Forthcoming Article
l,' Introduct1on
TBE TBE0RY 0F I
0RIHIRATI0R
by
Kenneth J. Arrow
Te fact that d1fferent roups of workers be they sk1IIed or
unsk1IIed bIack and wh1te or maIe and femaIe rece1ve d1fferent aes
Inv1tes the expIanat1on that the d1fferent roups must d1ffer accord1n .
to some character1st1c vaIued on the market. In standard econom1c theory
Mo th1nk f1rst of aII of d1fferences 1n product1v1ty. The not1on of
dIscrU1nat1on 1nvoIves the add1t1onaI concept that personaI character1st1cs
of the worker that are unreIated to product1v1ty are aIso vaIued on the
market. 8uch personaI character1st1cs as race ethn1c background and
sex have been freguentIy adduced 1n th1s context.
01scr1m1nat1on 1n th1s paper 1s cons1dered onIy as 1t appears on
the market. 0bv1ousIy one can have d1scr1m1nat1on 1n the same sense
whenever dec1s1ons are made that concem other 1nd1v1duaIs nameIy that
the1r personaI character1st1cs other than those properIy reIevant enter
1nto the dec1s1on. BeI1berate rac1aI sereat1on and d1scr1m1nat1on 1n
entrance to schooIs and coIIees depr1vat1on of the r1ht to vote aIon
soc1aI and sexuaI I1nes and d1scr1m1natory taxat1on are aII exampIes
of d1scr1m1nat1on 1n nonmarket d1scr1m1nat1on.
It may as weII be adm1tted that the term "d1scr1m1nat1on " has
vaIue 1mpI1cat1ons that can never be compIeteIy erad1cated thouh they
can be ster1I1zed for spec1f1c emp1r1caI and descr1pt1ve anaIyses.
2
I have spoken of personaI character1st1cs that are "unreIated to pro
d0ctIv1ty" and not "properIy reIevant. " These terms 1mpIy def1n1t1ons
ct pr0duct and of reIevancy wh1ch are themseIves vaIue ] udments or at
aa
7
rate dec1s1ons by the schoIar. The bIack steeI worker may be thouht
of as produc1n bIackness as weII as steeI both evaIuated 1n the market
,Ve are s1nI1n out the former as a spec1aI sub] ect for anaIys1s because
somehow we th1nk 1t appropr1ate for the steeI 1ndustry to produce steeI
and not for 1t to produce a bIack or wh1te work force.
Bowever the vaIue ] udments are 1ntr1ns1c onIy 1n determ1n1n wh1ch
wae d1fferences we reard as worth study1ng as an exampIe of d1scr1m1nat1on
1n the emp1r1caI or theoret1caI anaIys1s of any form of d1scr1m1nat1on
once spec1f1ed.
In the foIIow1n I w1II address myseIf spec1f1caIIy to rac1aI
d1scr1m1nat1on 1n the Iabor market. For the most part the anaIys1s
extend w1th no d1ff1cuIty to sexuaI d1scr1m1nat1on w1th some reserat1ons
to be noted. The other markets 1n wh1ch d1scr1m1nat1on has been most
observed, espec1aIIy hous1n but aIso 1nsurance and cap1taI are anaIyzed
by the same eneraI methods but the operat1on of these markets has Ied
to a reater deree of s1mpIe excIus1on and Iess of pr1ce d1fferent1aIs.
The bas1c a1m here 1s to use as far as poss1bIe neocIass1caI tooIs 1n
the anaIys1s of d1scr1m1nat1on. As u1II be seen even thouh the bas1c
aeocIass1caI assumpt1ons of ut1I1ty and prof1tmax1m1zat1on are aIways
reta1ned many of the usual assumpt1ons w1II be reIaxed at one po1nt or
another: convex1ty of 1nd1fference surfaces costIess ad]ustment
3
are
the eopIoyer ho m1ht sacr1f1ce prof1ts to reduce or eI1m1nste
empIoyment 1n h1s pIant or the k orkers ho m1ht accept a Ioer ae to
vork 1n a pIant 1th more k and Iess orkers. (It 1s aIso poss1bIe for
products soId on a facetoface bss1s that customers m1ht d1scr1m1nste
by be1n 1II1n to pay h1her pr1ces to buy from h1tes th1s case couId
be stud1ed aIon s1m1Iar I1nes but 1II not be desIt 1th here. ) Rot
aII d1scr1m1natory feeI1ns csn f1nd express1on 1n the msrket sn
entrepreneur ho has a d1staste for compet1n aa1nst f1rms 1th orkers
has no ay 1th1n the econom1c system at Iesst of express1n h1s tsstes
and therefore of 1nfIuenc1n se IeveIs.
I
AIthouh my concern here 1s 1th d1scr1m1nat1on and not 1th product1v1ty
d1fferences I must note my skept1c1sm about the freguentIy~msde srument
that bIacks have Iesser future~or1entat1on. For th1s d1sreards the eII
knom fact that of any roup 1ncome IeveI bIacks ssve at Iesst as much as
w1tes.
5
t
asse that 1ven the tastes the markets ork smoothIy. 0eneraI
equ1I1br1 regu1res fuII empIoyment of both 8 and k orkers} the aes
of both w1II ad] ust to cIear the market and the d1scr1m1natory tastes
vIII be refIected 1n ae d1fferences.
Let us ]ust cons1der the s1mpIest case that 1n h1ch the empIoyer
d1scr1m1nates. Then he accepts a tradeoff beteen prof1ts and the
+.`
abers of 8 and k pIoyees. That 1s e suppose he seeks to max1m1ze
. '

not prof1ts but a ut1I1ty funct1on 0 ( 8 k) . ke asse to et the
\
sImpIest case that there 1s onIy one type of Iabor 1n the short run
V6 aIso take cap1taI as 1ven so that output 1s f (d8) s1nce the to
k1nds of Iabor are perfect subst1tutes (at a II rat1o) . If e take output
as nuoer a1re, then prof1ts are iven by the express1on
(IJ
were w
k
and w
8
are the ae rates taken as 1ven by the empIoyers. If
ve proceed aIon convent1onaI I1nes the empIoyer eguates the mar1naI
product1v1ty of each hand of Iabor to the pr1ce to h1m. 8ut here the
"pr1ce" of 8 Iabor 1s the market pr1ce k
8
_ _ the pr1ce the empIoyer
1s w1II1n to pay 1n terms of prof1ts for reduc1n h1s 8 Iabor force by
one. Th1s second term 1s hat 8ecker has termed the ''d1scr1m1nat1on
coeff1c1ent" to be des1nated as d
8
} 1t 1s the neat1ve of the mar1naI
rate of subst1tut1on of prof1ts for 8 Iabor . If as e usuaIIy suppose
the oar1naI ut1I1ty of 8 Iabor 1s neat1ve then the d1scr1m1nat1on
coeff1c1ent d
8
, 1s pos1t1ve.
Ia suoIs,
dwere 4_ W
B
81m1IarIy
~6
(2)
(3)
were 1s neat1ve (or zero 1f the empIoyer has no pos1t1ve I1k1n
tor bav1g k workers) .
k
.
are 1aterchaneabIe 1n
from (2) and (3)
g
M
d
dk
g
8
d
8
,
or
W d " _ ^
k B B
0
.
the to types of Iabor

8
= H
_
say. Then
(4)
so tbat equ1I1br1m regu1res that k aes exceed 8 ages as m1ht be
expected.
y
or the moment assume that aII f1ms have the same ut1I1ty funct1on
0 (*, B k ) . It then appears reasonabIe to assume that aII h1re the
same unts of B and M (but e 1II return to th1s po1nt 1n the next
sect1on) . Then each f1m' s Iabor force 1s the same and the aIIocat1on
of Iabor 1s eff1c1ent. The effects of d1scr1m1nat1on are pureIy d1s~
tr1but1ve. The most obv1ous 1mpI1cat1on then 1s that 8 orkers are pa1d
Iess than the1r mar1naI product so that the k orkers and empIoyers
toether a1n. AIso the k orkers cIearIy a1n or at Ieast do not Iose
from (3) 1th _ . The effect on prof1ts hoever depends on the
exact aature of the ut1I1ty funct1on. 0nder the assumpt1on made 1t
foIIms from (1
,
3) aad the fact that p N
B
H
L
that
7
(5)
.
`
d1scr1mnt1on
f1ts ouId be
F
_
f(L)  (G
L
) L
*
"  * "
_ aad therefore the chane 1n prof1ts 1s s1mpIy
! "
(6)
Te r1hthand tem has a s1nIe 1nterpretat1on. If e cons1der an
h,_ `
1acrease 1n the f1m' s Iabor force 1th the proport1ons of k and 8 orkers
, , t'' '
coastant, then the neat1ve of the mar1naI rate of subst1tut1on of
.`. \
prof1ts for th1s baIanced 1ncrease 1s s1mpIy d
w
(W/L) +d
8
(8/L) th1s
I"
1s the f1m`s need for add1t1onaI prof1ts to compensate 1t for a baIanced
1acrease 1n s1ze. Th1s term may of course be pos1t1ve or neat1ve.
Bowever a pIaus1bIe hypothes1s h1ch e shaII ma1nta1n hereafter
1s that empIoyers' ssat1sfact1ons depend onIy on the rat1o of 8 to W orkers.
In that case
(7)
d (6) teIIs us that empIoyers ne1ther a1n nor Iose by the1r d1scr1m1natory
behav1or. The ent1re effect 1s that of a transfer from 8 to k orkers.
Let us no reIax the assumpt1on that ut1I1ty funct1ons are 1dent1caI
on f1rms. We cont1nue to assume that for each f1rm the ut1I1ty
depends onIy on the rat1o of U to 8 orkers but some f1rms may be more
d1scr1m1natory than others 1n the sense that the mar1naI rate of
subst1tut1on of prof 1ts for 8 orkers 1II be more neat1ve at any 1ven
rat1o 8/k. Eguat1on (4) and {7) hoId for each f1rm at Ieast each f1m
=
8
=
that'eploys both types of orkers. They can be rearded as a pa1r of
I1near eguat1ons 1n k and 8 to y1eId
51ae
d
8
* 1f there are both and k orkers 1t must be that
w
?
8
,
as
b1hest vaIues of d
8
have the h1hest rat1os of k to
_
. Thus an observa
t1on on aII the f1ms 1n ex1stence at egu1I1br1um 1II reveaI a d1spers1on
\
of kproport1ons 1n the Iabor force and these rat1os 1II measure the
vary1n derees of d1scr1m1nat1on. Thus a part1aI deree of sereat1on
appears the 8 orkers tend to be found 1n the Iess d1scr1m1natory
t1ms, the k orkers 1n the more d1scr1m1natory ones.
Boever further anaIys1s Ieads to 1mpI1cat1ons h1ch m1ht ra1se
sooe p1r1caI guest1ons. pec1f1caIIy eguat1on {2) st1II hoIds 1th
H
8
H
L
, Bence, accord1n to the modeI H
_
1s h1her for more
d1scr1m1nat1n f1rms. 8ut then 1f e asse d1m1n1sh1n mar1naI pro
duct1v1ty of Iabor 1t foIIos that the Iess d1scr1m1natory the f1rm the
Iarer 1t 1II be. Th1s accords 1th comon sense d1scr1m1nat1on 1s
costIy to the entrepreneur and acts I1ke a tax on h1m s1nce 1t sh1fts
his deoand for Iabor to the more costs component. Bence 1t restr1cts
his scaIe.
51nce K
_
1s no Ioner the same from f1rm to f1rm 1t foIIos that
product1on 1s no Ioner eff1c1ent. The prev1ous stron statements
about the 1nc1dence of d1scr1m1nat1on no Ioner hoId exactIy e1ther.
~9
Eowever, the1r eneraI thrust 1s st1II probabIy correct. Eff1c1ency
Iosses are not apt to be reat, and the ma1n red1str1but1on 1s st1II
1 w .
I1keIy to be from 8 workers to k workers.
.
It has been seen that compet1t1on tends to reduce the deree of
d1scr1m1nat1on 1n the market, 1n the sense that the une1hted averae
!
ot d1scr1m1nat1on coeff1c1ents of the d1fferent f1ms exceeds the averae
we1hted 1n proport1on to the number of workers.
T1s resuIt, h1ch may or may not be emp1r1caIIy reasonabIe, appears
wre stronIy and Iess I1keIy hen one pushes the anaIys1s 1n the Ion run.
Rov we are ass1g that cap1taI , wh1ch has been h1therto heId f1xed,
1s ad]usted opt1maIIy to the s1ze of the Iabor force. Then cap1taI 1II
fIow to th ore prof1tabIe enterpr1ses h1ch, 1n th1s context, are the
Iess d1scr1m1natory. In the Ion run, output 1s therefore s1mpIy
proport1onaI to Iabor (assum1n the product1on funct1on d1spIays constant
retums to qgp1taI and Iabor) . The mar1naI product of Iabor 1s then
constant. A a resuIt, the compet1t1ve effect ]ust stud1ed assumes an
aerated form. 0nIy the Ieast d1scr1m1natory f1rs surv1ve. Indeed,
1f there were any f1rms h1ch d1d not d1scr1m1nate at aII , these ouId
be the onIy one to surv1ve the compet1t1ve struIe.
1nce 1n fact
zac1aI d1scr1m1nat1on has surv1ved for a Ion t1me, e must assume
that the modeI ]ust presented must have some I1m1tat1on to h1ch e
w1II return 1n sect1on k beIow.
ke have deaIt extens1veIy 1th the assumpt1on of d1scr1m1nat1on by
empIoyers. 8ut, as e observed earI1er, d1scr1m1nat1on by coorkers 1s
`
. .
'r
~I0
The most stra1htforard extens1on of the preced1n
_
totaI payments to fIoor workers.
Then the foIIow1n has been shown 1n Arrow [I97I Techn1caI Rote 8.
W
{I0)
~I2~
The Ieft~hand s1de 1s the market wae d1fferent1aI due to d1s~
m1natory tastes of foren reIat1ve to the wae IeveI 1n the absence
ot d1scr1m1nat1on.
"
y
/
_
. That 1s, the more 1mportant the share of
toren 1n the output of the f1rm reIat1ve to fIoor Iaborers, the
reater the wae d1fferent1aI.
Te Ianuae of the preced1n anaIys1s has assumed that 1t 1s the
toren or other supez1sory empIoyees who d1scr1m1nate accord1n to
the compos1t1on of the fIoor workers. 8ut the anaIys1s 1tseIf 1s com~
pIeteIy abstract. It may be 1IIum1nat1n to reverse the roIes.
uppose
that product1on workers have stron d1scr1m1natory feeI1ns about the1r
superv1sors. 0erta1nIy the 1dea that wh1te workers stronIy resent
be1n bossed by bIack superv1sors or maIe workers by femaIe foremen
(foreIad1es! forepersons! ) 1s a comon one. Then 1f 1n {I0) we
understand by k and 8 those k1nds of superv1sory workers, by the totaI
number of such workers, and by
y
the fIoor workers, we have an exceIIent
expIanat1on of d1scr1m1nat1on aa1nst 8 superv1sory workers, for
y
then wouId be Very Iare 1ndeed compared w1th
_
.
Foremen may poss1bIy d1ffer 1n the1r tastes for d1scr1m1nat1on. 0ne
m1ht suppose that th1s w1II Iead to a reduct1on 1n market wae d1fferent1aIs
anaIoous to the s1tuat1on w1th empIoymr d1scr1m1nat1on. ut a fuIIer
anaIys1s of th1s case rema1ns to be done.

I3
Z+ Roacoavex1t1es 1a Ind1fference
` l
I

tcz
_
b
_
cont1nu1ty of demand funct1ons. 8ut th1s arent appI1es onIy
to 1adIv1duaI dand funct1ons.
tarr [
tor one expos1t1on, see Arrow and Bahn [I97I 0hapter 7.
.
_
t 1s true that
1nce th1s
O
curve 1s fIat at the po1nt A the pr1ce rat1o 1s determ1ned by the sIope
of 8C. Row retura to the 1nd1v1duaI who has the or11naI 1nd1fference
cuze I
_
At these pr1ces hew1II max1m1ze ut1I1ty at two d1fferent
po1nts 8 and C but not at any po1nt 1n between. If for exampIe A
1s baIf~way between 8 and 0 the.. market egu1I1br1um 1s reaI1zed by
bav1n haII the 1nd1v1duaIs at 8 and haIf at 0. If A 1s two~th1rds
of the way from 8 to 0 the market egu1I1br1um 1s reaI1zed by hav1n two~
th1rds of the 1nd1v1duaIs buy the bundIe 0 and one~th1rd the bundIe 8.
F1ure I
I6
Note that each 1nd1v1duaI 1s at a po1nt of max1mum ut1I1ty for h1m
suject to h1s budet constra1nt so that th1s 1s truIy a compet1t1ve
eq0II1br1uo and therefore eff1c1ent. (The earI1er reference to
"approximte egu1I1br1um'' 1s reIevant hen there are not enouh 1nd1v1duaIs
to split them 1n the r1ght proport1ons beteen 8 and U. Thus 1f A 1s . 7I
of the vay from 8 to 0 there are onIy 50 1nd1v1duaIs 1n the economy
there shouId be J5I/2 1nd1v1duaIs at 0 and I4I/2 at 8. Thus at 0 or
8 the d1screpancy beteen suppIy and demand cannot be reduced beIo haIf
aa 1ad1v1duaI. Th1s 1s reIat1veIy a m1nor d1screpancy beteen suppIy
and demand. )
Thus nonconvex1ty 1mpI1es the ex1stence of d1st1nct n1ches for
econom1c aents 1n a sense of the ord h1ch I take to be cIose to that
0sd 1n ecoIoy. 0ne obseres aents 1dent1caI 1n the1r econom1c data
eaaed 1n d1verse consumpt1on patterns or other econom1c act1v1t1es. Any
1ven aent may be 1nd1fferent beteen severaI of these a1ches but
eq01I1br1um regu1res the1r coex1stence. Th1s arument underI1es Adam
5o1th's d1scuss1on of spec1aI1zat1on as opposed to R1cardo's h1ch as
based on d1fferences 1n the product1v1t1es of 1nd1v1duaIs or nat1ons 1t
has been made expI1c1t 1n an 1mportant but neIected paper of outhakker

Let me no appIy these abstract concepts to rac1aI d1scr1m1nat1on. Ue
take up a modeI due to 8ecker [I959 , 0hapter 4 and 1n a d1fferent form
to keIch [I967 and not anaIyzed above. Ro e Iocate the d1scr1m1natory
tastes 1n the k orkers vho are perfect subst1tutes for the 8 orkers.
I7
To keep oatters as s1mpIe as poss1bIe assume there 1s onIy one k1nJ of
labor, Then anaIoous to the assumpt1on made about compIementary foms
of labor, we no assume that U orkers have an 1nd1fference map beteen
hIch w1II o1n1o1ze the cost of h1r1n hatever totaI number of orkers
V + L 1t does h1re. Th1s cost 1s
(I2)
8ut 1t 1s easy to see that a f1rm 1II aIays ach1eve m1n1m cost
WIth e1ther an aIIU or an aII8 Iabor force. The to m1ht be eguaIIy
cheap but certa1nIy any comb1nat1on w1th U and 8 both pos1t1ve 1II be
We costIy than at Ieast one extreme case and poss1bIy more costIy than
both. To see th1s cons1der to casest
(a) RecaII that U/L I means an aIIU Iabor force.
7
Then for any U 0 < U < L
U
{U/L) 7
U
{I)
8
and therefore
U
{U/L)U
w
8
w
8
(W8)
8
L hence an aII8 Iabor force 1th cost
8
L 1s
cheaper than the m1xture. A aIIU Iabor force has a cost
w
{I) L and then
?
w
k
(I)L
8
L so the aII~U Iabor force 1s at any rate no Iess costIy
1f w
k
(I)
8
the to extreme cases are eguaIIy cheap.
(b)
U
{I) <
8
t Then 1f U <L
U
{U/L) U w
_
b 7
U
{I) W
8
8 7
w
W
(I) (W8) w
w
{I) L} the aIIU Iabor force 1s cheaper than any other
Iabor force.

I8
' Beace 1f v
k
(I) < v
8
every f1rm 1II f1nd 1t cheapest to seIect
aII8 Iabor force and 1f
0
( I) ?
8
every f1m 1II m1n1m1ze cost
] bL
g
1a an aIIk Iabor force. 8ut egu1I1br1um regu1res fuII epIoyment
t
l
oI both types of vorkers. The egu1I1br1um then regu1res
w
(I)
8
g
t even then no f1m 1II h1re both k and 8 orkers. ke f1nd then that
at eg01I1br1 every f1rm 1s sereated but then the onIy observed ae
for k vorkers 1s for those 1n aIIk f1rms 1e.
w
(I) h1ch 1s eguaI to
8`
Therefore d1scr1m1nat1on by k orkers 1II not resuIt 1n market
vae d1fferent1aIs but 1nstead does resuIt 1n sereat1on.
In techn1caI terms the funct1on 0 (k 8) 1s not a convex funct1on
spec1f1caIIy the 1socost curves 1n k8 space are not concave to the or11n.
Convex1ty 1mpI1es a tendency to the m1ddIe to comprom1se but here e
have a r0sh1n to extremes. ke aIso have the character1st1c 1mpI1cat1on
oI aoaconvex1ty a d1spers1on of f1rms 1th bas1caIIy 1dent1caI market
opport0n1t1es 1nto d1screte n1ches.
Rov 1n o1n back over the anaIyses of sect1on 2 1t can be observed
that the case ]ust d1scussed of d1scr1m1natory tastes by a perfect
s0bst1t0te roup of orkers 1s str1k1nIy s1m1Iar to that of d1scr1m1natory
tastes by vorkers of a compIementary type the ''foremen" of our exampIe.
To0h a deta1Ied anaIys1s of the nonconvex1t1es 1n th1s case has not
yet been made 1t 1s cIear that the prof1t funct1cn def1ned by (9) 1s not
La eneraI a concave funct1on. Rather the surface def1ned by prof1ts as
a f0nct1on of k and 8 has hoIes scattered throuh 1t. hence 1t 1s at
Ieast poss1bIe that for certa1n vaIues of
k
and v
8
and some egu1I1br1um
I9~
relation. w_(W/
_
) there are severaI d1st1nct po1nts of max1mum prof1ts.
Equilibriu on the three Iabor markets (k 8 and F) may be ach1eved
by different amounts of these even thouh each f1m has the same
production funct1on and each faces the same waes for k and 8 and the
sae relation between w
F
and k/L. Thus there w1II be a part1aI
segregation by f1ms.
The reIat1on (I0) stated earI1er st1II hoIds for each f1rm so the
previous concIus1ons rema1n vaI1d.
Ve can now recoasUer the theory of empIoyer d1scr1m1nat1on. The
utility function, 0(, 8 k) depends 1t has been assumed onIy on the
ratio 8k. 8ut 1t 1s shown 1n the Append1x that such a ut1I1ty funct1on
2
cannot poss1bIy:have convex 1nd1fference surfaces everywhere. Therefore
it is poss1bIe and 1n fact I1keIy that 1n the short run egu1I1br1um w1II
require the coex1stence of f1rms of d1fferent s1zes w1th d1fferent U/8
ratios, even 1f aII f1ms have same ut1I1ty funct1on. Thus at Ieast
partial sereat1on 1s a I1keIy outcome of the ut1I1tyax1m1zat1on
theory. AII the f1ms w1II have to have the same ut1I1ty so that the
Iarer f1rms w1II be those w1th the Iarer proport1ons of kworkers
s1nce ut1I1ty 1ncreases w1th and w1th k/ 8.
In the Ion run 1ndeed 1t can be seen that w1th constant returns
to scaIe there must be perfect sereat1on and eguaI1ty of waes.
2
The nonconvex1ty that ar1ses when onIy rat1os matter 1s of 1mportance
in the theory of poIIut1on aIso. For the character1st1c s1tuat1on
there 1s that the poIIutee 1s faced w1th consum1ng a1r or water 1n wh1ch
the proport1on of poIIutants 1s 1ven to h1m.
!
20
For s0ppose there 1s not perfect sereat1on 1. e. there 1s at Ieast
one f1m h1r1n both 8 and U orkers. For that f1rm d
8
7
0 and .
51nce W 1s constant 1n the Ion run 1t foIIos from the preced1n
aect1on that aII 8 orkers 1II be 1n f1rms 1th the smIIest d
8
and aII
V workers1n f1ms 1th the (aIebra1caIIy) smaIIest _. Then aII
fIW have the same d
8
and same and therefore aII have the same U/8
rt1o. Te eg01I1br1um vaIues of
w
and
8
1II be N
L
 _ and N
L
 d
8
respect1veIy here K
_
1s the Ionrun mar1naI product of Iabor
a constant. 8ut then any f1m h1ch 1ncreases 1ts /U rat1o sI1htIy
can wke pos1t1ve prof1ts by 1ncreas1n 1ts scaIe 1t can make
1ndef1n1teIy Iare prof1ts 1th onIy a sI1htIy aItered U/8 rat1o.
It o0Id therefore have a h1her ut1I1ty so e have a contrad1ct1on
to the ex1stence of equ1I1br1um 1th at Ieast one 1ntegrated f1rm.
Bence aII f1rms are sereated. In an aII8 t1rm d
8
0 for an
1ncrease 1n 8 does not chane the U/8 rat1o and therefore Ieaves ut1I1ty
0nchaned. 1m1IarIy 1n an aII~U f1rm _ . It foIIos that as
1n the case of d1scr1m1nat1on by subst1tutes the Iong run equ1I1br1
1s one of perfect sereat1on and eguaI aes.
The correspond1n anaIys1s for d1scr1m1nat1on by compIementary
pIoyees has not been orked out. It can be con]ectured thouh that
sereat1on pIus poss1bIy compet1t1on amon foremen 1th vary1ng
d1scr1m1natory tastes 1II reatIy eaken ae d1fferent1aIs.
4
`
t
2I
3. Costs of Ad_ustment
0t1I1tymax1m1zat1on theor1es then prov1de a coherent and by no
means unreasonabIe account of the effect of d1scrU1natory tastes on
the market 1n the short run. Yet they become unsat1sfactory 1n the
Ion run. I propose as a poss1bIe expIanat1on that Ion run ad]ustment
processes do not ork as perfectIy as they are usuaIIy assumed to.
Wen there are s1n1f1cant nonconvex1t1es the ad]ustment processes
c&lled for must be very rap1d 1ndeed mar1naI ad]ustments are pun1shed
not rewarded. In the case of d1scr1m1nat1on by subst1tute orkers the
fir' would have to be 1II1n to f1re 1ts ent1re aIIU Iabor force and
replace 1t by an aII8 Iabor force or v1ce versa 1n response to a very
small chane 1n aes. It 1s not unreasonabIe to asse that there are
costs not ord1nar1Iy taken account of h1ch 1II restra1n the f1m
from be1n gu1te so free 1n 1ts ad]ustment behav1or.
Row the 1dea that ad]ustment 1s costIy 1s one that has apeared 1n
several d1verse areas of econom1cs. The costs of roth enter expI1c1tIy
1n some vers1ons of the dynam1c theory of the f1m (Penrose [I959
Harr1s [I964) . That 1s the roth of the f1rm 1mposes a cost h1ch
depends on the rate of roth and h1ch 1s add1t1onaI to the purchase
of cap1taI oods.
The same pr1nc1pIe that cap1taI costs of an unconvent1onaI k1nd
play an 1mportant roIe 1n econom1c behav1or and dec1s1ons has been
appI1ed to the study of Iabor turover a probIem more cIoseIy connected
with outs. 0perat1ons researchers 1n try1n to dra up pIans for the
t
bIrIn cf gerscnneI have 1ncorpcrated 1n the1r oodeIs a f1xed cost of
t: ` `
1rIn an Ind1vIduaI. 5ooetIoes 1t 1s aIso heId t0at t0ere 1s a cost
>
`
.
t
attac0ed tc fIr1n as weII. These costs are partIy 1n ado1n1strat1on
P
ttIy Ia tra1n1n. Even 1n the case of workers who have aIready been
eaezaIIy tra1ned 1n the k1nd cf work to be done there 1s a need for
;
IeatIn the ways of the part1cuIar f1m, Th1s approach 1t has been
'
'\
arued by sooe has 1oportant eneraI econoo1c 1opI1cat1ons 1t 1opI1es
t0at f1m shouId not ad]ust the1r Iabor force to cycI1caI sh1fts 1n
demnd s1nce they oay 1ncur both h1r1n and f1r1n costs 1f they do
4
ccsts that are avo1ded 1f the worker 1s reta1ned dur1n sIack per1ods.
Wrk
g
rs are be1n heId 1n eopIoyoent even thouh they contr1bute I1ttIe
to output to avo1d the costs of reh1r1n theo 1n the expected future
bcco. I do not know oyseIf whether th1s expIanat1on 1s 1n fact adeguate
but oereIy note that 1t 1s ser1ousIy cons1dered.
I suest that a s1o1Iar cons1derat1on expIa1ns why the ad]ustoents
wIch wcuId w1pe out rac1aI wae d1fferent1aIs do not occur or at Ieast
are reatIy retarded. ke have onIy to asse that the eopIoyer oakes
an 1nvestoent Iet us caII 1t a _ersonneI 1nvestoent every t1oe a worker
Is h1red. Be oakes th1s 1nvestoent v1th the expectat1on of oak1n a
coopet1t1ve return on 1t 1f he h1oseIf has no d1scr1o1natory feeI1ns
t0e wae rate 1n fuII egu1I1br1 w1II eguaI the oar1naI product of
Iabor Iess the return on the personneI 1nvestoent. Let us cons1der the
s1opIest of the above oodeIs that of d1scr1o1nat1on by feIIow pIoyees
vhc are perfect subst1tutes. If the f1ros starts w1th an aIIk Iabor
~)3
r ,:, torce It vIII aot fInd It prof1tabIe to f1re that force 1n w1ch 1ts
persoaneI capItaI has aIready been sunk and h1re an aII8 force 1n
)6~
econd 1t must be that the empIoyer must 1ncur some cost before he
c detem1ne the empIoyee's true product1v1ty. I 1t couId be detem1ned
t
' i

the Inc0r a personneI 1nvestment cost, as d1scsed 1n the Iast sect1on,
batore be can detem1ne the worker's product1v1ty. Th1s personneI
I0V68OouI oIbt, for expIe, 1ncIude a per1od of tra1n1n, onIy after
wIcb I It poss1bIe to ascerta
g
n the worker's product1v1ty or 1ndeed
.` '
I W be onIy a per1od of obserat1on Ion enouh for reI1abIe detem1nat1on
ot prod0ctIv1ty. In the absence of a personneI 1nvestment cost, after aII
tbe pIoyer couId s1mpIy h1re everyone who appI1ed and f1re those
wquaI1t1ed, or pay them accord1n to product1v1ty.
T1rd, 1t must be assumed that the empIoyer has some 1dea or at any
rate preconcept1on of the d1str1but1on of product1v1ty w1th1n each of
tbe to cateor1es of workers.
Tbe s1mpIest wdeI to br1n ouL the 1mpI1cat1on of these assumpt1ns
ses to be the foIIow1n.

g
. 0n the other hand 1f the orker h1red
tuts out to be unguaI1f1ed the epIoyer rece1ves noth1n. Bence the
expected return to a k worker h1red 1s 00

g
) p
Q
. If the epIoyer
1a r1skneutraI th1s must be eguaI to r.
1m1IarIy
and therefore
(I4)
were Q p
8
p
\l
Thus 1f for any reason p
8
< p
g
g
1s a e1hted
averae of w
8
and H
. '
ect1ve prcbab1I1t1ea wh1ch w1II auppIy an appropr1ate ]ust1f1cat1on
fcr b1a conduct.
F1naIIy one can aIso seek expIanat1ons 1n wh1ch p
and p
8
d1ffer
Ia reaI1ty even thouh the 1ntt1ns1c ab1I1t1es of U and 8 orkers are
1dent1caI.
uch an expIanat1on regu1res some further assumpt1ons.
5pec1f1caIIy whether or not a orker 1s guaI1f1ed 1s no taken to be
the resuI of a dec1s1on by h1m rather than some type of 1ntr1ns1c
ab1I1ty. Hore spec1f1caIIy a orker becomes guaI1f1ed by mak1n some
type cf 1nvesUent 1n h1mseIf. In accordance w1th the prev1ous assumpt1ons
th1a 1nvestment must not be observabIe by the emp1oyer. Bence the
1nveatoenta are not the usuaI types of educat1on or exper1ence h1ch
are obaezabIe but more subtIe types of personaI depr1vat1on and deferment
cf rat1f1cat1on h1ch Iead to the hab1ts of act1on and
t
u
that favor
cod performance 1n sk1IIed ]obs stead1ness punctuaI1ty respons1veness
d 1n1t1at1ve.
'`
)9~
F1naIIy 1t must be
assumed as 1s reasonabIe that the human cap1taI
,
needed to guaI1fy cannot be acgu1red on a perfect cap1taI market. It
' l
toIIows that the proport1on of e1ther roup (U or 8) ho quaI1fy 1s an
Increas1n funct1on of the a1n from guaI1fy1n. In accordance 1th our
1``
pec1f1caIIy Iet v
w

0
be the a1n to a
V worker from guaI1fy1n were
8

0
. Then e postuIate an 1ncreas1n funct1on
S(v), such that
{I5)
tet1_ be the mar1naI product1v1ty of unsk1IIed Iabor so that
{I6)
Rote that K
and K
0
are determ1ned by the suppI1es of sk1IIed and
unsk1IIed Iabor and these 1n turn are detem1ned by the proport1ons p
and p
8
. Bence the system cons1st1n of the eguat1ons {I3) {I5) and {I6)
pIus the eguat1on obta1ned from {I3) by repIac1n 8 by U const1tute a
.
syst of eguat1ons 1n the unknoms u
w
8
w
0
p and [
0`
y
rom the symmetr1c formuIat1on of the system 1t 1s cIear that there
1s a sy etr1c egu1I1br1 1.e. an egu1I1br1um 1n h1ch[
w
[
8
and
V
@
8
Boever 1t rema1ns an open guest1on hether th1s 1s the onIy
egu1I1br1. It has been possibIe hoever to ask hether or not the
suetr1c egu1I1br1um 1s stabIe. The anser turns out to be that 1t
J0
dapans on the paraoeters of the probI 1t 1s certa1nIy poss1bIe that
$B
1
e
0LI1br1 be unstabIe, a resuIt wh1ch stronIy suests, thouh
6 doea not prove, that there are egu1I1br1a other than the syetr1c,
nn~d1acr1o1natory, one.
Int01t1veIy, cons1der a poss1bIe seguence of events 1n h1ch
1n1t1aIIy g
g
sI1htIy exceeds p
B
for sooe reason. Then
g
sI1htIy
exceeds w
B
d therefore, froo (I5) p
g
tends to 1se reIat1ve to p
B
,
l
therefore re1nforc1n the or11naI d1segu1I1br1uo. Th1s verbaI aruoent
2
wrrIs K. 194 The Econom1c Theory of 'Hanager1aI' Ca_1taI1sm.
CrIde 0. K. t Cambr1de 0n1vers1ty Press.
Fearose E . 1959 The Theory of the 0roth of the
y
1rm. 0xfordt 0xford
0n1vers1ty Press.
Rothenber J. R^
5tarr . R^
5tarrett 0. 1971 A FundamentaI Ronconvex1ty 1n ExternaI 1seconom1es.
keIch F. 197 Labor Harket 1scr1m1nat1ont An Interpretat1on of
Income 1fferences 1n the RuraI 8outh. JournaI of PoI1tcaI Economy
75 t 22J240.
APPRIX
Ronconvex1ty of Ind1fference Hsps epend1n on st1os
ke suppose thst pIoyer d1scr1m1nst1on 1s determ1ned by s ut1I1ty
funct1on 0( 8 U) where muIt1pIy1n both 8 snd U by the ssme
_ pos1t1ve constant Iesves ut1I1ty unchsned. Ue sIso sssume thst 0 1s
. .
B 1ncreas1n funct1on of prof1ts . It w1II be shown thst the
1nd1fference osp def1ned by 0 csnnot hsve convex 1nd1fference surfsces
spec1f1caIIy a convex comb1nst1on of two 1nd1fferent po1nts 1s not
everywhere st Iesst ss ood ss e1ther.
Choose eny po1nt (
^
8 U ) e Then choose v
I
v
o
( ss cIose ss
o o o
needed) and 8
I
w
I
so thst
(I)
Froo the sssupt1ons msde (I) w1II cont1nue to ho1d 1f 8
I
snd U
I
sre
reduced
I/ 2^
o
8' I/2
I
I/28
o
U' 1/2U
I
I/ 2U
o
, k
I
k ) (r
, k ) .
O O O O O
ut and