Shine a light on journal self-citation inflation

Photo by pnc.

Photo by pnc.

Note: Welcome Inside Higher Ed readers. I’d be happy to hear accounts from disciplines other than sociology. Email me at

In my post on peer review the other day, I mentioned that a journal editor made this request — before she agreed to send the paper out for review:

“If possible, either in this section or later in the Introduction, note how your work builds on other studies published in our journal.”

A large survey on “coercive citation” practices, published in Science in 2012 (paywalled; bootlegged PDF) found that 20% of researchers had, in the previous five years, “received a request from an editor to add more citations
from the editor’s journal for reasons that were not based on content.” The survey, which was sent to email lists for academic associations, including the American Sociological Association, found sociologists and psychologists were less likely to report having experienced this practice than were economists and those in business-related disciplines.

The journal I named, Sex Roles, is high on the list of those most frequently mentioned — cited by four respondents, more than any journal outside of business, marketing, or economics. But there are a lot of other journals you know on the list.

Although I made the assumption that the Sex Roles editor was trying to increase the impact factor — the citation rate — for her journal, one could defend this practice as being motivated by other interests (I’ll leave that to you). It also seems likely that some requests are open to interpretation — for example, mixing in citations from different journals, or offering specific reasons for including particular citations.

Tell me about it

To look into this a little more, I’m asking you to send me requests for journal self-citation that you have received. I’ll keep them confidential, but if I get enough to make it interesting, I will post: (1) journal name, (2) the type of request, (3) the date (month and year), and (4) the stage in the publication process. Feel free to include extenuating details or other information you would like to share, and let me know if you want it disclosed. I assume most of you are sociologists, but I’ll include items from any discipline.

To be included on the list, I’ll need to see copies of the letter or email you received. I will not disclose your identity or information about you, or the specific article under review. I won’t use quotes that might identify the author or article under review.

I will also send the list to the current editors of journals named and give them an opportunity to respond.

My contact information is here.

Maybe there’s not enough here to go on, but if there is, I think shining a light on it would be a good thing, and might deter the practice in the future.


Filed under Me @ work

Our broken peer review system, in one saga

When at last Odysseus returns.

When at last Odysseus returns.

Everybody’s got a story. This is the story of publishing a peer-reviewed journal article called, “The Widening Gender Gap in Opposition to Pornography, 1975–2012.” The paper has now been published, and is available here in preprint, or here if you’re on a campus that subscribes to Social Currents through Sage.

Lucia Lykke, a graduate student in our program, and I began this project in the fall of 2012. We came up with the idea together. I did the coding and she wrote the text. Over the course of two years we sent the paper to four journals – once to Gender & Society, four times to Sex Roles, once to Social Forces, and twice to Social Currents, which finally accepted it in July 2015 and published it online on September 21.*

This story illustrates some endemic problems with our system of scholarly communication, both generally and in the discipline of sociology specifically. I discuss the problems after the story.


The gist of our paper is this: Opposition to pornography has declined in the U.S. since 1975, but faster for men than for women. As a result, the gender gap in opposition – with women more likely to oppose pornography – has widened.

That’s the finding. Our interpretation – which is independent of the veracity of our finding – is that opposition has declined as porn became more ubiquitous, but that women have been slower to drop their opposition because at the same time mainstream porn has become more violent and degrading to women. We see all this reflecting two trends: pornographication (more things in popular culture becoming more pornographic) and post-feminism (less acceptance of speaking up against the sexist nature of popular media, including porn). We could be wrong in our interpretation, and there is no way to test it, but the empirical analysis is pretty straightforward and we should accept it as a description of the trend in attitudes toward pornography. And for doing that empirical work we beg permission to tell you our interpretation.

The analysis is possible because the General Social Survey has, since 1975, asked a large sample of U.S. adults this question about every two years:

Which of these statements comes closest to your feelings about pornography laws: 1. There should be laws against the distribution of pornography whatever the age. 2. There should be laws against the distribution of pornography to persons under 18. 3. There should be no laws forbidding the distribution of pornography.

We tracked the rate at which people selected the first choice versus the others. It’s not very complicated (although we tried it half a dozen other ways, of course). Also of course it’s not perfect – it’s not a great question for today’s social reality, but it’s the only thing like it asked over such a long period. This is what’s great and what’s limiting about the General Social Survey. So, let’s agree to collect better data, and also use this. There, was that so hard?

Here is supporting detail on our particular saga. (We have left the typos from reviewers intact, because it makes us look smarter than they are. And these are selective excerpts to make various points – there was a lot, lot, more.)

Before and after

Just to be clear what the world gained by 13 reviews and two years of waiting, you can compare the abstract at the beginning to the one at the end. This was the original abstract:

In the last several decades pornography in the U.S. has become more mainstream, more accessible, and more phallocentric and degrading to women. Yet research has not addressed how opposition to pornography has changed over the past several decades. Here, we examine opposition to pornography and gender differences in anti-pornography attitudes, using the 1975-2012 General Social Survey. Our findings show that both men’s and women’s opposition to pornography have decreased significantly over the past 40 years, but men’s opposition has declined faster and women remain more opposed to pornography. This is consistent with both the growing normative nature of pornography consumption for men and its increasingly degrading content. We situate these trends within a cultural climate in which women are caught between postfeminism and pornographication – between cultural messages that signal the social acceptability of pornography and compel women’s acquiescence, on the one hand, and the increased presence of pornography many women consider offensive and harmful on the other.

And this was the abstract we ended up with:

In the last several decades pornography in the U.S. has become more mainstream, more accessible, more phallocentric and more degrading to women. Further, consumption of pornography remains a major difference in the sexual experiences of men and women. Yet research has not addressed how opposition to pornography has changed over the this period, despite shifts in the accessibility and visibility of pornography as well as new cultural and legal issues presented by the advent of Internet pornography. We examine gender differences in opposition to pornography from 1975 to 2012, measured by support for legal censorship of pornography in the General Social Survey. Results show that both men’s and women’s opposition to pornography have decreased significantly over the past 40 years, suggesting a cultural shift toward “pornographication” affecting attitudes. However, women remain more opposed to pornography than men, and men’s opposition has declined faster, so the gender gap in opposition to pornography has widened, indicating further divergence of men’s and women’s sexual attitudes over time. This is consistent with the increasingly normative nature of pornography consumption for men, increases over time in men’s actual consumption of pornography, and its increasingly degrading depiction of women.

The regression model we started with in 2013 had logistic regression coefficients showing a decline of .012 per year in the log odds of women favoring laws against the distribution of pornography, versus .022 for men. (That is, the decline has been almost twice as fast for men.) After all we went through with the other variables, we ended up with .012 and .023.


August 6, 2013: Submitted to Gender & Society

September 23, 2013: Rejected, with four reviews

Reviewer A was concerned about framing, and about the dependent variable.

if one takes this more complex and nuanced definition of postfeminism into account, the theoretical frame of does not work well for the paper … I also thought that the authors could have gone further in discussing broader cultural changes in sexuality in the media, especially the increasing sexualization and pornification in advertising and the media. …

an analysis of a GSS question concerning laws regarding the restriction of pornography, seem limited. In particular, that GSS question does not seem to get at the historical changes that have occurred in pornography distribution and consumption given its widespread internet usage.

Reviewer B was all about framing:

[I] appreciate your analysis of anti-pornography research and the effects of post-feminism on attitudes towards pornography … [but] I think the literature review needs to spend at least some time outlining feminist pro-pornography arguments. …

doesn’t it make sense to incorporate a discussion of the history of pornography regulation since the 1970s in the U.S? [… and …] While you bring up race in the analysis of your data, the literature review is surprisingly devoid of anything having to do with pornographic representations of gender and race.

Reviewer C thought we should have included a content analysis of pornography over time – done a different study, that is — and framed it differently:

Pornography needs to be defined … Cost, images and rejection of feminist view would clearly support a content analysis on pornography … The provided discussion of pornographication seems to more support the use of images and actual study of pornography, more so than people’s attitudes toward it … more justification to the existing literature needs to be added … Some legal gender studies should be included here … Gender is not one sided and the author should consider adding some agency to [men’s] role in the study and discussion.

Reviewer D concluded that the data weren’t good enough to support our interpretation:

The author, however, does not empirically demonstrate that the found decline in opposition is the result of either postfeminism or pornographication. … The General Social Survey is convenient, easy to access, and quick to run. This, however, does not necessarily make for good empirical evidence. … If the author wanted to investigate postfeminism and pornagraphication and the relationship to pornography, a much more nuanced empirical study would have needed to have been designed.

In a world with limited space for publishing research – which is not our world – this would be a good reason to reject the article.

October 7, 2013 (approximate): Submitted to Sex Roles

October 9, 2013: Returned by the editor

The editor, Irene Frieze, returned the paper almost immediately, saying: “major revisions are needed before we can move ahead in the review process.”

Some of what she asked for reflects the competitive climate of contemporary academic journals. For example, she asked us to pad the journal’s citation count: “If possible, either in this section or later in the Introduction, note how your work builds on other studies published in our journal.”

And she tried to make the journal seem more international:

Explain why your study is important to readers from many countries with a sentence or two. … Note what country each empirical study you cite was done in and explain how any cited studies done in other countries are relevant in understanding your sample.

She also asked for what appear to be standard requirements for the journal:

Add demographic information about the sample and explain more about how they were recruited. Add a table showing the demographic characteristics of the women as compared to the men in the sample in different time periods. … Add correlations computed separately for women and men as well.

And, the dreaded memo requirement: “Assuming you do wish to submit a revision, I would need a revised manuscript and a detailed list outlining the changes you have made in response to these comments.”

November 9, 2013: Resubmitted to Sex Roles, first revision

February 17, 2014: Revise and resubmit, based on one review (“major revisions”)

The reviewer had trouble with our statistical presentation:

I see that on Table 2, the difference between the women’s and men’s regression effect for year shows both women’s and men’s significant (-.012 and -.022). This suggests that for both female and male respondents the year is significant, but it doesn’t show statistically that men’s decline in opposition is steeper than is women’s. Where is the statistic showing a significant difference in slope? [The table had a superscript b next to the men’s coefficient, with the note, “Gender difference significant at p<.05.” Although we didn’t provide the details, that test came from a separate, “fully-interacted” model in which every variable is allowed to have a separate effect by gender.]

This reviewer – who stuck with this complaint for three rounds – also had trouble with the smallness of the coefficients:

Although the coefficient is twice as large for year among men than among women, it’s a very small percentage. With such a large sample size, almost anything will be significant. I’d like to see an effect size statistic.

She might have been confused because the variable here is “year” – a continuous variable ranging from 0 in 1975 to 37 in 2012, so the coefficient reflects the size of the average one-year change, which makes it look “very small.”

A common problem for authors responding to reviewers is the simultaneous demands for less and more. Sometimes that’s good – a healthy revision process. Here is a funny example of that: “There seems to be a much longer introduction than is needed for the findings, especially since what would be interesting to me is omitted.”

However she grasped the concise nature of the findings, which she somehow took as a weakness:

I would like to see how each of these control variables interacts with the changes over years. I believe that analysis is possible using time series analyses. The reader is left with only a few main conclusions: both men and women indicate less opposition over time to pornography, and that men’s opposition declines more than female’s, and men show less opposition to pornography control overall.

Exactly. Oh well.

May 17, 2014: Resubmitted to Sex Roles, second revision

July 8, 2014: Revise and resubmit, with two reviews

The editor now told us: “We were able to find a second reviewer, this time. We won’t continue to add new reviewers for additional drafts.” (This promise, sadly, did not hold.)

The dependent variable – that three-response question about laws regulating pornography – caused continuing consternation. The editor wrote:

none of us feels that the combining of the three categories of responses for the pornography acceptance variable is appropriate. You either need to omit one of the 3 categories from the analysis, or do something like a discriminant analysis to look at differences in those responding to each of the three categories.

And then this bad signal that the editor and reviewers did not understand the basic structure of the analysis:

Another issue that all of us agree on is that you have failed to provide statistical evidence supporting your assertion of evidence of a linear trend in support over time. Either do a real trend analysis, for women and men separately, or compare the data over several specific years using something like ANOVA by year and gender. This would also allow you to see if these is really the interaction you assert is present.

As you can see in the final paper – which was the case in this revision as well – we did a “real trend analysis, for women and men separately.”

We tried to make this as clear as possible, writing in the paper:

We use logistic regression models to test for differences on this measure between men and women across the 23 administrations of the GSS since 1975. We test time effects with a continuous variable for year, which ranges from 0 in 1975 to 37 in 2012. This coding allows for an intuitive interpretation of the intercept and produces coefficients equal to the predicted change in the odds of opposing pornography associated with a one year change in the independent variable (non-linear specifications did not improve the model fit). … The first model combines men and women, while models 2 and 3 analyze men and women separately, after tests showed differences in the coefficients by gender on six of the variables (marked with superscript ‘b’). … Comparison of Model 2 and Model 3 confirms that the decline in opposition to pornography has been more pronounced for men than for women, as the coefficient for the year variable is almost twice as large.

We thought that Reviewer 1, back from the previous round, was doubling down on misunderstanding what we did, and the editor thought this as well. The reviewer wrote: “I don’t agree that the years need collapsing in the analyses. I believe it is better to see the linear trend. Also, I don’t like to see data left out, in this case data from the individual years.”

In fact, we found out in the next round of reviews we found the she meant this is a disagreement with the editor! (“The authors misread my statement about collapsing the years. I was disagreeing with the editor who suggested collapsing the years. I did not suggest myself that the years should be collapsed. I agree that the years should not be collapsed. It’s not me who misread the paper, it’s the authors who misread my statement.”)

That said, she still did not grasp the analysis:

You state that ‘This coding allows for an intuitive interpretation of the intercept and produces coefficients equal to the predicted change in the odds of opposing pornography associated with a one year change in the indepenjdent variable.’ In the results section, please describe how your data fit an ‘intuitive’ interpretation and how the coefficients that are produced explain the one year change. There is a disconnect for me from this statement and the description of the data.

And she added:

Please carefully describe the statistical analysis and statistical findings that describe the difference between the declines in opposition for women vs. men. Is the beta for gender .78 and for year -.02, and how did you test for the difference in betas of -.01 vs. -.02? Mention the test you used to assess this. This doesn’t seem like much of a difference in slope. That one is twice as large as the other is fairly meaningless when it is .01 vs. .02.

And added again later: “P. 22, agvain when you say a coefficient for the year variables is “amost twice as large,” you are talking about .01 vs .02.”


The editor and Reviewer 1 had a long-running dispute about how to handle all of our control variables. The editor was sticking to the policy that we needed a table showing complete correlations of all variables separately by gender. And a discussion of every variable, with references, justifying its inclusion. The editor said in the first round:

You also need to explain each of the control variables you include in your regressions in the Introduction. Add at least a sentence for each variable explaining why it is important to the issues you are testing.

In response, we included a long section beginning with, “Various social and demographic characteristics are associated with pornography use and attitudes toward pornography, and we account for these characteristics in our empirical analysis below.”

But then Reviewer 1 said of that passage: “Much of the material in “Attitudes Toward Pornography” is not relevant. … Gender and gender differences are what you are studying.”

And in response to our gigantic correlation table of all variables separately by gender, Reviewer wrote: “I … strongly recommend deletion of Table 3. This is not a study of the correlates of attitudes toward pornography, and the intercorrelations of all the control variables are outside the range of your focus.”

Never mind.

Reviewer 2, the new reviewer, had some reasonable questions and suggestions. For example, s/he recommended analyzing the outcome with a multinomial logistic regresstion, which we did but it didn’t matter; and controlling for pornography consumption (“watched an x-rated movie in the past year”), which we did and it didn’t matter (in fact, basically none of the control variables affect the basic story much, but reviewers have a hard time believing this). S/he also had lots of objections to how we characterized various feminist authors and terms in the framing, and really didn’t like “pornographication” as a term, listing as a “major” objection:

the term ‘pornographication’ is problematic and should be removed from the paper in favor of a more academic description of increased access to sexualized media.

September 10, 2014 (approximate): Resubmitted to Sex Roles, third revision

October 11, 2014: Revise and resubmit, with one review

The editor now informed us that one reviewer just recommended rejecting the paper because we didn’t address her concerns, while the other called for “major revisions.”

Given this type of feedback, I would normally reject a paper already in its third revision. However, I would like to offer one more opportunity for you to make the requested changes. If you do resubmit, I may seek new reviewers and essentially begin the review process anew, unless it is clear that my earlier concerns are fully addressed.

Despite three drafts and as many memos, the editor still did not seem to understand that our outcome variable was a single question with three options. She wrote:

One of my basic requests has been that you consider the question about exposure of pornography to those under 18 as a separate dependent variable, or omit this entirely from the study. Conceptually, I feel this is quite different from the other two survey items and cannot be combined with them. This will require major changes in the analysis and rationale for predictions relating to each of these measures.

The reviewer, however, disagreed, voicing approval for our choice. The editor clarified, “If my requests conflict with those of the reviewer, it is my requests you need to follow, not those of the reviewer.”

They had no trouble agreeing, however, that they did not understand the linear time trend we were testing: “As the reviewer explains, we do need a clearer discussion of how the linear trend is being tested.”

Reviewer 1 wrote:

Regarding the analysis of the time trend, although the authors state [in the memo] that the starring of the coefficients on Table 4 demonstrate a significant linear trend, it was not apparent to the editor and reviewers. As one of the main points of the study, it should be made very obvious that there is a significant linear trend via statistics. If this means being more explicit in the text of the results section, it would be important to do. If there’s this much confusion, the statistical analysis needs clarification.

You can look at the table in the final publication for yourself to see if this remains unclear. And then the reviewer added:

As I previously mentioned, though significant, a change of -.02 vs .-.01 is not substantial. Thus, the authors should refrain from concluding one is twice as large as the other.

We decided to take our business elsewhere rather than submit another revision.

November 4, 2014: Submitted to Social Forces

December 29, 2014: Rejected, with two reviews

Reviewer 1 only had concerns about framing, such as, “expand their discussion of the broader cultural changes in sexuality in the culture,” and discuss “changes in gay and lesbian identities and visibility during this period.”

Reviewer 2 simply thought we couldn’t answer the questions we posed with the data we had:

The paper is motivated by a largely assumed cultural ‘pornographication’ process linked to post-feminism. Neither concept seems well-suited to explain public opinion formation or change, and greater specificity about these concepts would likely outstrip the operational capacity of the GSS to model how gender and sexuality attitudes may influence shifts in beliefs about pornography.

There were some other technical issues about specific variables that aren’t very important. Again, this is very reasonable basis for making the ridiculous judgment forced by the system of publishing in the limited pages of a print journal.

January 16, 2015: Submitted to Social Currents

April 9, 2015: Revise and resubmit, based on three reviews

The editors, Toni Calasanti and Vincent Roscigno, wrote:

While stated differently in each case, the overriding sentiment across the reviewers is that the paper needs better framing. … the potential contribution of this study is not realized because the theoretical framework is lacking, limiting your ability to discuss the implications of your findings.

Reviewer 1 wanted the “post-feminism” discussion put back in the front: “It’s not until the conclusion of the manuscript that we learn about a potential contribution to ‘postfeminism’ and current work there.”

Reviewer 1 also attempted to lead us into a common trap. S/he wrote:

The hypotheses don’t necessarily derive from a particular theory in sociology or test a specific argument about gender, public opinion theories, and pornography per se. Rather, the project is descriptive (divergence of male/female support for legal control, rate of change over time, etc.). That isn’t fatal. But a project that makes a more direct connection to advancing current theoretical work in feminism and sexuality studies, or current theorizing about the importance of public opinion and values about pornography, would strengthen the overall contribution of this research.

Making the paper more theoretical is not a bad suggestion, but in this context – since the data are so limited – it’s a sure setup for a future reviewer to complain that you have asked questions you can’t sufficiently answer with your data.

The three reviewers’ other concerns by this point were quite familiar to us. For example, “perhaps a line or two to strengthen the validity of measure could be added based on some of the studies cited.” And a worry about about collapsing the dependent variable into two categories. And the need to acknowledge debates within feminism about the meaning of “pornographication.” We dutifully beefed up, clarified, and strengthened. And wrote a memo.

May 20, 2015: Resubmitted to Social Currents, first revision

July 18, 2015: Accepted


Some of the problems apparent in this story are common to sociology, some are more general.

Sociologists care way too much about framing. Most (or all) of the reviewers were sociologists, and most of what they suggested, complained about, or objected was about the way the paper was “framed,” that is, how we establish the importance of the question and interpret the results. Of course framing is important – it’s why you’re asking your question, and why readers should care (see Mark Granovetter’s note on the rejected version of “the Strength of Weak Ties”). But it takes on elevated importance when we’re scrapping over limited slots in academic journals, so that to get published you have to successfully “frame” your paper as more important than some other poor slob’s.

The journal system gets in the way. When journals reject you they report the low percentage of papers they accept. This is supposed to make the rejected authors feel better, but it also shows the gross inefficiency of the system: why should you bounce from journal to journal with low acceptance rates – in our case, asking our colleagues to write 13 reviews – instead of being vetted once by a centralized system with reviewers who work to a common standard? The answer is because that’s the way they did it in the Dark Ages, when physically printing research papers at high cost was the only way of distributing scholarly output.

The system is slow. As a result of these and other systemic problems, we do a terrible job of advancing knowledge. From the time of our first submission to the publication date was 776 days. For 281 of those days it was in our hands, but for the other 495 days it was in the hands of editors, reviewers, and the publisher. Despite responding to 13 reviews, with a lot of tinkering, the basic result did not change from our first submission in August 2013 to our last submission in May 2015. The new knowledge was all created two years before it was published.

The system is arbitrary I don’t want to make Social Currents look bad here, with the implication that they are a lower quality journal because they published something rejected by three journals before. After all, Granovetter’s paper was rejected by American Sociological Review before getting 35,000 citations as an American Journal of Sociology paper. I also like the example of Liana Sayer and Suzanne Bianchi’s paper on economic independence and divorce, which was rejected by the Journal of Marriage and Family, the flagship journal of the National Council on Family Relations (NCFR), before promptly winning NCFR’s best-paper award after it was published in the Journal of Family Issues. That is, one small group of reviewers deemed it unpublishable in a top journal, and the next declared it the best article of the year. That’s a very wide spread. The arbitrariness of the review system we have now creates cases like this – and who knows how many others. It’s not a systemic problem that Sex Roles has a reviewer that won’t let you say .02 is twice as large as .01. The problem is that could happen anywhere – and cost people their careers – at the same time that bad stuff gets through for arbitrary (or pernicious) reasons. There is too much noise in the current peer-review system to trust it for quality control.


Consider an alternative system, for example, in which the paper – having passed a very low bar of basic quality – had been published after the first set of reviews and then subjected to post-publication review and discussion in the field. Another alternative is publishing it before any formal review process, and allowing post-publication review to do the whole vetting process.

Models exist. Sociology doesn’t have a central working paper system, but there are smaller systems. In my neck of the woods, the California Center for Population Research has a working paper archive, which houses papers from six population centers. Math types have arXiv, which has more than a million papers, with each new one “reviewed by expert moderators to verify that they are topical and refereeable scientific contributions that follow accepted standards of scholarly communication.” They also use a system of member endorsement to cut down on junk submissions. If papers are subsequently published the arXiv version is updated to link to the published version. Sociology should make something like this.

Another step in the right direction is rapid-response, open-access peer-review, with quick up-or-down decisions. In sociology this includes Sociological Science, run by an independent team and supported by author fees (often paid by university libraries or grants); and Socious, run by the American Sociological Association and subsidized by the for-profit publisher Sage in an attempt to pacify open-access advocates. These work more or less like PLOS One, which “accepts scientifically rigorous research, regardless of novelty.”

I’m happy to publish in such outlets, but many of us worry about the career implications for our students who risk having their CVs seen as sketchy by old-fashioned types. We need them to be institutionalized.

In the meantime, those of us in position to conduct peer review can do our part to be better reviewers (see this excellent advice). And we can make explicit decisions about which journals we will review for. The system runs off our discretionary contributions, and we shape it through our actions. That argument is for a separate post.

* We did the research together — and Lucia did most of the work — but blame me for the content of this post.


Filed under Me @ work

NYT magazine infographic: not just dumb and annoying

This graphic from the New York Times magazine is bad data presented poorly (and reproduced poorly, by my camera phone):


It’s presented poorly because those blood stains are impossible to compare since you can’t discern their edges, and it appears they don’t taper toward the edges at the same rate. Maybe they simply resized one of them to get the relative size, which would be wrong. Anyway, if they cared about communicating the data they probably would have used real data in the first place. (You could also complain that a red speckle-cloud is unfriendly to some color-blind people.)

It’s bad data because it’s an online NYT reader survey, which — although it’s from the “research and analytics” department (and no, I’m not going to add “analytics” to my Windows dictionary) — represents unknown sample selection effects on an undefined population. In other words, who cares what they think?

A survey like that would be a start if it was the only way you had to answer an important or hard-to-measure issue, and if you clearly stated that it was likely unreliable. But in this case there is good, nationally-representative data on this very question. So if NYT Magazine wanted to inform its readers of something, they could have used this.

Here’s the good data — from the General Social Survey — in a graph that is at least a lot better: this is good data in a chart that’s easier to read accurately, includes a breakout by strength of opinion, and uses more accessible colors (click to enlarge).

gss spank 2014.xlsx

I think the NYT Magazine graphics violations are not just dumb and annoying — here’s another post all about them — I think they harm the public good. Graphics like this spread ignorance and contribute to the perception that statistics – especially graphic statistics – are just an arbitrary way of manipulating people rather than a set of tools for exploring data and attempting to answer real questions. (If you want awesome real graphics, check out Healy and Moody’s Annual Review of Sociology paper.)

P.S., I wrote more about spanking here.

Leave a comment

Filed under Uncategorized

Weathering and delayed births, get your norms off my body edition

You can skip down to the new data and analysis — or go straight to my new working paper — if you don’t need the preamble diatribe.

I have complained recently about the edict from above that poor (implying Black) women should delay their births until they are “financially ready” — especially in light of the evidence on their odds of marriage during the childbearing years. And then we saw what seemed like a friendly suggestion that poor women use more birth control lead to some nut on Fox News telling Rebecca Vallas, who spoke up for raising the minimum wage:

A family of three is not supposed to be living on the minimum wage. If you’re making minimum wage you shouldn’t be having children and trying to raise a family on it.

As if minimum wage is just a phase poor people can expect to pass through only briefly, on their way to middle class stability — provided they don’t piss it away by having children they can’t “afford.” This was a wonderful illustration of the point Arline Geronimus makes in this excellent (paywalled) paper from 2003, aptly titled, “Damned if you do: culture, identity, privilege, and teenage childbearing in the United States.” Geronimus has been pointing out for several decades that Black women face increased health risks and other problems when they delay their childbearing, even as White women have the best health outcomes when they delay theirs. This has been termed “the weathering hypothesis.” In that 2003 paper, she explores the cultural dynamic of dominance and subordination that this debate over birth timing entails. Here’s a free passage (where dominant is White and marginal is Black):

In sum, a danger of social inequality is that dominant groups will be motivated to promote their own cultural goals, at least in part, by holding aspects of the behavior of specific marginal groups in public contempt. This is especially true when this behavior is viewed as antithetical or threatening to social control messages aimed at the youth in the dominant group. An acknowledgment that teen childbearing might have benefits for some groups undermines social control messages intended to convince dominant group youth to postpone childbearing by extolling the absolute hazards of early fertility. Moreover, to acknowledge cultural variability in the costs and consequences of early childbearing requires public admission of structural inequality and the benefits members of dominant groups derive from socially excluding others. One cannot explain why the benefits of early childbearing may outweigh the costs for many African Americans without noting that African American youth do not enjoy the same access to advanced education or career security enjoyed by most Americans; that their parents are compelled to be more focused on imperatives of survival and subsistence than on encouraging their children to engage in extended and expensive preparation for the competitive labor market; indeed, that African Americans cannot even take their health or longevity for granted through middle age (Geronimus, 1994; Geronimus et al., 2001). And one cannot explain why these social and health inequalities exist without recognizing that structural barriers to full participation in American society impede the success of marginalized groups (Dressler, 1995; Geronimus, 2000; James, 1994). To acknowledge these circumstances would be to contradict the broader societal ethic that denies the existence of social inequality and is conflicted about cultural diversity. And it would undermine the ability the dominant group currently enjoys to interpret their privilege as earned, the just reward for their exercise of personal responsibility.

But the failure to acknowledge these circumstances results in a disastrous misunderstanding. As a society, we have become caught in an endless loop that rationalizes, perhaps guarantees, the continued marginalization of urban African Americans. In the case at hand, by misunderstanding the motivation, context, and outcomes of early childbearing among African Americans, and by implementing social welfare and public health policies that follow from this misunderstanding, the dominant European American culture reinforces material hardship for and stigmatization of African Americans. Faced with these hardships, early fertility timing will continue to be adaptive practice for African Americans. And, reliably, these fertility and related family “behaviors” will again be unfairly derided as antisocial. And so on.

Whoever said demography isn’t theoretical and political?

A simple illustration

In Geronimus’s classic weathering work, she documented disparities in healthy life expectancy, which is the expectation of healthy, or disability-free, years of life ahead. When a poor 18-year-old Black woman considers whether or not to have a child, she might take into account her expectation of healthy life expectancy — how long can she count on remaining healthy and active? — as well as, and this is crucial, that of her 40-year-old mother, who is expected to help out with the child-rearing (they’re poor, remember). Here’s a simple illustration: the percentage of Black and White mothers (women living in their own households, with their own children) who have a work-limiting disability, by age and education:


Not too many disabilities at age 20, but race and class kick in hard over these parenting years, till by their 50s one-in-five Black mothers with high school education or less has a disability, compared with one-in-twenty White mothers who’ve gone on to more education. That looming health trajectory is enough — Geronimus reasonably argues — to affect women’s decisions on whether or not to have a child (or go through with an accidental pregnancy). But for the group (say, Whites who aren’t that poor) who have a reasonable chance of getting higher education, and making it through their intensive parenting years disability-free, the economic consequence of an early birth weighs much more heavily.

Some new analysis

As I was thinking about all this the other day, I went to check on the latest infant mortality statistics, since that’s where Geronimus started this thread — with the observation that White women’s chance of a baby dying decline with age, while Black women’s don’t. And I noticed there is a new Period Linked Birth-Infant Death Data File for 2013. This is a giant database of all the births — with information from their birth certificates — linked to all the infant deaths from the same year. These records have been used for analyzing infant mortality dozens of times, including in pursuit of the weathering hypothesis, but I didn’t see any new analyses of the 2013 files, except the basic report the National Center for Health Statistics put out. The outcome is now a working paper at the Maryland Population Research Center.

The gist of the result is, to me, kind of shocking. Once you control for some basic health, birth, and socioeconomic conditions (plurality, parity, prenatal care, education, health insurance type, and smoking during pregnancy), the risk of infant mortality for Black mothers increases linearly with age: the longer they wait, the greater the risk. For White women the risk follows the familiar (and culturally lionized) U-shape, with the lowest risk in the early 30s. Mexican women (the largest Hispanic group I could include) are somewhere in between, with a sharp rise in risk at older ages, but no real advantage to waiting from 18 to 30.

I’ll show you (and these rates will differ a little from official rates for various technical reasons). First, the unadjusted infant mortality rates by maternal age:

Infant Death Rates, by Maternal Age: White, Black, and Mexican Mothers, U.S., 2013. Infant death rates per 1,000 live births for non-Hispanic white (N = 1,925,847), non-Hispanic black (N = 533,341), and Mexican origin (N = 501,390) mothers. Data source: 2013 Period Linked Birth/Infant Death Public Use File, Centers for Disease Control.

Infant Death Rates, by Maternal Age: White, Black, and Mexican Mothers, U.S., 2013. Infant death rates per 1,000 live births for non-Hispanic white (N = 1,925,847), non-Hispanic black (N = 533,341), and Mexican origin (N = 501,390) mothers. Data source: 2013 Period Linked Birth/Infant Death Public Use File, Centers for Disease Control.

These raw rates show the big health benefit to delay for White women, a smaller benefit for Mexican mothers, and no benefit for Black mothers. But when you control for those factors I mentioned, the infant mortality rates for young Black and Mexican mothers are lower — those are the mothers with low education and bad health care. Controlling for those things sort of simulates the decisions women face: given these things about me, what is the health effect of delay? (Of course, delaying could contribute to improving things, which is also part of the calculus.) Here are the adjusted age patterns:

Adjusted Probability of Infant Death, by Maternal Age: White, Black, and Mexican Mothers, U.S., 2013 Predicted probabilities of infant death generated by Stata margins command, adjusted for plurality, birth order, maternal education, prenatal care, payment source, and cigarette smoking during pregnancy; models estimated separately for white (A), black (B), and Mexican (C) mothers (see Tab. 1). Error bars are 95% confidence intervals. Data source: 2013 Period Linked Birth/Infant Death Public Use File, Centers for Disease Control.

Adjusted Probability of Infant Death, by Maternal Age: White, Black, and Mexican Mothers, U.S., 2013. Predicted probabilities of infant death generated by Stata margins command, adjusted for plurality, birth order, maternal education, prenatal care, payment source, and cigarette smoking during pregnancy; models estimated separately for white (A), black (B), and Mexican (C) mothers (see Tab. 1). Error bars are 95% confidence intervals. (A separate test showed the linear trend for Black women is statistically significant.) Data source: 2013 Period Linked Birth/Infant Death Public Use File, Centers for Disease Control.

My jaw kind of dropped. Infant mortality is mostly a measure of mothers’ health. Early childbearing looks a lot crazier for White women than for Black and Mexican women, and you can see why the messaging around delaying till your “ready” seems so out of tune to the less privileged (and that really means race more than class, in this case). Why wait? If women knew they had higher education, a good job, and decent health care awaiting them throughout their childbearing years, I think the decision tree would look a lot different.

Of course, I have often said that delayed marriage is good for women. And delayed childbearing would be — should be — too, as long as it doesn’t put the health of the mother and her children at risk (and squander the healthy rearing years of their grandparents).

Please check out the working paper for more background and references, and details about my analysis.


Filed under Me @ work

How about we stop moralizing and end child poverty tomorrow?

How much would you pay to stop having to listen to rich people tell poor people how to run their families?

If my calculations are correct, we can end child poverty for $62 billion per year. Is that a lot? No, it’s not. It’s $578 per non-poor family — but (if Twitter analytics are to be believed) my typical reader will pay less because I’ll put it on a sliding scale for you. Details below.

Americans tend to think of poverty as a giant, intractable problem, combining intergenerational dynamics, complex policy tradeoffs, conflicting cultural values, and “personal responsibility” (not to mention genetics). For example, in her book Generation Unbound, Isabel Sawhill says, “If we could return marriage rates to their 1970 level, the child poverty rate would be about 20% lower.” She’s (wisely) not advocating that, because it’s impossible, but think of it — rolling back one of the major demographic trends of the last half century would be social reversal on an unprecedented scale. For a measly 20% reduction in poverty? Apple alone could eliminate 100% of U.S. poverty for two years with the money under its couch cushions. (One reason people think poverty is so hard to solve is they don’t understand the scale of the population and the economy. Because “millions and millions” of poor people sounds like an insurmountable problem, it’s very helpful to play around with real numbers to get a sense of the magnitudes we’re dealing with.)

In our system, the vast majority of poor people are those in hard-to-employ categories. As Matt Bruenig recently wrote, 83% percent of poor people are either children, old people, people with disabilities, students, people taking care of family members, or people who can’t find jobs. (Among the employed poor, most are sharing their income with family members who can’t work.) We are a “country that relies heavily on the market to distribute the national income,” Bruenig writes. But it’s actually the market via the family. If these vulnerable groups are people who need someone else’s labor to support them, at least temporarily, then the attitude written into our policies is that such support should come from their families. If your family can’t do it — or you don’t have a family — good luck. It doesn’t have to be this way.

Isabel Sawhill, incidentally, is behind a column the other day by Catherine Rampell, which makes the reasonable suggestion to increase access to contraception for poor women, for the unreasonable reason that such a policy would “fight poverty” and reduce spending on welfare. The poverty angle here is that poor parents have — wait for it — poor children:

Children brought into the world before their parents were financially or emotionally ready for them are … disadvantaged before they’re even born, no matter how loved they are.

That “financially or emotionally ready” line is from Sawhill, and its implication is clear, though its advocates are for some reason squeamish about saying it plainly: poor people should not have children. I hate this attitude.

Look: children usually (fortunately) don’t make money. Somehow income from someone else’s labor has to pay for their homes, schools, doctors, food and water. A lot of that money comes from the state (for rich and poor kids alike). But under our stingy welfare state, if their parents don’t have decent jobs they wind up poor. The mindset that sees our welfare system as a fixed entity looks at this and says, “These kids are poor because of their parents. They weren’t financially ready to have kids.” Wrong. They’re poor because we insist on it.

I would like to live in a society — in a neighborhood, a community — in which people without good jobs can still have children, while they’re young, and have happy families. And I’m willing to pay my share of the cost of that. Are you? It’s not as much as you think.

Here are the details

All I did was calculate how much below the poverty line all the poor families with children are. That is the amount we need to raise (each year) to end child poverty. Then I distributed that cost across the non-poor families, on a sliding scale. How hard would this program be? We already have all the infrastructure in place to move income around; it’s just a change in the tax code.

With the 2014 Current Population Survey data from, I can calculate how much each poor family is below their poverty threshold. I’m focusing on families with children for now. There are 6.5 million poor families with a child under 18, and on average they are $9,450 below the poverty line based on their family size and composition. So, to eliminate child poverty we need $61.6 billion dollars per year.*

Where are we going to get that kind of money? From non-poor families.

There are 107 million non-poor families, so we’re going to need about $578 per family per year to pay this bill and end the scourge of child poverty. Of course, $578 is a lot of money for some people, but on average the non-poor families have incomes $40,874 more than their poverty threshold. To ease the pain, I created a simple, continuously-graded scale. I broke the non-poor families into 10 equal-size bins from rich to less rich, and slid the tax rate from 1.8% down to 0% (that way there’s no penalty for moving just over the poverty line). Here’s how it works (click to enlarge):

eliminate child poverty

The tax is only applied on the surplus for each family — that is, the resources they have (after taxes, work expenses, health care and child care) over their poverty threshold. If we tax the surpluses of the richest 10% of non-poor families at the virtually painless rate of just 1.8% — and everyone below them at an even lower rate — we end child poverty in the U.S.

Here’s the chart that shows how much you have to pay, broken down by average income in each decile of non-poor people**:

eliminate child poverty.xlsx

Some people say the Pope should stick to religious matters, and not speak about politics. Some people also say a social scientist should stick to scientific analysis, and not make moral demands. You can ignore my moralizing, as long as you understand the fact that child poverty is a choice we make with our policies. Eliminating child poverty does not require restructuring American families, mass contraception campaigns, or a new ethos of shame. It just costs a little money.

* Technical note: To do the calculations, instead of the official poverty rate I used the Supplemental Poverty Measure. This measures resources versus needs for “resource units,” which are either families (including cohabitors, foster children, and other people that are normally considered “non-relatives”) or unrelated individuals. For every resource unit, the poverty threshold is based on the cost of food, clothing, shelter, and utilities, adjusted for geographic location, housing type, and family composition. In addition to money income, the resources for the calculation include non-cash assistance like food stamps, school lunch, housing and energy subsidies; and then they deduct from resources taxes, work expenses, child care expenses, medical expenses, and child support (it’s all described here). I call resource units “families,” although some of them are single people. The Stata code I used to analyze the data, which includes the variables you need from IPUMS, is here.

** Please consider making a contribution of at least twice this to help address the much larger problem of poverty in the poor countries of the world.


Filed under In the news

Pick your potion, Dalai Lama and Pope Francis edition

If you only like religious leaders who agree with what you already think, what do you need them for?

Sometimes people cheer for statements by religious leaders like sports teams: Yay when they agree with you, boo when they don’t. So what’s the leader for? When was the last time a religious leader made you change your mind about a core moral issue?

That time when you realize the Dalai Lama really thinks a female Dalai Lama would be "not much use" if she weren't attractive.

That feeling when you realize the Dalai Lama really thinks a female Dalai Lama would be “not much use” if she weren’t attractive.

The BBC has an interview up with the Dalai Lama, which focuses on the refugee crisis and other issues. Such as the gender of the next Dalai Lama. Starting at 4:52 of the video:

Reporter, Clive Myrie: Is there going to a 15th incarnation of the Dalai Lama?

Dalai Lama: The very institution of the Dalai Lama should continue, or not, up to the Tibetan people.

CM: So the people will decide. Could it be a woman?

DL: Yes! One occasion in Paris, one woman’s magazine, one reporter, come to see me, I think more than 15 years ago. She asked me, any possibility female Dalai Lama. I mentioned, why not? The female, has biologically more potential to show affection…

CM: And compassion…

DL: Yes, compassion. Therefore, you see now, today’s world, lot of trouble, troubled world, I think female should take more important role. And then, I told that reporter, if female Dalai Lama come, her face must be — should be — very attractive. [Laughs]

CM: [Laughs] Oh well. So you can only have a female Dalai Lama if they’re attractive. Is that what you’re saying? You can’t have…

DL: I mean, if female Dalai Lama come, that female…

DM: …will be…

DL: …must be…

DM: …must be very attractive. It’s just gonna…

DL: Otherwise, not much use.

DM: Really?!

DL: I think some people — my face…

DM: You’re joking, I’m assuming. Oh, you’re not joking.

DL: Oh? I mean, true!


All over right now there are conservative Catholics who are unhappy because the Pope is not saying the things that they already believe. Like the Federalist Staff, who are upset that:

During his remarks [to Congress], which were regularly interrupted by rounds of applause from the assembled lawmakers, Pope Francis condemned the death penalty, called for better environmental stewardship, and even talked about the ills of political polarization. He did not, however, mention Jesus Christ, whose life, death, and resurrection form the very foundation of the Christian faith.

Apparently, Francis’s faith in Jesus is not to be taken for granted. (Personally, I find it polite it when religious leaders from religions to which I don’t belong, when speaking in state-sponsored settings before audiences that include non-followers, don’t invoke their own gods.)

Some religious people (but not only them, of course) use their religion to prop up unsupported empirical assertions. Michael Strain, from the American Enterprise Institute, for example, recently wrote, “we must begin with the understanding that each of us is called to love God and to love others.” Beginning with an understanding — rather than coming to an understanding on the basis of evidence — is one hallmark of faith over reason. But what Strain really has faith in is free markets — which to him are the one-variable empirical solution:

free enterprise dramatically reduces extreme poverty. In 1970, over one-quarter of the world lived on less than one dollar per day. By 2006, about one in 20 people lived in extreme poverty — an 80 percent reduction. We have the adoption of free markets across the developing world to thank for this massive reduction.*

For Strain, His Holiness’s appreciation for this single-variable view of history is disappointing: “The effect of liberalizing markets on extreme poverty and the good this does for families is a fact I wish the Holy Father discussed more often.” Strain seems to prefer Pope John Paul II, who wrote that “Man is the image of God partly through the mandate received from his Creator to subdue, to dominate, the earth.” That mandate, for example, apparently includes the mandate to reform disability insurance to make more disabled people work.

This is not allegiance to religious leadership, but rather the political business of cheering for the expression of views one already holds. (For example, I like it when powerful people say good things about sociology — not because it makes me believe them, but because it’s a point for our side.)

Having a pope speak against their views must be especially disheartening to the people who specifically chose to be Catholic because they thought the Catholic church would tell them (and their neighbors) to believe what they already believe.

As an atheist, I find some of this mystifying. However, I do appreciate the way people use religion to provide institutional support to values they support (especially when I support those values, too). That’s just building a social infrastructure to satisfy collective needs. (And yes, I know that the values I hold are partly the result of religious influence on me and those who taught me right from wrong. But citing religion isn’t the same as having faith in it.)

What I find even more mystifying is religious authority. And especially people going out of their way — like changing religions — to follow a religious authority. This seems sad to me; it’s an affirmation of one’s impotence. But odder still is people complaining about the views expressed by the religious authorities they choose to follow. I guess it’s like being misled by a movie preview and finding yourself stuck in the kind of movie you hate. You’ve already bought the ticket, and now you’re sitting there. Grr.

* That link Strain uses is to an NBER paper that seems to be an outlier in poverty analysis. The World Bank had 13.5% of people living at under $1 per day in 2008 (you can see various estimates here), but they prefer a measure of $1.25 per day, by which 22% of people were that poor in 2008.


Filed under In the news

Lifetime chance of marrying for Black and White women

I’m going to Princeton next week to give a talk at the Office of Population Research. It’s a world-class population center, with some of the best trainers and trainees in the business, so I figured I’d polish up a little formal demography for them. (I figure if I run through this really fast they won’t have time to figure any mistakes I made.)

The talk is about Black and White marriage markets, which I’ve written about quite a bit, including when I posted the figure below, showing the extremely low number of local same-race, employed, single men per women Black women experience relative to White women — especially when they have less than a BA degree.

This figure was the basis for a video we made for my book, titled “Why are there so many single Black women?” For years I’ve been supporting the strong (“Wilsonian“) case that low marriage rates for Black women are driven by the shortage of “marriageable” men — living, employed, single, free men. I promised last year that Joanna Pepin and I were working on a paper about this, and we still are. So I’ll present some of this at Princeton.

Predictions off

Five years ago I wrote about the famous 2001 paper by Joshua Goldstein and Catherine Kenney, which made lifetime marriage predictions for cohorts through the Baby Boom, the youngest of whom were only 30 in the 1995 data the paper used. That’s gutsy, predicting lifetime marriage at age 30, so there’s no shame that they missed. They were closer for White women. They predicted that 88.6% of White women born 1960-1964 would eventually marry, and by the age 49-53 (in the 2013 American Community Survey) they were at 90.2%, with another 2.3% likely to marry by my estimates (see below). For Black women they missed by more. For the 1960-1964 cohort, they predicted only 63.8% would ever marry, but 71.3% were already married by 2013, and I’m projecting another 7.5% will marry. (I also wrote about a similar prediction, here.) If they actually get to 79%, that will be very different from the prediction.

Their amazing paper has been cited another 100 times since I wrote about it in 2010, but it doesn’t look like anyone has tried to test or extend their predictions.

Mass incarceration

Interestingly, Goldstein and Kenney undershot Black women’s marriage rates even though incarceration rates continued to rise after they wrote — a trend strongly implicated in the Black-White marriage disparity. This issue has increased salience today, with the release of a powerful new piece by Ta-Nehisi Coates in the Atlantic (my old job), which exposes the long reach of mass incarceration into Black families in ways that go way beyond the simple statistics about “available” men. The large ripple effects implied by his analysis — drawing from his own reporting and research by Deva Pager, Bruce Western, and Robert Sampson — suggest that any statistical model attempting to identify the impact of incarceration on family structure is likely to miss a lot of the action. That’s because people who’ve been out of prison for years are still affected by it, as are their relationships, their communities — and their children in the next generation.

Some new projections

I should note that some readers unfamiliar with demographic analysis may find parts of what follows morbidly depressing.

To set up the marriage market analysis I’m doing with Joanna — which isn’t ready to show here yet — I’m going to introduce some marriage projections at the talk. These use a different method than Goldstein and Kenney, because I have a different kind of data. This is a lifetable approach, in which I use first-marriage rates at every age to calculate how many women would get married at least once before they die if they lived 2010 over and over again from birth to death. I can do this because, unlike Goldstein and Kenney in 2001, I now have the American Community Survey (ACS), which asks a giant sample of people if they have married in the previous year, and how many times they’ve been married before, so I can calculate a first-marriage rate at every age. To this I add in death rates — making what we call a multiple-decrement life table — so that there are two ways out of the birth cohort: marriage or death. (Give me marriage or give me death.)

The way this works is you start with 100,00 people, and each year some of them die and some of them get married — according to the rates you have measured at one point in time. For example, in my tables, of 100,000 Black women at the start of year 0, only 98.7% make it to age 15, the first year they can be counted as married in the data. By the time you get down to age 30, there are only 67,922 left, as 2,236 have died and 29,843 have married for the first time. And so on down to the bottom. In the last row of the table, when they are all dead, you calculate how many got married before dying.*

The bottom line: 85.3% of White women, and 78.4% of Black women born and stuck in 2010 forever are projected to marry before they die — a surprisingly small gap. The first figure shows you that basic result:

NHBW life tables 2010.xlsx

Note that my projections of 85.3% of White women and 78.4% of Black women ever marrying are lower than, for example, the roughly 96% of White women and 91% of Black that were actually ever-married at age 85+ in 2010 (reported here), for several reasons. First, I count dead people against the ever-married number (additionally, married people live longer, not necessarily because they’re married). Second, today’s 90+ year-olds mostly got married 70 years ago, when times were different; my estimates are a projection of nowadays.

A very interesting age pattern emerges here, which is relevant to the incarceration and “available men” question. If you look back at the figure, notice that the big difference in marriage opens up early — peaking at 28 points by age 33, before narrowing to 7 points at the end.The big difference in marriage is that White women marry earlier. In fact, as the next figure shows, after age 33 Black women are more likely to marry than are White women. I don’t think I knew that. Here are the number marrying at each age:

NHBW life tables 2010.xlsx

Specifically, although White women are twice as likely to marry in their mid-twenties, of our fictional 100,000 women stuck in 2010, just 15.6% of White women, compared with 36.8% of Black women end up marrying after age 33.

The other way of looking at this — and an answer to a common question about marriage rates — is to see the chances of marrying after a given age if you haven’t married yet. This figure shows, for example, that a White women who lives to age 45 without marrying has a 26% chance of someday marrying, compared with a whopping 49% for Black women.

NHBW life tables 2010.xlsx

It is surprising that Black women, with lower cumulative odds of marrying at every age in the cohort, are so much more likely to marry conditional on getting to their 40s without marrying. Maybe you’ve got a better interpretation of this, but this is mine. Black women are not against marriage, and they are not ineligible for marriage in some way (even though most of these single women are already mothers**). Rather, they have not married earlier because they couldn’t find someone to marry. That’s because of all the Black men who are themselves dead, incarcerated or unemployed (or scarred by those experiences in their past) — or married to someone else. So within their respective marriage markets (which remain very segregated), the 45-year-old single White woman is much more likely to be someone that either doesn’t want to marry or can’t marry for some reason, while the 45-year-old single Black woman is more active and eligible in the marriage market. This fits with the errors in the earlier predictions, which failed to pick up on the upward shift in marriage age for Black women — marriage delayed rather than foregone.

What do you think of that interpretation? If you have a better idea I’ll mention you at Princeton next week.

Note: I found so many mistakes as I was doing this that it seems impossible there are any more. Nevertheless, caveat emptor: This analysis hasn’t been peer reviewed yet, so consider it only as reliable the latest economist’s NBER paper you read about on the front page of the every newspaper and website on earth. (And if you’re a journalist feel free to refer to this as a new working paper.)

* Technical notes: I used death rates from 2010 (found here), and marriage rates from the five-year ACS file for 2008-2012 (which has 2010 as its midpoint), from I adjusted the death rates because never-married people are more likely to die than average (I told you this was depressing). I had to use a 2007 estimate of mortality by age and marital status for that (found here), which is not that precise because it was in 10-year increments, which I didn’t bother to smooth because they didn’t have much effect anyway. The details of how to do a multiple-decrement lifetable are nicely described (with a lot of math) by Sam Preston here (though if you really want to replicate this, note one of his formulas is missing a negative sign, so plan to spend an extra few days on it). To help, I’m sharing my spreadsheet here, which has the formulas. (Note that survival in the life table doesn’t refer to being alive, it refers to being both alive and never-married.) The mortality and marriage rates are for non-Hispanic women; the never-married adjustment is for all women. For the marriage rates I used all Black and White women regardless of what other races they also specified (very few are multiple-race when you exclude Hispanics).

** In 2010, 63% of never-married Black women who lived in their households had at least own of their own children living with them.


Filed under Me @ work