I then worried that maybe a formal model would disprove my thinking, so I decided to do a little modeling myself, enough to convince me that I am not crazy. Since it involves, math, it goes below the fold.

The intuition that I have is this. We have an industry (agriculture) that faces somewhat inelastic demand. A technology comes along that raises the average product of labor but lowers the marginal product of labor. Hence wages and employment fall in that sector.

I want to compare wages and employment in a pre-innovation and post-innovation setting. In the pre-innovation setting, I assume a linear relationship between output and employment.

1. Q = L

In the post-innovation setting, I assume a diminishing-returns relationship between output and employment.

1'. Q = dL^{1/2}

where d is a constant greater than one, so that for relevant values for the amount of labor, output is higher under 1' than under 1.

What is strange here, and perhaps a swindle, is that this is an *industry* production function. It is not the production function of a representative firm. A representative firm with diminishing returns to scale won't help me if the industry as a whole can scale by adding firms. Somehow, I want to imagine lots of competitive agricultural firms, but because land is a fixed factor, there are diminishing returns to labor for the industry as a whole. Maybe the best way to come to terms with this is to assume lots of entrepreneurs bidding to farm on the marginal plot of land. Thus, prices and wages are set competitively, and for a given level of industry output, everyone faces the same marginal cost.

Next, assume that labor supply is an upward-sloping linear function of the wage.

2. L = aW

where a is a constant

Next, let demand be given by a downward-sloping linear function

3. Q = b - cP

At this point, if you do not want to bother with the calculus and algebra, feel free to skip to the numerical example.

Assume that the industry is competitive, then the price is equal to the wage times the marginal product of labor, Q'.

4. P = WQ' = (L/a)Q'

where the second equality was obtained by solving the labor supply function (2) for the wage rate in terms of L and substituting.

Next, we need to find the marginal product of labor, by taking the derivative of the production function. For the pre-innovation case, this is

5. Q' = 1.

For the post-innovation case, this is

5'. Q' = (d/2)L^{-1/2}

Substituting (5) into (4) into (3) and setting equal to (1) gives, for each case respectively:

6. b - cL/a = L; L = ab/(a+c)

6'. b - cdL^{1/2}/2a = dL^{1/2}; L = (2ab/[2ad+cd])^{2})

__Numerical Example__:

Suppose that we let a = 1, b = 12, c = 3 and d = 4. Then, in the pre-innovation case, we have:

L = 3, Q = 3, W = 3

In the post-innovation case, we have

L = 1.44, Q = 4.8, W = 1.44

So, we have the desired result, namely that after the innovation, there is more output, but less labor input and lower wages.

This is a partial-equilibrium story. The (urban) consumers of our farm products are wealthier than they were before, and I have not shown what they do with this extra wealth. In the long run, I assume that they spend some of it on goods that can employ the agricultural workers. But in the short run, there may be nothing that the agricultural workers can produce that interests the urban consumers, so the urban consumers, too, may increase their leisure. At least, that is the story I aimed to tell in Who Will Write This Paper? And, of course, I have to translate "leisure" into involuntary unemployment, as I pointed out in Who Will Write This Paper, No. 2?. If you look at some of the references suggested by Tyler and Alex in their comments on that post, they talk about "indivisibilities" of labor, which is pretty much what I am talking about when I say fixed cost of labor. Basically, on your farm you would much rather have one worker for 10 hours than 10 workers for one hour each.

(As an aside, I suspect that in a slump, this fixed cost of labor improves the bargaining power of firms, lowering wages more than would be the case without such fixed costs. But I could have that wrong. It needs another modeling exercise.)

Note that if for some reason wages are sticky, then relative to the solution presented above the wage rate will be higher and employment will be lower.

Economists need to add the complexity to your modeling that in many lines of work workers are not homogenous. Some are useful hands that are effectively leverage for others, and technology sometimes comes around to allows employers to leverage the high value workers with technology rather than low value workers. Which impacts wages, profits, etc.

My guess is that within the college-educated category (and perhaps within the other as well), dispersion has also increased.]My impression of the engineers and scientists universe is that you are right on about that. Especially within larger companies. A fair slice of our work has been farmed out, some to small to mid-sized companies and an even bigger slice has gone overseas. Result is that some job categories have been eliminated and others down graded in value.

Sorry I can't critique your equations. The math looks simple enough but I have zero background in how economists model things. Maybe someday I'll take the time to learn a little bit about it.

First thought is that the probablistic methods commonly used in things like control systems and wireless signal design/analysis, could be readily adapted to economic models. But this is just off the cuff. Do economists typically learn enough math to crank those kinds of equations?

What technology produced the unemployment among auto workers, wall street bankers and house builders?

PS, in the ag example you have reduced the direct labor input with mechanization, but what about indirect labor? Labor had to build the tractors. And labor had to make the steel. And labor had to mine the iron ore.

That's why Hayek sometimes views it all as labor. Some labor is tied up for longer periods of time before the final product, so he divides labor into short term labor and long term labor.

But my main point is that no less labor is used in mechanized farming than before. Indirect labor substitutes for direct labor.

I'm confused. Aren't you modeling inelastic *supply* here, with your sqrt(L) output equation?

Also, d must be > sqrt(L) to imply 1' > 1, right?

Fundamentalist,

Re auto workers, etc.

There is a long-standing trend of automation displacing manufacturing labor, not only in the U.S. but all over the world. Manufacturing employment is declining in almost every country (even China, as I recall), while as manufacturing output goes up worldwide.

I suspect that the Internet is causing obsolescence in a lot of areas. The current excess supply of lawyers is really remarkable, for example.

Re your PS:

1. I said this was partial equilibrium, so I did not account for the production of capital for farming.

2. Still, on net, the labor requirement to produce a bushel of wheat should go down. Otherwise, why choose the roundabout production method?

3. A main point here is that I want to get away from the notion of homogeneous labor, particularly in the short run. So I

don'twant to think in terms of "it's all labor."4. I think that ten years from now we will be much better off because of technological change. In the short run, however, it seems plausible to me that some types of workers are going to suffer unemployment and low wages. That is what I am trying to illustrate here.

"What is strange here, and perhaps a swindle, is that this is an industry production function. It is not the production function of a representative firm."

No problem. Just replace "d" with "N**0.5", where N is amount of land. Constant returns to scale for the individual farm which can vary both labour and land, but diminishing returns to labour in aggregate, since land is fixed in aggregate.

Don't understand why you need nonlinearities and diminishing marginal output

Q = xL

L = aW

Q = b - cP

P = L/a Q' = L/a x

Now, technological breakthrough makes x -> X, X>x.

b - cL/a x = xL leading to L = b/(x(1+c/a))

L1 = b/(X(1+c/a))

W1 = L1/a

P1 = L/a X = b /(1+c/a) = P, so prices remain the same.

If you need "non-" story I would vote for nonequilibrium considerations, not necessarily nonlinear: it takes _time_ to adjust to the new equilibrium. In your recalculation story time factor cannot be hypotesized away.