Last month, I examined one of the studies embraced in the current Psychology Today article, “The Ideological Animal” (Dixit, 2007). That study asserted that conservatives, among numerous other deficits, are lower in openness to experience and integrative complexity than liberals, and that people choose conservatism because it serves to reduce their inherent fear and anxiety (Jost, et al., 2003). The poor dears.
As I demonstrated in some detail, constructing a study that debases one’s ideological opponents so thoroughly requires a good measure of abhorrent research methodology. It also requires a body of equally fallacious and substandard literature from which to draw. That literature has been growing for some time now, with no signs of stopping. For example…
The same PT article also made use of a 2006 longitudinal study, “Nursery School Personality and Political Orientation Two Decades Later” (Block & Block, 2006). That study is the topic of this week’s column.
Nursery School Blues
In the nursery school study, researchers followed a group of preschoolers into adulthood. When the participants reached age 23, the researchers determined their political orientations and compared them against the personality inventories administered 20 years earlier. Here’s how PT summarized the results:
“As kids, liberals had developed close relationships with peers and were rated by their teachers as self-reliant, energetic, impulsive, and resilient. People who were conservatives at age 23 had been described by their teachers as easily victimized, easily offended, indecisive, fearful, rigid, inhibited, and vulnerable at age 3” (p. 83).
That captures the general flavor (liberals strong; conservatives weak), but let’s allow Block & Block to describe their own findings, starting with what they discovered about the participants at age 23 (p. 745):
- As adults, the liberal study participants are “bright, distinctive, having a wide range of interests, being aesthetically responsive to the world about them.”
- The conservative adults are “uncomfortable with uncertainty, conventional, traditionally sex-typed, constricted in their behaviors, judging self against conformist standards, and moralistic.”
- Liberal men are “relatively introspective, reflective about existential problems, and intellectually oriented.”
- Conservative men “display an egocentric self-image, with an orientation toward the virtues of power, a willingness to offer advice, and a concern about their status within the pecking order” (p. 745).
- Liberal women are “interpersonally oriented, perceptive of others and socially instrumental, aware also of sensuality.” (I’m sure the liberal men were happy to hear this.)
- Conservative women are, well, a lot like conservative men: egocentric, constricted, moralistic, etc.
Here’s what Block & Block discovered about the participants as three-year-old preschool students:
- Liberal children of both genders “were viewed as resourceful, autonomous, expressive, and self-reliant” (p. 745).
- Conservatives of both genders were characterized by their teachers “in psychologically unflattering terms but more markedly for little girls than for little boys” (p. 745).
- Liberal boys were “resourceful and initializing, autonomous, proud of their blossoming accomplishments, confident and self-involving” (p. 739).
- Conservative boys were “visibly deviant, feeling unworthy and therefore ready to feel guilty, easily offended, anxious when confronted by uncertainties, distrustful of others, ruminative, and rigidifying when under stress” (p. 740).
- Liberal girls were characterized by “a coherent host of qualities: self-assertiveness, talkativeness, curiosity, openness in expressing negative feelings and in teasing, bright, competitive, and as having high standards” (p. 740).
- Conservative girls were “indecisive and vacillating, easily victimized, inhibited, fearful, self-unrevealing, adult-seeking, shy, neat, compliant, anxious when confronted by ambiguity, and fearful” (p. 740).
As if that weren’t bad enough, Block and Block determined that the more intelligent participants tended to become liberals [1].
As with any study of this type, these results are assumed to be generalizable to you and me [2]. Given the vast and diverse range of characteristics present in a population of hundreds of millions of people, one must wonder how we can be so neatly divided into two groups – one of which is emotionally and cognitively inferior to the other. Could there be other factors at play in this study?
|
I am instantly suspicious of any study that divides the population along some demographic line and concludes that my half is smart and groovy while your half is deficient. |
In my deconstruction of the Jost et al. article, I was forthright about my political orientation in the interest of full disclosure. Allow me to be forthright about another bias:
I am instantly suspicious of any study that divides the population along some demographic line and concludes that my half is smart and groovy while yourhalf is deficient.
In this article we’ll examine the means by which Block & Block arrived at just such a conclusion. While this study is much more polished than others in this line of research, there are severe methodological problems [3]. Without such problems, I believe it would be impossible to divide liberals and conservatives so neatly into good guys and bad guys. Populations simply don’t work that way.
In the words of a great professor, let the commencement beginulate.
An Elegant Looking Study
How do liberals and conservatives differ in their early childhood years, before they become political beings?
That is the question asked and answered in the article, “Nursery School Personality and Political Orientation Two Decades Later.” This elegant looking study began in 1969 with the Block and Block Longitudinal Study of Cognitive and Ego Development. That study followed 128 preschoolers, assessing them during childhood, adolescence, and adulthood. Over the years, the study has provided a good deal of clinically useful information about human development. For example, it helped us understand primary factors in adolescent substance abuse.
In this most recent paper based on the longitudinal study, Block & Block identified the political beliefs of the 95 remaining adult subjects, then correlated those beliefs with 1) the personality assessments administered when the participants were three years old, and 2) the personality assessments administered twenty years later. You’re already familiar with the results.
I’m not the first to question the methodology of this study. Others have noted the possibility of rater bias – specifically, the possibility that the preschool teachers who administered the first personality assessments might have unknowingly tainted the end results. And, as the authors acknowledge, the initial assessments took place in the leftist community of Berkeley, California and that may have colored the data.
But having reviewed this study extensively, it seems to me that speculating about rater bias is like complaining about squeaky propellers on the Hindenberg. There is a much bigger problem with this study, so let us not waste time on trivialities.
The LIB/CON Scale
In order to determine whether personality correlates with political orientation, the authors needed to develop a method of quantifying each participant’s political orientation. A simple number would do. They devised a scale called LIB/CON from which each participant would receive a score describing their political orientation. Each participant’s LIB/CON score is a composite of six different measures (p. 736-737):
- A five-point scale in which participants were asked to identify their political ideology.
- A survey in which participants were asked to indicate their position on ten issues that separate the Democrat and Republican parties.
- A conservatism scale written by Herbert McClosky in 1958 and updated for this study (the authors did not describe the details of the update).
- The Kerlinger Liberalism Scale.
- The Kerlinger Conservatism Scale.
- Questions regarding political activism.
We will discuss these. Oh yes, we will discuss them all.
|
The LIB/CON scale is perhaps the single most important aspect of this study’s design. |
In fact, the remainder of my critique will focus on these measures, as the LIB/CON scale is perhaps the single most important aspect of this study’s design.
This study is an attempt to correlate personality with political ideology. The LIB/CON scores are the only link between the two. If the LIB/CON scale was well-constructed by the authors, then any correlations between personality and ideology will probably be valid, barring other methodological problems.
On the other hand, if the LIB/CON scale has holes, then the study could be rendered meaningless – there will be no way to know what it measured. While the authors invested a good deal of effort in explaining and justifying the personality scales that they used (which have been repeatedly used and validated in the past), they were vague regarding the details of the LIB/CON scale.
In order to understand the construction of the LIB/CON scale, I had to do some digging. Let’s look at each of the six measures.
LIB/CON Measure #1: Self-Identification
According to the authors, “participants self-identified their position on a 5-point continuum ranging from ‘very liberal,’ ‘liberal,’ ‘middle of the road,’ ‘conservative,’ to ‘very conservative’” (p. 736).
On its face, this measure appears to be straightforward and dependable. However, a recent study suggests that self-identification questions like this one tend to proffer inaccurate results.
Sanders, Burton & Kneeshaw (2002) discovered that when party identification questions include a non-identity option (e.g., “I don’t consider myself a member of a political party”), the number of people who identify with a party drops substantially. They discovered that fewer than 50% of respondents in a sample of the British electorate chose a party affiliation when a non-identity option was available. Moreover, the traditional wording of such questions – as appears to have been used in this study – does not accurately capture true party identification.
Granted, there is a difference between party affiliation and political ideology, but the general concern here is that a forced-choice question increases the appearance of partisanship. Block & Block did not address this, nor did they provide the specific phrasing of the question. And while they admitted that Berkeley is a far-left cultural enclave, they did not address the possibility that participants in a Berkeley study might experience a certain social pressure to endorse liberal views.
These observations do not constitute a fatal blow to the LIB/CON scale, but they do raise concerns about the accuracy of each participant’s composite LIB/CON score.
LIB/CON Measure #2: Stand on Issues
In this measure, “participants indicated their direction of agreement regarding 10 issues generally viewed as then distinguishing the Democrat and Republican parties: abortion rights, amount of money spent on welfare, national health insurance, income tax rates, environmental protection, affirmative action, extent of funding for national defense, support of use of military force to remove a hostile foreign government, government job guarantees, and civil rights for suspected criminals” (p. 736).
|
Two questions leap instantly to mind: How were the questions worded, and how were the answers scored? |
Two questions leap instantly to mind: How were the questions worded, and how were the answers scored?
It is no secret that the phrasing of survey questions can lead respondents toward desired responses. And, of course, the responses themselves are subject to rater bias. (See, for example, Wolf, 1990.) Unfortunately, Block & Block did not list the questions or their scoring criteria, so I contacted Dr. Block via email and asked him for the details. In the interest of full-disclosure, I also sent a link to my previous methodology deconstruction.
Dr. Block politely informed me that, due to personal circumstances that I won’t disclose here, he would be unable to consult the data. When I asked if someone else might be able to answer my questions, he explained that the data had been shipped to Harvard and would be unavailable for a couple of years. He thanked me for my interest.
What can I say? Without an explanation of the methodology used in measure #2, there is simply no way to know how and why participants were assigned this portion of their composite LIB/CON score. Given the results of the study, this mystery factor should be viewed with great skepticism.
LIB/CON Measure #3: McClosky’s Dimensions of Political Tolerance
In 1958, Herbert McClosky (University of Minnesota) proposed that personality helps mold political ideology. In the article “Conservatism and Personality,” he identified the personality traits he believed to be most characteristic of conservatives and the scale he designed to measure them.
|
“The data show clearly… that the most articulate and informed classes in our society are preponderantly liberal in their outlook.” |
Beginning with a sample of more than 1200 people from the Twin Cities area, McClosky and his team started with a pool of 2300 statements designed to identify conservatives. For example, someone who agreed with the statement, “duties are more important than rights,” would be classified as conservative.
The statements did not include specific policies over which liberals and conservatives tend to disagree. McClosky intentionally avoided specific attitudes toward “free enterprise, toward trade unions, toward expansion of government functions, toward the New Deal and its welfare measures, toward tariffs, farm supports,” etc. (p. 30). The stated purpose of avoiding specific issues was to obtain a measuring instrument that would capture a pure distillate of ideological conservatism, free from “secondary and unstable correlates of liberal and conservative tendencies.”
Indeed. Why pollute theory with empirical data?
McClosky and his team pared the 2300 items down to 12. Conservatives were found to agree with statements such as (p. 30):
- You can’t change human nature.
- The heart is as good a guide as the head.
- Few people really know what is in their best interest in the long run.
You can’t change human nature? The conservatives I know might agree that this characterizes conservatism; changing human nature is a leftist cause. But some of the others are completely off base. The heart is as good a guide as the head? My conservative friends would joke that they have no hearts and point to the rational empiricism of Ayn Rand as their guide. People don’t know what is in their best interest? They would most decidedly protest this statement. Isn’t it the left that consistently tries to abdicate personal freedom in the name of safety?
These ideas are certainly debatable, particularly with the passage of five decades since McClosky designed his conservatism scale. That may be an inherent drawback to using a 50-year-old assessment tool.
|
“By every measure available to us, conservative beliefs are found most frequently among the uniformed, the poorly educated, and so far as we can determine, the less intelligent.” |
However, I don’t wish to get caught up in such peripheral debates. They aren’t germane beyond the possibility that liberals and conservatives might have been incorrectly classified using this scale. The point is, Block & Block found McClosky’s conservatism scale to be a suitable measuring stick.
And what are the implications of this assessment tool? Let’s look at the personality traits that McClosky found to be typical of conservatives, as that will shape each participant’s LIB/CON score and, ultimately, the outcome of the present study. In the results section of McClosky’s article, we learn that,
“By every measure available to us, conservative beliefs are found most frequently among the uniformed, the poorly educated, and so far as we can determine, the less intelligent” (p. 35).
Why, the little scamp. I think he’s flirting!
By the way, McCloskly didn’t use empirical measures, such as IQ tests, to assess intelligence. He relied on the demographic information of people he had already labeled as undesireable and, therefore, conservative. There’s more to this love-fest:
“The data show clearly… that the most articulate and informed classes in our society are preponderantly liberal in their outlook” (p. 36).
“Uniformly, every increase in the degree of conservatism shows a corresponding increase in submissiveness, anomie, sense of alienation, bewilderment, etc.” (p. 37).
“Conservatism, in our society at least, appears to be far more characteristic of social isolates, of people who think poorly of themselves, who suffer personal disgruntlement and frustration, who are submissive, timid, and wanting in confidence, who lack a clear sense of direction and purpose, who are uncertain about their values, and who are generally bewildered by the alarming task of having to thread their way through a society which seems to them too complex to fathom” (p. 37).
And the world was much simpler back then, what with the womenfolk confined to the barn and the young ’uns at work in the mine. In today’s complex world, it’s a wonder that conservatives don’t trip over their own knuckles. There’s more.
“In many ways hostility is a principal component of the conservative personality as it is a principal component of conservative doctrine” (p. 41).
Paging Dr. Irony, you’re needed in 1958.
More? OK.
“Although aggressively critical of the shortcomings of others, [conservatives] are unusually defensive and armored in the protection of their own ego needs” (p. 37).
“…education trains people in some measure to demand greater precision in speech and thought, to be more open-minded and tolerant, to be intellectually flexible and receptive to scientific modes of discourse, to reject mystical or non-natural explanations, and so on. These are traits which we find to be negatively correlated with conservatism; or, put another way, we find that individuals who learn to think in the ways just described are, other things being equal, far more likely to become liberals than conservatives” (p. 41).
Block & Block chose to include in each person’s LIB/CON score the results from an outdated scale that portrays conservatives as pissed-off morons. Put more accurately, they chose to include a scale that labels pissed-off morons as conservative.
With McClosky’s scale as one of the measures comprising LIB/CON scores, conservatives have no chance of emerging from the Block & Block study resembling anything other than frightened, damaged idiots. Already, the deck is stacked, the game is rigged, and the door is welded shut.
LIB/CON Measures #4 and #5: Kerlinger’s Liberalism and Conservatism Scales
Fred Kerlinger’s 1984 book, Liberalism and Conservatism, is the most even-handed, empirically supported, rational treatment of political ideology I have encountered in the weeks that I have spent examining this branch of the professional literature.
This book explores both liberalism and conservatism in a non-dogmatic, rationally curious manner. The methodology and statistical operations are impeccable and are clearly presented. Kerlinger offers methods of assessing both liberalism and conservatism, and if he has a bias toward one or the other, it is difficult to detect. Of the tools used in LIB/CON scoring, this one is clearly the most rigorous.
Unfortunately, it appears as though Block & Block have missed the point of Kerlinger’s work entirely, misused his scales, and likely confounded this portion of each participants’ LIB/CON score.
One of the most obvious differences between Kerlinger’s scales and McClosky’s (aside from an unabashed hatred of conservatives) is the fact that Kerlinger relies on specific, real-world issues to quantify liberal and conservative tendencies, rather than relying on abstract and debatable ideals. Put differently, he relies on quantifiable behavior rather than the opinion of other academics.
He noticed, for example, that liberals tend to support the United Nations more so than conservatives. (To be more specific, he noticed that some people display U.N.-lovin’ behavior, and U.N.-lovin’ behavior tends to occur in concert with other specific behaviors. That cluster of behaviors might reasonably be labeled liberalism. This is not a trivial distinction; it constitutes a major methodological advantage.)
He refers to these real-world issues as “referents” and he reasons that “any social recurrence can be the referent of an attitude” (p. 31), meaning that a person’s opinion an a particular issue can be statistically tied to an overarching attitude. A few examples of referents include private property, industry, moral standards, money, business, labor unions, equality, freedom, and birth control.
Kerlinger noted that these referents do not exist on a single, continuous dimension such as the LIB/CON scale devised by Block & Block. He explains:
“The unidimensional assumption is clearly incorrect. The available evidence indicates that several factors underlie both liberalism and conservatism. The evidence seems to indicate that liberalism and conservatism are separate and distinct dimensions that are relatively orthogonal [uncorrelated] to each other” (p. 49).
Kerlinger drew two conclusions from this. First, liberalism and conservatism should be measured separately, using separate scales. Second, items at each end of the scale should not be reversed on the assumption of a single continuum. Liberal referents are not necessarily the opposite of conservative referents.
Yet, Block & Block went against Kerlinger’s recommendations: “A composite score, termed LIB/CON, for each participant was then generated by averaging the standard scores of the 6 convergent variables” (Block & Block, p. 737). In other words, the LIB/CON scheme mashes Kerlinger’s Liberalims and Conservatism Scales (along with the four other measures) into one continuous range in which, apparently, liberalism and conservatism are opposites. This is the one thing that Kerlinger asked them not to do.
Here’s the problem. Suppose a person – take me, for example – believes that A) government should engage in minimal interference with business, and B) women and men should receive equal pay for equal work. That person would receive a point each on the Conservatism Scale (statement A) and the Liberalism Scale (statement B).
How could Block & Block compress these two distinct referents into one meaningful score? Would they classify me as liberal? Am I a conservative? Would they toss out the scale that didn’t fit?
Worse yet – and this appears to be the case – would they “average” the two referents, thereby classifying me as “middle of the road” when, in fact, I hold strong beliefs? There is no logical answer, and there is no way to use Kerlinger’s scales without misclassifying me on the LIB/CON scale.
How did they reconcile this problem? The article doesn’t say.
LIB/CON Measure #6: Political Activism
In this measure, “participants were questioned regarding their personal Political Activism, whether the subject had written letters to express political views, attended political rallies or demonstrations, or boycotted companies and products singled out as politically aversive” (p. 737).
|
“Activists are found to be intellectually gifted, academically superior, and politically radical young people from advantaged homes…” |
Huh?
What is the criterion here? Does political activism correspond with one end of the LIB/CON scale but not the other? Who singled out the aversive companies and products – the researchers or the participants? Is political activism taken to be indicative of ideology? If a participant doesn’t attend demonstrations, does that affect his score? What, exactly, is being measured?
Once again, I asked Dr. Block for clarification. What questions were asked, and how were the responses rated? Once again, he was unable to help me or direct me to someone who could. And, once again, I found myself rummaging through the literature in search of an explanation.
In 1969, Jeanne Block was the lead author of a paper titled “Socialization Correlates of Student Activism,” which the authors referenced in the present study. In that article – an examination of parenting styles – the authors reported that,
“In the studies completed to date, activist students have been defined variously as members of the Students for a Democratic Society, marchers in peace parades, protestors against university compliance with ranking students for the draft, dissenters against university cooperation with the House un-American Activities Committee, organizers of the Vietnam Summer Project, and Participants in the Berkeley Free Speech Movement. Despite these different selective criteria for student activism, the results across studies have been remarkably coherent. Activists are found to be intellectually gifted, academically superior, and politically radical young people from advantaged homes in which the parents are successful in their careers, comfortable in their economic position, and are liberal in their political orientations” (pp. 143-144, citations have been omitted).
What’s the implication of this? I dunno. Your guess is as good as mine, since the authors offer no guidance or explanation regarding this portion of the LIB/CON score.
The passage does, however, shed light on the authors’ attitude toward activists in the 1960s. Political activists, who came from advantaged, liberal homes (and participated in decidedly liberal causes) were judged to be gifted and academically superior. Might that mean that participants in the present study who reported a history of activism were judged to be liberal, gifted, and superior? We can only speculate, but there is certainly a striking resemblance to the present study’s conclusions.
A LIB/CON Job?
The LIB/CON scale, which is absolutely pivotal to the integrity of this study, appears to be a shambles of bias, emotion, and muddled application. Factors that appear to have contributed to each participant’s composite LIB/CON score include:
- Seething contempt for conservatives (measure #3)
- Misuse of two scales resulting in probable misclassification of participants (measures #4 and #5)
- Mystery Factor X (measures #2 and #6)
- A self-identification measure that is contraindicated by the current literature (measure #1)
- Outdated, inaccurate ideological descriptions (measure #3)
Additionally, there appears to be a strong potential for the effects of social desirability pressure and rater bias (measures #1 and #2), as well as a possible affinity for and bias in favor of liberal political activists (measure #6). The authors reported nothing in the way of controlling for these factors.
|
I’m a selfish man. I don’t like explaining to friends and family why my profession seems to despise half the nation. |
It appears as though the correlations between ideology and personality cannot be supported. In fact, it is entirely unclear what this study actually measured.
Trickle-Down Histrionics
I’m a selfish man. I don’t like explaining to friends and family why my profession seems to despise half the nation. And I certainly don’t like the idea of people suffering through psychological difficulties because they think they are unwelcome at my door.
Of course, most clinicians I know are perfectly willing and able to work with patients of different political orientations. The problem is, studies like this one are leapt upon by reporters and ideologues hungry for a sound bite, and they make a big impression.
These editorials-masquerading-as-research are taken seriously by mainstream outlets such as Psychology Today and the Toronto Star, which ran a fawning account of the findings last March (Kleiner, 2006). And, naturally, the simplified findings of this study were joyfully replayed on blogs such as The Daily Kos (2006). None of them questioned the inexcusable methodology. Why would they?
That leaves me wondering how to justify my profession to people who are becoming increasingly and understandably suspicious. In the third and final installment of this series, I’ll try to come up with a coherent answer. I better get to work.
-IS
Footnotes:
1. Block & Block reported that, “it is of interest to note that the LIB/CON score does not relate to intelligence (WPPSI IQ) at age 4 for either boys or girls (r’s of .07 and .01). However, LIB/CON correlates positively with intelligence in the following decade for both boys and girls. WISC IQ at age 11 correlates with LIB/CON .30 and .28 [p<.10], and WAIS IQ at age 18 correlates .36 and .24 [p<.10] for boys and girls, respectively” (p. 743; italics in original). Translation: participants were administered intelligence tests at various ages and, as they matured, higher scores correlated with liberalism.
2. When challenged on the topic of external validity (generalizability), Dr. Block has argued elsewhere that within the sample, “the results hold” (Kleiner, 2006). In their article, Block & Block admit that Berkeley is “appreciably different from much of America” and that the “San Francisco Bay Area provides a context that unsurprisingly and unembarrassedly encourages liberalism and looks askance at much of conservatism” (p. 744). Why, then, devote two pages to a general discussion titled “some implications of the results” (see section 4.1, pp. 745-746), beginning with the phrase, “the thing speaks for itself”? Why boast about the bulletproof representativeness of the LIB/CON construct (p. 737)? Why not protest, rather than defend, when media outlets pounce on oversimplified results? Why publish the article at all if it only applies to 95 people? Despite their diplomatic qualifications concerning external validity, one could easily walk away with the impression that they have no question whatsoever regarding the generalizability of the results.
3. On a personal note, I take no pleasure in writing this particular deconstruction. Dr. Jeanne Block passed away recently and in addition to losing his wife, Dr. Jack Block has had other challenges. One of my wisest clinical professors studied under the guidance of Professor Block, making the man a distant intellectual uncle of sorts. Nevertheless, this study represents one more in a series of pugilistic articles aimed at ideological opponents, produced under the guise of high-minded, legitimate research. Someone needs to speak up. It may as well be me.
References:
Block, J. & Block, J.H. (2006). Nursery school personality and political orientation two decades later. Journal of Research in Personality, 40(5), 734-749.
Block, J.H., Haan, N., & Smith, M.B. (1969). Socialization correlates of student activism. Journal of Social Issues, 25, 143-147.
The Daily Kos (11/29/2006). Downloaded on January 26, 2007, from: http://www.dailykos.com/comments/2006/11/29/155510/54/264#c264.
Dixit, J. (2007). The ideological animal. Psychology Today, 40(1), 81-86.
Jost, J.T., Glaser, J., Kruglanski, A.W., & Sulloway, F.J. (2003). Political conservatism as motivated social cognition. Psychological Bulletin, 129(3), 339-375.
Kerlinger, F.N. (1984). Liberalism and Conservatism: The Nature and Structure of Social Attitudes. Hillsdale, NJ: Lawrence Erlbaum.
Kleiner, K. (2006). How to spot a baby conservative. Toronto Star, March 19, 2006, p. D03.
McClosky, H. (1958). Conservatism and personality. The American Political Science Review, 52(1), 27-45.
Sanders, D., Burton, J., & Kneeshaw, J. (2002). Identifying the true party identifiers: a question wording experiment. Party Politics, 8(2), 193-205.
Wolf, F.M. (1990). Methodological observations on bias. In Wachter, K.W., Straf, M.L. (Eds) The Future of Meta-Analysis (pp. 139-151). New York: Russell Sage Foundation.