Disconnect between aggregate earnings and payrolls

Usually Claudia Sahm is answering the hard questions in economics. She does it better than anyone else. But yesterday she asked a question, which automatically warrants some thinking. 

Dr. Sahm pointed out the disconnect between total wage and salary income and payrolls. While it’s clear that transfer payments explain the higher levels of personal income during the COVID-19 recession, it’s less clear how total wages and salaries could have nearly returned to their pre-recession level with 10 million fewer jobs. 

In November 2020, workers received wages and salaries equivalent to an annual rate of $9.6 trillion, an increase of two percent from November 2019. Yet over the same period, the total number of jobs fell by 9.2 million or 6.1 percent. This disconnect, pointed out by Dr. Sahm, may suggest that wage and salary data are painting an overly-rosy picture of the economy. While the staff of stats agencies hold a special place in my heart, wage and salary data for November 2020 are preliminary estimates from incomplete sources. As new data become available, the November 2020 estimates, produced by the Bureau of Economic Analysis (BEA), will be revised. Two years after the previous recession ended, BEA economists were still lowering their estimates of the wages and salaries received during the recession. 

So Dr. Sahm is right; the disconnect between earnings data and payrolls data is concerning. The most plausible explanation for this gap comes from John Vorheis, who posits that jobs permanently lost were disproportionately low wage, while jobs that have come back or were not lost have more wage growth. To test this, I’ll try to estimate the gap.

First, let’s estimate what total wages and salaries might have been in 2020 for the people who were employed in 2019, if they had all remained employed. For an upper-bound estimate, the average wage growth for individuals was around six percent in 2019. Six percent growth in wages and salaries would result in an annual rate of $10 trillion in November 2020, around $400 billion more than the current estimate, or $43,500 for each missing job. The result produced by assuming six percent growth is plausible, but I’m not sure that six percent growth is a realistic assumption.

As an alternative estimate, in the first two months of 2020 total wage and salary growth was 4.5 percent while payroll growth was 1.5 percent, meaning individual wage and salary growth was around three percent. A more-realistic continuation of three percent growth would result in annual wages and salaries of $9.7 trillion in November 2020, around $100 billion above the current estimate, or $10,900 per missing job. 

The more-realistic estimate, based on three percent individual wage and salary growth, suggests to me that the preliminary wage and salary data and the payrolls data do not match. It’s worth thinking therefore about possible reasons for revisions to the data. In 2019, the BEA increased total wage and salary figures by $111 billion to adjust for under-reporting of tips in taxes returns, and added $430 billion to adjust for employees paid while bypassing unemployment insurance (NIPA table 7.18). The adjustments for 2020 are not available yet, but I could imagine a situation where tips are far lower and more likely to be reported and fewer people are trying to bypass unemployment insurance. If this is the case, the current estimates of wages and salaries could be revised downward.

Measuring Earnings from Work

Despite the ongoing pandemic, the US Bureau of Labor Statistics report that a typical full-time worker earns the equivalent of more than $52,000 a year. This post discusses earnings from wages and salaries and presents two broader measures that are designed to include people with little or no earnings.

As a starting point for thinking about a person’s earnings: survey data show that wages tend to increase with age. The typical older worker has more work experience and is paid more than the typical younger worker. Among full-time wage and salary workers, earnings increase steadily from $25,000 per year at age 16 to more than $60,000 per year around age 45, then plateau and remain near their peak until the social security retirement age (see chart 1). 

Measuring the earnings of full-time workers, however, does not tell us about people who are either working part-time, unemployed, or work-limited in another way. Typical earnings are much lower if we include the earnings of all people in the survey rather than just those who are currently working full-time. The median wage falls from $52,500 for current full-time workers to $15,290 for anyone in the survey aged 16 or older. Broken down by age, the broader measure shows earnings are zero for a typical 16 year-old, around $30,000 for the typical 25 year-old, and peak at around $42,000 for a typical person in their mid 40s (see chart 2). Median earnings past retirement age are zero.

By including people working part-time as well as people with zero earnings, the second chart reverses a major result from the first chart. Specifically, during the pandemic many people lost jobs or became part-time and therefore are removed from the 2020 measurement of full-time workers’ wages. Since jobs lost during the pandemic were more-often below the 2019 median wage, removing these jobs from a wage measurement makes the new median wage of the remaining jobs higher. In other words, the chart 1 result showing that the wage of the typical full-time worker is higher in 2020 than it was in 2019 is because the group of people being measured has changed and the typical worker is a different person. Measuring the same group of people in both 2019 and 2020 shows that earnings have instead fallen substantially during the pandemic. The reduction in earnings is confirmed by data on total wage and salary income.

Importantly, despite far worse economic conditions in 2020 than in 2019, chart 2 shows that several features of the relationship between earnings and age persist across both years. As one persistent feature, in both 2019 and 2020 the typical person who has reached retirement age has stopped working and has zero earnings. The US does a reasonably good job of providing welfare payments to people who reach retirement age, through Social Security, which facilitates retirement.

A second persistent feature is that children have no earnings in either year. With some exceptions, including newspaper delivery and family businesses, wage labor from children under the age 16 is uncommon and it is not measured in the survey. Also, many teenagers above the age of 15 are not working because of school.

Lastly, some portion of working-age adults are work-limited at any given point in time. The work-limited include people with an injury, disability, or illness, as well as students, and people with additional responsibilities at home such as caregivers and new parents. US rules around welfare of children and persons with disabilities are more strict and payments are meager, if available at all, relative to the help retirees receive from Social Security. One consequence of a work-focused welfare system is that many who become work-limited are expected to rely on the earnings of their family members to get by.

Family Earnings

Living with family is the way many people, including children, receive resources. Living together allows people to pool their income, share costs, and use resources like a car or a kitchen more efficiently. Economists sometimes calculate a concept called “equivalized” family earnings by adjusting the total earnings of a family for the size of the family (see footnote of chart 3 for details). The result is a per person measure of family wage and salary earnings that accounts for the efficiency of the size of a person’s family. Typical equivalized family earnings are $21,620 per person per year in 2020. 

By age, median equivalized family earnings are around $32,000 for newborns, $40,000 for 20-year olds, more than $50,000 for 50-year olds, and zero for those age 70 or older (see chart 3). The drop in family earnings starting at age 60 is from the combination of retirement and from people in these age groups being less likely to live with someone else who works.

Interestingly, the decrease in median earnings from 2019 to 2020 is less severe in the family earnings measure from chart 3 than in the personal earnings measure from chart 2. The resilience of equivalized family earnings to the recession is due in part to more people moving in with family. In 2020, a larger share of people utilize the resource efficiency of living with others, which is captured by an equivalized measure such as family earnings but not by the personal earnings measure.

Families with No Earnings

While the above sections focus on the typical person, measured as the median or middle member of the group, the effects of the COVID-19 recession have fallen more heavily on people nearer to the bottom of the wage distribution, an effect already evidenced in chart 1. One way to see how job losses hurt families is to look at the share of people whose family have no earnings. During a recession, the share without earnings is larger because there are fewer jobs. In 2020, 37.8 percent of people have no family earnings in the survey week, compared to 35.4 percent in 2019. By age, around 15 percent of newborns have no family earnings, compared to less than 10 percent of 45 year-olds and more than half of 70-year olds (see chart 4). 

Chart 4 can help to illustrate the relationship between poverty, age, and welfare. By definition, families without earnings will be in poverty unless they receive sufficient income from welfare or from private assets. Because the elderly are far less likely to have family earnings from work, the age group relies heavily on other sources of income to avoid poverty. For elderly people without much private wealth, Social Security payments facilitate independent living and result in much less poverty for elderly people and their families. That is, social insurance makes up for a lack of family earnings and reduces elderly poverty to the same rate or lower as the overall population. Other countries go further than the US and apply this same technique to other groups of people who do not have earnings, such as children. Including people who do not have earnings when measuring earnings can be useful for thinking about how the US should treat someone when they are work-limited.

Who gets credit for the wage growth of low wage workers?

Last month, I pointed out 7% wage growth at the first decile (10th percentile). The CEA noted this same figure in a recent report, claiming credit for the uptick at the bottom of the wage distribution. My piece, in contrast, had assigned credit to the state, local, and company minimum wage increases. AEI has a piece that thanks both the long expansion and the minimum wage increases.

The AEI piece points to Atlanta Fed analysis showing that states without state minimum wage hikes have seen increases in the median wage of the first quartile (bottom 25%) of wage earners. This makes a strong case that state minimum wage increases only explain part of the story. But it’s worth testing the argument that if additional states had raised their minimum wage those states would have seen larger wage gains for low-wage workers. That is, by making the wrong decision the no- and low-minimum wage states fared worse, but despite making the wrong decision, they still fared reasonably well.

A test of my claim comes from comparing total hours worked in different wage groups and in two state groups: those that raised the state minimum wage from 2016 to 2019 and those that didn’t. To better capture total wage income, aggregate hours are used instead of number of workers. In the baseline data which cover 2016, slightly more than half (55.4%) of hours worked in the US were from people in the 27 states with minimum wage increases during 2016-2019. This is how the total hours worked are distributed by wage group and state group in 2016:

result2016

Notably, the share of hours that are paid the federal minimum wage or less is low; several times as many hours of work were paid between $7.26 and $12.00.

In the latest year of data, which ends November 2019, the story has changed:

resultlatest

The share of work hours paid less than $12 has fallen substantially since 2016. The effect is more pronounced in states that increased the minimum wage but also apparent in those that didn’t. As a result, states that didn’t raise the minimum wage now disproportionately have low wage work relative to the states that did.

It’s worth noting that total hours of work increased by five percent over the period. The concern that raising the minimum wage will cause job losses doesn’t show up in the data. The states that raised their minimum wage maintained a fairly constant share of total hours of work, claiming 55.2% of total hours over the past year.

Looking more closely at the data, the long economic expansion does deserve some credit for wage growth at the bottom of the distribution. Importantly, states that acted to raise minimum wages better shared the economic expansion with low wage workers.

Wage growth is in the pipeline

If you haven’t looked at the US wage distribution recently, you might be surprised to see nominal growth of 7.0 percent in 2019 Q3 and 6.7 percent in Q2 in first decile usual weekly earnings. My own calculations show an even stronger 8.7 percent year over year increase in October 2019. The increase in the first decile wage seems to be coming from state, local, and company increases to the minimum wage. It is also an indication that higher wages may soon be coming to the workers in the middle of the income distribution.

First, the data:

First decile wages and growth

The dark blue lines are the BLS quarterly data on first decile usual weekly earnings for full-time wage and salary earners. The light green lines (bd CPS) calculate the same series on a monthly basis.

The BLS and bd CPS series both show first decile wage growing at its strongest rate since the late 1990s. One conventional story here is that a tight labor market, as measured by a low unemployment rate, makes it harder for employers to find qualified replacements for employees who quit, which makes existing employees more likely to get raises. As a result, wage growth is usually strongest when the labor market is tight.

But there’s a catch. BLS uses the same sample to report nominal median wage growth of 3.6 percent in Q3 and 3.7 percent in Q2. BLS data on real average hourly earnings, which adjusts nominal earnings data for inflation, shows 1.2 percent real wage growth in October (nominal wage growth of 3.0 percent and inflation of 1.8 percent). This isn’t terrible, but it’s no seven percent. With the unemployment rate historically low, why isn’t median and average wage growth stronger?

Leaning on unemployment to fully explain wage growth is clearly not working. An alternative measure of whether a labor market is tight is the employment rate for people age 25-54. This measure is currently at it’s highest level since 2007 but is still the equivalent of more than a million jobs away from its late 1990s and early 2000s level. In other words, there’s evidence that the labor market is still over a million jobs away from full employment.

Maybe it’s better to ask instead why wages are growing so rapidly at the first decile and whether it means anything for wage growth more broadly. The data show that the number of full-time workers earning $400 per week or less has fallen to 7.5 million in the latest three months from 14.7 million in the same three months of 2014. The story here seems to be higher local and state minimum wages and higher company minimum wages at large employers. Several areas and employers moved the minimum wage to above $10 per hour ($400 for a 40-hour week) during the past five years.

Higher minimum wages could eventually translate into higher median wages. For example, the first decile and median wage tend to move together:

First decile and median wage growth

In the period of faster wage growth and tighter labor markets during the late 1990s, first decile wage growth accelerated and median wage growth accelerated shortly after. If employment trends continue, it seems reasonable to expect an increase in the median wage of five percent or more in 2020.

But the current period is not the late 1990s. Business fixed investment and labor productivity growth are both particularly weak now compared to then. Negative net capital investment and increasing payrolls suggest an aggregate situation akin to fewer tools per worker, which casts doubt on forecasts of sustained real wage growth.

Technical Note

Unlike the first two charts, I use some additional steps to smooth out the third chart comparing median and first decile wage growth. I calculate the wage in each month as the wage in the previous three months combined, so October 2019 is based on combined microdata from August, September, and October 2019. Additionally, I use the Census X13as program to seasonally adjust the results. The growth rate is based on this resultant seasonally-adjusted nominal wage.

In both sets of charts, I replicate the BLS process of taking a “binned median” to reduce how much the data reflect breaks around certain round numbers (like $400 per month).

Do regional inflation differences affect nationwide real wage estimates?

Some basic facts about the US economy point to the possibility that inflation is overstated in measures of real wage growth for workers at the very bottom of the wage distribution. If this is true, it would mean that low-wage earners have actually seen more wage growth than published estimates suggest.

First, the places that have a higher state or local minimum wage than the federal minimum wage tend to be in the west region of the US, and sometimes in the northeast or midwest regions, but rarely in the south. The regional differences in minimum wage create large differences in the regional distribution of low wage earners. In November 2018, the population (including children) is divided by region as follows: midwest: 20.8%, northeast: 17.6%, south: 37.8%, and west: 23.8%. In contrast, people earning $8.00 per hour or less are distributed as follows: midwest: 19.2%, northeast: 16.3%, south: 51.3%, and west: 13.2%. In other words, low wage earners are disproportionately in the south and disproportionately not in the west.

Second, in recent years inflation has been higher in the west region, compared to the south or midwest regions. This is largely because inflation has recently been driven by housing shortages, which are particularly severe in the west region. Price growth from December 2017 to December 2018 was 1.9% nationwide, 1.3% in the midwest, 1.7% in the northeast, 1.5% in the south, and 3.1% in the west.

The above creates a potential issue for estimates of real wage growth for low wage earners. This is because nearly all such estimates apply the nationwide rate of inflation (for all urban consumers) to workers who disproportionately live in the south, which has less inflation than the nation as a whole. Likewise, low-wage workers are not nearly as likely to live in the west, which has higher inflation than the nation as a whole.

To test this, I used CPS microdata to calculate the real wage for the fifth percentile wage earner (nationwide) using both the usual CPI-U (the CPI-U-RS is preferred, but that would have complicated analysis) and using the regional CPI for each of the four regions. That is, in the regional CPI estimate, each wage observation is adjusted using the price index for the region where the person lives.

rw_by_cpi

The results show that the story above does apply, but that the effects are very minimal and not particularly cumulative. The total cumulative difference since 1989 is less than three percent. In other words, it’s true that low wage earners did better than published estimates suggest, but the effect is not very big. It’s important to point out that adjusting for age, or including only full-time workers, eliminates most of the cumulative impact. Usually findings this trivial are not written up as a blog post, but they might be of interest to people who have the same question. Here’s the jupyter notebook used to calculate the results.

Comments and feedback are always welcome. Please let me know if I did something wrong, so I can learn!