May 2016 Job Growth – a Paltry 38,000 (Preliminary) Net New Jobs
Unemployment Rate Falls from 5.0 Percent in April 2016 to 4.7 Percent in May 2016
So what happened to the May 2016 employment data, which added just 38,000 net new jobs versus a consensus estimate of 160,000 or more? The decline was one of three things (or a combination thereof):
- The economy had the wheels fall off in May (with some in the press blaming the ongoing presidential contest – no kidding)
- May 2016 and the prior two months’ job growth samples were not truly representative of the overall population of total job growth – perhaps bad methodology or a sampling error
- Someone is manipulating the data (for you conspiracy theorists out there)
So what do the U.S. employment numbers look like? The following table shows jobs numbers and gains by month since December 2015, expressed in thousands. Forget the measly 38,000 net new jobs (preliminary) for May 2016, and look at the restatements for March and April. March job growth, which as of April 2016 was estimated to be 208,000 net new jobs, was restated down to 186,000 – a 10.58 percent decline. April, which as of a month ago was thought to be 160,000 dropped to 123,000 – a restatement change of -23.13 percent. Huge. Massive. Material. That’s like flying to the moon and missing it by 55,257 miles. Yep – that’s a big miss.
When you see a 23 percent downward restatement as in the April job growth numbers, something is really amiss.
Admit it or not, our daily inflow of information, news and politics is based on sampling. When the FDA completes a study of the effectiveness and safety of drugs in trails it is based on two relatively small groups, one which is given the actual tested drug and another that receives a placebo. Compared to the U.S. population, that sample size is miniscule.
Do not believe everything you hear or read is actual and carved in stone – particularly if it is based on a sample. A sample is merely intended to be a likely outcome of the total population. This could be voters, defective light bulbs or even incomes by age groups or gender, for example. Look at the newspapers that declared Dewey as winning the presidency in 1948 when he actually had lost. That’s the error we see periodically in sampling.
Given the upcoming election season, every day now we read or hear of poll results on who is ahead and who is behind. A good example is the latest Reuters-Ipsos poll which is premised on people who are not voting for a person for president, but rather who they are voting against. The current results show that 47 percent of those that will vote for Trump do so because “I don’t want Hillary Clinton to win.” On the other side, 46 percent of those voting for Clinton will do so because “I don’t want Donald Trump to win.”
First, I find the power of statistics and sampling amazing that, by sampling less than 600 people voting for each of the two (469 for Trump and 599) for Clinton, the results can likely be statistically valid within a margin error of less than 5.5 percent. The margin of error for those voting for Trump is 5.3 percent and 4.7 percent for Clinton, based on the sample size of each.
From a statistics perspective, the resulting likely outcome margin of error is predicated on several assumptions:
- The sample is truly random and representative of the overall population
- The estimator must follow a normal distribution – we’ve all heard of the bell curve. With a minor in statistics, I have been surprised to find out how much data are not normally distributed, which voids the statistical validity of the results and the corresponding confidence interval. Bet we have some this going on the jobs surveys.
So how does the U.S. government estimate total employment numbers and unemployment rates? There are two major surveys: one for overall employment (known as the Payroll Survey) and another for unemployment (the Household Survey).
The Payroll Survey sample includes approximately 146,000 private sector businesses and government agencies which covers about 623,000 total jobs. When put in perspective, with an estimated total 143.894 million jobs in the U.S., the sample size is just 0.43 percent of the total employed workforce. New firms are not included immediately in the Payroll Survey, so if you have, for example, a large increase in employment in relatively new firms, the Payroll Survey would miss those jobs. The BLS also tweaks the data using a birth-death model. Earlier this year the BLS stated that, given a growing workforce, 160,000 net new jobs are needed each month just to break even, so to speak. Each year about one-fourth of the businesses included in the sample are rotated out and replaced by other firms.
The Household Survey is a monthly survey of approximately 60,000 households responding. With an estimated 124.59 million total households in the US, the survey size is a miniscule 0.048 percent of the total workforce– less than 1/20th of 1 percent. Overall, approximately 72,000 homes are included in the sample from 824 sample areas. When a home (household) is included for the first time in the survey, a worker goes onsite to collect initial data. After that, computer assisted telephone surveys are completed monthly interspersed with periodic personal interviews by data collectors.
To read the latest details on these two surveys conducted by the Bureau of Labor Statistics click http://www.bls.gov/web/empsit/ces_cps_trends.pdf
For an overall summary of these methodologies click http://www.bls.gov/bls/empsitquickguide.htm
As both of these survey methodologies can yield noisy data results, other statistical smoothing techniques are employed.
So what can I say about May 2016 employment data? I do not believe them. Let’s wait a month and see what changes. Just look at the other strong economic data summarized in a Reuters article http://www.reuters.com/article/us-usa-economy-idUSKCN0YV1D7
- Initial claims for unemployment fell 4,000 last week to a seasonally adjusted 264,000 versus an expected increase to 270,000
- Unemployment claims dropped 30,000 since rising to 294,000 in early May, with below 300,000 considered a threshold for a strong economy. The US has now delivered 66 straight weeks below the 300,000 hurdle rate – the longest streak since 1973
- Wholesale inventories notched the largest increase in 10 months in April, with economists now raising their forecasts for Q2 GDP growth
- Job openings hit a 10-month high in April, with layoffs falling to the lowest seen since September 2014
- The number of Americans receiving unemployment benefits fell to the lowest seen since October 2000 in the week ending May 28
These are all strong economic indicators, not at all in synch with just 38,000 new jobs in May. After all, how can the total number of new jobs rise by just 38,000 and the unemployment rate shrink? My gut reaction is sampling issues and methodologies.
And for those that believe in conspiracies, I do have to admit that you might have a point given recent issues and lack of transparency at the Justice and State Departments and the IRS.
As for the jobs numbers, let’s wait a month and see what surfaces. So much for the No-Jobs May 2016 Jobs Report.
Ted