Wednesday, September 20, 2017

The asterisk in real median household income

 - by New Deal democrat

This is a follow-up to the post I wrote last week about the latest data on real median household income.

One of the things I notes is that "households" includes the millions that are composed of retirees, a burgeoning demographic due both to healthier longevities and the demographics of the Boomer generation.

This morning Jared Bernstein helpfully includes a graph of real median household income excluding those over age 65:

Households headed by working age adults did finally surpass their 2007 income, but were still 3.4% below the all-time highs of incomes of 2000.

But mainly I wanted to follow up on that break in the graph in 2013.  It was caused by a change in methodology by the Census Bureau.

Here's the graph I ran last week of real median household incomes at various quintiles and deciles:

So I was surprised a few days later to see another, more pessimistic graph floating around, purporting to show that only the top 20% of households had higher incomes than in 2000:

Note that this graph doesn't have any break. 

I traced the information back to its source, which turned out to be the Economic Policy Institute. And at the bottom of their article, I found a footnote explaining thus:

In other words, as best I can tell the E.P.I. simply took the 3.2% difference in the two methodologies in 2013 and projected it forward.

Now, let me state right up front that I don't know whether the data as represented by E.P.I. is correct or not.

But, neither does the E.P.I.

In fact, neither does anybody else, apparently including the Census Bureau itself.

That's because the Census Bureau hasn't provided data in any year since 2013 as to what the numbers would have been under their old methodology, so that we can form a basis of comparison.

By contrast, when the methodology for "real retail sales" was updated in the 1990s, the old data series continued to run for nearly 10 years, giving us an excellent basis for comparison, and confirming that the YoY changes were virtually identical, meaning that we can stitch the two series together confidently, giving us 70 years worth of data:

The Census Bureau didn't do that with real median household income, so there will *always* be a disconnect and a lack of ability to reliably and directly compare data from before 2013 with data afterward.  We'll always be guessing. 

This is a major problem for this important data series, and the Census Bureau should take steps, to the extent possible, to correct it.

Tuesday, September 19, 2017

Hurricane workarounds for industrial production and housing

 - by New Deal democrat

Hurricane Harvey has already affected some of the August data releases.  Irma has already started to affect some weekly releases, and will undoubtedly affect the September monthly releases.

I have already begun to adjust for the hurricanes in the case of initial jobless claims.  But what of the monthly data?

While there is nothing so timely and precise as backing out affected states from the initial jobless claims report, there are workarounds that can at least tell us if there has been any significant change in trend for both the industrial production and housing reports.

I will put up separate posts, but to cut to the chase, we can use the Regional Fed reports (minus Dallas, and adding the Chicago PMI) to give us a reasonable estimate of industrial production in the non-hurricane affected areas. Similarly, we can make use the regional breakdowns in the housing report by subtracting the South and determining the trend in the remaining 60% of the country outside of that census region.  I have already looked at this morning's housing report, and it turns out the effect is not what you would think!  I'll have that post up by tomorrow.

Unfortunately there is no regional or state-by-state breakdown of retail sales or regional consumption expenditures on any sort of timely basis, so we're kind of stuck there.

Saturday, September 16, 2017

Weekly Indicators for September 11 - 15 at

 - by New Deal democrat

My Weekly Indicators post is up at

The effects of Hurricane Harvey have shown up in a number of data series, but the underlying trends are intact.

Friday, September 15, 2017

2.5 cheers for 2016's new high in real median household income!

 - by New Deal democrat

Given that I consider jobs and wages for average Americans my #1 focus, it's only fair that I write about this week's release of the real median household income for 2016, don't you think? 

A few years ago I wrote that real median household income was the most misused statistic in the entire econoblogosphere.  That's because: 

  • it is NOT a measure of real wages.  It includes all income -- things like pensions, dividends, and social security.
  • it includes ALL households, not just those headed by workers. A household of full time students is included.  Your 85 year old grandparents are included. A household where one or more persons is unemployed is included.  A household where one spouse stays home to raise the children is included.
  • it is subject to distortions from demographics.  In particular, since retirees tend to have only about half the income of households headed by wage or salary earners, as the percentage of households headed by retirees rises (due to both healthier longevities and Boomer retirements), downward pressure is placed on the median. Below is the graph of current US demographics, showing the barbell effect of young Millennials and old Boomers bracketing the smailler Gen X:

Another problem with the metric is that it is only reported once a year, and with a nine month delay to boot. So this week's data release tells us what household income was for the period of 9 to 21 months ago! Not exactly timely.

Fortunately, there are several ways to get more timely data.

First of all, a decent back of the envelope estimate can be made by taking average hourly wages, which are reported monthly, multiplying by the number of hours worked, also indexed monthly, and dividing that by the entire population age 16 and over. It's an average, rather than a median, and it won't give you an exact number but will give you the overall trend.

Even better, Sentier services calculates an estimate of actual median household income monthly based on the Household Survey reported each month by the Census Bureau (that's the report that gives us, e.g., the unemployment rate). Doug Short has done excellent work putting this in graphic form each month.

Back before the big decline in gas prices had the result of driving real, inflation-adjusted wages to new highs, I would constantly read how real median household income showed that wages were actually declining. That was hogwash.  Now that real wages are at new multi-decade highs, there's a whole new round of hogwash!

But, enough background. Let's turn to the actual numbers.  

As you no doubt already know, real median household income at long last made a new all-time high in 2016, finally surpassing its previous highs in 2007 and 1999:

That is unadulterated good news.

Further, the official Census Bureau number was very much in line with Sentier's monthly reports for 2016. Here's Doug Short's aforementioned graph:

The average number for the 12 months of 2016 was higher than any previous year. This is the second year in a row that the very timely monthly reports by Sentier showed strong increases in real median household income, and were later confirmed by the official numbers. In other words, Sentier has a very trustworthy record and you should be paying attention if you care about this metric.

Digging a little deeper, here is how real median household income as broken down by quintiles, as well as the top and bottom 10%, since its previous 1999 peak (h/t Wall Street Journal):

The bottom two quintiles have still not made up all of the ground lost since 2000, although the top 50%-60% have.  On the contrary, the very top 10% have been doing extremely well (which has led to some earth-shaking political consequences on both the left and right).

Another legitimate caveat is that the Census Bureau changed its methodology 3 years ago, and regrettably has not published its results for both methodologies in order to create a baseline correlation for future use.  We know that three years ago, the change in methodology made the number *for that year* about 3.2% higher. Hence the break in the lines at that time period.  Is there a similar issue with regard to this year? Nobody knows for sure. But, yes, because of this we do have to put an asterisk next to the claim that 2016 was a new all-time high.

Yet another legitimate caveat is that real median household income for men only is still below its peak from the early 1970s (h/t Mike Shedlock):

All of the peaks since then, including the new peak for 2016, are in part because women entered the workforce, and thus the income of households where both spouses work has gotten higher.  But of course this is a real tradeoff if one of the spouses would prefer to stay home and raise the kids, but feels they have to work to make ends meet.

It is also perfectly fair to point out that the new highs in income last year don't do anything to erase the many years of lesser income in particular since the beginning of the Great Recession.  This means that median household *wealth* is still below what it would have been had we been at or near full employment for all of those years. In other words, the long-term well being of the median American household is still compromised.

Finally, it wouldn't be a true positive report without its very own batch of Doomer hogwash.  This year's hogwash is that the increase only happened because people had to work longer hours.  I'm not sure who originated it, but here are two links:  here and here, with the relevant quote:
[T]he Census Bureau data show that the bulk of the gains in real income in 2016 was explained by one factor: higher employment. In other words, hours worked rose but wages did not. The members of American median households are working harder at more jobs to finally get an increase in incomes.  
In other words, hIs logic runs: 
1. real median household income mainly rose because of more employment.
2. that means hours worked rose.
3. that means people were working more hours.
4. that means that household income really only increased because people had to work more hours for the same hourly pay.

Did you see what he did there?  He backed out the fact from line 1 when he got to line 3. Line 3 is true  in the aggregate, but he treated it as if it was true for most individuals in the set. The true line 3 is as follows:
3. that means that aggregate hours increased, due to unemployed people finding work, part-timers getting more hours, and some part-timers getting full-time employment.

Oh, that's a little different.

The bottom line is: 2.5 cheers for the new high in real median household income for 2016. That a majority of US households are earning more than that group ever did before is great news.  But, as evidenced by the last 15 years of this statistic, it by no means erases the long-term decline in the health of what used to be called the American middle class.

Thursday, September 14, 2017

Hurricane adjusted initial claims for week of Sept. 2: 239,000

 - by New Deal democrat

Last week I promised I would repeat an exercise I first undertook in 2012 when Superstorm Sandy disrupted the initial claims data: estimating what the initial jobless claims would have been, but for the hurricane.

In 2012 I created that adjustment by backing out the affected states (NY and NJ) from the non-seasonally adjusted data.  That gave me the number of initial claims filed in the other 48 states.  I compared that with the same metric one year earlier, and multiplied by the seasonal adjustment.

What that does is give me the number if the affected states had the same relative number of claims during the given week, as all of the unaffected states.  In 2012, it showed that Sandy was not masking any underlying weakness in the economy.

The state by state data is released with a one week delay.  So what follows is the analysis for the week of September 2, the number for which was reported one week ago. This week I only had to back out Texas.  Next week I will undoubtedly have to back out Florida as well.

Here is the table for the Week of September 3 in 2016 vs. September 2 this year:

Metric                              2016                   2017
Seasonally adjusted:       257,000              298,000
Adjustment for total:       1.18%                1.19%
Not seasonally adjusted: 217,715              250,621 
Texas claims:                     15,707                63,788
NSA claims ex-TX           202,008              186,833
TX as % of total:              7.2%                   n/a
2017 w/ TX adjustment:  n/a                      201,405

If we use the 2016 weekly seasonal adjustment of 1.18% for the adjusted 201,405 total, this gives us ~238,000.

If we use the 2017 weekly seasonal adjustment of 1.19% for the adjusted 201,405 total, this gives us ~240,000.

Thus the hurricane-adjusted initial jobless claims number for the week of September 2, 2017 is 239,000.

The underlying national trend in initial jobless claims remains very positive.