Why So Many Jobs Are Crappy

I’ve been thinking about it. Work, being a core part of life, is meant to be interesting, engaging, and meaningful. Otherwise, why are we wasting our time on this planet? Yet, for many, work is not living up to its name. Work of the good kind is less and less on offer in the jobs being created. I’ve been reflecting on possible reasons why, and decided it’s really simple. The problem is not the jobs. It’s us. Most humans are simply not the kind of people a boss would want to hire.

Take yourselves as a case in point. I’m guessing you’re the kind of people who’d prefer to feel needed rather than expendable. Well, that kind of attitude won’t do. Bosses want to keep your wages down, and that would be harder to do if you were given opportunities to make yourselves invaluable and near on irreplaceable. Bosses need to keep their options open in case some of you get ideas about better pay and conditions, or just generally become ‘difficult’ or, dare I say, ‘bolshie’.

You know it’s true. A boss needs to be able to dump you at the drop of a hat. Maybe it’s to boost profits. Maybe it’s to cut costs. Or maybe it’s just because it feels good.

And a boss needs to be able to dump you without it having detrimental effects. There must be ready replacements, eager to crank it up, the moment you’re out the door. And if morale suffers, because the buddies you left behind miss you, the boss will want to send them packing too, and bring in a fresh batch of ready recruits.

Quite simply, there is little place for satisfying roles, the kind where you get to learn more and more interesting stuff over time. The only good on-the-job learning is no learning at all. Or, if you must learn, thirty minutes tops to master a dead-end role.

Although the point seems obvious, I don’t believe it has been treated with the kind of gravity that only the economics blogosphere is fully equipped to deliver. It’s high time the situation was spelled out in painstaking – okay, not painstaking – analytical terms. Then we can all lower our expectations and knuckle down to a lifetime of short-lived McJobs and frequent sackings. If we’re lucky.

One way to characterize a job is by the learning that occurs in it. This can be described by a learning curve:

Single LC

In the diagram, u(t) stands for the unit labor cost that a worker – let’s say you – achieves at a given point in time, after you have built up an amount, t, of experience. It is how much you cost the employer per unit of output produced, at a given moment in time. We can call this the ‘instantaneous unit labor cost’, or sometimes just ‘efficiency’ for short. If learning occurs on the job, you get more and more efficient, and your unit labor cost falls over time. The curve is drawn assuming a particular wage level. If the wage increases, the curve will shift up.

On first being hired, you were green, and cost the bosses m + c per unit of output. Eventually, through learning on the job, you will get this down almost to m, which is your ‘potential efficiency’, given current pay and conditions. Some jobs will allow more learning than others, which will be reflected in the amount c, which is the ‘scope for learning’.

Here is one possible algebraic representation of the learning curve:

LC Eqn1

The second term is the one that captures the learning process. When you just start the job, t = 0, and the second term equals c. Just as shown in the diagram, you cost the bosses m + c per unit of output. The longer you stay in the job, the larger t gets, and the closer the second term gets to zero, which it approaches asymptotically. Your unit labor cost converges on m, as indicated in the diagram.

The rate at which you reduce your unit labor cost from m + c down to m depends on how fast learning can take place on the job. The ‘rate of learning’ is represented by λ. When λ is large, the learning curve will be very steep initially, and almost all the learning will occur just after being hired. When λ takes intermediate values, the learning process is steadier and longer lasting, reflected in a more gradual curvature in u(t).

McJobs are those where learning is either nonexistent or extremely rapid but short-lived. If there is no learning, c = λ = 0 and the learning curve would just be a horizontal straight line showing a constant unit labor cost of m. Rapid but short-lived learning would mean the learning curve slopes down almost vertically until m is nearly reached, then stays almost flat after that. It would probably take most of us a half shift to master flipping burgers, but after that we’d have it down pat. Bosses love McJobs. They make us readily replaceable.

Satisfying jobs – let’s call them ‘good jobs’ – will generally be ones where learning occurs at a steady pace more or less indefinitely, probably as part of a defined career path. Bosses would prefer not to offer these, and will always be looking for ways to deskill roles that, for now at least, need to allow workers greater autonomy, ingenuity, and scope for on-the-job learning.

Once you gain experience in a good job, you will soon become much more efficient in the role than an inexperienced replacement would be. This might remain true even if you happen to win a pay rise, work less hours, or start operating at a more leisurely – let’s say human – pace. Any of these things would shift your learning curve up, because you would now have a higher unit labor cost at any given level of experience. Even so, you might still be more efficient than a prospective replacement.

In fact, let’s say you do win a pay rise, plus a longer lunch break. Thanks to your experience on the job, you have realized that you can afford to be more bolshie. The boss knows that you can’t be replaced without some cost, at least in the short term.

Your new situation is illustrated below. You have switched from your original learning curve, now called u2, to a higher learning curve, u1, after an amount τ of experience.

Incidentally, that greek letter tau is meant to look the same as the greek letter tau in the diagram. I have no idea why it doesn’t match. Probably some bolshie worker at Microsoft is the culprit. Typical.

Anyway, you are sitting at point A and the boss is not happy, since before your improvements in pay and conditions, you were on a point almost directly below it on curve u2. Extensions to the boss’s country house are in jeopardy thanks to your militant demands for higher pay and a full half-hour for lunch.

Learning Curves B

As it happens, your immediate boss is sufficiently annoyed to call in his bosses. They’re having a liquid lunch for the express purpose of dealing with the situation you’ve placed them in. They need to decide whether to kick your sorry butt out of the joint and find a replacement, somebody who will accept a lower wage and be subservient, like you used to be before you went all bolshie. The bosses want someone who is willing to be positioned on learning curve u2. Unfortunately, such a worker would have no experience, so would be back at point E at time zero.

Algebraically, the two curves look like this:

LC Eqn2

The diagram shows that m2 < m1. This means that a replacement worker’s potential efficiency would be superior to yours. It’s also possible that a replacement worker would learn a bit faster, now that you are insisting on operating at less than sweatshop pace. This is reflected in the equation for u2, where the rate of learning is vλ rather than λ, and v is assumed to be greater than 1.

Right at this moment, with the bosses plotting your downfall, you happen to have a lower instantaneous unit labor cost (at point A) than a replacement would have on first starting the job (at point E). But the bosses wonder whether, given enough time, a replacement would save the firm money and get the country house extensions back on track.

How patient are the bosses? In the diagram, they have a ‘time horizon’ of h. If a replacement would cost less over time period h than you would, the bosses will gleefully liberate themselves from your services. Given that amount of time, the replacement’s experience on the job would increase from 0 to h. You’ve had a ‘head start’ of τ, so you would be able to build up your experience from τ to τ + h.

The bosses can see that the replacement would get down to point F after time h on the job, which represents a lower unit labor cost than you would offer them at point B. However, for much of the period, you would outperform the replacement. What matters to the bosses is the total accumulated costs over the entire period. In the diagram, if area ABCD is larger than area EFGH, you will be on your bike. Otherwise, you’ll be safe, even though the bosses are itching to see the back of you.

It is self-evident that you are more likely to be fired the bigger the gaps in potential efficiency and scope for learning (the m’s and c’s), the larger the difference in learning rates (determined by v), and the longer the time horizon, h. On the other hand, the more experience, τ, you already have under your belt, the harder it will be for the bosses to get rid of you.

More interesting, though, is the influence of learning speed on the decision of the bosses. As indicated at the outset, bosses don’t like jobs that provide an opportunity for indefinite learning at an intermediate rate.

We can see this by looking more closely at the areas ABCD and EFGH. Call the first area C1. It represents your accumulated unit labor costs from time τ to τ + h. Similarly, the second area can be called C2, and refers to the corresponding measure for a replacement worker employed from time zero to h.


LC Eqn3

Let’s define the ‘firing payoff’, Z, as:

LC Eqn4

This measures the cost savings the bosses stand to reap if you are fired and replaced by somebody else. If Z > 0, the firing payoff is positive and the bosses will replace you. Otherwise, you win out, even though you’ve become intolerably bolshie!

The behavior of the firing payoff, Z, as the rate of learning, λ, is varied is always the same. For very low and high rates of learning, it is more likely that firing will pay off for the bosses. But for intermediate rates of learning, there is a greater chance that you will survive. What qualifies as ‘intermediate’ values for the rate of learning varies with the choices for other parameters, but the pattern is always the same.

The following graph shows the firing payoff as a function of the rate of learning when it is assumed that m1 = 100, m2 = 75, c1 = 200, c2 = 150, τ = h = 1, and v = 4/3.


For λ < 0.6779, Z > 0, and you are fired. For λ > 4.3889, again Z > 0, and you are hanging out in soup kitchens. But there is a sweet spot. You get to stay in the job with your better pay and conditions in between those two cutoff points.

Fortunately, the bosses weren’t the only ones able to zero in on the behavior of the firing payoff. Unbeknownst to them, you had already made the appropriate calculations before formulating your most recent demands for a pay rise and longer lunch break. You had the bosses right where you wanted them. They were screwed from the outset, and didn’t even realize it. No wonder they would prefer to eradicate all the good jobs and leave us all fighting for scraps in McDonalds.


I tried to keep the model as simple as possible without losing the key relationship between the firing payoff and the rate of learning. Various extensions are possible, but they only make the situation worse for the bosses when it comes to good jobs.

One obvious extension, though not very interesting, is simply to introduce a discount rate. If the bosses value cost reductions in the present more highly than cost reductions in the future, it is appropriate to include a positive discount rate. This doesn’t alter anything much other than to make being bolshie even more likely to pay off.

Another extension is to take into account that the traits of the replacement worker are not completely known. The bosses will worry that a replacement will become bolshie, just like you did, before your removal has a chance to pay dividends. As with a positive discount rate, incomplete information concerning the traits of a replacement worker stack the odds more in favor of the incumbent.

The incompleteness of information adds an additional wrinkle, which is that now even large gaps between potential efficiencies will not necessarily hurt you as the incumbent. The reason is that there will now be a chance that the replacement will simply approach the same unit labor cost as you. Basically, for a big gap in potential efficiencies to matter, the probability of the replacement becoming bolshie can’t be too high.

So, next time we hear people complaining about the crappiness of their jobs, we can sooth them with the knowledge that it is better for the bosses that we all remain unchallenged and expendable.

Background Reading

The idea for this post comes, in a roundabout way, from a 1981 paper by A. M. Spence entitled, ‘The learning curve and competition’, which was published in the Bell Journal of Economics. In the paper, Spence argues that early entrants into an industry will be more likely ultimately to dominate the market when production is subject to a learning curve.

The idea is also linked, again in a roundabout way, to a 1994 paper by Flora Gill entitled, ‘Inequality and the Wheel of Fortune: Systemic Causes of Economic Deprivation’, and published in Australian Economic Papers.

A different application of the argument presented here would be to consider the circumstances under which instances of bad luck in competition for employment could have long-lasting effects on the earnings of individual workers. Workers who luckily gain appointment into good jobs, perhaps initially contrary to merit, will get the opportunity to learn on the job and lock in their advantage, whereas unlucky workers consigned to bad jobs, again perhaps contrary to merit, might have a tough time reversing their fortunes.

I like the present application, though, in which capitalist deskilling emerges partly as a response to the effects on worker discipline that are potentially created by on-the-job learning.