-pleasing as the dreaded Rachel Maddow? How about his holy chart of per-capita medical expenditures by nation, which he says is evidence of "looting" by powerful interests? Is he as dumb as he says Rachel Maddow is because his "looting" explanation goes beyond what he knows to be true? Is he willing to apply the same hairsplitting, pin-dancing analysis to data sets he likes? Or is he just grinding an incredibly tiny axe to a ridiculous sharpness?

    ReplyDelete
  5. Or, to put it another way, Bob has said time and again that some of the reason why black students lag behind white students on standardized tests (even though the trend in performance is, happily, upwards) is that African Americans were systematically discouraged from learning for decade after decade. But by the Maddow Standard he articulates in these posts, would he deem it appropriate to offer such an explanation? Or would it be ignorantly "tribal"?

    ReplyDelete
  6. But of course, Somerby doesn't use NAEP as a snapshot to "prove" discrimination -- nor does he really say that the gap snapshot proves anything at all!

    It's crucial to exactly none of his arguments whether you believe that past discrimination is reflected in current results.

    For Bob, the NAEP shows what its data trend shows: that the tested gap in performance has been narrowing as both groups increase compared to previous years.

    The issue is that no one mentions these gains. Or that they deny them outright.

    You seem to want to imply either that Somerby's doing something unseemly in use of NAEP, or that he ought to grant Maddow the same leniency on wages.

    But you haven't come close to making a showing.

    A possible "Maddow standard" violation might occur if you found Bob asserting that every difference in education was due to race discrimination, then refusing to look at what other reasons might explain some differences. Good luck with that.

    ReplyDelete
    Replies
    1. I don't think Bob would take a week to grind out a rebuttal of how someone who said that black kids on average do worse than white kids on the same test was actually incorrect because some, but not all, of the disparity was due to discrimination. But that's the heart of why he thinks Maddow is wrong: he's not arguing with the stats, he's arguing with the implication that the difference between men's and women's pay is due to systemic discrimination. Fine. But that means his judgments about disparate health care spending due to "looting" are similarly out on a limb, and if someone were to say that the reason black kids do worse on standardized tests includes many other factors, Bob would have to defend that claim, rather than haranguing the world about how no one else but him cares about the children. He's playing games that are beneath him as someone who claims to care about rigorous quantitative analysis and avoiding overreach in making conclusions.

      Delete
    2. And he _regularly_ points to the criminalization of educating black people as a major factor in why black children lag behind on these tests. Which he is right to do! But in his zeal to attack Maddow, he's pretending that she's doing something wrong. Whatever he thinks she's doing wrong, he has himself done repeatedly. Think of how many times he has nailed education "reformers" like Rhee for saying that what black kids need is higher standards. That's what Castellanos is doing by saying the pay disparity due to gender can be explained by factors other than discrimination. But Bob, rightly, doesn't have a lot of patience for bullshit like that when it comes to education. And yet here he is, tolerating immense levels of bullshit, all to the higher, glorious cause of nitpicking Rachel Maddow for accurately reporting a social science finding. He's becoming Captain Ahab.

      Delete
    3. But 77 cents on the dollar is NOT an accurate social science finding, because it is social science that has explained much of that difference due to factors other than dscrimination. Castellanos was basically backed up by Sylvester and Hartmann who also explained the gap.

      You may disagree with their analysis, or with Somerby's treatment of it, but it is better to argue the merits of this post rather than bring up a number of red herrings. I have no idea what you are talking about here. If you are gonna make a valid point, you should at least cite something from the incomparable archives.

      You don't think it is BS to say the gap is 23 cents when the unexplained gap is really only 5-10 cents?

      Delete
    4. Nah, he just hates Somerby and this is his daily "got him."

      'whig knows what Maddow asserted about pay and discrimination is not equivalent to what Somerby's asserted about about testing and race, but he pretends it is.

      Delete
    5. Maddow didn't say the gap was entirely due to discrimination. Maddow pointed out a gap that is attested by many studies. Is there a gap? Yes. Is it legitimate to investigate some reasons for the gap? Sure. Those are built into the study being quoted. It controls for "the same work." But Bob doesn't like it because he's playing gotcha games.

      Delete
  7. I'll tell you something I can say from personal experience: you do NOT correct Bob Somerby.

    ReplyDelete
    Replies
    1. Especially if your "correction" is a load of horse manure...

      Delete
    2. One of the Howlerettes in action.

      Delete