A Helpful Mnemonic in the Minimum Wage Debate
I am working on a paper for the Fraser Institute on the minimum wage debate. Both Krugman and Daniel Kuehn stressed the importance of picking a “treatment” versus a “control” group, in order to see that the minimum wage really doesn’t have much impact on employment (at least for modest hikes). Of course the pioneering study in this tradition is the famous Card-Krueger paper, which argued that the 1992 minimum wage hike in New Jersey didn’t affect fast-food employment relative to the control group in neighboring Pennsylvania, where the minimum wage wasn’t raised.
Now I’ve been immersed (again) in this literature for weeks, and what you need to realize is that the prima facie regression results almost all show that yes, making unskilled labor more expensive causes employers to hire fewer unskilled workers. It’s only when you add sufficient “controls” that the result disappears.
In that light, on a hunch I decided to take 5 minutes and check FRED. Why look at this:
So, if you ever forget the year when New Jersey raised its minimum wage and thus gave the “natural experiment” for the famous Card-Krueger study, you can just look up to find the first time in over a decade when New Jersey’s unemployment rate surpassed Pennsylvania’s.
I know, I know, I couldn’t get this published in Econometrica. But this literature is chock FULL of misleading coincidences just like this, ones which all make it LOOK LIKE employers cut back on hiring when workers get more expensive, even though we have been told the empirical debate is settled.
(Also, less sarcastically, I realize there is a distinction between the unemployment rate and the level of employment. However, as I explain in this EconLib article, I don’t think we should so casually toss aside the welfare implications of rising unemployment.)
The unemployment rate in New Jersey seems to have surpassed that in PA in February, before the April minimum wage hike.
When was it announced then?
Presumably most business plans ahead to the best of their ability.
Card and Krueger even had partly found positive effects, would be nice if you can elaborate on those as well. I mean if you find positive effects this really starts to look somewhat fishy, doesn’t it? Or have we turned low skilled workers into Giffen Goods due to MW hikes?
skylien – so a simple Micro 101 monopsony model will give you a positive employment effect. Undergrad stuff. Very not fishy. Whether it’s *true* is a different question. I don’t think many people think you’re likely to get a positive effect, simply because that doesn’t pop up often. Maybe in New Jersey’s case, though, who knows. It’s a high wage region and minimum wages tend to be less burdensome there so it’s possible you might hit that sweet spot I guess.
Somebody ought to run the unemployment rates for, say, Mercer and Buck county for this period. Unless Bob thinks sticking with border counties is recklessly throwing in more controls, and that we’re better off with Pittsburgh in the mix.
Monopsony is hardly uncontroversial, and at best just an ad hoc explanation which you cannot prove anyway for this instance.
If I find a study that says replacing oil with sand in a gearbox was partly found to have a positive effect on the life time of the gear box, then this should put up a red flag naturally. I don’t see a problem with questioning that.
I’m just responding to your point that it looks fishy. It doesn’t. Very straightforward ideas explain it.
I don’t think you’re using “ad hoc” correctly.
I don’t know what you mean by “this instance”. Do you mean “given the data available for this particular paper”? Because there’s lots of empirical work identifying whether a labor market is monopsonistic or not.
Your conclusion about the alternative study makes sense to me but I have no idea why you think it has anything to do with the minimum wage and monopsony.
Well, I am sorry if I offended you with the word “fishy”, maybe I should have used “extremely unlikely to impossible”, so also the rest of the study should be doubted as well.
With “ad hoc” I mean that “monopsony” is just used as an opportune explanation no matter if you have any prove if it is true or not for those specific data points (instances) at which Card and Krueger said it was even positive, although the very fact that they found positive effects could be just as well flashing signals that something with the data or interpretation of the data is wrong (wrong controls and/or swamped by other factors).
Did Card and Krueger at least prove in your sense that whenever they found positive effects in their study it was due to monopsony? And does e.g. Bob accept the “empirical work identifying whether a labor market is monopolistic or not” as true that first monopsony is real and is having exactly the effects you ascribe to it? I guess not.
Strange, I am sure I did make copy and paste, but now instead of “monopsonistic” I have “monopolistic” in the text I quoted from you. I guess that was due to autocorrection…
Wait what is offensive about it?
My point is just that it’s not fishy (and I seem to mean what you mean by fishy). There’s very reasonable, standard explanations. They might not be right explanations but that doesn’t make it fishy.
I thought that you thought that I wanted to imply bad intent by using that word..
Well, maybe for you it is not but for me it is a fishy result for exactly the same reason as it would be in my gearbox example. I don’t accept monopsony as an “ad hoc” explanation.
Somebody ought to run the unemployment rates for, say, Mercer and Buck county for this period.
The FRED unemployment rate data for Buck and Mercer counties only goes back to 1990, so I couldn’t replicate the same time period. However, when I looked at the data for the 1990s the same swap occurs, albeit not as pronounced.
Weird. When I do it there’s not much of a difference and Mercer’s unemployment rate isn’t higher until three years after the minimum wage increase (and it’s lower for a year and a half after the increase): Can you share your data?
https://research.stlouisfed.org/fred2/graph/?chart_type=line&recession_bars=on&log_scales=&bgcolor=%23e1e9f0&graph_bgcolor=%23ffffff&fo=verdana&ts=12&tts=12&txtcolor=%23444444&show_legend=yes&show_axis_titles=yes&drp=0&cosd=1990-01-01%2C1990-01-01&coed=2015-05-01%2C2015-05-01&height=445&stacking=&range=&mode=fred&id=PABUCK5URN%2CNJMERC1URN&transformation=lin%2C&nd=%2C&ost=-99999%2C&oet=99999%2C&lsv=%2C&lev=%2C&scale=left%2C&line_color=%234572a7%2C&line_style=solid%2C&lw=2%2C&mark_type=none&mw=2&mma=0%2C&fml=a%2C&fgst=lin%2C&fgsnd=2007-12-01%2C&fq=Monthly%2C&fam=avg%2C&vintage_date=%2C&revision_date=%2C&width=670
It looks to me like Buck county was hit much harder by the 1990 recession. There’s a huge difference in the pre and post recession unemployment numbers for the two counties.
Unemployment was 1.3 pct higher in Buck when the law took effect. That gap was gone within 15 months.
I’m of course not the one arguing that an unemployment series gives you the effect of the minimum wage, but however you interpret it you need to interpret symmetrically with Bob’s graph. You haven’t said what you think of Bob’s graph, but if you’re just going to attribute it to the recession (which I think is fair – and that’s why we should control for pair-specific time effects like Dube, Lester, and Reich do) then you really need to look at Bob’s graph and say “it looks like NJ was hit harder by the recession”.
Like I said I think it’s fair to attribute the higher Buck unemployment rate to the recession (at least in part), but one thing this definitely doesn’t show is an observable negative impact on Mercer due to the minimum wage, and this comparison is definitely a better comparison than Bob’s (which includes all of western Pennsylvania).
Maybe we aren’t using the same starting point (April 92). Based on my graph link below, I think it’s fair to say that the gap narrowed considerably (buck almost caught up with Mercer) over the 18 months following the April 92min wage increase, which supports Bob’s position better than the graph he used.
No, I don’t think broad unemployment is a good measure of the policy impact, but I don’t think narrow fast food is either.
https://research.stlouisfed.org/fred2/graph/?chart_type=line&recession_bars=on&log_scales=&bgcolor=%23e1e9f0&graph_bgcolor=%23ffffff&fo=verdana&ts=12&tts=12&txtcolor=%23444444&show_legend=yes&show_axis_titles=yes&drp=0&cosd=1992-04-01%2C1992-04-01&coed=1993-10-01%2C1993-10-01&height=445&stacking=&range=&mode=fred&id=PABUCK5URN%2CNJMERC1URN&transformation=lin%2C&nd=%2C&ost=-99999%2C&oet=99999%2C&lsv=%2C&lev=%2C&scale=left%2C&line_color=%234572a7%2C&line_style=solid%2C&lw=2%2C&mark_type=none&mw=2&mma=0%2C&fml=a%2C&fgst=lin%2C&fgsnd=2007-12-01%2C&fq=Monthly%2C&fam=avg%2C&vintage_date=%2C&revision_date=%2C&width=670
Did you mean this graph?
https://research.stlouisfed.org/fred2/graph/?g=1vFQ
DK wrote:
and this comparison is definitely a better comparison than Bob’s (which includes all of western Pennsylvania).
This is one of the big issues under dispute though. I understand why you might think the set of *all* contiguous counties (that straddle a state border) would be better than a set of the same number of counties that weren’t contiguous.
However, that’s not the choice we face.
Look in this specific blog post example. *All* of NJ and *all* of PA were hit by the treatment/control effect. So I am looking at all of the “patients” if you will.
You, however, want to just look at two of the patients because they were fraternal twins.
Is that how you’d test a new drug? You’d have data with it administered to 100 people and placebo given to another 100 people, the treatment group has (on average) the effect the researchers expect, but you throw all that data out and just pick one person from each group because you think they are more similar?
Bob – I’m not sure it’s like a reduction from 100 cases to 1 case at all, but medical studies will absolutely reduce the sample size if treatment is non-randomly assigned so that the treatment group looks more like the comparison group. Medical statisticians use the same non-experimental techniques as economists after all.
But, with medical studies, you would not assume the results applied to the larger group. You might compare only patients with obesity, epilepsy or Her2 positive tumors but you would then only claim that the result is applicable to the same subpopulation.
Right you don’t draw inferences beyond the population you’re sampling from at least not without caution.
Threatening consenting adults with extortion if they don’t agree to trade at a particular minimum dollar rate is the kind of caution you’re talking about
Card and Krueger kind of, sort of make their study a monopsony model by using only the fast food industry. When you look at the economy, rather than cherry picking industries, a monopsony model in a state, or in a populous county, isn’t a very useful model.
“skylien – so a simple Micro 101 monopsony model will give you a positive employment effect.”
-As well as a positive unemployment effect.
“Monopsony is hardly uncontroversial, and at best just an ad hoc explanation which you cannot prove anyway for this instance.”
“…the very fact that they found positive effects could be just as well flashing signals that something with the data or interpretation of the data is wrong (wrong controls and/or swamped by other factors).”
E.Harding,
You know to answer this just with monopsony is like you find a man in his 30s lying on a park bench not moving and then you call the doctor and he determines without actually examining the person „Oh he is dead and died because of a stroke, would be perfectly possible according to Med 101, case closed“.
With the quite important difference that at least a stroke actually is a really uncontroversial cause of death. And even then it is obviously not a satisfying answer, the men could just be sleeping.
E.Harding,
Sorry, I realize that I misinterpreted your comment..
I remember the days when you had and used a “Daniel Kuehn” tag. What’s happened to us Bob?
Fixed! My omission of your name tag was pretty fishy, wasn’t it…
🙂
Interesting issue regarding the positive effect of minimum wage on employment, and the “backward bending” supply curve of labour.
http://fellowshipoftheminds.com/2015/06/13/unintended-consequences-of-new-seattle-minimum-wage-workers-requested-reduced-hours-to-stay-in-subsidized-housing/
So there’s a number of key points here:
* price controls: magical thresholds for access to government wealth transfer implicitly involve a price control
* employment metrics: all of the commonly announced statistics ignore changes in number of hours, but pay great attention to changes in number of employees.
* backward bending supply: it can happen for surprising reasons (more than just one reason).
Ah, the old “we tried to raise low-skilled workers’ wages but made them worse off because they actually face a greater-than-100-percent implicit marginal tax rate” problem. This is what happens when you apply every intervention someone ever thought was a good idea all at once.
Not so fast. You presume there’s a “problem” here, but if we instead look at it as government optimising policy to the statistics that get reported, then they are doing exactly what the incentive would suggest they should do.
More “employed” people (working 1 hour per week still counts as employed) means lower unemployment. When workers have an incentive to reduce the hours they work, employers must hire someone else to fill in the difference, even for just a shift here and there. The minimum wage just created a job.
Fantastic theoretical paper on the income effect and backward bending supply curves.
https://mises.org/system/tdf/rae9_1_4_2.pdf?file=1&type=document
What is mean’t by “doesn’t have much impact on employment?” Didn’t the CBO run a study which showed that a wage increase to $10.10 would cause around 500,000 job losses? Is that considered modest? Also, didn’t Neumark and a collegue do a survey of the MW literature and conclude that 2/3 of the studies show disemployment effects? That would seem to indicate that the empirical record is anything but settled.
Do you mean this?
https://mitpress.mit.edu/books/minimum-wages
It is really funny if you read the summerry of the Card and Krueger piece directly after it:
http://press.princeton.edu/titles/5632.html
Great that all did an amazing Job in showing the exact opposite of the other economist.
Don Boudreaux has a good post on this topic: http://cafehayek.com/2015/07/pooling-the-effects-over-time.html
The idea that monopsony can explain the real-world effects of the minimum wage is problematic. In the market process view (i.e. the more realistic view), the presence of monopsony creates a profit opportunity. Whether or not that opportunity is exploited depends on whether or not other state action inhibits (or prohibits) it.
Besides that, the US has had a federal minimum wage policy for several decades. It’s absurd to think employers don’t expect continual increases in the minimum wage every few years.
Any minimum wage law that causes just a single person to lose their jobs, is a just as totally unacceptable and immoral action to take, as would if many people lost their jobs.
If the cost of living keeps increasing too much for the poor, then the solution is more liberty, not less.
Excess Teen Unemployment
02/16/13 – AEI Ideas by Mark Perry [edited]
( http://www.aei-ideas.org/2013/02/lets-review-the-adverse-effects-of-raising-the-minimum-wage-on-teenagers-when-it-increased-41-between-2007-and-2009/ )
=== ===
The minimum wage rose 41% in three stages between 2007 and 2009. This had a disastrous effect on teenagers. The jobless rate for ages 16-19 increased from about 16% to more than 26% (10 percentage points). The overall US jobless rate also increased from about 5% to 10%.
The graph attempts to isolate the effect on teenagers by plotting “excess teen unemployment” the difference between the teenage and overall jobless rates.
=== ===
See the graph at the link. Excess teen unemployment closely traced increases in the minimum wage. Teens are a measurable segment of all people of low experience, knowledge, and productivity.
Nearly all economic models predict that the higher minimum wage reduces teenage employment, even beyond what you would expect in a recession
WSJ: “Studies confirm that when teens work during summer months or after school they have higher lifetime earnings than those who don’t work. Raising the minimum wage may inadvertently reduce lifetime earnings.”
Defenders of the minimum wage say it accomplishes a social good without much harm. They are wrong. The worst effect applies to people of low ability who may never earn more. They are thrown out of work or receive fewer non-wage benefits and accomodations until inflation lowers the real burden of the minimum wage.
Proposals to index the minimum wage to inflation would permanently unemploy people of low ability or who now benefit from non-wage accomodations which lower their productivity.
We should oppose the minimum wage on behalf of the low skilled, whether or not they are later able to learn more and advance in productivity.
I did something similar to what Perry did, but took a more conservative approach: a regression using the inflation adjusted minimum wage all the way back to 1948 (when the unemployment series begin) and prime age (25-54) unemployment. I tried various approaches: deflate using the consumer price index (which is probably inappropriate, labor is not a consumer good) a kind of pseudo gdp deflator (more or less the consumer price index, multiplied by the ratio of the gdp deflator to cpi (this is not exactly what I did, but what exactly I did would take too long to explain. On a timescale greater than 3 months, this is essentially what I did)) I also tried removing the means of all series first, though this didn’t make any real difference. In each case I confirmed that a higher real minimum wage correlates with higher youth unemployment. The effect I found was smaller than Perry suggests here, but I’m probably underestimating the effect because: A) by using prime age unemployment as a separate explanatory variable, I effectively assume prime age unemployment is not effected by the minimum wage, which is probably not true and B) Unlike Perry, I am assuming that the higher volatility of youth unemployment over the business cycle is not itself a consequence of the minimum wage, especially the strong tendency to increase it during recessions. These are conservative assumptions. However it appears that even with these assumptions: a one dollar increase in the real minimum wage (ie one “current” dollar) increases youth unemployment by about 1.2% points. Again I think this is a conservative estimate. If anyone’s interested I can send more details.
Andrew
Did you try estimating it in logs? I’d be interested in seeing results from different combinations of log and linear specifications.
The Excel plugin I’m using does linear regressions but not log regressions. If you want to play around with the data, I’ve uploaded the spreadsheet with some comments that explain what I did:
https://devoidofnulls.files.wordpress.com/2015/07/minimumwageyouthunemployment.xlsx
Can’t you just create a column of log values and then do the linear regression on that? Get the results and apply an exponential to bring them back again.
I sense that you never grew up with a slide rule around the house?
You caught me, I am of the Texas Instruments generation.
Also thanks for the tip, I did actually think that might be the correct way to do it but wasn’t sure.
A quick shot at it appears to produce similar results: the coefficient on the natural log of the prime age rate is about 0.626 and on the simulated gdp deflator deflated minimum wage is about 0.516-This translates to an effect of basically the same magnitude.
Thanks Andrew. Very interesting stuff!
The conservative UK Govt recently announced a significant increase in minimum wage over the next few years – from £6.50 now to £9 by 2020. They have out-done Labour, who promosed £8/hour by 2020 in the election campaign. The cost in jobs is estimated at 60,000. Tories do not usually support minimum wages, bit here there is something we could call the “benefit interaction effect”. Welfare payments subsidise employers who can pay lower wages that hey otherwise would. The answer may be to cut welfare – they are doing that as well. Maybe they think that the increased minimum wage is just anticipating the natural wage rise that would follow welfare reduction.
http://www.bbc.co.uk/news/uk-politics-33437115
“Maybe they think that the increased minimum wage is just anticipating the natural wage rise that would follow welfare reduction.”
If they think that then the only reason they do this is to gain votes at the next elections, else there is no reason to prescribe something via law if you expect it to happen anyway. In short they would be grubering the public.
Actually, it’s because they hate legal immigrants.
To Gruber is now a verb.