Two things to share today. We have new data. And we also have a response from the Bureau of Labor Statistics that raises some new, important concerns about the reliability of the data.
Monthly change in the labor force: -1,900
Monthly change in the number of unemployed: -21,100
Monthly change in payroll employment: +19,200
So this is good news for North Carolina. Although year-over-year numbers continue to point to the view that labor force dropouts have caused the sharp drop in unemployment, December's numbers (above) were not favorable to that view.
I heard from Patrick Carey, a branch chief at the U.S. Bureau of Labor Statistics' LAUS division. His email, verbatim and in full, and then my own remarks follow:
At 60,000 households, the CPS sample size is not large enough to yield reliable monthly estimates below the national level, so the LAUS program uses signal-plus-noise models to produce estimates for states. For example, North Carolina only has about 1,200 households in the monthly CPS sample, which would result in very volatile estimates if we relied on the CPS alone. In brief, we have bivariate models for states that use (1) payroll employment estimates from the Current Employment Statistics survey and (2) unemployment insurance claims counts from the state workforce agencies as secondary inputs to mitigate the volatility of the monthly CPS-based employment and unemployment tabulations at the state level. For more information on the LAUS modeling procedures, see chapter 4 of the BLS Handbook of Methods and the technical paper Model-Based Labor Force estimates for Sub-National Areas with large Survey Errors.So this is sort of crazy. The monthly CPS sample size for North Carolina -- 1,200 -- is close to a Gallup tracking poll. There are good reasons why we don't do statistical inference with that little data. And now I see where the LAUS data comes from. It's an estimate calculated from the CPS data with adjustments, not an independent data set.
We can’t speak to error measures on the CPS data that the authors processed and seasonally adjusted themselves for the population ages 16–65 years old (note that our universe is 16+). We can say of the seasonally adjusted LAUS data for North Carolina that the approximate thresholds for statistically significant changes (90-percent level) over a 6-month period (using the change from June-December 2013) are as follows: 68,000 for civilian labor force; 61,000 for employment; 32,000 for unemployment; and 0.7 for unemployment rate. So, our official estimates indicate that unemployment (both in terms of the level and rate) in North Carolina has gone down significantly since June, but neither the labor force nor employment level changes were significant. We produce and publish standard errors for 1- and 12-month changes here for state employment and unemployment levels and unemployment rates, but not labor force levels. For internal, diagnostic purposes, we do have standard errors for state labor force, employment, and unemployment levels and unemployment rates for 1-, 2-, 3-, 6-, and 12-month changes, but not 5-month changes as appear in the paper.
Again, these error measures are for LAUS estimates. We caution against the application of our internal error measures to the researchers’ own dataset. Also, please note that we will be revising LAUS estimates for the latest 5 years on February 28th as part of our annual processing cycle, at which point the standard errors calculated for the 2013 forward estimates will no longer be applicable, nor will they be available on a revised basis.
I'd also raise some concerns about the reliability of LAUS data from what Carey says. We know that the payroll numbers from CES (which has a larger sample size than CPS) show stable growth (graph here). However, the other input -- unemployment insurance claims -- raises questions when you're trying to estimate the effect of unemployment insurance cuts on the labor force. I've emailed Carey back to ask if that could be introducing a bias into the data -- it's not every day that a state cuts the weekly number of claims in half, after all.
Update (1/24): Marcus Hagedorn, Fatih Karahan, Iourii Manovskii and Kurt Mitman have a new review of the labor data on North Carolina. They find that data from the "household survey" do not show the same sharp contraction in the labor force that is found in the "local-area unemployment statistics" data.
That's interesting to me, as I hadn't seen the household-survey labor-force numbers until yesterday. The North Carolina labor-force data we have seems insufficient to come to strong conclusions. I should have been clearer on that point in this piece and regret that. I do still expect to see a drop in labor-force participation nationally and in North Carolina as the result of the expiration of unemployment insurance, but what I need to revise is that the data do not yet show this with sufficient certainty. The contradiction of the two data sets has put me back in "wait-and-see" mode.
Kan Zhang of the Brookings Institution notes via Twitter that the household survey data they use are not considered reliable for inference regarding the state-level labor force. It turns out that the varying sources and specifications of obscure state-level labor market data are not my area of specialty, so I've sent in a few questions to the nice people at the Bureau of Labor Statistics. I'll post their response when I receive one.
As we get more data on the labor force in North Carolina and at the national level, I'll come back to this issue and see if we can get any clarity.
Update (12/20): We have new data from the Bureau of Labor Statistics. In one graph, here's why I am standing by my contention that the cuts to unemployment insurance are not helpful to employment growth and are likely doing some damage to labor-force participation.
Note: Blue line = unemployment rate in North Carolina. Red line = year-over-year percentage change in N.C. labor force. Green line = year-over-year percentage change in N.C. payroll employment.
Ok, I guess I'll respond to Scott Lincicome's piece. I have to give him credit for going through the data. Let me just run through my comments:
1. I have been very careful, in this most recent post as in others (1, 2), to note the limitations of this analysis. Trying to get a sense of the "treatment effect" here is just not always going to lead you to solid conclusions.
2. It's true that the labor-force drop began before the forfeiture of federal support for unemployment insurance. It's also true that it has continued, and that the magnitude of the decline is unprecedented.
3. The data are not strong enough yet to suggest that cuts to unemployment benefits are going to cause huge declines in labor-force participation. I'm not saying that, and maybe one place Scott might be mistaken is in reading a stronger version of my position than I really am taking.
4. So, then, what is my position? It might be helpful to restate. I don't think that, given high unemployment and few job openings, that unemployment compensation is a meaningful disincentive against work search. Further, I think that it is likely that, on the margin, the "active search" requirement is more powerful in keeping people in the labor force than is desperation.
5. I wish Scott would look at the drop in initial and continuing claims. What is so striking about his piece is the idea that, so long as it appears to him that employment is doing OK, then we really shouldn't care at all about the hollowing-out of this part of the safety net. That hollowing-out is by far the most significant consequence of these cuts. Keep that in mind when he talks about how I'm wrong to claim "abject despair" in the state. North Carolina's UI program is half the size it was in July and may be still shrinking. I think that's both bad economic policy and a moral outrage.
6. Let's not get into an argument about what the household survey data say. I really don't like explaining the standard errors of various government datasets are. I had to do that over the summer on the issue of part-time employment and Obamacare, and it's just not a fun game to play. The only thing I want to say is that the changes in N.C. employment are consistent with the national economic recovery.
7. Scott seems to think I just looked at one article on North Carolina and then threw it up as anecdotal evidence. This is not the case, as would be evident from the fact that all of the quotes besides the one attributed are original to my reporting. I spent maybe three hours on the phone with various people (not just those quoted) who would have some insight into the trends of the private side of the safety net in their state. They all -- each and every one -- said they've seen a significant jump in the demand for assistance following the loss of UI. Some of those people (Pringle, for instance) oversee hundreds of food banks and collect data, also.
8. This is not relevant to Scott's critique, but more to his argument that forcing people into "less-than-ideal jobs" is good public policy: Daron Acemoglu and Robert Shimer wrote a paper back in 1999 making a strong case that giving people the flexibility to hold out for good jobs contributed to productivity growth. It's just something to think about if "take a job, any job" is really what we should be doing.
9. I don't want to seem overly sensitive, but I can't help but be a little bit frustrated by the way that Scott writes about me. He wrote a similar response piece last week that said (wrongly, as he later conceded) I was naïve about Obamacare's rents for the healthcare industry. This week, it's that I'm swayed by confirmation bias. Now, I don't know Scott personally, and he doesn't know me. I certainly wouldn't write such things about him. I just don't think it's the way to go about things, and I'd ask him to be a bit more careful, and this tone thing is something I've noticed at The Federalist a bit more generally.
10. Let's review the argument we've seen from conservatives about UI. It's been that the UI tax is a meaningful disincentive against hiring, and the benefits are a meaningful disincentive for work search. The UI tax floats around one percent of payroll. The actual benefits replace on average about 40 to 50 percent of the earlier salary. I wrote this piece, and the ones before it, because I think this is a weak argument coming from conservatives with no empirical support. In fact, as I wrote, there's room to reason that cuts to UI could reduce labor-force participation due to the end of the "active search" requirement. The data on N.C.'s labor force, I'll agree with Scott, don't prove this yet. But I think we're now far away from the intellectual position that cutting unemployment insurance is a good idea.