What happened to the distribution of real earnings during the recession?

[Note: I noticed an error in some of the graphs a few minutes after publishing the original post.]

[Update: Everyone should read Kevin Milligan's take using the SLID data. I'll get back to this in another point soon.] 

I noted on Econowatch at Maclean's a few weeks ago that average and median real earnings increased over the recession, and the point was surprising enough to me that I decided to dig deeper. I was also inspired by this post that reproduced a couple of fascinating graphs from this paper by David Green and Benjamin Sand plotting real wage changes against wage percentiles. This way of representing data is so good that I can't help but steal it, and that's what I'm going to do here. Their paper stops with the 2006 census, and since the 2011 National Household Survey data are unusable (thanks again, Tony Clement!) so I've decided to work with the public use microdata files from the Labour Force Survey instead. The weekly nominal wage data are constructed by multiplying reported usual weekly hours by the reported usual hourly wage. The real earnings data are the nominal earnings divided by the CPI.

LFS earnings data are available starting from 1997, so I'll break down the sample into three 5-year blocks: 1997-2002, 2002-2007 and 2007-2012. That last block covers the recession. Here are is the graph of the per cent change in men's real weekly earnings over those three intervals: 

The data for 1997-2002 and 2002-2007 look like the Green-Sand graphs - growth in the top few deciles, and no growth elsewhere. The data after 2007 have the same pattern in that the highest growth rates are in the top part of the distribution, but are different in that lower percentiles saw measurable increases in real earnings.

Here are the data for women:

The vertical axis is the same as in the previous graph. 

Canadian women seem to have done reasonably well during the last five years.

And here are the data for both sexes, still with the same vertical axis:

[Frances can take credit for the fact that I broke out the wage data out by sex before looking at the total. It wouldn't have occured to me to do so in the early days of WCI.]

That last graph is particularly interesting in light of the discussion about whether or not the middle class is 'hollowing out'. The middle part of the distribution saw the slowest growth, but if the trend over the last five years continues, the income distribution will be more concentrated in the middle as those in the lower part of the distribution catch up to the middle. (I've done the simulations.)

Finally, here is that last graph for the periods before and since the Conservatives came to power. This isn't really all that interesting as far as the economics goes, but it's the sort of thing that should be better-known in political circles:

 

[Click here for a version with the same vertical scale as the other graphs.]

I don't think that the Conservatives can claim much credit for the broad-based increase in real earnings that have occurred since they came to power - much/most of it can be attributed to the Bank of Canada's failure to keep inflation from drifting below target. Lower-than-expected inflation means higher-than-expected real wages.

On the other hand, if the opposition parties want to use the theme of stagnant wages to attack the Conservatives, they would do better to look at the data first.

Comments

[Nick Rowe]

My first thought was: so real wages increased faster during the recession than previously. OK, so that confirms that wages are stickier than prices.

But there's something a little strange about the third and fourth graphs. The fourth graph shows 2006-2012, which isn't very different from 2007-2012. But the blue 2006-2012 graph is so strongly above the red 1997-2006 graph. Was there a big real wage increase in 2006-2007?

[Stephen Gordon]

Well, the vertical scale *is* different. But I think it's more that the pre-2002 data are dragging down the 1997-2006 data.

eta: I just added a link to a graph with the same vertical axis.

[rsj]

This is why more charts should be presented as animated gifs, one curve or a 3 yr moving average, per second, with the year flashed at the bottom right as you see the distribution evolve in time.

[rsj]

Stephen, did the micro-data you cut-off very high and low earnings? Whenever I looked at PUMS I had to discard the ends because of this.

[Joe]

To avoid misunderstandings: This is about wages, right? So unemployed drop out of the distribution? Then isn't the increase in the lowest percentile possibly driven by, say, very-low income workers becoming unemployed during a recession so that the rest of the former distribution is stretched over the whole range now?

[Robert McClelland]

Does this include pension and benefits?

[Stephen Gordon]

rsj: I know the census PUMF truncate income at the top, but I don't recall seeing anything about that in the LFS documentation (it may be, I just don't remember seeing it). But even if they aren't, the point is still well-taken. The numbers at the extremes - say, the top/bottom 2% or 3% - should definitely be taken with a grain if salt. But the real story IMV is what's happening across the entire distribution.

Joe: Yes, it's about wages.

Robert: No. I wanted to see what was happening in the labour market.

[Kevin]

Hi,

Green and Sand do the following:

- fulltime workers only
- weeks worked between 1 and fullyear
- total income>0
- divide annual earnings by weeks to get weekly earnings
- use log(weekly earnings)

The picture looks quite different when you actually do what Green and Sand do. Myself, I would be much more hesitant to make definitive statements in comparing some LFS results using a different measure to what Green and Sand did.

I will post some charts using the SLID shortly.

[Stephen Gordon]

I wasn't actually trying to reproduce the Green and Sand results, especially insofar as they only looked at full-time workers. The reduction in earnings from cutting back on hours worked by part-time workers during the recession is something I wanted to take into account.

[Kevin]

Here is my post:
http://blogs.ubc.ca/kevinmilligan/2013/12/15/earnings-growth/

Three conclusions:

* Median earnings growth is greater than zero for males since 1996. This is good news.

* Earnings growth continues to be skewed, in particular in the top 10 percent. Male earnings polarization appears to be continuing up to 2010.

* It is important to use similar data and methodology when comparing results to those in the literature.

[Kevin]

Hi Stephen. What's the difference between this one you just posted
https://twitter.com/stephenfgordon/status/412313038520197121

and the ones in the blog post above? Thanks.

[Stephen Gordon]

The graphs above are the growth rates of a given percentile over 2 points in time: eg, change in 75th percentile between 2007 and 2012.

The graph I posted on twitter took the average of a given percentile over 1997-2000 and compared it to the average of the same percentile over 2006-10.

[Kevin]

Hi,

Are you averaging the percentage growth, or averaging the earnings then calculating the percentage growth?

I just posted here https://twitter.com/kevinmilligan/status/412328192527958016 2010 vs 1996, no 5-yr averaging. Still see big take-off of earnings at the top.