Tag Archives: jobs

A Comparison of the Net New Job Estimates of the ADP and the BLS: The Differences are Large

Posted on by

Chart 1

With the government shutdown, the Bureau of Labor Statistics (BLS) is no longer issuing its often eagerly awaited monthly report on employment growth.  Institutions such as the Fed are flying blind, even though the Fed is now reducing interest rates out of concern the job market is weak.  Private markets know even less.

In the absence of the BLS reports, the reports issued by the private firm ADP (a large processor of payrolls for firms) have taken on greater importance. The ADP has long issued a report (normally two days before the BLS issues its monthly report) that provides a first look at what the employment numbers might be that month.  It is now producing these reports in a collaboration with the Digital Economy Lab of Stanford University.  ADP bases its estimates on the payrolls it processes for firms, now covering 26 million employees.  While not as large as the sample of employers who report their payrolls to the BLS each month (roughly one-third of all workers – including government employees – or around 53 million currently), those are both still huge “samples”.

But the ADP does not have the advantages the BLS has.  The BLS obtains its estimates from a stratified random sample of employers, while the ADP must base its sample on the firms that have contracted with it for payroll processing services.  While both the BLS and the ADP try to correct their samples to reflect a proper weighting of different types of firms (e.g. by size of the firm, by sector of activity and region of the country, by type of ownership, and by other factors that may be relevant), there can still be biases in the data they have to work with.  Such biases are likely significantly more of a factor for ADP.

It is also important to recognize that while the focus of the media is typically on the number of “new jobs created” in the month of the report, that is a confused term on many levels.  There is constant turnover in the job market, with workers starting new jobs and leaving old ones (mostly by resignation, but also from layoffs, retirement, and so on).  The estimates of neither the ADP nor the BLS are directly of “new jobs”.  Rather, they both estimate the total number of workers who were employed (i.e. were on the payroll) in some specific period of each month (for the BLS, for the payroll period that included the 12th day of the month).  The number of “new jobs created” is then the difference between the estimated total number of workers on a payroll in the given month less the separately estimated total number of workers on a payroll in the prior month.  This difference is thus a net figure, and is estimated by taking the difference between two total employment estimates.

Both ADP and the BLS also revise their monthly estimates each year in a rebenchmarking exercise, based on results from the Quarterly Census of Employment and Wages of the BLS.  This report is comprehensive  – a census – that provides a good count of the number of workers employed as of a given month each year (mostly from available data on who is covered by unemployment insurance).

Are the ADP monthly estimates on net employment growth then close to what the BLS estimated?  Not really.  The chart at the top of this post shows the difference between net private employment growth as estimated by ADP and as estimated for the month by the BLS.  The currently issued ADP series on private employment begins in January 2010, and hence one can find the change in employment starting in February.  The last estimate issued by the BLS before the government shutdown was for August of this year.  The BLS figures used here are its estimates for private employment (to be comparable to the ADP estimates), not the more commonly cited figures for overall employment – including government – that the BLS also provides.

I have removed from the chart the figures that would apply for the period from March 2020 to July 2020.  This was during the peak of the Covid disruptions, and the differences in the monthly net new jobs estimates were huge.  The largest was for April, where ADP estimated private employment fell by 6.1 million while the BLS estimated it fell by 19.6 million – a difference of 13.5 million.  But these were far from normal circumstances, and including these months on the chart would have distorted the scale.  They were instead set to zero.

Leaving out the March 2020 to July 2020 figures, the average absolute deviation (i.e. the average deviation regardless of whether higher or lower) was 108,000 workers.  That is, on average the ADP and BLS estimates of net new private jobs in a given month differed by 108,000.  To put this in perspective, the average monthly growth in net new jobs over this period was 152,000 in the BLS figures.  The average deviation between the ADP and BLS estimates (of 108,000) was more than two-thirds of this.

The standard deviation was 180,000, meaning that (in the long run) in two-thirds of the months the deviation can be expected to be less than that, and hence that in one-third they can be expected to be more.  These monthly differences in the employment estimates between the ADP and the BLS are large.  It would not be safe to assume the ADP figures provide a good estimate of what the BLS figures would have been.

The chart at the top of this post covers the full period from February 2010 to August 2025, a period when there were other major disruptions and not just those of the Covid period.  If one limits the analysis to the period since 2022, when growth has been relatively smooth and unemployment has been low, one has:

Chart 2

While the deviations between the two sets of estimates are less than over the full period going back to 2010 (with an average absolute deviation of 68,000 workers and a standard deviation of 99,000), the differences are still large.

It is commendable that ADP issues employment estimates based on the payroll data they have access to.  With no figures from the BLS, there is currently nothing else to go on to ascertain whether employment is growing or not.  But the ADP figures are not a terribly good approximation to what the BLS provides in its monthly net new employment estimates, and users should be aware of the often large differences.

Not a Good Jobs Report – And Firing the Messenger Will Not Help

Posted on by

Chart 1

Update – August 5, 2025:  In the initial version of this post, I mistakenly said that businesses are required by law to participate in the Current Employment Statistics (CES) survey of the Bureau of Labor Statistics (BLS) – the survey that the BLS jobs numbers are based upon.  This is not correct.  While participation is required in certain states under state laws, there is no federal law on this.  The post below has been corrected to reflect this and has added material on the participation rates.  

On August 1, the Bureau of Labor Statistics (BLS) released its estimate of net job growth in July.  The news was not good.  Net job growth in July was just 73,000, where growth in the number of jobs in health care and social assistance alone came to 73,300.  That is, net job growth for everything else was essentially zero, indeed negative.

But more surprising and concerning than the disappointing growth in July, the updated BLS estimate of total job growth in May was reduced by 120,000 to just 5,000, while the estimate for June was reduced by 133,000 to just 14,000.  The initial job growth estimates for May and June were a healthy 139,000 and 147,000, respectively.  Now, both figures are close to nothing.

Revisions to the monthly job estimates are both automatic and routine.  The BLS always revises the estimates for the most recent two months as each new monthly report is released, as it updates its estimates based on more complete data sent to it by employers who are covered in its Current Employment Statistics (CES) survey.  The CES sample includes approximately 121,000 businesses and government agencies at approximately 631,000 individual worksites – covering a total of about one-third of all nonfarm payroll jobs.  The BLS estimates then rely on timely reporting from these employers.

Participation by employers who have agreed to take part in the survey is high.  While not mandated by federal law (although it is in certain states by state law), approximately 94 to 95% who have agreed to be part of the survey then respond by the time of the final release of the employment estimates (based on response rates over the past year).  But not all respond immediately, and some cannot.

The figures reported are of the number employed by the establishment at some point during the payroll period that covers the 12th day of each month.  These are reported along with figures on the compensation paid by the firm (in dollars) and the total hours worked (and separately the total overtime hours worked).  The firm will not know what these will be until the payroll period is over.  If the payroll period is one week, or even usually two weeks, the firm should be able to file its report (which is normally done online) in time for its data to be included in the initial estimate of employment that the BLS issues each month.  But if the payroll period is for the full month, then by definition the BLS will not have that firm’s data in time for its initial employment estimate of that month.  The BLS thus issues a revised estimate the next month, and then also the following month, as further data arrives.  This is all standard.

The BLS does, however, arrive at an estimate each month for total employment – despite still having only partial reports from its survey – by imputing values for the missing data based on past patterns and relationships.  On average those initial estimates are very good, with the later revisions sometimes higher and sometimes lower.  The average revision since 1979 has been less than 0.01% of the number employed.  In absolute value terms (i.e. in terms of just the revisions themselves, not whether they were positive or negative), the average revision was a still very small 0.05%.

The BLS thus updates its job estimates for any given month at the time of its next monthly report and then again at the time of the monthly report after that, based on the more complete reports it has received by then.  There is also an annual revision of the monthly figures, issued each January, based on updated control totals for overall employment and its composition broken down by such factors as the size of the firm, the sector, and so on.  These control totals are obtained from periodic census information.  The process followed for that annual revision was described in an August 2024 post on this blog.

All this is standard and routine, and follows a methodology that was developed years ago.  Furthermore, it is automated and done by computer, with only a cursory review of the numbers by staff at the end.  The BLS is also fully transparent on the revisions it makes.  One can find on this webpage on the BLS site all the monthly revisions (from both the initial estimate to the second, and from the second to the third), month-by-month, back to January 1979.

Trump paid no attention to this.  Faced with the reality of disappointing jobs numbers for the last several months, Trump decided that the best course of action was to shoot the messenger.  He fired the well-respected and professional head of the BLS, Dr. Erika McEntarfer, asserting (with a clear lack of understanding) that McEntarfer was somehow manipulating the numbers for political reasons.  Trump had also asserted in August 2024 that the BLS under Commissioner McEntarfer had manipulated the annual revision of the benchmark control totals released at that time (which happened also to be a downward revision).  The control totals are released every August, prior to the detailed sectoral and monthly figures then released in the coming January.  See my August 2024 blog post on that episode.

Trump also does not understand that the revisions – while significant – are far from unprecedented in size.  The chart at the top of this post shows the revisions to the monthly figures (defined as the difference between the third estimate and the first, with the sole exception for the June 2025 estimate which is between the second estimate and the first), as a share of the overall number employed in the economy.  It is shown as a share of total employment in order to make meaningful comparisons over time, as there are now 80% more employed in the US than there were in 1979.  The figures for May and June 2025 are shown in red at the far right in the chart, although the red may be hard to see.

The first thing to note from the chart is how small the revisions mostly are, especially over the last quarter century.  Unless there are major economic disruptions underway or soon to be underway (such as from the Covid crisis in 2020 and the recovery from it in 2021, from the 2008/09 Great Recession, and earlier in the years at the end of the Carter presidency and the first term of Reagan, when the economy went through two separate recessions), the range of the revisions is almost always well less than +/- 0.1% of the number employed.  This is extremely small.  Keep in mind that the BLS arrives at its estimates of total employment each month from a zero base, where it asks its (rather large) sample of business establishments how many people they currently employ.  The estimates are not based on a survey asking, for example, what the change in employment at their establishment may have been.

The May and June 2025 revisions were reductions in estimated employment of 0.075% and 0.083% of total employment, respectively.  These were major revisions, but certainly far from unprecedented.  Plotting the revisions onto a histogram with bins of 0.05%, we have:

Chart 2

The May and June revisions fit into the -0.10% to -0.05% bin.  There were a total of 37 monthly cases that fit in that group in the 558 months from January 1979 to June 2025, and there were also 27 monthly cases where the reductions were greater.  That is, revisions similar in magnitude to, or greater than, those in May and June have happened in about 11% of the monthly cases since 1979.

There is no evidence that the BLS or its head Erika McEntarfer somehow manipulated these estimates to arrive at figures that Trump did not like.  Indeed, it would probably be impossible, given how automated the process is.  That does not mean, however, that a new Trump appointee to head the BLS would not be able, over time, to redesign the process in order to give Trump something closer to the numbers he wants.  This will need to be watched closely.

Until recently, the first groups that would be advised of any changes in its methodology that the BLS was considering would have been two advisory panels of outside professionals.  The panels were made up of individuals from universities, research institutes, and private businesses, who were impartial professionals and were not paid (other than for travel expenses).  Those panels – the BLS Technical Advisory Committee and the BLS Data Users Advisory Committee – were, however, dismissed in mid-March by the then new Trump administration.

There were concerns when the panels were dissolved that the Trump administration had plans to politicize the process by which basic economic data is gathered, with the aim of ensuring only flattering figures are released.  The firing of the Commissioner of the BLS appears to be a further step in that process.

Real Wages of Individuals Under Obama, Trump, and Biden

Posted on by

There have been repeated assertions by Trump during the presidential campaign (as well as by Vance in the October 1 debate between the vice presidential candidates) that people’s wages were higher under Trump than they now are under Biden.  What has in fact happened?

The chart above shows how indices of the real wages of individuals have moved during the last two years of Obama’s presidency, the four years of Trump’s presidency, and Biden’s presidency through to August 2024 (the most recent data available as I write this).  There is much to note, but first a few words on the methodology.

The primary data comes from the “Wage Growth Tracker” website provided by staff at the Atlanta Fed.  It makes use of data generated as part of the Current Population Survey (CPS) of the Bureau of Labor Statistics.  From the way the survey is designed, they can obtain data on the wages earned by each household member at a point in time and again for that same individuals twelve months later.  From this raw data, staff at the Atlanta Fed calculate for the individuals in the matched households how much their wages changed over those twelve months.  Since the CPS also collects information on the individuals themselves, they can also then determine what the average (as well as median) changes in wages were for individuals grouped by various characteristics, such as age or gender, race, education, occupation, and more.  The chart above shows both how real wages changed for workers as a whole, as well as the changes with wage-earners grouped by quartile of wage income, from the lowest to the highest.  The figures shown here are for the medians in each category.

The Atlanta Fed wage data goes back to December 1997, is presented in terms of the 12-month percentage changes, and is in nominal terms.  I converted the data to real terms based on the change over the same 12-month periods in the overall CPI (formally the CPI-U, produced by the BLS), converted this to an index number, and then rebased this to set January 2021 equal to 100.  The result is the chart at the top of this post.

In interpreting these figures, it is critically important to recognize that they reflect what households actually experience in terms of the changes in their individual wages.  This differs from what one will normally see when reference is made to changes in mean (i.e. average) or median wages.  The figures in the chart track the experience of individuals, and individuals will normally see their wages start relatively low – when they are young and inexperienced – and then grow over time as they gain skill and experience.  That is the normal life cycle.

Statistics on wages as normally presented, in contrast, measure not what the experience is of individuals, but rather movement in the overall mean or median wages of all those in the labor force at the time.  Changes in such wages will normally be less than what one observes for individual wages, as the labor force is dynamic, with young people entering (at normally relatively low wages) while older people retire and leave the labor force (at normally relatively high wages).  This will reduce the measured growth in average wages as higher-wage workers have left while lower-wage workers have entered.  While this change in the average wage of all those employed at each point in time is a useful statistic to know, it does not reflect the lived experience of individuals, who normally see their wages grow over time (at least in nominal terms) as they gain experience and hence ability.

One sees a consequence of this in the chart above.  Those in the lowest quartile of the distribution of wage earnings have seen growth in the wages they earn as individuals that is greater than the percentage increases of those in the higher quartiles.  This is because those starting out in the labor force – and entering at relatively low wages – generally see a relatively fast rate of wage growth as they gain skills and are promoted.  This slows down over time, with older workers still receiving annual wage increases (in at least nominal terms) but not as large in percentage terms as young workers do.

Tracking the real wages of individuals is therefore of interest, but cannot then be used to track over long periods of time what has happened to average (or median) wages.  But for periods of several years, as well as for a comparison of growth in some early period to growth in a similar later period, tracking as in the chart above is of greater interest than what has happened to average or median wages of an always changing labor force with young workers entering and older workers leaving.  It is useful in comparisons of the growth in wages between presidential terms.

With this understanding, a number of points may be noted on individual wage growth in recent years:

a)  Individual wages in real terms were rising at a reasonable rate in the last few years of the Obama administration.  They then grew at a similar rate (not a faster rate) during the first three years of the Trump administration prior to the disruptions due to Covid.  In fact, the growth rate of overall individual wages (as measured at the medians) was 1.4% per annum in real terms during the final two years of the Obama administration (January 2015 to January 2017), and then the exact same 1.4% per annum in real terms during the first three years of the Trump administration (January 2017 to January 2020).

Trump has repeatedly claimed that wage growth (as well as many other things) were the highest ever during his administration, but that is not the case.  The most that Trump can rightfully claim is that he did not mess up the growth path that Obama had put the economy on following his reversal of the economic and financial collapse that began in 2008, in the last year of the Bush administration.

b)  With the onset of the Covid crisis in early 2020, individual real wages in fact rose despite the chaos of the lockdowns.  This might appear perverse, but in fact makes sense.  First of all, the rate of unemployment shot up to 14.8% – the highest it has been since the Great Depression (so Trump now owns this record).  But 85.2% remained employed, and were employed under often difficult personal circumstances given the easy spread of Covid and a lack of preparation by the Trump administration for the approaching pandemic.  (Trump instead repeatedly stated that all would be fine; that the virus would quickly disappear; and that banning flights from China had been a great success in stopping the virus.)

Those who remained in their jobs during this difficult period were often compensated well for their willingness to do so.  They received significant increases in their wages and/or bonuses.  The alternative of unemployment was also not as bad as it normally would be.  Aside from the safety aspect of protecting yourself from exposure to Covid, programs for the unemployed at the time were more generous and more easily available than they normally are, due to special legislation passed to address the exceptional circumstances of Covid.  Workers had this alternative, and firms had to respond.  Firms also received often generous support through various special programs during this period, that enabled them to pay higher wages to the employees who remained on the job.

Thus one sees in the chart above that individual real wages in fact rose in 2020, despite of (or perhaps one should say because of) the Covid disruptions.

c)  The Covid disruptions continued into 2021 and the first half of 2022, while the special support programs for firms and the unemployed were scaled back to normal.  But supply chains had been radically disrupted globally due to the crisis, did not start to recover until vaccines became widely available, and then required time to catch up and normalize.  And while supply was constrained, demand rose more quickly starting in 2021 as shoppers returned.  This demand was especially high both because of pent-up needs or desires for items not purchased in 2020 due to the lockdowns as well as caution due to the easy spread of the disease, while personal savings were exceptionally high and could now be spent.  Savings (and bank accounts) were high due both to the lack of spending in 2020 and to the extremely generous financial support packages passed under both Trump and Biden.

Global supply chains then worked themselves out by mid-2022.  The rate of inflation had been relatively high before then due to the high demand confronting limited supply, but inflation as measured by the CPI index for all items other than shelter then fell dramatically from July 2022 once supply was no longer constraining.

This inflation was then reflected in the decline in real wages from early 2021 to the trough in June 2022, as seen in the chart above.  From January 2021 to June 2022 the overall individual real wage fell at a rate of 3.5% per annum.  But probably a more appropriate measure would be for the period from January 2020 (immediately before the Covid crisis) to June 2022.  Over this period, the overall individual real wage fell at a rate of 1.2% per annum.

d)  Once Covid and its related impacts were largely over in mid-2022, real wages immediately began to grow again.  And indeed, they have grown since then (through at least to August 2024 – the most recent data available as I write this) at a rate of 2.5% per annum.  This is substantially faster than the pace they had grown under Trump (as well as under Obama before him), although this can be attributed in part to a recovery from the decline in the period ending in June 2022.

With this recovery, the overall individual real wage is now back on average to where it was in January 2021.  And the real wages of those in the lowest quartile and in the second quartile of the wage income distribution are now significantly higher than they have ever been.  But the levels as of August 2024 should not be seen as especially significant in themselves.  August is simply the most recent data available.  Rather, what is significant is the strong growth seen in real wages since June 2022, with no sign yet that that strong growth is abating.  Eventually that growth will likely return to the longer-term growth seen under Obama and then in the first three years of Trump, but it is not there yet.

e)  One should also note that all these figures are for the medians over a diverse population.  While the overall figures (whether measured at the means or the medians) have gone up and down, the actual real wages of any given individual can be quite different.  While the median individual real wage is now back to where it was in January 2021, this will not be true for everyone.  That diversity in experience needs to be recognized and acknowledged.

 

Biden inherited an economy that had suffered the sharpest downturn and highest unemployment since the Great Depression.  Managing the onset of the Covid pandemic in 2020 would have been difficult for even the most competent of administrations, but the Trump administration was far from the most competent.  The impacts of that crisis – on supply chains among other effects – continued into 2021 and the first half of 2022, and they led to falling real wages over this period.  But as supply chains normalized, real wages began to recover.  As of August 2024, overall individual real wages are back to where they were in January 2021.  But more importantly, those real wages have been growing at a rapid pace since mid-2022 and as yet show no sign of slowing down.