Monthly Archives: February 2023

Roland Fryer – His Life Story, His Work on Education and on Police Use of Force, and Harvard

Posted on by

Marie Curie, Nobel laureate:  “Nothing in life is to be feared, it is only to be understood.  Now is the time to understand more, so that we may fear less.”

Roland Fryer grew up in difficult circumstances; rose to the top of his profession; did fundamental work on education that was not, however, always welcomed by the education establishment; found (contrary to all expectations, including his own) that there was not an anti-Black racial difference in police shootings in the data but rather, if anything, the opposite; and was soon thereafter suspended without pay for two years from his tenured position as a full professor at Harvard, with the research lab that served as the vehicle for his work permanently closed.

[Note:  Sources and references used, with links, are all gathered at the end of this post.]

A.  Roland Fryer

Roland Fryer is Black.  He was born in 1977 in Daytona Beach, Florida, but moved to Lewisville, Texas (about 20 miles northwest of Dallas) when he was very young.  Fryer’s mother left him (perhaps more left his father) when he was still an infant, and they did not meet again until he was in his 20s.  His father had been a math teacher but then became a copier salesman, was extremely abusive when he was drunk (which was often), and beat him often (especially when he was drunk).  And when Fryer was 15, his father was convicted of a rape and sent to prison for eight years.

Fryer spent his summers with his grandmother in Daytona Beach, Florida, where much of his extended family still lived.  And he remained close to his grandmother for many years (she passed away just a few years ago).  She was a school teacher (of sixth graders), and clearly a major influence.  But his family in Florida had difficulties.  In an interview, he noted (when asked for a count) that 8 of his 10 closest family members either died young or spent time in jail.  This was also true for many of his friends.  When staying with his grandmother in Florida, he would also frequently visit an aunt, who ran a profitable crack-cocaine operation from her home.  One day he went to visit his aunt, but arrived later than normal.  When he did he saw the home surrounded by police.  Nearly everyone there was arrested, as he likely would have been had he arrived a bit earlier.

Fryer was certainly not separate from this.  He made money by selling marijuana as well as selling items he systematically shoplifted from stores.  Given the nature of his clientele, as well as the all-cash transactions involved, he carried a gun (a .357 Magnum).

He did not do well in his studies at school.  But he was active in athletics, and must have been quite good as he was awarded a basketball scholarship to play at the University of Texas at Arlington – an NCAA Division 1 school.  It was not one of the top basketball programs in the country (it only had 11 winning seasons out of 48 as of 2004, a few years after he was there), but with that scholarship he got into a decent college.  And as he later related in an interview, while he had to take the SAT in order to qualify for the basketball scholarship, he was only interested in getting a minimum score of at least the 700 required to participate in NCAA athletics.  He got that (I am not sure by what margin – he said he did not do well in part as he was still drunk from the night before).  But keep in mind that a score of 700 (out of a possible 1600, and where one cannot get less than 400) should not be all that difficult.  Based on the more recent scores for 2020 (which are easier to find than earlier scores, but should be similar enough), a score of 700 will place you at the 3rd percentile.  Note this is percentile – that is, the 3% mark, not 30%.

He attended UT-Arlington but in the end never played on the basketball team there.  His basketball skills might not have been at that needed for the college-level game.  But he enrolled in an economics class and loved it.  He was also very good at it, despite what must not have been very good preparation in high school.  But the professor teaching the class supported him, provided tutoring as needed, and with the support of a dean he was switched from his athletic scholarship to an academic one.  He excelled in his studies, and graduated Magna cum Laude in just 2 1/2 years – not the normal four.

From UT-Arlington he enrolled in the economics Ph.D. program at Penn State – a solid school with a number of good faculty members, although not an economics program one would rank as among the top 20, say, in the US.  He received his Ph.D. in 2002, but already from 2001 had obtained a post-doc position in the Department of Economics at the University of Chicago – one of the very top schools in the country for economics.  Jim Heckman (a Nobel laureate) was a sponsor and a major influence, and at Chicago Fryer co-authored papers with Steven Levitt (of Freakonomics fame, although that came later).

He began to author or co-author a series of influential papers (some of which will be discussed below), moved to Harvard in 2003, and began to accumulate a series of honors and awards.  At Harvard he entered as a Junior Fellow in the prestigious Harvard Society of Fellows (2003-06), became an Assistant Professor of Economics in 2006, and received tenure as a full Professor of Economics in 2007 at the age of 30 – the youngest African-American ever to receive tenure at Harvard and one of the youngest ever of any race.

As further evidence of Fryer’s work at the top levels as an economist, and the public recognition of it:

a)  As I write this, Fryer is listed as the author or co-author of 51 papers distributed by the National Bureau of Economic Research (NBER);

b)  In the most recent c.v. I have been able to find – which was as of February 2018 – Fryer listed 69 papers either published or as working papers in progress;

c)  He has had papers published in many of the top economics journals, including the AER, QJE, JPE, REStat, and JPubEcon, as well as numerous other peer-reviewed journals;

d)  He received a MacArthur Fellowship in 2011 (more commonly known as the MacArthur “Genius Awards”);

e)  He is a fellow of the American Academy of Arts and Sciences;

f)  He received the John Bates Clark Medal in 2015, which is awarded annually (prior to 2009 biennially) by the American Economic Association “to that American economist under the age of forty who is judged to have made the most significant contribution to economic thought and knowledge” (from the AEA website on the award).  Many of the recipients of this award later went on to win a Nobel Prize in Economics.  The first recipient, in 1947, was Paul Samuelson.

There have been numerous other awards as well, in addition to very substantial financial support that Fryer has received to enable his research from a number of foundations as well as through government grants.  A vehicle Fryer set up in 2008 for this work was the Harvard Education Innovation Lab (or EdLabs), which employed Harvard students (undergraduate and graduate), post-docs, and others (many, if not most, recent students).  At one point at least (and based on a photo in a video), there were close to 20 people working at EdLabs.  It was a major operation, with office space just off-campus.

Roland Fryer is certainly a solid economist.

B.  Work on Education

A focus of Fryer’s work, from the start, has been on education.  I will summarize some of that work here, as it is of interest in itself in addition to giving a sense of how Fryer approaches issues.

Among his earliest papers were several co-authored with Steven Levitt of Chicago on the Black-White test score gap – its magnitude, how it changes as kids age, and underlying causes.  They found that there was no gap at the very youngest ages (9 months – although I have no idea how one assesses intelligence or abilities at such an age), but that a substantial gap had opened up by the time the children had reached kindergarten age.  They found that just a few measures of social background (socio-economic status, birth weight, poverty status) could account for this fully.  That is, Blacks from a given social background scored the same as Whites from a similar social background at kindergarten age.  After that, however, the Black-White gap continued to widen, and it no longer could be fully accounted for by social background.  Something was happening as the kids entered into their school years, with what was happening in the schools themselves a possible cause.

At around this time, Fryer also looked at the issue of “Acting White” – work that he later became well known for.  While many politicians and others asserted it to be a fact and that it was holding back exceptional Black students, there was very little if any evidence.  Some cited surveys of students on whether they felt their popularity suffered if they got good grades, and the kids generally said no.  If true, then there was no such issue as “Acting White”.  But such self-reporting is not a reliable guide.

Fryer had the insight that one could instead use a very large survey (of 90,000 students in 175 schools) undertaken in 1994, that asked the students who their best friends were.  They could list up to five males and five females.  The insight was not simply to rely on who a student said his or her best friends were, but rather count only those who also named you as one of the best friends.  If it was not reciprocal, it did not count.

Once pointed out, it is a pretty obvious insight.  But no one had pointed it out before.

Fryer found that the popularity index values he constructed were pretty similar across races at the low grade point average (GPA) of a D.  It then grew steadily for Whites (i.e. the students scored higher on the popularity index as their GPAs went up), with no drop-off among those with the highest GPAs.  But while it also at first went up for Blacks (although not as steeply as for Whites), it then dropped off following a GPA of 3.5 (i.e. halfway between an A and a B average).  “Acting white” did seem to matter.  Black students with exceptional GPAs were less popular among their peers than those with a lower GPA.

Going a bit further, Fryer found that such a relationship did not exist for Blacks in private schools.  Nor did he find it in schools that were almost entirely Black (80% or more).  But, disconcertingly for those of us who believe in integration (as I do, and as I believe Fryer does), he found the penalty on popularity of the high-achieving Black students was much larger in the more integrated schools.  That is a problem for society.  And Fryer did not try to bury the result.  While disconcerting, it came out of the data.

At some time around 2006, Fryer became involved with New York City schools.  Joel Klein (appointed by then Mayor Michael Bloomberg) was the school chancellor and trying many things.  Some of the first work Fryer did was to examine whether one could have an impact via financial incentives.  He was a frequent visitor to a school in the Bronx, and on his own at some point bought pizzas for a class to celebrate the achievement of some goal.  This was repeated a few times, but he then wanted to look more systematically at whether rewards of some sort could make a difference.

His work built up by steps from there, as reformers in other school districts sought to have various proposals tested.  With financial support from foundations, Fryer developed a series of experiments in several school districts across the country to see whether financial incentives would have an impact on student achievement.  In the end, randomized control trials were organized in five cities (New York, Chicago, Dallas, Houston, and Washington, DC), at a total cost of just $9.4 million, and with 36,000 students (18,000 receiving the financial awards and 18,000 in the control groups) in 250 schools.  The schools chosen were inner-city schools of predominantly Black and Hispanic students, with high concentrations of students whose low family income qualified them for free lunches.  Three of the school systems were headed at the time by nationally-prominent school reformers (Joel Klein in New York, Michelle Rhee in Washington, DC, and Arne Duncan – later Obama’s Secretary of Education – in Chicago).

The work was controversial, and the education establishment did not like it.  Some argued that paying financial incentives to students would “destroy their love of learning”.  But there was no evidence of this in the results.  It was also noted that it would not be much different from the practice among many middle-class parents of rewarding their kids financially if they got good grades, or did their homework on time, or read certain books, and so on.

Fryer’s findings are well worth reading (see sources cited at the end of this note), but a brief summary of the main result is that financial incentives to students on the side of inputs (do your homework, read books, attend school as you should and behave well when there) had positive effects, although sometimes with marginal statistical significance.  The positive effects in just one year relative to the control groups (based on regressions controlling for various factors) were equivalent to accelerating student progress by the equivalent of one to two months of school.  While that in itself would not close Black-White education gaps, the rate of return given the low cost was enormous.  They paid 2nd graders in Dallas just $2 for each book they read, and 5th graders in Houston just $2 for each math homework assignment completed.  But such small financial rewards led to gains of one to two months of schooling (as a statistical best estimate) relative to the control groups.

Note also that the effects found in months of schooling gained are averaged over all who were in the group offered the financial incentives -whether they took up the offers fully, took them up only partially, or did not take them up at all.  They are therefore estimating the compounding of two effects:  whether the students will undertake the activity in response to the incentive, and if they do, whether this activity will then lead to a gain in an educational outcome.  One wants to measure the two together as even if some activity might lead to fabulous gains in terms of student outcomes, it will not matter if no one takes it up in response to the incentive.

In contrast to the benefits seen on the input side, financial rewards on the output side (i.e. tied to grades received on their report cards, or scores on tests) did not have a positive effect.  Indeed, while not at all close to statistical significance, the estimated coefficients were slightly negative in most of the cases.

Fryer also examined a New York City program that provided financial incentives to teachers.  He found that they did not matter.  This might, however, have been a consequence of how the specific program was designed (working with the agreement of the teacher’s union).  The bonuses were paid in some amount per teacher, but then as a total sum to the schools meeting certain criteria on student gains.  The schools would then decide how to divide that pot of money internally.  Individual teachers did not receive a financial incentive based on the gains of their specific students.

Financial incentives to students could be considered the demand side of education.  There is also the supply side – the schools themselves and how they teach.  As just noted, financial incentives to teachers (at least in the program examined) did not appear to matter.  But Fryer also took a close look at what was being done at several chartered schools run as part of the Harlem Children’s Zone project.  Those schools have been highly successful.  Fryer looked at this systematically by comparing the outcomes of students who won the lottery to get into the over-subscribed schools and the otherwise similar students who lost in that lottery.

The schools were amazing.  Fryer found that students starting in middle school (6th grade), while far behind White peers, were able in three years (i.e. by 8th grade) when attending the Harlem Children’s Zone schools to fully close the gap in math test scores and close about half of the gap in reading scores.  And children enrolled in their elementary schools were able to fully close the gap in both scores by their 3rd grade.  Those who had applied to try to enter the Harlem Children’s Zone schools but had not won the lottery performed similarly to others in the city system.  That is, they did not do well.  And six years later, those who won the lottery to get into these schools compared to those who had not were more likely to have entered college, less likely (if female) to have been pregnant, and less likely to have been incarcerated.

The question then was whether the model of the schools being run by the Harlem Children’s Zone could be replicated elsewhere.  Fryer contacted a large number of school districts around the country, but none were interested in working with Fryer on trying to copy the model until the school district for Houston expressed a willingness.  The State of Texas was preparing to take over control of a number of Houston’s schools that were deemed to be failing, so as a last alternative Houston was willing to allow the group Fryer had organized to take over management of the failing schools instead.  They were a mix of twenty elementary, middle, and high schools.

Major changes were made:  The school day (for middle and high schools) was lengthened by an hour plus summer vacation was shortened by two weeks – this increased the number of instructional hours by about 20%; they replaced 19 of 20 principals and close to half of the teachers; added tutoring for students who were lagging (in just the 4th, 6th, and 9th grades due to funding limitations); used frequent tests to assess progress during the school year and to decide if certain lessons should be repeated; and inculcated a culture of high expectations.  Fryer calculated the extra financial cost per student of all these measures was about $1,800, or an extra 15%.

The results were largely similar to those found at the Harlem Children’s Zone for math, but less for reading.  Gains in math for students in the elementary schools were highly significant statistically and such that the racial gap was eliminated in three years.  The results on reading were also positive but less and not always statistically significant.  The gains in math for the students in the middle and high schools were almost as large, but there were almost no changes in the reading scores compared to the controls.

And an anecdote:  Fryer related in an interview that one time when teaching his undergraduate class at Harvard, one of the students came up to him after the class.  The student told him that he had gone to one of the high schools in Houston that Fryer had worked with, and wanted to say thanks.  Fryer had not been aware of this until that moment.  It was a failing school, that the State of Texas was getting ready to take over due to its poor record, and now a graduate from that school had been admitted to Harvard.

Despite these highly positive gains, school districts around the US have largely ignored this demonstration of effective schooling.  One suspects that the fact 19 of 20 school principals were replaced, as well as close to half the teachers, may be a factor.

Fryer’s work on education is revealing.  It shows Fryer is hands-on, wishes to work out practical approaches, follows the data, and is both careful and able to use rigorous statistical approaches to examine results of real-world significance.  He has followed the same approach in looking at the issue of police use of force.

C.  Racial Differences in Police Use of Force

With the increased attention being given to police shootings in the 2010s following a number of prominent cases, Fryer initiated a study to determine the degree to which there were differences across races in police shootings in the US that might reflect racial biases.  It was assumed there were – the question was by how much.

The problem is that there was very little data.  It was well known that databases kept by the FBI were far from adequate as reporting shootings to the FBI is voluntary and many police departments choose not to.  More fundamentally, reports that are just tallies of numbers do not tell much as they do not provide data on the circumstances of the shootings.

Some police departments do keep such records, but often they are just on paper and not in a machine readable form.  After contacting police departments around the country, Fryer was able to obtain data on incidents where officer weapons were discharged from 10 large police departments (the cities of Austin, Dallas, and Houston; Los Angeles county; and 6 large Florida counties), each for some subset of the years between 2000 and 2015.  They got 1,316 cases and could code for 65 different variables, but the data included only incidents when an officer’s weapon was discharged.  Hence that data could not be used to address whether there were similar incidents when an officer’s weapon might have been fired but was not.

For this, Fryer was able to obtain detailed data from the City of Houston.  They were provided access to the written files – which could vary between 2 pages and 100 pages for each incident.  From the files for incidents of serious crimes where lethal force might have been justified (defined as attempted murder of a police officer, aggravated assault of a police officer, and resisting, evading, or interfering in an arrest), they combined the incidents where an officer in fact fired a weapon with a random sample of other cases where they did not.  They ended up with 1,532 cases, and had their researchers read the files and code for 300 variables.  To help ensure accuracy, they had each file read by two researchers, acting independently, and then checked whether both coded the same for the given variables given their reading of the files.

Fryer’s key, and surprising, finding was:

“we find, after controlling for suspect demographics, officer demographics, encounter characteristics, suspect weapon and year fixed effects, that blacks are 27.4 percent less likely to be shot at by police relative to non-black, non-Hispanics” (emphasis in original).

Similarly, with the data from the 10 large police departments (including Houston, but also the nine others they were able to collect data from), Fryer was able to examine the impact of race on whether the officer or the suspect fired their weapon first.  Taking all factors into account, he found in his regression analysis that an officer firing first in the encounter was 46% less likely when the suspect was Black.  And 44% less likely if the suspect was Hispanic.

There were important limitations, of course.  For starters, the data was only for Houston (for the first analysis) and only for Houston and nine other major cities/counties (for the second analysis).  The results might well be different elsewhere.  They also had to rely on case records as recorded by the officers involved.  That could add a particular slant.  More fundamentally, the results are conditional on instances where there was some form of interaction with the police (within a specified type of serious cases examined).

For these reasons, further work on police use of force to confirm (or refute) Fryer’s findings is certainly warranted.  But for reasons that will become clear in the next section below, many academics who would otherwise research this further will likely, for understandable reasons, choose to stay away.

While there are these limitations, this is the only statistically rigorous study I am aware of that has examined such police use of force.  Furthermore, in a separate short paper, Fryer looked at how the data sets he constructed on officer involved shootings compared to those others have assembled.  That data is unfortunately limited, but Fryer identified four sets:  1)  Data assembled by The Washington Post on officer involved fatal shootings and the victims; 2)  Similar data on fatal shootings assembled by The Guardian; 3)  Data assembled by VICE News from 47 of the 50 largest police departments in the US, that includes both fatal and non-fatal shootings; and 4)  Data assembled by Cody Ross, then at University of California – Davis, as the sole academic analysis he could find. 

Fryer compared the measures of bias that each of these four data sets presented to what that measure would be for his own data from the 10 large police departments (the 10 cities and counties).  The results across the resulting five data sets were mostly similar.  For example, the Washington Post reported the percentage of Black civilians among the unarmed men killed by police.  This was 40% in the Post data, and varied between 34% and 53% in the others.  In the Fryer data it was 44%.  All of these are, of course, far higher than the Black share of the population (which is 12%).  And similarly for the measures used by each of the others.

Fryer’s data therefore does not appear to be out of line with what others have found.  But what is different is that he was able, with his much richer set of observations on other factors, to take into account in his statistical analysis the impact of these other factors and hence could isolate the role of race.  And when he did, he found no evidence of racial differences.  Indeed, in the data he was able to assemble the racial differences went the other way from what was expected.  As Fryer later described it (in a New York Times article on the research), it was “the most surprising result of my career”.

Fryer’s results were certainly controversial.  As he recounted later, many friends advised him to bury his findings and not publish them.  But he insisted he would, and his paper appeared in the Journal of Political Economy in June 2019 – one of the top economics journals in the US.

D.  Consequences

The first draft of his paper on the police use of force appeared in July 2016; a revised draft is dated January 2018; and it was published in June 2019.  During this time, in mid-2017, Fryer decided that his personal assistant at Harvard EdLabs would be let go as she was just not able to perform the job adequately.  A severance package was negotiated and agreed upon.  But an administrator at Harvard decided it should be reduced by $25,000 (I have never seen from what to what).  Had that administrator not insisted on this, the matter would have ended there.

The staff member was upset with this $25,000 reduction.  A short time later, she had lawyers file a case against Roland Fryer (as well as Harvard) claiming there had been sexual harassment at EdLabs.  In response, the Harvard office responsible for examining such claims launched an investigation.

The filing prepared by the lawyers for the assistant listed 38 complaints.  The Harvard office investigating the charges dismissed 6 immediately, leaving 32.  Of those 32, they then concluded following their investigation that a further 26 should be dismissed.  Several included what they were able to show were fabrications by the assistant – that is, they could demonstrate the assistant had lied.  That left 6 complaints.  They concluded that these could be valid, even though in some they took the assertion of the assistant over the denial of Fryer (despite knowing that the assistant had fabricated a number of the other complaints).

From what has come out (these are all based on leaks, as the documents have not been officially released) the findings were that some of the banter in the office was considered inappropriate due to sexual innuendo.  One example was Fryer complaining at some point about some Harvard administrator, and saying that the administrator was acting as if they “had not had sex since Blacks were slaves”.  The other instances were similar.

In no case was there an assertion that Fryer had made an improper advance on any individual.  Rather, there was banter in the office that some later said they considered to be inappropriate.  Only one complaint was filed (by the personal assistant let go), but during their investigation the Harvard investigators were able to find three other women (including one who had been let go in 2008, the year EdLabs opened) who were willing to say that some of what Roland had said when they worked there they now considered to be inappropriate.  The number of staff who had worked at EdLabs by that point were in the dozens – overwhelmingly strong supporters of Fryer and glad that they had worked with him, and mostly women.

That there had been such banter was not denied.  Indeed, it was open and not a secret, although apparently not common.  And while it should not be belittled if such talk was seen by anyone in the office as inappropriate, there had been no such complaints made in the ten years since EdLabs had opened until the personal assistant filed her complaint after being fired (and a Harvard administrator decided that the negotiated settlement package should be reduced by $25,000).  During those ten years, no one had ever told Fryer something along the lines of “it is inappropriate to say that and it makes me uncomfortable”.  Had someone said that, including by any of the four who later told the Harvard investigators that they now considered some of the remarks to be inappropriate, the Harvard investigators would presumably have highlighted it.  The investigators certainly would have asked all the EdLabs staff they interviewed whether they had ever said themselves to Fryer, or ever heard someone else say to Fryer, that they considered some remark in the banter to be inappropriate.  And had someone told the investigators that, the investigators would certainly have emphasized in their conclusions that Fryer continued with such banter even after some member of the EdLabs staff had told him that it made them uncomfortable.  But there was evidently no such finding.

The conclusion of the Harvard office investigating the issue was that Fryer should be required to take sensitivity training.  And that was it.  Nothing else.

But that was then deemed insufficient by the then Dean of the Faculty.  And in her position as dean she would be the one making the final decision on the penalty, if any.  She decided to establish a special committee to advise.  That committee was then made up of members she personally chose, and she included herself on it.

Following that review, the dean then decided not only that Fryer would be suspended without pay for two years – with no contact allowed with students or others at Harvard (so students being supervised would need to switch to someone else) – but also that EdLabs would be closed, and closed permanently.  All EdLabs staff would be dismissed and all its research programs ended.

She also recommended to the then president of Harvard – Lawrence Bacow – that Fryer’s tenure as a university professor be revoked.  Bacow, however, declined to make such a recommendation to the Harvard Corporation (the trustees of Harvard), who would have had the final say.  No professor at Harvard has had their tenure revoked in more than a century.

Needless to say, many of those involved with EdLabs and Fryer’s research were shocked.  It was not only that Fryer was being punished severely (where even the Harvard office charged with investigating such issues only recommended sensitivity training to address the complaint), but all of EdLabs was in effect being punished as well.  By her edict, the valuable and important work of the lab – with most of it still centered on the education of minority groups – was being stopped.

This was, however, extremely convenient for those who approached racial issues in a different way, with different conclusions reached.  That included the Dean of the Faculty herself, whose own research focused on racial issues (with an appointment in the Department of Government and jointly in the Department of African and African-American Studies).  If Fryer’s findings were correct, then aspects of her own work would be undermined.

She was born in New York to Haitian immigrants, but had a comfortable upbringing.  It could not be more different from Fryer’s.  Her father was an engineer, and the family spent several years of her childhood in Saudi Arabia when her father was posted there.  She attended Phillips Exeter Academy – perhaps the most elite prep school in the US – and then received her BA from Stanford and Ph.D. from Harvard.  She taught first at Stanford and then returned to Harvard as a full professor.  She was later appointed Dean of Social Science in 2015, and Dean of the Faculty of Arts and Sciences in 2018.

That Dean of the Faculty was Claudine Gay.  And on December 15, 2022, Harvard announced that Claudine Gay had been selected to be the next president of Harvard, with her appointment to start on July 1, 2023.

E.  Conclusion

Glenn Loury, a Professor of Economics at Brown University and a co-author of several papers with Fryer early in Fryer’s career, summed it up well:

To do the kind of work Roland does, you have to be more than brilliant. You have to be fearless. And I cannot help but suspect that now Roland is paying the price for pursuing the truth wherever it leads.  Several years ago, he was accused of sexual harassment by a disgruntled ex-assistant.  In my opinion and that of many others, those accusations are baseless.  But Harvard has used them as a pretext to shut down Roland’s lab, to curtail his teaching, and to marginalize him within the institution.

I’ll not mince words.  Those at Harvard responsible for this state of affairs should be utterly ashamed of themselves.  They have unnecessarily, heedlessly tarnished the career of an historically great economist.  Again, I can’t help but suspect that they have effectively buried vital research not because it was poorly done but because they found the results to be politically inconvenient. “Veritas” indeed.

Postscript

As of July 2021, Roland Fryer was allowed by Harvard to return to teaching – at both the graduate and undergraduate levels.  However, he would not be allowed to supervise students or participate in an advisory role for a further two years.

 

Sources and References

Links are provided to readily accessible sources, when possible.  Articles published in The New York Times are unfortunately only available behind a paywall.  For academic articles, the link will generally be to the version of the paper that appeared on the NBER site, as these are openly available online while subscriber access is generally required for the final version that appeared in the journals.  The articles as published are normally identical to the final NBER versions, or at least are very close.

a)  Biography and Life Story:

The first article below is a profile of Roland Fryer that appeared in The New York Times Magazine in March 2005, written by Stephen Dubner.  Dubner had heard of Fryer from Steven Levitt, with whom he co-authored Freakonomics (that appeared also in 2005).  This profile by Dubner did much to make Fryer a known figure outside of the economics profession.  Much of the article is on Fryer’s upbringing.

Stephen Dubner, “Towards a Unified Theory of Black America”, The New York Times Magazine, March 20, 2005.

Other sources on Fryer’s life and work:

Announcement of the award in 2015 to Roland Fryer of the AEA’s John Bates Clark Medal (with substantial material on Fryer’s research).

Roland G. Fryer biography on JRank.

b)  Curriculum Vitae and Publications:

This cv dated February 2018 is the most recent I have been able to find.  The academic papers on the list of Fryer’s NBER papers (51 as I write this) can be accessed in full from this page in most and possibly all cases (at least for the papers I have tried to look up thus far):

Roland Gerhard Fryer, Jr. “Curriculum Vitae”, February 2018.

National Bureau of Economic Research (NBER), Roland G. Fryer page.

c)  Work on the Black-White Test Score Gap, Co-Authored with Steven Levitt:

The first article is for a more general audience, and the following two the academic papers.

“Falling Behind” (with S. Levitt). Education Next 4 (August 2004): 64-71.

“Understanding the Black-White Test Score Gap in the First Two Years of School” (with S. Levitt). The Review of Economics and Statistics 86, no. 2 (May 2004): 447-464.

“The Black-White Test Score Gap Through Third Grade” (with S. Levitt). American Law and Economic Review 8, no. 2 (July 2006): 249-281.

d)  Work on “Acting White”:

The first article is for a more general audience, and the second the academic article.

“Acting White”. Education Next, Winter 2006.

“An Empirical Analysis of ‘Acting White’” (with P. Torelli).  Journal of Public Economics, 94, no. 5 (2010): 380-396.

e) Work on Financial Incentives for Students and for Teachers:

The first article is again for a more general audience, and the others for academics. 

“The Power and Pitfalls of Education Incentives” (with B. Allan), Discussion Paper 2011-07, The Hamilton Project, September 2011.

“Financial Incentives and Student Achievement: Evidence from Randomized Trials,” Quarterly Journal of Economics, 126, no. 4 (2011): 1755-1798.

“Teacher Incentives and Student Achievement: Evidence from New York City Public Schools,” Journal of Labor Economics, 31, no.2 (2013): 373-427.

f)  Work on the Chartered Schools Managed by the Harlem Children’s Zone:

“Learning from the Successes and Failures of Charter Schools,” Policy Brief 2012-06, The Hamilton Project, September 2012.

“Are High Quality Schools Enough to Increase Achievement Among the Poor?” Evidence from the Harlem Children’s Zone” (with W. Dobbie). American Economic Journal: Applied Economics 3, no. 3 (2011): 158-187.

“The Medium-Term Impacts of High-Achieving Charter Schools,” (with W. Dobbie),  Journal of Political Economy, vol 123, No. 5, October 2015.

g)  Results from New Management of a Group of Low-Performing Public Schools in Houston:

“Injecting Charter School Best Practices into Traditional Public Schools: Evidence from Field Experiments,” Quarterly Journal of Economics, August 2014, vol. 129, issue 3: 1355-1407.

h)  Work on Police Use of Force:

Link to the final NBER version of the paper:

“An Empirical Analysis of Racial Differences in Police Use of Force”, NBER Working Paper #22399, revised draft, January 2018.

Link to a pdf file of the paper as it appeared in the Journal of Political Economy:

“An Empirical Analysis of Racial Differences in Police Use of Force”, Journal of Political Economy, June 2019, vol 127, no. 3: 1210-1261.

The paper comparing statistics on racial differences in police shootings in the Fryer data to what they would be when defined as in four published sources:

“Reconciling Results on Racial Differences in Police Shootings”, AEA Papers and Proceedings, May 2018, vol. 108: 228-233.

i)  The Harvard Investigation:

  1)  The New York Times article reporting on the conclusions of the Harvard investigation of Roland Fryer:

Jim Tankersley and Ben Casselman, “Star Economist at Harvard Faces Sexual Harassment Complaints”, The New York Times, December 14, 2018.

  2)  Roland Fryer, letter to the New York Times in response:

Roland Fryer, “At a Harvard Lab, the Accused Responds”, The New York Times, December 20, 2018.

  3)  Article of Stuart Taylor, Jr., on the Harvard investigation and the bias in the New York Times report on it:

Stuart Taylor Jr., “Harvard, the NY Times and the #MeToo Takedown of a Black Academic Star”, Real Clear Investigations, January 29, 2019.

  4)  New York Times article reporting on Harvard’s decision to suspend Fryer and permanently close EdLabs:

Ben Casselman and Jim Tankersley, “Harvard Suspends Roland Fryer, Star Economist, After Sexual Harassment Claims”, The New York Times, July 10, 2019.

  5)  Glenn Loury’s assessment of the saga:

Glenn Loury, “The Truth about Roland Fryer”, March 13, 2022.

The Unemployment Rate, the Growth in Employment, and Productivity

Posted on by

A.  Introduction

The January jobs report (more properly the “Employment Situation” report) released by the Bureau of Labor Statistics (BLS) on February 3, was extraordinarily – and surprisingly – strong.  The unemployment rate fell to 3.4% – the lowest it has been since May 1969 more than a half-century ago.  And despite the low unemployment rate, the number of “new jobs created” (also a misnomer – it is actually the net increase in non-farm payroll employment) was a surprising 517,000.  But it was not only this.  The regular annual revisions undertaken each January to reflect revised population controls and weights for the employment estimates led this year to significantly higher labor force and employment estimates.  With the new industry weights, the increase in the estimated number of those employed in 2022 (the number of `”new jobs”) rose to 4.8 million.  The earlier estimate had been 4.5 million.

All this is an extraordinarily strong jobs report.  However, one should not go too far.  It is important to understand what lies behind these estimates, as well as some of the implications.  For example, strong growth in the total number employed while GDP growth is more modest implies that productivity (GDP per person employed) went down.  That could be a concern, except that when viewed in the context of the last several years we will see that productivity growth has in fact been rather good.

This post will first examine the new figures on unemployment and then on employment growth.  We will then look at the change in productivity – both in the recent past and from a longer-term perspective.

B.  The Unemployment Rate and Its (Non)-Impact on Inflation

The unemployment rate in January fell to 3.4%.  This is the lowest it has been since May 1969.  And if it falls a notch further to 3.3% in some upcoming month, it will have fallen to the lowest since 1953.

A 3.4% unemployment rate is certainly low.  But what is more significant is that the unemployment rate has been almost as low for most of the past year.  It fell to just 3.6% in March 2022, and until last month varied within the narrow range of 3.5 to 3.7% – hitting the 3.5% rate several times.  It is now at 3.4%, but what is most significant is that it has been at 3.7% or less for almost a year.

One needs to recognize that the unemployment rate is derived from a survey of a sample of households (implemented by the Census Bureau) called the Current Population Survey (CPS).  The CPS sample includes approximately 60,000 households each month, in a rotating panel, and from this they derive estimates on the labor force participation rate, the unemployment rate, and much more.  It complements the Current Employment Statistics (CES) survey, which covers a much larger sample of 122,000 businesses and government agencies representing 666,000 individual worksites (with each employing many workers).  Hence employment figures are generally taken from the CES as there will be less statistical noise.  But the employers surveyed for the CES cannot know how many workers are unemployed (they will only know how many workers are employed by them), so the smaller CPS needs to be used for that.  (A brief explanation of the CPS and CES is provided by the BLS as a “Technical Note” included in each of the monthly Employment Situation reports.)

Due to the size of the sample, the estimated unemployment rate is actually only known within an error limit of +/- 0.2 percentage points, using a 90% confidence interval.  That is, simply due to the statistical noise a change in the unemployment rate of 0.1 percentage point from one month to the next should not be considered statistically significant, and 10% of the time even a 0.2 percentage point change may have just been a consequence of the statistical variation.  However, repeated observations over several months in a row of an unemployment rate at some level will be a measurement one can have much more confidence in.  That can no longer be a consequence of simply statistical noise.  Thus one should not place too much weight on the January change in the unemployment rate to 3.4% from 3.5% the month before.  But the fact that the unemployment rate has consistently been within the relatively narrow – and extremely low – range of 3.4 to 3.7% since March 2022 is highly significant.

An unemployment rate anywhere close to a range of 3.4 to 3.7% is also far below the rate at which economists used to believe would be possible without the rate of inflation accelerating – i.e. without inflation going higher and higher.  This was given the acronym name of “NAIRU” (for Non-Accelerating Inflation Rate of Unemployment).  It was held that at an unemployment rate of less than the NAIRU rate, the rate of inflation would rise from whatever pace it was at to something higher.  This was viewed as unsustainable, and hence the proper goal of economic policy was, in this view, to manage macro conditions so that the unemployment rate would never fall below the NAIRU rate.  That rate was also sometimes called the “full employment rate of unemployment”.

The question then is what the NAIRU rate might be.  While different economists came up with different estimates, estimates generally fell within the range of 5 to 6%.  An unemployment rate of less than this would then (under this theory) lead to a rise in inflation.

But that did not happen.  The unemployment rate fell to below 5% in 2016, and inflation remained low.  It fell to below 4% in 2018 and inflation remained low.  It fell to 3.5% in 2019 and into early 2020 and inflation remained low.

With the once again very strong labor market – with unemployment hitting 3.4% – has this now changed?  The rate of inflation did rise in 2021 and into 2022.  But if one looks at this chart, one sees that the timing is wrong:  Inflation rose earlier – in 2021 – when the unemployment rate was still well over 6% early in the year.  Furthermore, nominal wages only rose later:

Inflation (measured here by the consumer price index – the CPI – for all goods and services) can be volatile, but the upward trend began already in the second half of 2020 (although in part this was initially due to a recovery in prices from depressed levels earlier in 2020 due to the Covid crisis).  The chart shows the rates in terms of 3-month rolling averages (at annual equivalent rates and in arrears, so the figure for a January, say, would be for the months of November through January).  The pace of change in nominal wages (also as 3-month rolling averages and at annual rates) did not start to rise until mid-2021.  The increase in nominal wages appears to be more in response to the prior increase in prices – as firms found it profitable to employ more workers in an economy that grew strongly in 2021 – rather than a cause of those higher prices.  This is consistent with the view that the inflation was primarily due to demand-pull, rather than cost-push, factors.

[Technical Note:  The figures on changes in the nominal wage come from data assembled by the Federal Reserve Bank of Atlanta, drawing on data that can be obtained in the underlying micro-data files of the CPS.  The rotating panel of households in the CPS are interviewed for four months, not interviewed for the next eight months, and then interviewed again for four months.  New households are added each month and then removed after month 16 for them.  This allows the researchers to match individuals with their reported wages to what they had earned 12 months before.  It also allows them to examine the wage changes broken down by individual characteristics – such as age, gender, race, education level, occupation, where they are in the income distribution, and more – as these are all recorded in the CPS.  It is all very interesting, and worth visiting their website where they make it easy to see the impact on the measured changes in wages of many of these different factors.

The matching of wage changes by individuals also provides a much more reliable index than the commonly cited changes in average wages provided in the monthly Employment Situation report.  The latter comes from what employers report in the CES survey on the average wages they are paying.  Those averages will be affected by compositional effects.  For example, the reported average wages will often jump at the start of an economic downturn – such as it did in 2020 – as the less experienced and lower-wage workers are generally laid off first.  This leaves a greater share of more highly paid workers, which will lead the reported average wage to rise even though the economy had entered into a downturn.]

Not only did the rise in inflation precede the more modest increase in the pace at which nominal wages rose, but since mid-2022 the rate of inflation has come down while the job market has, if anything, become tighter.  The unemployment rate, as noted above, has been in the 3.4 to 3.7% range since March 2022, and is now at 3.4%.  Despite this, the three-month average increase in the seasonally adjusted CPI fell from 11.0% (at an annual rate) in the three months ending in June 2022, to just 1.8% in the three months ending in December.  If a tight labor market was driving inflation, one would have expected inflation to have kept going up rather than fall – and certainly not to fall by such a degree.

Furthermore, growth in nominal wages fell slightly from a peak of over 6.7% in the three months ending in June and also July 2022 (at an annual rate), to 6.1% as of December.  One would have expected the pace of change in wages to have continued to go up, rather than start to ease.

It is still early to be definitive on any of this.  Trends could change again.  Importantly, a significant part of the sharp fall in inflation in the second half of 2022 (when measured by the full CPI) was due to a fall in the prices of oil and other energy products.  However, while more recent, there are also early indications that core inflation (where food and energy prices are left out) is also falling.  In terms of the core CPI (again the seasonally adjusted index), the pace of inflation fell from a peak of 7.9% (at an annual rate) in the three months ending in June 2022, to just 3.1% in the three months ending in December.

That measure of inflation – the core CPI, which is often taken to be a better measure of underlying inflationary trends than the overall CPI as food and energy prices are volatile and go down as well as up – is now falling despite unemployment at the lowest rate it has been in more than a half-century.  If a tight labor market was driving inflation, then one would expect the pace of inflation to be rising, not falling.

C.  Employment Growth

The January jobs report was also noteworthy for its figures on employment growth.  Nonfarm payroll employment rose by 517,000 – far higher than most expected.  It is not that an increase in employment of a half million in a month is unprecedented.  It is rather that there was such an increase even though the unemployment rate was already at an extremely low 3.5% in the prior month.  (And while nonfarm payroll employment excludes those working in agriculture, that number is now small at only 1.4% of the labor force – based on estimates from the CPS and including those in agriculture who are self-employed.  It also excludes the self-employed outside of agriculture – a more substantial 5.6% of the labor force according to the CPS – but still not that large.  In terms of changes in the numbers from one period to the next, the impact on the employment estimates will be small.)

In addition, the January report also reflected revisions – undertaken every January – where new weights are used to generalize from what is found in the sample in the CES of firms and other entities (such as government agencies) that employ workers to what is estimated for the economy as a whole.  The re-weighting is based on a comprehensive count of payroll jobs in March of the year, with this then used to revise the estimates for all of the year (2022 in this case).

Due to the new weights, the increase in the number of jobs in the economy rose from the earlier estimate of 4.5 million in 2022 (i.e. from December 2021 to December 2022) to 4.8 million.  Between January 2022 and January 2023 the increase was an estimated 5.0 million additional jobs.  That is, between January 2022 and January 2023, the number employed increased by an average of 414,000 per month.

The 4.8 million growth in the number employed in 2022 was remarkable not only because it is a big number, but also because it came after the even stronger growth in employment in 2021.  Employment grew by 7.3 million in 2021.  In absolute terms, the 4.8 million figure in 2022 is higher than that of any year (other than 2021) in the statistics going back to when they started to be collected in the present form in 1939 (using BLS data).  Such a comparison is more than a bit unfair, of course, as the US economy has been growing and there are far more people employed now than decades ago.  But taking 2021 and 2022 together, the percentage growth over the two years – at 8.5% – was exceeded since 1951 only by greater increases in 1977-78 (10.2%), in 1965-66 (9.7%), and in 1964-65 (8.7% – that is, there was strong growth in the three straight years of 1964, 1965, and 1966).  Joe Biden was right when he said job growth in the first two years of his presidency (of 12.1 million) was greater than that of any other president, but it is not really a fair comparison as the economy is now larger.  But even in percentage terms, his record is excellent.

But such growth in the number employed cannot continue much longer.  To put this in perspective, the total adult population in the US (as reflected in the CPS, and with the new population controls) rose by only 1.8 million between January 2022 and January 2023, or 150,000 per month on average.  And the labor force figure, as estimated in the CPS, grew by only 1.3 million over that period, or 111,000 per month.  One cannot keep adding 414,000 per month to the number employed (as we saw in the year to January 2022) when the labor force is only growing by 111,000 per month, when the unemployment rate is already at a historical low of 3.4%.

[Note that one cannot simply subtract the January 2022 figures reported from the new January 2023 figures, since in the CPS they do not go back and revise the previous year figures to reflect the new population controls.  But they do show what the impact would have been on the December 2022 figures, and I assumed that they would have had the same impact on the January 2023 numbers.  The impacts should be similar.  One can then do the subtractions on a consistent basis.]

An increase in the number employed of an estimated 414,000 per month when the labor force was growing by only an estimated 111,000 per month was possible in 2022 in part because the unemployment rate came down (from 4.0% in January 2022 to 3.4% in January 2023), and in part because the labor force participation rate went up slightly (from 62.2% in January 2022 to 62.4% in January 2023).

But also a factor is that these are surveys from two different sources (households for the CPS and firms and other employers for the CES), and the sample estimates will not always be fully consistent with each other.  As was discussed in an earlier post on this blog, the estimates can differ from each other sometimes for significant periods of time.  However and importantly, over the long term the two estimates will eventually have to approach each other.  The population estimates used for the CPS will yield (for a given labor force participation rate) figures on the labor force, and hence growth in the adult population will yield figures on growth in the labor force.  For a given unemployment rate, the number employed – within the bounds of the statistical estimates – cannot grow faster than this.

With the unemployment rate now at 3.4%, one should not expect much if any further fall.  Indeed, the general expectation (and the more or less openly stated hope of the Fed) is that it will start to rise.  It is possible that the labor force participation rate will rise, but changes in this are generally pretty slow, driven mostly by demographics and social factors (the share of people aging into the normal age of retirement; the share of the young entering into the labor force given their decisions on whether and for how long to enroll in colleges and universities; decisions by households on whether one or both spouses will work; and similarly).

While there will be uncertainty in what will happen to the unemployment rate and the labor force participation rate, for given levels of each of these, employment cannot grow any faster than the labor force does.  (Indeed it is slightly less:  At an unemployment rate of 3.4%, employment will only grow at 96.6% of what the labor force grows by.)  With the labor force growing by 111,000 per month in the year ending in January 2023 (with this already reflecting a small increase in the labor force participation rate from 62.2% to 62.4%), it will not be possible for the monthly increase in employment to grow by much more than this.

Looking forward, one should not, therefore, expect growth in the number employed to be sustained at a level that is anywhere close to the 517,000 we had in January.  There will be month to month fluctuations, but one should not expect an average increase over several months that would be much in excess of the 111,000 figure for the growth in the labor force seen in the year ending in January 2023.

D.  Productivity

Politicians like strong job growth.  It is indeed popular.  But the flip side of this is that while the number employed grew rapidly in 2021 (by 3.2% December to December), GDP growth was less (1.0% from the fourth quarter of 2021 to the fourth quarter of 2022, based on the most recent estimates).  With the number employed growing faster than GDP, the mathematical consequence is that GDP per person employed went down.  That is:  Productivity fell in the year.

Higher productivity is ultimately what allows for higher living standards.  Falling productivity would thus be a problem.  However, in the context of the last several years, productivity growth has in fact been pretty good:

We are once again seeing the consequences of the highly unusual circumstances surrounding the Covid crisis.  With the onset of a downturn, firms will lay off workers.  But they may often lay off more workers than their output falls.  This might be because of uncertainty on how much the demand for whatever they make will fall in the downturn (and they will wish to be careful and if anything to overcompensate, given the difficulty of obtaining finance in a downturn and the very real possibility of bankruptcy); or because special government programs during the downturn reduce the cost to them and their workers of these layoffs (for example through the common response of extending unemployment benefits and making them more generous); or because the first workers being laid off are the least productive ones (possibly because they are relatively new and do not yet have as much experience as others working there) so that they end up with a workforce which is on average more productive.  Or, and very likely, it could be a combination of all three factors.  It looks very much like Schumpeter’s “creative destruction”.

The consequence is that productivity can in fact jump up in a downturn.  One sees such a clear jump in the chart in 2020, at the time of the sharp collapse due to the Covid crisis.  One also sees it in 2008-09, with the financial and economic collapse in the last year of the Bush administration and then the turnaround that began in mid-2009.  In terms of the numbers:  Real GDP fell by 1.3% between the first quarter of 2020 and the third quarter of 2020 (in absolute terms – not annualized).  But employment over this period fell by 7.4%.  As a result, productivity (real GDP per person employed) jumped by 6.6% in this half year.  In 2008/2009, real GDP was basically flat between the last quarter of 2008 and the last quarter of 2009 – rising by just 0.1%  But employment over this period fell by 4.1%, leading to an increase in productivity of 4.4%.

Following these brief periods where businesses are scrambling to survive the downturn by producing more (or perhaps not too much less) with many fewer workers, firms then enter into a more normal period where, as the economy recovers, they are able to sell more of their product.  They hire additional workers who are, by definition, less experienced in the work of that firm than their existing workforce.  The new workers might also be less capable or have a less applicable skill mix.  Productivity may then level off or even go down.  The latter situation is in particular likely when the economy recovers quickly and firms scramble to keep up with the increased demand for their product.

The latter fits well with what we saw in 2021.  GDP in 2021 rose by 5.9%, the highest of any year since 1984.  And the Personal Consumption component of GDP rose by 8.3% in 2021, the highest of any year since 1946.  This was spurred by the series of Covid relief packages passed in 2020 (under Trump) and in 2021 (under Biden), which totaled $5.7 trillion in the two years, or 12.8% of GDP of 2020 and 2021 together.  Personal savings rose to an unprecedented level as a share of GDP (other than during World War II, with data that go back to 1929), which then supported the strong growth in personal consumption in 2021.  This is consistent with a demand-led inflation that got underway in late 2020 or early 2021 (discussed above) – a risk of inflation that Larry Summers had warned of in early February 2021 when Biden’s $1.9 trillion Covid package was first proposed (and eventually passed, largely as proposed).

But what matters to long-term living standards is not so much the changes in average productivity in the periods surrounding economic downturns, but rather the trends in productivity growth over time.  A ten-year moving average is a useful metric:

The chart shows rolling ten-year averages starting from 1947/57 through to 2012/22 of the growth in GDP, in employment, and in productivity (GDP per person employed).  Productivity growth was relatively high at about 2% per annum in the 1950s and through most of the 1960s.  But it then started to fall in the 1970s to less than 1% a year before recovering and returning to about 2% a year in the ten-year period ending in 2004.  It then fell to roughly 0.8% a year since about 2017 (in terms of the ten-year averages), with some sharp fluctuations around that rate associated with the 2020 Covid crisis.  As of the end of 2022, the most recent ten-year average growth rate for productivity was 0.80%.

This has important implications for GDP growth might be going forward.  The labor force grew by 0.8% in 2022 (the adult population grew by 0.7%).  With unemployment close to a record low, employment will not be able to grow faster than the labor force – as discussed above.  And the labor force cannot grow faster than the adult population unless labor force participation rates increase.  But while there major disruptions in labor force participation in 2020 and 2021 surrounding the Covid crisis – with its lockdowns, economic collapse and then recovery, as well as health concerns affecting many – labor force participation largely returned to previous patterns in 2022.  Labor force participation rates have been slowly trending downwards since the late 1990s, and while it is possible this pattern might be reversed, it is difficult to see why it would.  There might well be short-term fluctuations for a period of a few years, but longer-term patterns are driven mostly by demographics (the age structure of the population) and social customs (e.g. whether women decide to enter into the paid labor force).

What follows from this is that if the labor force continues to grow at 0.8% a year (as it did in 2022 – and it grew only at a lower rate of 0.6% a year in the ten-year period ending in 2022), and productivity grows at 0.8% a year (as it did in the ten-year period ending in 2022), then GDP can at most grow at 1.6% a year on average.  This would be disappointing to many.  While there certainly can be and will be significant year to year variation around such a trend, faster growth would require either higher productivity growth or more entering into the labor force.

E.  Summary and Conclusion

The January jobs report was strong.  The unemployment rate is now at the lowest it has been in more than a half-century, and the number employed grew by more than a half million – a very high figure when the unemployment rate is so low.  While these are still preliminary figures and are subject to change as additional data become available, they present a picture of an extremely strong labor market.

The fall in the unemployment rate by one notch to 3.4% from the previous 3.5% should not, in itself, be taken too seriously.  That is well within the normal statistical error for this figure.  But what is indeed significant is that the unemployment rate has been within the narrow range of just 3.4 to 3.7% since March 2022.  That is low.  And it was in this low range during a period (in the second half of 2022) when inflation was coming down.  While changes in the price of oil have been a major factor in driving the inflation rate in 2022, the core rate of inflation (which excludes energy prices as well as those for food) has also started to come down.  The rate of change in nominal wages did start to grow in mid-2021, but this appears more to be a consequence of the rising prices rather than a cause of them.  And there has been a slight reduction in the pace of change in wages in recent months.

One does not see in this any evidence that a tight labor market with extremely low unemployment (the lowest in more than a half-century), has led to higher inflation.  The opposite has happened.  Inflation has come down at precisely the time the labor market has been the tightest.

GDP grew rapidly in 2021, but then slowed to a more modest 1.0% rate in 2022 (from fourth quarter to fourth quarter).  Coupled with rapid employment growth in the year, productivity (as measured by GDP per employed person) fell.  However, this appears more to be a continued reaction to changes surrounding the disruptions resulting from the 2020 Covid crisis.  During that crisis, GDP fell but employment fell by much more, leading to a jump in productivity despite the downturn.  As the economy recovered and the situation normalized, workers were hired to bring workforces back to desired levels.  Viewed in a longer timeframe, productivity growth has been similar to what it has now been since the mid-2010s.

That productivity growth is not especially high.  It was 0.8% at an annual rate in the most recent ten-year average.  Coupled with a labor force that grew at 0.8% in 2022, and going forward might grow by even less (it grew at 0.6% a year in the ten-year period ending in 2022), the ceiling on GDP growth would be 1.6% a year, or less.  That is not high, but expectations need to adjust.

That is also a ceiling on what GDP growth might be.  Many expect that there very well could be a recession either later in 2023 or in 2024.  Much will depend on whether the government will be able to respond appropriately if the economy appears to be heading into a downturn.  But with Republicans now in control of the House of Representatives, and threatening to force the US Treasury into default on the nation’s public debt if their demands for drastic spending cuts are not met, one cannot be optimistic that the government will be allowed to respond appropriately.