Chapter4
TAXES AND LABOR SUPPLY*
JERRY A. HAUSMAN
NationalBureau of Economic Research and
Massachusetts Instituteof Technology, Cambridge, MA
1. Introduction
The effect of taxes on labor supply introduces interesting questions in economic
theory, econometrics, and public finance. Since the greatest share of federal tax
revenue, approximately 50% in 1980, is raised by the individual income tax, we
are certainly interested in its effects on economic activity. The federal income tax
is based on the notion of "ability to pay"; and its progressive structure has
received wide acceptance. The income tax has not been thought to induce large
economic distortions so that it has been generally accepted as probably the best
way to raise revenue where an unequal distribution of income exists. At the same
time we finance social security by FICA (Federal Insurance Contributions Act)
taxes which is a proportional tax with an upper limit. As both the tax rate and
limit have grown rapidly in recent years, FICA taxes have become the subject of
much controversy. In 1980, FICA taxes represented 28% of total federal tax
revenue. In Table 1.1 the income tax and payroll tax revenues are given for the
period 1960-1980. It is interesting to note over that same period while the
marginal income tax rate of the median taxpayer remained constant, the FICA
tax rate more than doubled. At the same time the earnings limit rose about 220%
in constant dollars. Over the same 20-year period the corporate income tax has
decreased from 24% to 13% of federal tax revenues. Likewise, excise taxes have
decreased from 13% to 5%. Thus, taxes on labor supply currently amount to
about 75% of federal taxes raised. Their potential effects on labor supply and
welfare are important because of the large and increasing reliance on direct
taxation.
To measure empirically the effect of taxes on labor supply, problems in
economic theory and econometrics need to be treated. First, the effect of
*The NSF and NBER provided research support; and A. Auerbach, S. Blomquist and J. Poterba
made helpful comments on an earlier draft of this chapter.
1Of course, not all income tax revenue is a tax on labor supply because of the taxation of capital
income which was about 12% of adjusted gross income in 1980. Also, a portion of the incidence of
FICA taxes fall on the employer although the amount is likely to be small.
Handbook of Public Economics, vol. I, editedby A.J. A uerbach and M. Feldstein
©1985, Elsevier Science Publishers B. V. (North-Holland)
Table 1.1
Revenues from income and payroll taxes (billions).
Income tax % Payroll tax % Median Earnings limit
Income tax Payroll taxa of federal of federal marginal Tax rate for for payroll
Year revenues revenues revenues revenues tax rate
b
payroll tax tax
1960 $40.7 $11.3 45% 12% 20% 3.0% $4800
1965 48.8 17.6 43 15 20 3.626 4800
1970 90.4 39.5 47 21 20 4.8 7800
1975 122.4 75.6 45 28 20 5.85 14100
1980 244.1 139.3 49 28 20 6.05 25900
'Includes old-age, survivors, disability, hospital insurance, and railroad retirement taxes.
bFrom Steverle and Hartzmark (1981) with interpolation to have years match up.
progressive taxation is to create a convex, non-linear budget set where the net,
after-tax wage depends on hours worked. Since most of consumer theory is based
on constant market prices which are independent of quantity purchased, theoretical
notions such as the Slutsky equation need to be modified to assess the effect of
a change in the tax rate. Theoretical problems increase in complexity when we
realize that other provisions of the tax code such as the earned income credit, the
standard deduction, and FICA together with transfer programs such as AFDC
create important non-convexities in the budget set. Then certain portions of the
budget set cannot correspond to utility-maximizing points. Little definite knowledge
can be gained by a theoretical analysis of the effect of taxation. In fact, we
cannot usually tell whether an increase in tax rates will increase or decrease hours
worked. Nor can we decide how an increase in exemptions or other similar
changes will affect hours worked. Thus, only empirical investigation can determine
the sign and magnitude of the effect of taxation.
Appropriate econometric techniques to measure the effect of taxation also need
to treat the non-linearity of the budget sets which taxation creates. Other
problems such as components of the stochastic specification, limited dependent
variables, and unobserved wages for non-workers arise. Econometric procedures
to handle these problems, many of which have only recently been developed, have
been used to estimate labor supply functions. We review these results and discuss
the possible effects on labor supply of various tax reform proposals which have
been enacted or have been discussed in the U.S.
The other important aspect of the taxation of labor supply is the effect on
economic welfare. If Hicksian deadweight loss (excess burden) is accepted as the
appropriate efficiency measure of the distortion created by taxation, we know that
the deadweight loss is proportional to the square of the tax rate.2
The ratio of
deadweight loss to tax revenue raised rises approximately with the tax rate. In
2
See Auerbach (this volume) for a discussion of appropriate welfare measures in the presence of
taxation.
214 Jerry A. Hausman
Table 1.1 it can be seen that the marginal tax rate for the median taxpayer is 26%,
while the top marginal tax rate on labor supply is 50%. If compensated labor
supply elasticities are non-zero, even though small, the deadweight loss from the
income tax is likely to be substantial. The important redistributive aspect of the
income tax must not be lost sight of, but the cost of the current means of doing so
is an important consideration. Again, we will consider various tax reform proposals
and their possible effect on economic welfare.
The plan of the paper is as follows. Section 2 considers the theory of labor
supply with taxes. The effect of the non-linearity of the budget sets complicates
the analysis so few definite conclusions can be reached. In Section 3 we develop
an econometric model of labor supply so that the problems created by convex and
non-convex budget sets can be solved. Section 4 discusses the various tax systems
in the United States. The federal income tax, FICA tax, and state income taxes all
are used to develop the appropriate budget sets. We also discuss AFDC, social
security benefits, and a negative income tax to determine how they affect labor
supply budget sets. In Section 5 we present empirical estimates for husbands and
wives labor supply functions. We also calculate the economic cost of the tax
system for certain individuals. Because of small numbers in cross-section samples,
and measurement problems, high-income individuals are difficult to treat within
the context of a labor supply model. Thus, in Section 6 we review the individual
questionnaire data for high-income people. It is interesting to note that it agrees
broadly with the econometric evidence. In Section 7 we review the evidence from
the negative income tax experiments and from samples of social security beneficiaries.
These individuals face extremely high marginal tax rates so that
interesting evidence of the effect of taxes is produced in these situations. The
purpose of this chapter is to concentrate on the effects of taxes. A vast empirical
literature in labor economics exists which considers labor supply without explicit
consideration of taxes. A recent survey of this literature is given by Killingsworth
(1983).
2. The theory of labor supply with taxes
In a world without taxes, the theory of labor supply is characterized by the same
conditions which characterize the theory of consumer demand. That is, the
Slutsky conditions completely exhaust the theoretical restrictions on consumer
response to a price change. Thus, in most previous work on the effect of taxation
on labor supply, the authors consider taxes as lowering the net, after-tax wage.
Using the Slutsky equation
dh ah hah
dw__ a(2.1)
dw aw -= y
Ch. 4: Taxes andLabor Supply 215
we decompose the change in hours into the substitution effect and the income
effect. Since labor is supplied while leisure is demanded, the sign of the substitution
effect is positive, while the sign of the income effect is negative if leisure is a
normal good. We can conclude that the sign of the sum of the effects is
indeterminate. It might then be considered the goal of empirical analysis to
determine the sign and magnitude of the effect of taxation.
However, this approach is seriously misleading in all cases except one. Consider
the two-good diagram of Figure 2.1. The composite good is used as numeraire, so
consumption is measured on the vertical axis with hours supplied on the horizontal
axis. Non-labor is denoted by y. The original pre-tax market wage is w and
preferred hours of labor are h*. The effect of a proportional tax is then to lower
the net, after-tax wage to w,= w(1- t). Depending on the individual's preferences,
the desired hours of work h* can either increase or decrease according to
equation (2.1). Thus, in the case of proportional taxation, the traditional analysis
is correct. But, only for proportional taxes is the analysis so simple. What makes
the proportional tax case so special is that non-labor income y is unaffected by
the tax which is implicitly assumed to be only a tax on labor income. If y were
also subject to taxation at rate t, we would have to take account of another
income effect which would cause h* to rise. Equation (2.1) would then need to be
modified to account for taxation of y to
dh ah dw ah ahdy
Oh u= drtt+ hy (2.2)
dt Ow uut y dt
When we consider the effect of taxation, the income and substitution effect of a
change in the wage as well as the change in non-labor income must be accounted
Y
Figure 2.1
216 Jerry A. lHausman
Ch. 4: Taxes and Labor Supply 217
Wi
$
Y3
Y2
Yl
-H H2 h^
H1 U
Figure 2.2
for. This equation becomes the key device in analyzing the effect of taxation on
desired hours of work. The total effect of taxation is still indeterminate but a
complication has been added since changes in both w and y must be considered.
In cases of progressive taxation or government tax and transfer programs, both w
and y are affected. The traditional analysis has neglected to account for the effect
on y of the tax system. We now consider how the analysis changes when
non-proportional tax systems are considered.
Let us first analyze the simplest case, that of a progressive tax on labor income
so that the marginal tax rate is non-decreasing. In Figure 2.2 three marginal tax
rates are considered, t, t2 and t3, which lead to three after-tax net wages, w, w2
and w3, where wi = w(1 - ti). H1 and H2 correspond to kink point hours which
occur at the intersection of two tax brackets. But an important addition to the
diagram are the "virtual" incomes Y2 and y3, which follow from extension of a
given budget segment to the vertical axis. They are denoted as virtual income
because if the individual faced the linear budget set B2 = (w2, Y2), he would still
choose hours of work h* as in Figure 2.2. In assessing the effect of taxation on
labor supply, two questions arise. How does h* in Figure 2.2 differ from the
no-tax situation of Figure 2.1? And how is h* in Figure 2.2 affected by a change
in the market wage w or the tax rates ti?
To consider the first question we combine Figures 2.1 and 2.2 in Figure 2.3. We
see that no general effect can be identified. If the individual's h* falls on the first
budget segment B1 = (wl, yi) we are back in the case of Figure 2.1 with offsetting
income and substitution effects. Alternatively, if h* falls on either B2 or B3, then
the net wage is lower than w which leads to an income and substitution effect,
which virtual income Y2 or y3 exceeds y, and a further income effect from
217Ch. 4: Txes and Lor Supply
\
,,
W1
W
-H
JerryA. Hausman
$
Figure 2.3
equation (2.1) is created which would reduce labor supply.3
One result which does
follow is that on the budget segment B2 or B3 labor supply is less than it would
be if the analysis were based on (w2, Yl) or (w3, yl); that is, if the effect of the
virtual income were ignored.
To answer the second question we initially consider an increase in the market
wage from w to w'. In Figure 2.4 we see that this wage change leads to a
clockwise rotation of the budget set. The effect of the rotation is to raise the w,,
but it also leaves the virtual incomes unchanged. For instance, the virtual income
Y2 is
where E1 is the earnings limit for the first tax bracket, i.e., E1 =y, + wlH1. Thus,
the virtual incomes depend only on the tax system and non-labor income y,.
Therefore, so long as the individual's preferred hours of work h* remain on the
same budget segment Bi, the effect of a wage change can be analyzed using the
traditional local analysis which is contained in the Slutsky equation (2.2)
The effect of a change in a tax rate tj depends on which ti changes. To take the
simplest case, suppose t3 rises so that in Figure 2.2 the w3 segment rotates
counterclockwise. The virtual income y3 also rises. We have the same effect as
3 It may well be this latter income effect, which creates the appearance of a backward bending labor
supply curve which has been found in many empirical studies. The important point here is that not
only do we have a income effect from the change in wage, but virtual income also rises due to the
effect of the tax system.
218
Ch. 4: Taxes and Labor Supply
Figure 2.4
before where the change in wage alone induces both an income and substitution
effect and the change in virtual income induces more labor supply from equation
(2.2). It is important to note that a person, whose preferred hours were previously
on the third budget segment B3 so that h* > H2, may now decrease his preferred
hours to H2 if the substitution effect is large enough, but will not decrease the
preferred hours down to the second budget segment. Individuals, whose preferred
hours were less than H2 before the change, will not be affected. However, if the
tax rate were to decrease, we could again have people shifting from the second
segment to the third segment because of the substitution effect. For these cases,
we need "global" information on the individual's preferences, since the local
information in the Slutsky equation is not sufficient to analyze the possible
changes. Now if either t or t2 were to change, the situation is more complicated
since all later budget segments are also affected. However, the later budget
segments are affected only by a change in their virtual income since the net wage
remains the same. Thus, if t rises, for those individuals with h* > H a, the effect
of the tax change is to cause their preferred hours to rise. For people whose initial
h* < H1, only w changes (although Y1 may change also) so that the Slutsky
equation can be used. Lastly suppose one of the tax bracket limits Ei changes. If
Ei is lowered, all virtual incomes on later budget segments fall. Therefore if
initially h* > HI, we have a similar qualitative effect to a rise in t. Preferred
hours of work will rise. For an individual whose initial h* < H1 but the E[ have
H' < h*, the analysis is more complicated. They may switch to B2 with its lower
net wage and higher virtual income, or they may decrease their desired hours of
work so that h* < H' and they remain on the first segment.
From the analysis of the progressive tax case we see that very few general
propositions can be deduced about the effect of taxes on labor supply. The
219
JerryA. Hausman
-H 0
Figure 2.5
piecewise linear progressive tax system is defined by a sequence of budget
segments B, = (wi, y,) of net wages and virtual incomes for the individual over a
set of hours (Hi, H,,+). Some limited results are possible for changes in ti and E,
for individuals whose initial hours of work are on a subsequent budget segment
Bi+j, j > 0. But to assess the effect of taxation adequately we really need to know
the individual's preferences or, equivalently, his utility function. We will show
how knowledge of his utility function arises in the process of estimating his labor
supply function so that numerical computations of the effect of taxation can be
carried out.
When we do not have a progressive tax system, matters become more complex
since the budget set is no longer convex.4
Non-convex budget sets arise from the
presence of government transfer programs. The three most important programs of
this type are AFDC, social security, and a negative income tax (NIT) program. In
Figures 2.5 and 2.6 we show the two most common types of non-convex budget
sets.5
In the first type of budget set used in the NIT experiments, and in the
majority of AFDC programs, non-labor income is raised by the amount of the
government transfer. The individual then faces a high marginal tax rate, usually
0.4 or higher, until he reaches H, the breakeven point at which all benefits have
been taxed away. Beyond the breakeven point, the individual rejoins the federal
tax system, here taken to be convex. Figure 2.6 has one additional complication
which arises as an earnings disregard in social security benefits or as a maximum
4
As we will discuss subsequently, even the federal tax system is not truly convex because of the
effect of social security payments, the earned income credit, and the standard deduction. However, it
may well be the case that treating taxes in a convex budget set is a sufficiently good approximation for
empirical work.
A non-convexity may also arise, not from the tax system, but due to fixed costs to working; e.g.,
Hausman (1980). We will discuss fixed costs in the next section.
220
Ch. 4: Taxes and Labor Supply
-H H H O
Figure 2.6
payment amount in some AFDC programs. 6
Hours up to H, are taxed only at
FICA rates where H1 is determined by E1 . Thus earnings up to a set amount are
"disregarded" or not taxed by the program which is giving the transfer. Beyond
this point, the individual faces the high marginal rates until breakeven hours are
again reached. On a priorigrounds, almost nothing can be said about the effect of
taxation in the non-convex budget case. The added complication arises from the
possibility of multiple tangencies between indifference curves at the budget set.
Figures 2.7 and 2.8 demonstrate two cases of multiple tangencies although
actual cases may be even more complex due to the possibility of skipping entire
budget segments. The possibility of having multiple optima as in Figure 2.7
because w1 < w3 while Y2 <y3 . In the convex case this possibility does not arise
because as w, falls y, is rising. To determine the global optimum we need to have
knowledge of the utility function. Figure 2.8 demonstrates the case of a joint
tangency, the possibility of which arises with each non-convex segment. Small
changes in the wage or any parameter of the tax system can then lead to large
changes in desired hours of work.
In the convex budget case, we must always have a tangency which is unique
and which represents the global optimum if desired hours are positive. For if we
had two tangencies we could connect the two points, and the connecting lines
which would lie inside the original budget set would represent preferred points by
the assumed concavity of preferences.7
Furthermore, the effect on h* of a change
in the market wage, taxes, or the earnings limits is "smooth" in an appropriate
6
See Hanoch and Honig (1978) for a theoretical analysis of the social security case. Burtless and
Moffitt (1982) discuss the social security budget set within a model of labor supply and retirement. See
also Blinder, Gordon and Wise (1980) for a treatment of intertemporal considerations.
7
See Hausman (1979) for further analysis and implications of this case.
221
A_
JerryA. Hausman
$
-H 0
Figure 2.7
mathematical sense that the change is continuous and differentiable. For the
non-convex case, this reasoning no longer follows since the line connecting the
multiple tangencies no longer lies within the budget set. Thus, multiple tangencies
may occur. Likewise, the effect of changes in the budget set are no longer smooth
since a small change may cause a jump in desired hours from an initial tangency
to the neighborhood of another initial tangency. Thus, it seems that no general
propositions hold. The extended Slutsky equation (2.2) is not usable since the
possibility of a jump from one budget segment to another is always present.
-H
Figure 2.8
222
We briefly consider the cases where we could say something definite in the
convex case: a rise in t or a drop in El for individuals not on the first segment.
For individuals who remain on the convex budget segments like w2 and w3 in
Figure 2.7, virtual incomes again fall while wi remains constant so that the local
effect is again a rise in desired hours of work. But one cannot rule out the
possibility of a non-local jump down to the first segment or even withdrawal from
the labor force entirely. Similar possibilities exist if E1 is decreased. Thus, the
analysis of the non-convex case cannot proceed without knowledge about the
form of the individual's utility function.
Important potential shortcomings exist in this theory of labor supply which we
now discuss briefly. Future work on econometric models will need to incorporate
these problems into the theory and estimation.8
First, individuals may face
quantity restrictions in labor supply. That is, h* may not be possible for
systematic reasons. Certainly "involuntary" unemployment falls into this category.
In principle, even if quantity restrictions exist, we can estimate the underlying
demand function or preference structure and analyze the effect of taxes. But a
more difficult problem is to ascertain if individuals are actually constrained.
Endless debates on the possibility of involuntary unemployment highlight this
problem. Furthermore, survey questions on the ability of a person to work more
hours are very untrustworthy. To date, only limited progress has been made on
this problem.9
Better data seems to be required to put quantity constraints into
an empirical model in a totally satisfactory manner.
This type of labor supply theory also does not adequately treat the type of jobs
people take or their intensity of work while on the job.'O An effect of taxes is to
make non-pecuniary rewards more attractive so that a measure of earnings may
seriously misrepresent the preference comparisons being made among jobs.
Academics need hardly be reminded of this fact in the present world of falling
real academic wages. Yet it is doubtful that this problem will ever be completely
solved. "Perks" from a job could be evaluated monetarily and included in
earnings. But we cannot hope to measure adequately certain types of non-monetary
rewards to jobs.
A last consideration is intertemporal aspects of the model. We have considered
a static world devoid of human capital considerations and intertemporal factors
'Considerable research is currently undertaken in these areas.
9
Ham (1982) has introduced quantity constraints into a labor supply model without taxes. See
Deaton (1981) and Deaton and Muellbauer (1981) for further research on quantity restrictions. Brown
et al. (1982) have attempted to incorporate quantity restrictions in a short-term labor supply model
which does incorporate taxation.
" To the extent that wages reflect intensity of work, this problem may not be too serious. However,
for many jobs wages may be only loosely related to current effort with longer-run goals important. We
discuss this issue subsequently. See Rosen (1980) for a discussion of these problems.
Ch. 4: Taxes and LaborSupply 223
such as savings. But intertemporal issues may be quite important for new entrants
into the labor market and for individuals close to retirement. The eighty-hour
weeks put in by young lawyers will be rewarded in the not-too-distant future so
that current compensation in an inadequate measure of earnings. Furthermore,
issues of on-the-job training may be important. To date, research on issues of
intertemporal labor supply have indicated only limited empirical importance of
this problem. But further empirical research based on less restrictive models may
find a more important role for intertemporal considerations.
Another potentially important intertemporal consideration is that the models
that we have specified are static labor supply models. Intertemporal considerations
may render the static approach inapplicable except under very strong
separability assumptions. How would our approach change if life cycle considerations
are accounted for? Blundell and Walker (1984) have demonstrated that only
a minor change is required. Under the assumption which is typically made in life
cycle models, a two-stage budgeting approach makes the labor supply decision
conditional on planned life cycle asset accumulation." At the first stage the
lifetime budget is allocated to equalize the marginal utility of money in each
period. At the second stage the labor supply decision is made conditional on this
allocation. The second-stage decision is therefore analogous to our static approach
except that non-labor income must be replaced by net dissaving in the
period. MaCurdy (1981) discusses the interpretation of the estimated coefficients
from the labor supply model in a life cycle context. However, note that the tax
system still only taxes the income component of this dissaving. The static
specification then becomes correct only if assets remain constant over the life
cycle.
The specification of a static labor supply model may thus lead to biased
estimates because the incorrect measure of virtual income is being used in the
labor supply equation. Note, however, that most of the variation in virtual
income, which enters the labor supply equation occurs because of the tax system.
The variation in net dissavings compared to the variations caused by the tax
system may be small. More research is required here. What is primarily needed is
a cross-section data set which contains both labor supply information and also
carefully constructed information on net dissavings over the period of observa-
tion.
MaCurdy (1983) takes a different approach to the problem of incorporating life
cycle considerations. He estimates the marginal rate of substitution function
between consumption arid hours of work. Thus, data on both labor supply and
consumption is needed. Unfortunately, it is not straightforward to see how
constraints such as zero hours of work can be applied in the MaCurdy formulation.
Further econometric problems exist with respect to errors-in-variables
problems in representing the tax system. In its initial application to the Denver
1
iDeaton and Muellbauer (1980) discuss two-stage budgetary approaches.
224 JerryA. Hausman
Income Maintenance Experiment, MaCurdy estimates labor supply elasticities
which are extreme outliers with respect to previous results which will be discussed
subsequently. Further research will hopefully determine whether MaCurdy's
results arise because of the life cycle considerations which previous studies have
ignored or because of the econometric specifications which he uses.
In this section we have considered from a theoretical point of view the effect of
taxes on labor supply. The Slutsky equation, which has been traditionally used to
analyze the problem, is inadequate except for the case of a proportional tax.
Progressive taxation results in a convex budget constraint which leads to a
multiplicity of net wages and virtual incomes. We see that except for a few cases
the effect of a change in the tax rate cannot be determined on a priori grounds
even if reasonable assumptions are made such as leisure being a normal good.
Government tax and transfer programs result in non-convex budget sets which
are even more difficult to analyze theoretically. Thus, we now turn to the
econometrics of the problem so that models can be estimated. From the estimated
models we can assess the effect of taxation. However, as with all models, we
discuss certain aspects of the problem which have not been included. The results
should be interpreted with these limitations in mind.
3. Tax systems
In the previous section we have discussed the theory of labor supply with taxes.
We will now describe the type of tax systems which exist in the United States. To
give a historical perspective on the problem, we will outline the evolution of the
income tax rate over the century. We will also provide data on actual marginal tax
rates since 1960 when the data become available. We shall discuss federal and
state income taxes first. It turns out that, even though the basic federal income tax
is progressive, the resulting budget set that an individual faces is not convex.
FICA payments, the standard deduction and the earned income credit all
introduce non-convexities. These additions to the basic progressive tax system will
be explained. Next we discuss AFDC tax systems for each state. Lastly we briefly
discuss tax systems for social security beneficiaries and negative income tax (NIT)
recipients. All of these tax systems have very large non-convexities together with
quite high marginal tax rates.
We first outline the basic federal income tax system in 1980 by twelve
brackets.1
2 The first bracket is $1,000 wide with succeeding brackets falling at
intervals of $4,000. Since we are interested in the effect of taxes on labor supply,
we consider only taxes on earned income. Table 3.1 lists the brackets along with
12
Here we discuss our procedure for joint returns. We follow similar procedures for single persons
and heads of households, but do not report the details here.
Ch. 4: Taxes and LaborSupply 225
JerryA. Hausman
Table 3.1
Basic federal tax rates for 1980 on earned income for married couples.
Average rate
Taxable income Marginal rate at midpoint
0-2,000 0.14 0.140
2,000-4,000 0.16 0.147
4,000-8,000 0.18 0.160
8,000-12,000 0.21 0.174
12,000-16,000 0.24 0.189
16,000-20,000 0.28 0.204
20,000-24,000 0.32 0.222
24,000-32,000 0.37 0.250
32,000-42,000 0.43 0.287
42,000-56,000 0.49 0.331
56,000 + 0.50 the
marginal tax rates and average tax rates at the midpoint of the bracket. It is
interesting to note that the average tax rate remains significantly below the
marginal tax rate until quite high levels of earned income are reached. Thus a
theory which stated that individuals react to average after-tax income when
making marginal decisions, might come up with rather different results. However,
the theory of individual behavior with respect to progressive taxation contains
both the marginal net wage and the appropriate virtual income which reflects
average tax rates up to the current tax bracket. In a certain sense, the entire
characteristics of the tax system are accounted for in this way.
In determining taxable income, personal exemptions need to be accounted for.
An exemption of $1,000 per person was allowed in 1980. The standard deduction,
or zero bracket amount, was $3,400 for married couples in 1980.13 Itemized
deductions in excess of $3,400 could also be subtracted from gross income. They
were approximately 9% of adjusted gross income in 1980. The standard deduction,
i.e., no itemized deductions, was used on approximately 70% of all tax
returns in 1980. Next, the earned income credit grants of 10% below $5,000 of
gross income. From $6,000 to $10,000 the credit is reduced by 12.5%, so that the
breakeven point is reached at $10,000 when the credit has been completely
exhausted. A non-convexity is created at $10,000 because the tax rate falls by the
12.5% payment when the breakeven point is reached. Lastly FICA contributions
were 6.05% up to a limit of $25,900 in 1980. Thus, in the appropriate bracket
when the FICA limit is reached, the marginal tax rate falls from about 0.38 to
about 0.32, which also creates a non-convexity. l4
We provide some historical data
13
Tables exclude the zero bracket amount for the standard deduction.
14
However, empirical work by Hausman (1981a) did not indicate that the non-convexities created
by the earned income credit and FICA had an important influence on the econometric estimates.
226
Table 3.2
Federal income tax: Selected marginal rates.
Taxable
income 1950 1969 1970a 1980 1984b
(1000's) (1.0,3.3)C (1.23,5.6) (1.61,9.8) (3.42,21.0)
2-4 22 22 19 16 12
6-8 30 30 26 21 16
10-12 38 38 33 26 20
16-18 50 50 43 40 30
20-22 56 56 49 43 23
26-32 62 62 54 50d 38
38-44 69 69 59 50 42
50-60 75 75 64 50 46
60-70 78 78 66 51 48
70-80 81 81 68 50 48
80-90 84 84 70 50 49
90-100 87 87 71 50 50
100-150 89 89 72 50 50
150--200 90 90 72 50 50
aIncludes 2.5% surtax.
bThe 1984 rates reflect the entire 25% tax reduction pased by Congress in 1981. The tax will
then be indexed.
CFirst entry is CPI in 1950 dollars and second entry is median family income in thousands of
current dollars.
d Maximum tax on earned (labor) income was 50% beginning in 1972 under the Tax Reform Act
of 1969.
on tax rates and actual marginal rates to provide a historical perspective on the
income tax system.15
In Table 3.2 we provide a summary of marginal tax rates for the period
1950-1984, according to current legislation.1
6 These rates are for single taxpayers,
with no exemptions or deductions accounted for. We also give the CPI and
median family income, so that valid comparisons across different years can be
made. First, note that the tax system between 1950 and 1980 was only imperfectly
indexed for inflation. The median income faced a marginal tax rate of 22% in
1950, but multiplied by the change in the CPI, this amount faced a marginal rate
of 26% in 1980. Similarly, $10,000 of earned income in 1950 had a marginal tax
rate of 38% in 1950, but adjusted for inflation, this marginal tax rate increased to
43% in 1980. Similar increases in marginal tax rates occurred over the periods
1960-1980 and 1970-1980. Of course, this imperfect indexation corresponds to
greater progressivity which may have been the intent of Congress over the period.
However, note that under the tax reform of 1981, marginal rates will drop
15Tax law changes in 1981 provide for exclusion from taxation of 10% of the secondary worker's
earnings up to $30,000 beginning in 1983. This change greatly increased the neutrality of the tax
system towards married persons.
16
The tax rates are taken from Tax Foundation (1981).
C3h. 4: Taxes andLaborSupply 227
substantially by 1984 due to the 25% tax reduction, with the exact amount
depending on inflation over the 1982-1984 period. Much of the "bracket creep"
of the past decade will be eliminated. Under current legislation, the tax system
will then be indexed after 1984. Another interesting finding which emerges from
Table 3.2 is the significantly higher marginal tax rates faced by the median earned
over the period. Besides the effect of inflation and imperfect indexation, real wage
growth also led to higher marginal taxes. Lastly, note the remarkable decline in
maximum taxes on earned income which arose with the Tax Reform Act of 1969.
To determine the effect of these tax changes we now consider the actual marginal
rates faced by given segments of the population.
We present marginal tax rates from a sample of returns in Table 3.3, calculated
by Steverle and Hartzmark (1981) in a very useful paper. Of course, the tax rates
correspond to total income rather than just labor income which was considered in
Table 3.2. The significant rise in the progressivity of the income tax in the
1960-1980 period is evident in Table 3.3. Note that for those households which
paid tax, the marginal rate was between 0.18 and 0.26 up through the 95%
percentile. In fact, 59% of all taxpayers who had a non-zero marginal rate had a
rate of 18%. While the marginal rate for the median return increased by 10%
between 1961 and 1979, the difference in rates on the interquantile range
increased by 33%. This considerable increase in the progressivity of the marginal
tax rates will be decreased by the tax legislation changes of 1981.
Another historical comparison of marginal tax rates is provided by Seater
(1982) who based his estimates on the StatisticsofIncome rather than a sample of
individual returns. Except for the years 1964-1967 when the Kennedy tax cut
Table 3.3
Marginal rates of taxation on personal income.
Marginal rates Average marginal tax ratesb
Percentile Payroll
of tax
returns 1961 1969
a
1974 1979 Year Rate included
1% 0.00 0.00 0.00 0.00 1950 15.2 -
5% 0.00 0.00 0.00 0.00 1955 16.3
10% 0.0 0.00 0.00 0.00 1960 16.4 19.4
25% 0.18 0.15 0.15 0.14 1965 14.0 17.6
50% 0.18 0.23 0.20 0.20 1970 17.2 22.0
75% 0.22 0.25 0.22 0.24 1975 17.4 23.3
90% 0.22 0.28 0.28 0.32
95% 0.26 0.32 0.32 0.38
99% 0.38 0.47 0.47 0.50
aIncludes an approximation for surtax changed in 1969.
bFrom Seater (1982).
JerryA. Hausman228
lowered tax rates, the average marginal tax rate increased over the period
1950-1980. When the effects of the payroll tax are included, the increase is from
an average marginal tax rate of about 15% in 1950 to an average rate of over 23%
in 1975, which is an increase of 43% in the 25-year period. Therefore, the
increases in the payroll tax over the period have a large effect on the marginal tax
rates.
State income taxes (including the District of Columbia) should also be briefly
mentioned. In 1980, nine states did not tax earned income, but the other
forty-two states had either progressive of proportional tax systems. Sixteen states
permitted deduction of federal income taxes. Among the states with progressive
tax systems Delaware had the highest overall marginal tax rate of 19.8%.
However, at $15,000 after personal exemptions, the marginal rate in California
was 10%, in Hawaii 10%, in Minnesota 14%, and in New York state 10%. In
Oregon the marginal tax rate was 10% above $15,000, and in Wisconsin the
marginal rate was 11.4% at $15,000. Nebraska, Rhode Island and Vermont were
the only states which took a constant percentage from the federal taxes paid.
Rhode Island took the highest proportion, 17%. Among states with proportional
rates after personal exemptions, Illinois had a rate of 2.5%, Massachusetts a rate
of 5.4%, and Indiana and Pennsylvania rates of 2%. State governments increasingly
turned to direct taxation as a source of revenue over the past 20 years.
Beside the operation of the Federal tax system, another potentially more
important influence on labor supply of female heads of household is the AFDC,
Aid for Dependent Children, tax and transfer system. It has often been contended
that AFDC presents a significant disincentive to labor supply, and its replacement
by NIT, Negative Income Tax, could significantly decrease the work disincentive.
The basic design of AFDC programs is a transfer payment which depends on
family size, accompanied by a tax rate of 67% until the breakeven point is reached
and the person returns to the federal tax system. A sizeable non-convexity is
created because at the breakeven point the marginal tax rate decreases from 0.67
to approximately 0.16. Thus, the potential disincentive effect is quite large.17
States differ in the size of the transfer payment and also in the exact operation of
the AFDC tax system. The majority of the states permit $30 of earned income per
month before starting to levy the 0.67 tax. Thus, in Figure 3.1 we show the basic
outline of the AFDC budget set. Breakeven hours H may not be reached even by
women who work full time at the level of wages which AFDC recipients typically
receive.
The workings of NIT tax systems resemble AFDC as in Figure 3.1, although no
earnings disregard exists. Major differences are eligibility, since all families would
qualify, and benefit and tax parameters. The NIT guarantee is a function of the
17Under current legislation, in certain cases the tax rate is 100%. An important distinction exists
between the statutory tax rate and the effective tax rate because of various allowable deductions.
Moffitt (1981) estimates the effective AFDC tax rates over a sample of recipients.
Ch. 4: Taxes andLaborSupply 229
JerryA. Hausman
$
-H H H1
Figure 3.1
poverty limit which depends on family size and the local cost of living. The
guarantee has been set at between 0.75 and 1.25 times the poverty limit in the
NIT experiments. For instance, in Indiana 0.75 times the poverty limit was 28%
higher than the AFDC payment for non-labor force participation for a family of
four. Thus the NIT guarantee is typically more generous than the AFDC
payment. The marginal tax rate up to breakeven hours is also lowered from 0.67
to a value between 0.4 and 0.6. The budget set has the non-convex form of Figure
3.2, where beyond breakeven hours, H, the individual returns to the federal tax
system. At breakeven hours, the marginal tax rate falls from 0.4 and 0.7 to around
0.25 when federal taxes and FICA are accounted for. For male heads of
households with good jobs on the less generous plans of a low guarantee and high
tax rate, breakeven hours will be reached at about 120 hours per month of work.
For males on very generous plans or those with low wages, breakeven costs will
not be reached even for high hours of work. Likewise, for female heads of
household the majority will not reach breakeven hours because of their relatively
low wages. Thus the position of the first tax segment and the non-convexity
created at Hf hours may have a significant influence on labor supply decisions.
The last tax system we consider is the operation of the social security earnings
test for individuals between 62 and 70 years old who are receiving social security
benefits. The budget set has exactly the same form as the operation of AFDC in
Figure 3.1. A level of benefits is determined by the earnings history covered by
social security and by family composition. An "earnings disregard" then exists up
to an amount which determines H1 hours. Beyond H1 hours earnings are taxed at
a rate of 0.5 until breakeven hours H are reached. Thus, we again seem to have a
230
Ch. 4: Taxes and Labor Supply
G
Y1
Figure 3.2
possibly large disincentive to working. 8
But this diagram leaves out a potentially
important effect which Blinder, Gordon and Wise (1980) point out. The effect is
that current earnings will replace lower previous earnings, which are used to
compute average monthly earnings which the benefit level is partly based on.
Especially with the low levels of previous FICA amounts, current earnings could
replace the $3,600 level in force in 1951-1954 and for about 20% of near-retirement
workers replace previous zero FICA earnings years. Thus, if individuals
understand the admittedly extremely complex social security benefit formulas, the
work disincentives can be greatly diminished. Blinder, Gordon and Wise actually
give an example where the earnings test is more than compensated for as a work
incentive exists. Thus, empirical studies which use historical data may have great
difficulty in adequately representing the correct budget set. The indexing provision
of the 1977 Social Security Amendments greatly lowers the quantitative
significance of earnings replacement. However, the disincentive effect of the
earnings test is still diminished. The intertemporal aspects of the interaction of
social security and the retirement decision probably require a more complex
model than our essentially one-period representation of the budget set. While the
problem is quite difficult to represent in a model, social security may have a
significant effect on retirements
In this section we have discussed the effect of federal and state tax systems on
the budget set. While federal tax rates are uniformly progressive, non-convexities
1
8If the individual is eligible to receive benefits but continues working without receiving benefits,
his future benefits are increased by an approximately actuarily fair amount between the ages of 62 and
65. The adjustment for ages 65-70 is considerably less.
19
The Blinder et al. conclusions have been challenged by Burkhauser and Turner (1981); a reply is
given by Blinder et al. (1981). For empirical estimates of the effect of social security on retirement, see
Gordon and Blinder (1980), Boskin and Hurd (1982), Burtless and Moffitt (1982) and Diamond and
Hausman (1982). These results are reviewed in the last section of this paper.
231
I -
still exist in the budget set due to the presence of the standard deduction, earned
income credit, and FICA contributions. State income tax and AFDC programs
are also discussed. Next the NIT tax system and its relation to AFDC is
considered. Lastly, the budget set for the social security earnings test and the
complex intertemporal aspects of retirement are outlined. In this last area further
work seems required to extend the labor supply model to account for intertemporal
decisions.
4. The econometrics of labor supply with taxes
The essential feature which distinguishes econometric models of labor supply with
taxes from traditional demand models is the non-constancy of the net, after-tax
wage.2
Except for the case of a proportional tax system, the net wage depends on
hours worked because of the operation of the tax system. Also the marginal net
wage depends on the specific budget segment that the individual's indifference
curve is tangent to. Thus, econometric techniques need to be devised which can
treat the non-linearity of the budget set. However, it is important to note at the
outset that a simultaneous equation problem does not really exist, even though
the net wage received depends on hours worked.21
Given a market wage which is
constant over hours worked and a tax system which is given exogenously by the
government, the non-linear budget set faced by the individual in deciding on his
preferred hours of work is determined exogenously to his choice.2 2
An econometric
model needs to take the exogenous non-linear budget set and to explain the
individual choice of desired hours. We first describe such a model for convex and
non-convex budget sets. As expected, the convex case is simpler to deal with. We
then consider other issues of model specification, such as variation in tastes, fixed
costs to working, and quantity constraints on available labor supply.
Econometric estimation is quite straightforward in the case of a convex budget
set. Since a unique tangency or a corner solution at zero hours will determine
desired hours of work, we need only determine where the tangency occurs. To do
2 0
Non-constant prices do exist in the demand for other goods, e.g., electricity with a declining
block rate. A general treatment of econometric techniques for non-linear budget sets is given by
Hausman (1982a).
21
In initial work on introducing taxes into labor supply models, Hall (1973) used the observed
after-tax wage which creates simultaneous equation bias in the estimated coefficients. Wales (1973),
Hausman and Wise (1976), and Rosen (1976) introduced instrumental variable techniques to take
account of this problem.
22If the market wage depends on hours worked, the same reasoning holds since the budget set is
still exogenous. A more fundamental problem exists over the question of whether labor supply models,
with or without taxes, are identified. Unobserved individual effects may exist in both the wage
equation and hours equation. If these individual effects also influence variables such as education
which appear in the wage equation but not in the hours equation and thus serve to identify the hours
equation, we would lose identification.
JerryA. Hausman232
so we begin with a slight generalization of the usual type of labor supply
specification,
h =g(w, y,z, f)+e=h*+ e, (4.1)
where w is a vector of net wages, y is a vector of virtual income, z are individual
socioeconomic variables, f is the unknown vector of coefficients assumed fixed
over the population, and e is a stochastic term which represents the divergence
between desired hours h* and actual hours. The typical specification that has
been used in g( ) is linear or log-linear and scalar w and y corresponding to the
market wage and non-labor income. The stochastic term is assumed to have
classical properties so that no quantity constraints on hours worked exist.
However, 0 < h < H, where H is a physical maximum to hours worked. We also
assume that, when the fi's are estimated, the Slutsky conditions are satisfied so
that g( ) arises from concave preferences.
The problem to be solved is to find h* when the individual is faced with the
convex budget set, B for i = 1,..., m.2 3
To find h* we take the specification of
desired hours on a given budget segment B,
h* =g(w, yi,z, ). (4.2)
Calculate h" and if 0 <h* .<Hi where the Hi's are kink point hours in Figure 2.2,
then h* is feasible and represents the unique tangency of the indifference curves
and the budget set. However, if h* lies outside the interval (0, Hi) it is not
feasible, so we move on to try the next budget segment. If HI < h I < H2, we
again would have the unique optimum. If we have bracketed the kink point so
that h > H, and h < H, then h = H so that desired hours fall at the kink
point. Otherwise we go on and calculate h3'. By trying out all the segments we will
either find a tangency or find that h < H,l1 for all i in which case h* = 0, or
h* > HI* for all i in which case h* = H. Then a non-linear least squares
procedure or Tobit procedure to take account of minimum hours at zero should
be used to compute the unknown parameters. The statistical procedure would
basically minimize the sum of =l(hi - hj*)2
, where j represents individuals in
the sample.2 4
Perhaps a better technique would be to use Tobit, which enforces
the constraint that h 2 0.
The case of the non-convex budget set as in Figure 2.5 or Figure 2.6 is more
complicated because equation (4.2) can lead to more than one feasible tangency
which leads to many potential h's. How can we decide which of these feasible
23
The technique used here is more fully explained in Hausman (1979b). See also Hausman (1981a,
1982a).
24
A potential problem exists in the asymptotic expansions used to compute the standard errors of
the coefficients.
Ch. 4: Taxes and LaborSupply 233
h*'s is the global optimum? Burtless and Hausman (1978) initially demonstrated
the technique of working backwards from the labor supply specification of
equation (4.2) to the underlying preferences which can be represented by a utility
function.25
The basic idea is to make use of Roy's identity which generated the
labor supply function from the indirect utility function v(wi, yi),
av (w,, y) /v (w,, y)
= hi = g(wi, y, z, p), (4.3)awi ayi I
along a given budget segment. As long as the Slutsky condition holds (w, yi)
can always be recovered by solving the differential equation (4.3). In fact, v( )
often has a quite simple closed form for commonly used labor supply specifications.
For the linear supply specification
hh* = oaw + y, + zy, (4.4)
Hausman (1980) solved for the indirect utility function
v(wi, yi) = e yi+ 2+ (4.5)
Given the indirect utility function, all of the feasible tangencies can be compared,
and the tangency with highest utility is chosen as the preferred hours of work
h*.26 Then, as with the convex budget set case, we can use either non-linear least
squares or a Tobit procedure to estimate the unknown coefficients. While using a
specific parameterization of the utility function seems upsetting to some people, it
should be realized that writing down a labor supply function as in equation (4.2)
is equivalent to writing down a utility function under the assumption of utility
maximization. To the extent that the labor supply specification yields a robust
approximation to the data, the associated utility function will also provide a good
approximation to the underlying preferences. The utility function allows us to
make the global comparisons to determine the preferred hours of labor supply.
The convex case needs only local comparisons, but the non-convex case requires
global comparisons because of the possibility of multiple tangencies of indifference
curves with the budget set.
25
Their work was done in the framework of labor supply and a composite consumption good. The
technique can also be used in the many-good case, although it is more difficult to apply. Alternatively,
one can begin with a utility function specification and derive the labor supply function as Wales and
Woodland (1979), Ashworth and Ulph (1981a), and Ruffell (1981) did.
2 6
The indirect utility function can be used to evaluate tangencies on both budget segments and at
kink points so that the direct utility function is unnecessary. See Hausman (1980) or Deaton and
Muellbauer (1981) for techniques to be used here. As Figure 2.8 shows, a tangency will not occur at a
non-convex kink point, but it may occur later on a convex portion of the budget set.
234 JerryA. Hausman
Ch. 4: Taxes and Labor Supply
- -
-H
Figure 4.1
We next introduce the possibility of variation in tastes.2 7
In the labor supply
specification of equation (4.1), all individuals are assumed to have identical fly's so
that the variation of observationally equivalent individuals must arise solely from
E. However, empirical studies seemed to do an inadequate job of explaining
observed hours of work under the assumption of the representative individual.
Burtless and Hausman (1978) allowed for variation in preferences by permitting 3
to be randomly distributed in the population. Their results indicated that variation
in 8/ seemed more important than variation in a.28
They also found that
variation in 8 represented approximately eight times as much of the unexplained
variance as did variation in e. Hausman (1981) also found parameter variation to
be an important part of his econometric specification. Blomquist (1983) tests and
rejects the constant preference (no parameter variation) model. However, his
results in terms of policy interest are quite similar whether or not preference
variation is specified. An even more satisfactory procedure would be to allow all
the taste coefficients to vary in the population. At present the requirement of
evaluating multiple integrals over non-rectangular regions for the more general
specification has led to the use of the simple case of one or two taste coefficients
varying. Further research is needed to determine whether this more complex
specification would be an important improvement over current models.
Another consideration which can have an important effect on the budget set for
women's labor force participation is fixed costs to working. Transportation costs,
2 7
For many linear regression specifications where the effect of taxes are not accounted for, variation
in preferences leads only to an efficiency issue for the econometric estimator. However, taxes create an
essential non-linearity in the problem so that variation in preferences can be quite important. A
similar issue arises in the specification of discrete choice models; e.g., Hausman and Wise (1978).
Greenberg and Kosters (1973) seemed to be the first paper that allowed for a dispersion of preferences
to affect their model in an important way.
28
It is interesting to note that Greenberg and Kosters had a similar type of variation in preferences.
However, they did not allow for the effect of taxes so that the results cannot be compared.
235
the presence of young children, and search costs of finding a job, all can lead to a
fixed cost element in the labor supply decision. The basic effect of fixed costs is to
introduce a non-convexity in the budget set at the origin. Thus, even if the
original budget set is convex as in Figure 2.2, the presence of fixed costs leads to a
minimum number of hours Ho, which depends on the wage below which an
individual will not choose to work. In Figure 4.1 non-labor income is y, with the
original convex budget set drawn by the dotted line. However, presence of fixed
costs lowers the effective budget set to the point y, - FC. The individual would
not choose to work less than Ho hours because she would be better off at zero
hours. This non-convexity invalidates the simple reservation wage theory of labor
force participation since hours also need to be accounted for. Hausman (1980,
1981a) has found average fixed costs to be on the order of $100 per month. The
importance of fixed costs could explain the often noted empirical fact that very
few individuals are observed working less than ten or fifteen hours per week.2
We now turn to the question of quantity constraints which seems to enter labor
supply models in two possible ways. The first type of quantity constraint might
arise if an individual has the choice of working either full time, say forty hours
per week, or not working at all. We can still estimate the parameters of his labor
supply function by discrete choice models which allow a distribution of preferences;
e.g., Hausman and Wise (1978), Zabalza et al. (1980). For example,
suppose we begin with the linear labor supply specification h* = awi + /3yi - zy,
along with the associated indirect utility function of equation (4.5). To compare
indirect utility at zero and forty hours we need to specify wi and y, that would
lead to the appropriate number of hours being chosen in an unconstrained
setting.30
But w and yi can be solved for by using the desired hours supply
equation and the linear equation through the point that gives net, after-tax
earnings for that number of hours of work. For forty hours, the equation is
E4 = wi- 40 + y,, where E40 arises from the budget set, w, is the net wage at forty
hours, and y, is the corresponding virtual income. We can solve the two equations
in two unknowns for wi and y and use the values for the required comparison so
that a, /i, and y can be estimated. It turns out that this procedure is equivalent to
solving for the direct utility function, where only quantities appear so that
quantity constraints enter in a straightforward manner. For instance, the direct
utility function for our example is
u(h,x)= hl exp x (4.6)
29Similar results in a model without taxes were found by Cogan (1981). Hanoch (1980) and
Nakamura and Nakamura (1981) have also included fixed costs in models of female labor supply.
3 0
Neary and Roberts (1980) and Deaton and Muellbauer (1980) discuss this technique in a general
demand setting. However, they do not consider the effect of taxes.
236 JerryA. Hausman
where x is consumption of the composite commodity. However, the direct utility
function need not exist in closed form, in which case the previous solution
procedure can be used with the indirect utility function. Of course, specification
of a direct utility function could be done ab initio, but it might not be easily
combined with the labor supply functions of unconstrained individuals.
The other type of quantity constraints which people seem to have in mind is the
choice among jobs, each of which comes with a distinct market wage and hours of
work combination. However, if the individual takes a given job he is constrained
to work the given number of hours which come with the job. Again a discrete
choice framework seems appropriate to model this situation. Use of either the
indirect or direct utility function would allow the appropriate utility comparisons
to be made. We would need to know the range of choices which a given individual
faces. But the choice set might be either established by survey questions or
estimated from a data set of choices of similar individuals. At this point we have
strayed rather far from our original theory of flexible hours of work. In our
empirical estimation we have not accounted for the possibility of quantity
constraints. It seems unclear how important an empirical problem quantity
constraints are. As we discuss later, even conditional on working in a given week,
the standard deviation of hours worked for prime age males is around fourteen
hours. Thus, the model of flexible labor supply with fixed costs may provide a
reasonably good approximation, especially in the long run.3 1
A question of some interest might be: what are the directions of biases in
estimated labor supply models which do not account for taxes? Given the
complexity of a model which incorporates taxes, the answer is not straightforward
although a partial solution is possible. Consider the linear labor supply specification
of equation (4.4). The net wage for individual i on budget segment j, wii,
and the corresponding virtual income, yij, are determined simultaneously with
the unknown coefficients a, /3, and y. Suppose that market wage w and observed
non-labor income xi are used instead. If xi is measured subject to error, which it
almost surely is in any survey, then the estimated coefficient for P will be subject
to errors in variables bias towards zero. As an approximation assume that the
contribution of virtual income, 3yij, is omitted so that it enters the stochastic
disturbance. Since the net wage wi and virtual income yj are positively correlated
due to progressive taxes and /3< 0 if leisure is a normal good, the estimate
of the wage coefficient will have a negative bias. The positive correlation between
the net wage and income arises under progressive taxes because higher wages put
the taxpayer into higher brackets which will have higher virtual incomes. In fact,
empirical studies of males which do not account for taxes typically estimate a to
31
Ham (1981), in a model without taxes, attempts to provide evidence on quantity constraints by
considering the response to a survey question on the possibility of additional work. Deaton (1981)
considers quantity constraints by a rationing model of the consumption function.
Ch. 4: Taxes andLaborSupply 237
be negative and substantial.3 2
On the other hand, estimates which account tor
taxes [e.g., Burtless and Hausman (1978), Wales and Woodland (1979), Hausman
(1981), Ashworth and Ulph (1981), Blomquist (1983), and Hausman (1982)]
estimate a to be much nearer to zero. These latter studies also find considerably
more evidence in support of economic theory than do studies which ignore taxes
and often find compensated demand curves which slope in the wrong direction.
For labor supply estimates for wives, the husband's before-tax income is often
used for y in equation (4.4). Then two counteracting biases are present in
estimates of the wage parameter a. If the wife's wage is positively correlated with
husband's income, then a negative bias of the estimate of a is created. However,
the bias from the income term turns out to be positive so that the net effect
cannot be determined. To the extent that husband's before-tax income is fairly
close to the wife's virtual income, the effect of the bias should not be as important
as in the husband's case. The empirical evidence to date supports this tentative
conclusion.
In this section we have demonstrated how the non-linearity of the budget set
which taxes create can be accounted for in an econometric model. The labor
supply (leisure demand) curves are still the focus of model specification. For the
convex budget set case the only new complication is to search for the budget
segment on which h* falls. When the budget set is non-convex, we need to solve
for the indirect utility function which is associated with the labor supply specification.
Then the multiple tangencies of the budget set and indifference curves can
be compared to find the h* which corresponds to maximum utility. We also
emphasized the potential importance of allowing for variation in preferences and
fixed costs to working. Previous empirical studies indicate the potential importance
of both considerations. Lastly, we discuss techniques to handle quantity
constraints within the context of our approach. However, unless on a priori
grounds we know who in the sample is quantity constrained, it is not clear that
these procedures can be applied in a given sample.
5. Results
In this section we summarize the results of studies of labor supply which take
account of taxes. The effect of taxes on both labor supply and economic welfare is
considered. However, difficulties arise in providing convenient summary measures
for the effect of taxes. Elasticity measures for labor supply, which are most often
32
Heckman and Borjes (1979) present a range of estimates. Despite its title this paper should not be
used for policy purposes since the studies reviewed ignore taxes in their models of labor supply.
238 JerryA. Hausman
used as summary measures in demand studies, are not fully adequate to assess the
effect of taxation for the following reasons: (1) Non-linearity of the budget sets
can lead to large changes in labor supply with small changes in taxes. The
non-convexity of many of the budget sets leads to this result. (2) About 50% of all
women are not labor force participants. Because of the non-tangency of their
utility functions with the budget sets at zero hours, small changes in taxes will not
affect most non-workers. (3) When taxes are changed, both the change in the net,
after-tax wages and the virtual incomes must be taken account of. Equation (2.2)
demonstrates the correct relationship. (4) If variations in preferences are specified
[e.g., Burtless and Hausman (1978), Hausman (1981a, 1982a), and Blomquist
(1983)], behavior of the "mean" individual may differ from the mean population
response.3 3
This difference arises from the non-linearity of the budget set. To
some extent problems which arise with the first and last reason are decreased by
aggregation from individual responses to the population. However, the middle
two problems remain.
5.1. Prime-agemales
These individuals are usually taken to be from 25 to either 55 or 60 years old.
Labor force participation among this group is nearly 100%, especially when
disabled individuals are not considered. Unemployment is typically low among
this group in a non-recession year. Most studies therefore do not account
specifically for unemployment or constraints on labor market activity.3 4
An
integration of behavior when unemployed and hours of work should be a goal of
future research, but theoretical advances as well as better data would be required.3 5
Another needed advance is an integrated model of family labor supply with taxes
to take account of wives labor market activity and its possible effect on husband's
labor market behavior. Hausman and Ruud (1984) present results from a family
model of labor supply. Significant interaction among the husband's and wife's
labor supply behavior is found to be present.3 6
The most natural interpretation of the labor supply results estimated on
cross-section data is an equilibrium model where actual hours differ from desired
hours because of stochastic reasons. One should not maintain the incorrect image
of the prime-age male labor uniformly at work for 40 hours per week and 2,000
3 3
Hausman (1983) investigates this issue.
34Hausman (1981a) does account for zero hours considerations. Ham (1981) considers constraints
on further work at the given wage.
35
Moffitt and Nicholson (1982) consider this problem for a special sample of individuals for which
data are available.
36
Ashworth and Ulph (1981b) also estimate household models of labor supply.
Ch. 4: Taxes and LaborSupply 239
hours per year. On the contrary, significant variation exists in both normal hours
per week and weeks worked per year. Hours per week of work, conditional on
being employed, typically have a mean of about 42 hours with a standard
deviation of 10-15 hours in typical cross-section data.3 7
Men presumably choose
jobs which have the number of hours which most closely correspond to their
desired hours taking account of overtime and possible layoffs. However, for a
significant proportion of the prime-age male population, changes between employers
is fairly rare, see Hall (1982). How much of the year-to-year variation in
labor supply for this group arises from fluctuations in their market wage is
problematical. Therefore, the models of labor supply and empirical results
presented here are probably less relevant for short-term labor supply response to
business cycle conditions.
We consider from sets of results for prime-age males: Wales and Woodland
(1979), Ashworth and Ulph (1981), Hausman (1981a), Blomquist (1983), and
Hausman and Ruud (1984).38 The results are given in Table 5.1. First, note that
the uncompensated labor supply elasticity is much closer to zero than is typically
found in labor supply studies which ignore taxes. This result concurs with the
econometric bias arguments given in the last section. The next difference is that
the income elasticities vary from -0.04 to -0.17 which imply that leisure is a
normal good in contrast to many studies which ignore taxes and find the opposite
sign. Given the magnitude of virtual income with progressive taxation, the clear
implication is that taxes will affect the labor supply decision. The combination of
these two results leads to the last result, which is perhaps the most satisfying. All
five studies imply a positive compensated wage elasticity so that the compensated
labor supply curve is upward sloping. These results are in stark contrast to models
which ignore taxation, and very often estimate a compensated elasticity of the
wrong sign. This finding is difficult, if not impossible, to justify even when more
general models of labor supply are considered. Since a non-negative compensated
elasticity is the only implication of economic theory for models of labor supply, it
is satisfying to find that the results become acceptable to the theory when the
effect of taxes is taken into account.
We now turn to the effect of taxes on labor supply. As equations (2.1) and (2.2)
demonstrate, the theoretical effect is indeterminate. Most models for prime-age
males which ignore taxes estimate a backward-bending labor supply curve.3 9
Therefore, a reduction in tax rates which has recently occurred in the U.S. and
U.K. would lead to a reduction in hours of work. A contrary view has been put
forward by "supply side" advocates in the U.S. who have argued that a reduction
37
This variation is calculated after the self-employed and farmers have been eliminated from the
sample.
38We do not use the earlier result of Hall (1973), Wales (1973), and Brown, Levin and Ulph (1976)
because of difficulties of interpretation and econometric technique.
3 9
Some models find a backward-bending curve only for medium- and high-wage males.
240 JerryA. Hausman
Table 5.1
Prime-age male labor supply results.
Wage Income
Authors Data Model elasticity elasticity
Wales-Woodland PSID
a
CES 0.09
b
-0.11 b
Ashworth-Ulph U.K. Generalized CES - 0.13 0.05
Hausmarf PSID Linear 0.00 - 0.17
Blomquistc Sweden Linear 0.08 - 0.04
Hausman-Ruudd PSID Generalized linear - 0.03 - 0.10
aPanel Study of Income Dynamics.
bResults are approximate since means of data were not given.
CSpecification permits variation in preferences. Mean results are given.
d Estimated from a model of family labor supply.
in tax rates will lead to such a large increase in labor supply that government
revenue would actually increase.
We first present some results of Hausman (1981a, 1981c) for the U.S. He
found, using 1975 data that compared to a no-tax situation, desired labor supply
was 8.2% lower because of the U.S. tax system, including FICA taxes and state
income taxes. In Table 5.2 the results are given by wage quintiles from the PSID
sample. In the second row the change from the no-tax situation is given. Note
that the effect of the progressiveness of the tax system is to cause high-wage
individuals to reduce their labor supply more from the no-tax situation than do
low-tax individuals. The higher marginal tax rates lead to higher virtual income
and a greater reduction in desired labor supply. Of course, this pattern of labor
supply has an adverse effect on tax revenues because of the higher tax rates that
high-income individuals pay tax at. In the second and third rows of Table 5.2 we
present the expected change in labor supply for tax cuts of 10% and 30%. Note
that desired labor supply increases with a tax reduction. We find the expected
pattern that the effect on high-wage individuals is greatest since the linear labor
supply model used has an increasing elasticity with virtual income. The effect of a
30% tax cut is roughly three times as large as a 10% cut, but the ratios are not
Table 5.2
The effect of taxes on prime-age male labor supply in the U.S.
Market wage
$3.15 $4.72 $5.87 $7.06 $10.01
Change in labor supply -4.5% - 6.5% - 8.5% - 10.1% -12.8%
10% tax cut +0.4% +0.5% +0.9% +1.7% +1.47%
30%tax cut +1.3% +1.6% + 2.7% + 3.1% +4.6%
Ch. 4 Taxes and LaborSupply 241
Table 5.3
The effect of tax rate changes on prime-age male labor supply in the U.K.
Quintiles
1 2 3 4 5 Total
15% tax cut -0.3% +0.7% +0.8% +1.6% +2.1% +1.8%
7% tax cut -0.1% + 0.3% +0.3% + 0.9% +0.9% +0.8%
7% tax rise +0.1% -0.5% -1.0% - 0.9% - 0.8% -1.2%
15% tax rise +0.3% -1.1% - 2.3% - 2.6% -2.1% - 2.9%
exact. However, neither of the two tax cuts is nearly self-financing as Hausman's
(1981c) results indicate.
Lastly, we consider two types of radical tax reform. We consider a progressive
linear income tax with all current deductions removed, e.g., interest deductability.
Therefore, we have broadened the tax base considerably and then determined the
tax rate which would raise the same amount of tax revenue as the current U.S. tax
system using 1975 data. With a zero exemption level so that a flat tax results, the
required tax rate is 14.6%. Desired labor supply for the prime-age males rises
about 8.1%. For a progressive tax with an exemption level of $4,000 (1975
dollars), the required tax rate rises to 20%. Desired labor supply increases by
about 7.7%. Therefore, a decrease in marginal tax rates does lead to an increase in
desired labor supply of significant amounts although much of the progressivity of
the tax system is lost with such a proposed tax reform.
Ashworth and Ulph (1981a) also considered the effect of tax changes on labor
supply. They considered changing the standard rate of tax in the U.K. from its
present value of 30% to four other rates representing changes of plus or minus 7%
and 15%. The standard rate of tax is the marginal tax rate for almost 90% of
prime-age males in the U.K. In Table 5.3 the percentage change in labor supply is
given for the entire sample as well as for each quintile of the income (not wage)
distribution.40
Note the qualitative similarity between the Ashworth-Ulph results and the
Hausman results. A much larger change in labor supply is forecast from the
higher-income quintiles. The magnitude of the predicted changes also do not
differ too much, although Ashworth and Ulph find that the income effect
dominates in the lowest quintile, leading to a small decrease in labor supply when
taxes are lowered. The labor supply changes given in Table 5.3 are not sufficient
to make a tax cut self-financing. The rise in labor supply would offset about 10%
of the fall in revenues from the tax cut which is again fairly close to what
Hausman (1981c) found.
40Note that a distribution ordered by wages is probably better, since labor supply choice enters the
income measure.
242 JerryA. Hausman
Ch. 4: Taxes and Labor Supply
Table 5.4
The effect of taxes on prime-age males in Sweden.
Market wage
10.0 Skr 20.03 Skr 40.0 Skr Total
Change in labor supply - 4.7% -13.6% - 27.1% -13.1%
Proportional tax - 1.9% + 6.2% +11.4% + 6.9%
243
The last set of results which we consider are Blomquist's (1983) estimates for
Sweden. Using 1973 data, he calculates the effect of taxes on labor supply for the
mean individual at wage rates of 10.0 Skr, 20.3 Skr, and 40.0 Skr which
correspond to a low wage, the average wage rate in the sample, and a higher wage
rate, respectively. In Table 5.4 the first row estimates the change in labor supply
from the no-tax situation. Note that the results are almost twice as large as the
estimates for the U.S. in Table 5.2. Much of this difference arises from the
considerably higher level of taxation in Sweden. In the second row of Table 5.4 an
equal yield proportional tax is considered for each of the "representative" males.
The corresponding tax rates are 27.8%, 39.1%, and 47.8%, respectively. For the
entire sample the equal yield proportional tax is 34% with desired labor supply
increasing 6.9% from its current level. Blomquist's estimates indicate a substantial
effect of taxes on labor supply in Sweden.
5.2. Economic welfare
The welfare cost of the distortion created by the imposition of a tax is measured
by use of deadweight loss (excess burden). We briefly sketch the theory of the
deadweight loss measure, and then we present estimates which arise from labor
supply studies.4 1
The first component of a welfare measure is the effect of the tax
on individual utility. Here the measure long used by economists has been some
form of consumers' surplus. Consumers' surplus corresponds to the concept of
how much money each individual would need to be given, after imposition of the
tax, to be made as well off as he was in the no tax situation. Measurement of
consumers' surplus often is done by the size of a trapezoid under the individual's
demand curve or here it would be the labor supply curve. But Hausman (1981b)
has demonstrated that in the case of labor supply this method can be very
inaccurate. Instead the theoretically correct notion of either the compensating
variation or equivalent variation should be used.42
These measures, set forth by
41
Auerbach (this volume) contains a more detailed discussion of deadweight loss.
42These measures correspond to the area under the compensated demand curve, which is determined
by the substitution effect in the Slutsky equation. For further discussion, see Hausman
(1981b) or Diewert (1982).
Sir John Hicks, are probably best defined in terms of the expenditure function.
The expenditure function determines the minimum amount of money an individual
needs to attain a given level of utility at given levels of wages and prices.4 3
Its
form is determined by either the direct utility function U(H,Y) or the labor
supply function. Consider the simple example of the wage tax for which the
compensating variation equals
CV(w, w', U) = e(w', U) - e(w, U). (5.1)
Equation (5.1) states that the welfare loss to the individual, measured in dollars of
the consumption good, equals the minimum amount of non-labor income needed
to keep the individual at his original utility level U minus his non-labor income in
the no-tax situation y. Since utility is kept at the pre-tax level U, the compensating
variation arises solely from the substitution effect in the Slutsky equation
(2.1). The income effect is eliminated because the individual is kept on his initial
indifference curve. In the more complicated case of progressive taxes, the only
difference is that we use virtual non-labor incomes in equation (5.1) rather than
actual non-labor income.44
We need one more ingredient to complete the measure of the welfare loss from
taxation. The government has raised tax revenue, and we need to measure the
contribution to individual welfare which arises from the government spending the
tax revenue. The assumption commonly used is that the government returns
the tax revenue to the individual via an income transfer. Here it would correspond
to increasing the individual's non-labor income by the amount of tax revenue
raised. Then the total economic cost of the tax is given by the deadweight loss (or
excess burden) as
DWL(w,w', U) = CV(w, w', U) - T(w, w', U)
= e(w', U) - e(w, U) - T(w, w' U). (5.2)
Equation (5.2) states that the deadweight loss of a tax equals the amount the
individual needs to be given to be as well off after the tax as he was before the tax
minus the tax revenue raised T(w,w', U).4 5
Deadweight loss is greater than or
equal to zero, which makes sense given that we expect taxation always to have an
economic cost. Of course, if no tax revenue is returned, the compensating
43For a more formal treatment, see Varian (1978) or Diewert (1982).
44The alternative measure of the equivalent variation uses post-tax utility U' as the basis for
measuring welfare loss. For labor supply in the two-good set-up, the equivalent variation typically
gives a higher measure of welfare loss than does the compensating variation.
45Here we follow Diamond and McFadden (1974) and use taxes raised at the compensated point.
Kay (1980) has recently argued in favor of using the uncompensated point. As with CV and EV
measures the problem is essentially one of which is the better index number basis.
244 Jerry A. Hausman
Ch. 4: Taxes and Labor Supply
Y
y
Figure 5.1
variation gives the welfare loss to the individual. In Figure 5.1 the compensating
variation and deadweight loss are shown in terms of our simple wage tax example
of Figure 2.2. Here the effect of the tax is to reduce labor supply from H* to H'.
The compensating variation is measured by the line segment yy'. We then
decompose the compensating variation into its two parts. The line segment CD
measures tax revenue collected, while the line CE measures the deadweight loss of
the tax. Since the taxpayer has been made worse off but no one has benefited
from the amount of the deadweight loss, it represents the economic cost of raising
the tax revenue.
In Table 5.5 we present Hausman's results from deadweight loss of taxation of
labor supply in the U.S. The first row gives the deadweight loss in each of the
wage quintiles. Note that the deadweight loss to tax revenue ratio rises sharply
because of the progressivity of the income tax. The "triangle formula" for
deadweight loss demonstrates that the loss is proportional to the square of the tax
rate so that higher-wage individuals face a higher economic distortion. Overall,
Hausman estimates the mean ratio of deadweight loss to tax revenue to be 22.1%,
which has important implications given the large proportion of the U.S. tax
revenues which are raised via the income tax.46
The next two rows of Table 5.5
calculate the deadweight loss under two tax cut proposals to give an indication of
the size of the marginal change. Lastly, we consider the deadweight loss under an
equal yield proportional tax. With no exemption and a tax rate of 14.6%, the
46
As a historical note it is interesting to point out that Harberger's (1964) seminal calculation of the
deadweight loss from the income tax used an income elasticity quite close to the estimate of Hausman
(1981). However, he took the uncompensated wage elasticity to be large and negative. Therefore, the
Slutsky equation led to a near-zero compensated wage elasticity so that Harberger's estimate of the
deadweight loss was very small. On the contrary, Hausman (1981a) finds the uncompensated wage
elasticity to be near zero.
245
Table 5.5
Deadweight loss estimates for prime-age male in the U.S.
Market wage
$3.15 $4.72 $5.87 $7.06 $10.02 Total
DWL/tax revenue 9.4% 14.4% 19.0% 23.7% 39.5% 22.1%
10% tax cut 8.5% 13.3% 17.4% 21.8% 36.1% 19.0%
30% tax cut 6.8% 10.9% 14.5% 17.9% 29.5% 15.4%
deadweight loss to tax revenue ratio declines to 7.1%. A progressive linear income
tax of 20.7% with an exemption level of $4,000 leads to a ratio of 14.5%. Both sets
of calculations indicate the size of the welfare cost which arises from the
progressivity of the U.S. tax system.
Blomquist (1983) does similar deadweight loss calculations for Sweden. For the
average individual in his sample who earns 20.3 Skr, he calculates the deadweight
loss to tax ratio to be 14%. The marginal tax rate faced by this individual is 62%.
An equal yield proportional tax would be 39%, and the deadweight loss to tax
ratio would decline to 5.5%. Over the entire sample Blomquist calculates that the
deadweight loss ratio is 19%. An equal yield proportional tax of 33.7% would
lower the ratio to 4%.
Increased attention in the U.S. and European nations has focused on the
incentive effects of the tax systems. Most of the attention has been on output
effects. The labor supply results for prime-age males reviewed here do demonstrate
that income taxes reduce desired labor supply. The answer to the question
of whether taxes increase or decrease desired labor supply is what most policymakers
worry about. Yet, the deadweight loss effects may be more important
from an economist's viewpoint. Since deadweight loss is a measure of the
economic cost or efficiency effect of the income tax, it provides the appropriate
measure within which to frame questions about the "optimal" progressivity of the
tax system or the cost of marginal government expenditure.4 7
Also, deadweight
loss calculations are central to proposals for tax reform measures. And the
deadweight loss question stands apart from labor supply effects since large
deadweight loss exists even when counteracting income and substitution effects
lead to small labor supply changes. The size of the deadweight loss associated
with the income tax is perhaps the most important finding of the recent labor
supply literature which considers the effect of taxes. Deadweight loss calculations
are likely to influence future discussions on tax changes among economists and
perhaps policymakers as well.
4 7
0f course problems still exist due to the necessity of interpersonal comparisons; cf., Atkinson and
Stiglitz (1976) and Stiglitz (1982). The optimal income tax literature begins with Mirrlees (1971). For
surveys of the literature, see Mirrlees (1982) and Atkinson and Stiglitz (1980).
246 JerryA. Hausman
Table 5.6
Wives labor supply results.
Wage Income
Authors Data Model elasticity elasticity
Ashworth-Ulph U.K. Generalized CES 0.19 - 0.14
Hausman PSID Linear 0.91 - 0.50a
Nakamura-Nakamura U.S. Census Linear - 0.16b
- 0.05
b
Nakamura-Nakamura Canada Census Linear - 0.30b - 0.19
b
Rosen Parnes (NLS) Linear 2.30 - 0.42C
Hausman-Ruud PSID Generalized linear 0.76 -0.36
aEvaluated for mean woman who works full time at virtual income which includes husband's
earnings.
bFor the age group 35-39; younger age groups have larger elasticities, while older age groups have
smaller elasticities.
cEstimated from results given in the paper.
5.3. Wives
Income taxation is generally thought to have important effects on wives labor
force behavior. Wives labor force participation in the U.S. is approximately 50%.
When the labor force participation decision is made by a woman whose husband
is employed, the tax rate which enters her decision is calculated from the
husband's earnings.48
Since this marginal tax rate is around 25% on average, taxes
should be expected to be important in wives labor supply decisions.4 9
Since
relatively large uncompensated wages elasticities are often estimated for wives, the
effect of various tax reform proposals may have important effects. However, it is
important to remember that since 50% of wives do not work, their reaction to
marginal changes in taxes will be zero to a great extent. Wives already at work
will change their desired hours and some wives will decide to enter the labor force
in response to a tax cut, but most non-participants will remain so. Therefore,
elasticity estimates should be used with caution in considering tax changes.
We consider five sets of estimates for wives labor supply behavior which
consider the effect of taxation: Ashworth and Ulph (1981a) for Great Britain,
Hausman (1981a) for the U.S., Nakamura and Nakamura (1981) for Canada and
the U.S., Rosen (1976) for the U.S., and Hausman and Ruud (1984) who use a
model of family labor supply for the U.S. The results are given in Table 5.6. The
estimates vary widely with the Ashworth-Ulph and Hausman results in the range
48 The child care credit reduces the tax rate, but its effect on the participation decision is decreased
because of fixed costs. Also note that beginning in 1983 the appropriate marginal tax rate will decrease
because of a credit included in the 1981 tax reform legislation.
49Given these facts it is surprising that labor economists who work on female labor supply have
largely ignored taxes. The recent book edited by Smith (1980) has only one mention of taxes in all of
the papers. This omission is even more surprising in light of the substantial estimated labor supply
elasticities.
Ch. 4: Taxes and Laor Supply 247
Table 5.7
The effect of taxes on wives labor supply in the U.S.
Market wage
$2.11 $2.50 $3.03 $3.63 $5.79 Total
Change in labor supply + 31.2% - 14.2% 20.3% - 23.8% --22.9% 18.2%
DWL/tax revenue 4.6% 15.3% 15.9% 16.5% 35.7% 18.4%
of studies which do not account for taxes. Rosen's estimated wage elasticity of 2.3
seems quite high. Econometric problems which include incorrect treatment of
virtual income and an inconsistent estimation technique may explain the divergence.
In subsequent analysis Feenberg and Rosen (1983) have used wage
elasticity estimates of 1.0. The most surprising result is that of Nakamura and
Nakamura (1981) who find significant negative uncompensated wage elasticities
which range from -0.39 for the 25-29 age group to -0.06 for the 50-54 age
group. 50
Almost no other econometric study of wives labor supply, whether or not
taxes are considered, estimates a negative wage elasticity. The NakamuraNakamura
paper has an incorrect treatment of virtual income together with other
econometric problems. Yet their finding that higher wages lead to lower labor
supply for wives, together with the implication that the effect of income taxation
is to increase wives labor supply, is so at odds with previous studies that it should
perhaps be disregarded unless further studies provide additional confirmation.
The estimated income elasticities are all negative so that taxation has two
counteracting effects: by decreasing husbands net earnings it increases wives labor
supply, but by its effect on virtual income it decreases labor supply. The overall
effect of the income tax seems quite clearly (apart from the Nakamura-Nakamura
results) to decrease wives labor supply because of the sizeable uncompensated
wage elasticity which is typically found.
In Table 5.7 we present results of Hausman (1981a) for a sample of wives by
quintiles. 5
' Note that the effect of taxation is to increase labor supply for the
lowest-wage quintile but to decrease labor supply for the other. quintiles. The
overall effect compared to the no-tax case is a reduction in labor supply by 18%.
Hausman also found substantial deadweight loss to tax revenues with the ratio
about 18%. Given the magnitude of the estimated elasticities the deadweight loss
estimate might seem small. However, when the fact that non-participants in the
absence of taxation will generally remain non-participants when taxes are levied
5 0
The large negative elasticity estimates are incompatible with economic theory because they imply
a downward-sloping compensated labor supply curve.
51 These estimates take into account the labor supply response of husbands and their change in net,
after-tax income.
JerryA. Hausman248
and remain at the same utility level is taken into account, the finding is
reasonable. The deadweight loss ratio would more than double, i.e., exceed 40%,
for labor force participants. Hausman (1981c) considers a 10% tax cut and
estimates that wives labor supply would increase by 4.1%. For a 30% tax cut he
estimates the increase in labor supply to be 9.4%. Deadweight loss is decreased
significantly. Hausman's results demonstrate an important influence of taxes on
wives labor force behavior together with a substantial economic cost of the
current tax system.
Feenberg and Rosen (1983) simulate the effects of numerous proposed changes
in the tax law on wives labor force behavior. We consider two cases: a tax credit
of 10% on the first $10,000 of earnings and taxation of wives as single individuals.
For the tax credit case they find only a very small effect on hours of work. The
impact of the lower marginal tax rate on some individual is just about cancelled
by the income effect it has on other workers. However, the effect of the tax change
on current non-participants is not treated altogether correctly. Income splitting
leads to a predicted increase in labor supply of about 5.5% for wives. Thus, the
effect of lower marginal tax rates leads to increased labor supply, although some
of the progressivity of the income tax is lost when it is judged at the family level.
The results for wives demonstrate that income taxation has an important effect
on wives labor supply decisions. The economic cost of the tax system is also
substantial. Because of the joint treatment of family income, wives typically face
substantial marginal tax rates on their earnings. No consensus has been reached
on the proper tax treatment of the family in the presence of progressive taxation.
Further econometric work will focus on more of a joint decision framework for
husband and wives, yet it is unlikely that the major findings of an important effect
of taxes will change drastically. Various tax reform proposals such as tax credits
reduce the effect of progressivity on wives. However, important issues will remain
unless a tax system with constant marginal tax rates is adopted.
6. High-income groups
Considerable interest has arisen over potentially large work disincentive effects on
two economic groups: very-low-income and very-high-income groups. Both groups
face high marginal tax rates on earned income; usually the marginal rate is 0.5 or
higher. Our knowledge of the effect of the high marginal tax rates on low-income
groups have been increased considerably by government-constructed cross-section
data sets and most importantly by the four negative income tax experiments. The
results of these NIT experiments will be considered in the next section. Yet very
little reform of the tax system and its treatment of low-income individuals has
been accomplished. On the other hand, our knowledge of the effect of high
Ch. 4: Taxes andLaborSupply 249
marginal tax rates on high-income groups has advanced little in the past decade.52
Yet significant changes in the tax systems as they affect earned income for
high-income groups have taken place. The United States lowered the maximum
marginal tax rate on earned income from 0.7 to 0.5 in 1969, and the Thatcher
government in England has also significantly reduced the highest marginal rates
in 1979. Furthermore, earnings at which the maximum tax level is reached have
increased dramatically in the U.S. under the tax reform legislation of 1981. In
1983, the 50% rate will be reached at $109,000 on a joint return; in 1984, the
maximum rate will be reached at $162,000. The change in rates from Table 3.2 is
remarkable. While high-income groups certainly complain loudly about taxes,
none of the surveys which we will summarize have found a significant disincentive
effect of the high tax rates. Thus we might conclude that a convincing efficiency
argument does not exist for lowering the marginal rates of high-income groups,
but vertical equity considerations have probably been foremost in legislators'
deliberations. 5 3
Almost all of our empirical knowledge of the effect of taxation on the labor
supply of high-income groups arises from interview surveys. An important sample
selection problem exists which has remained almost unnoticed [Holland (1976)].
Since we would expect on average high-income groups work more, those individuals
who are led to work less by the disincentive effect of the tax system are less
likely to be surveyed. Thus, a sample selection bias exists for the finding of a
small disincentive effect. And a small disincentive effect has been the overwhelming
finding of the interview surveys. Yet the empirical results have been so
striking, that it is probably safe to conclude that the sample bias is not giving a
spurious result. For instance, it does not appear that within the surveys that the
highest-income groups are affected to a lesser extent than lower-income groups.
Thus, the primary finding of the survey literature is that, while a disincentive does
exist, its likely magnitude is not especially large.
The classic study in disincentive effects on high-income groups is Break's (1957)
survey of lawyers and accountants in Great Britain. Break conducted 306 interviews
on a group of individuals both familiar with and having the ability to react
to the disincentive effect of the high marginal tax rates, which existed in Great
Britain at that time. Break found that the majority of the respondents were not
significantly affected by the tax system on their work effort. Of the 49% who
reported an effect, only 18% cited disincentive effects, while 31% cited an
5 2
The last significant survey is Holland (1969). Also, the most recent extensive survey of the
literature is Holland (1976).
53
Certainly large amounts of economic resources are used to lessen the burden of taxation by using
the capital gains provisions and other tax preference provisions of the tax laws. But this observation
has little bearing on the work effort of the high-income groups themselves.
250 JerryA. Hausman
incentive effect from the tax system. Thus, the overall income effect dominated
the substitution effect for these individuals.54
Using a much more stringent
criterion, where the interpretation of the sample responses was clearest, Break
concluded that 14% of the sample were significantly affected by taxation. The tax
incentive effect still predominated, with 8% of the original sample working harder
because of the tax effect. Still, Break concluded that a small net disincentive effect
might exist, because the 6% who reported a significant disincentive effect had
higher earnings than the 8%who reported an incentive effect.
Break's original study has been repeated by Fields and Stanbury (1970, 1971).
Fields and Stanbury find a significantly higher percentage of respondents report a
disincentive effect than did Break. They concluded that the disincentive effects
had become more important over time as individuals had adjusted their labor
supply slowly to the continued high marginal tax rates. But, on the other hand,
the 6% who showed significant disincentive effects in Break's survey had fallen to
only 2%, while those individuals with significant incentive effects had also
declined markedly. Both studies do find that disincentive effects increase with
income, yet we might well conclude that this finding primarily arises from an
income effect, not a substitution effect. The single important quantitative finding
in the Fields and Stanbury survey is that no significant difference exists between
average number of hours worked among groups of individuals who reported
disincentive effects, incentive effects, or no significant tax effects. Thus, whatever
net effect may exist, its likely empirical magnitude is small.
Similar interview surveys of American business executives have been conducted
by Sanders (1951) and by Holland (1969). From his interviews of 135 business
executives and 25 professional men, Sanders found the effects of taxation to be
quite small. Sanders concluded that important non-financial incentives more than
outweigh the change in financial incentives that taxation creates. Probably the
most important effect of taxation that Sanders found was the amount of time
used in creating responses to taxation through investment and tax avoidance
programs. The economic cost of this type of response is probably substantial and
has undoubtedly increased in magnitude since Sanders' survey.5
54From a social welfare analysis point of view, little comfort arises from these findings. It is
important to remember that only the substitution effect creates deadweight loss. Thus, even if the
income effect is large enough to outweigh the substitution effect, considerable deadweight loss may
still exist.
55
Executive compensation through stock options and other non-wage compensation become an
effective and important method to partly avoid the high marginal rates. But the combined effect of the
0.5 tax limit on earned income in the 1969 Tax Reform Act and the 1976 Tax Reform Act provision
for stock option plans decrease the importance of non-wage compensation. The tax legislation of 1981
increased the attractiveness of stock options to their pre-1976 status.
Ch. 4 Taes and LaborSupply 251
Holland (1969) conducted interviews of 125 business executives, in which he
attempted to isolate the substitution effect by considering a hypothetical tax on
potential income. The amount of the tax would be about the same as the tax paid
currently. However, it appears to me that the effort is not totally successful
because of the non-linearity of the budget set discussed in Section 2. There we
pointed out that the Slutsky equation does not adequately describe the tax
response because of the presence of virtual incomes different from non-labor
income. Thus, Holland's technique would seem to be exactly correct only in the
case of a proportional tax system. Holland's findings are much in line with
previous results. The hypothetical change in the tax system would have no effect
on 80% of the sample. Fifteen percent of the sample indicated they would work
harder, while one individual claimed he would work less hard. Holland seems to
conclude that on average a tax incentive effect exists, at least in the substitution
effect. But he concludes also that the magnitude is likely to be small. Thus, his
results accord well with the Break results and Sanders results.
The last sample interview we consider is Barlow, Brazer and Morgan (1966).
They conducted 957 interviews with individuals who had incomes exceeding
$10,000 in 1961. They also attempted to include a disproportionately large
number of very-high-income people in the sample. Their results are again very
similar to previous findings. Approximately 88% of the sample individuals responded
that the income tax did not affect their work effort. Among the 1/8 of
the sample which reported disincentive effects, Barlow et al. concluded that the
actual magnitude of the disincentive is likely to be very small. In fact, they
estimated the total effect on the economy to be of the order of 0.3% in 1963.
Given the rather different sample coverage, the Barlow results seem quite similar
to the results found in the other studies.
From these results we should not reach the too sanguine conclusion that high
marginal tax rates may not have a significant economic cost. We have already
mentioned the large amount of effort that goes into shifting ordinary income into
capital gains which are taxed at a much lower rate. Evidence of considerable
economic waste appears periodically from these schemes. But, the important
point to note here is that these machinations seem to have very little effect on
work effort. Presumably, for most people it is very difficult, if not impossible, to
shift compensation from working directly into capital gains. Furthermore, the
sensitivity of their work response is low to a given marginal tax rate. Perhaps
these results are not too surprising. For low-income individuals, the (uncompensated)
work disincentive effect is found to be small in econometric studies.
Previous findings that, if anything, the income effect predominates, are in accordance
with Break's findings. Recent econometric studies have found the income
effect to be the most important determinant in behavior toward taxation. Thus, in
terms of work response it does not appear that the rich are different than the rest
of us. But, they do have more money.
252 Jenry A. Hausman
7. Evidence from NIT's and social security effects
Four negative income tax (NIT) experiments have been conducted by the
government to produce information about the likely effects on labor supply of
replacing the current welfare system by an NIT. Three urban experiments took
place in New Jersey, Gary, Indiana, and Seattle-Denver. A rural experiment also
took place in Iowa and North Carolina. We review only the urban experiments,
since we have excluded farmers from our previous analysis.5
In principle, the
NIT experiments might seem to be an ideal laboratory in which to determine the
effect of taxes on the labor supply response of low-income workers.5 7
Observations
were recorded on individuals before the experiment began, and during the
three-year period of the experiments two groups were observed. The experimental
group was subject to an NIT plan, while a control group received nominal
payments to participate in the experiments. Yet the initial results were not clear
cut. Analysts found the results disappointing. A. Rees, in his summary of the New
Jersey results, concluded that "the differences in work behavior between experimentals
and controls for male heads... were, as we expected, very small.
Contrary to our expectations, all do not show a clear and significant pattern;
indeed they show a discernable pattern only after a great deal of refined
analysis."5 8
Unforeseen problems did arise which, in retrospect, is not surprising
since the New Jersey NIT experiment was the first social experiment ever
conducted. Statistical problems which arise in conducting experiments with
human subjects over time had not been accounted for.5
For instance, the
attrition problem in the New Jersey experiment almost certainly accounts for the
anomalous results found for black and hispanic males. Subsequent analysis of
the New Jersey and the other two urban NIT experiments has led to more definite
conclusions about the labor supply response. We will give a brief review of the
evidence.
We first consider the evidence for male heads of households. Two important
differences from the non-NIT framework arise for the analysis. Contrary to the
usual case of analyzing the effect of taxes on labor supply where the substitution
effect is considered to be much more important than the income effect, both the
income and substitution effects are important for an NIT. The expected ad-
5
6'A further problem exists since the results from the rural experiment seem extremely difficult to
interpret. Problems with the rural experiment are discussed in Palmer and Pechman (1978).
57
All the experiments were designed basically to measure labor supply elasticities. However, other
interesting areas of research, such as family consumption patterns and family stability, have been
studied. A possible problem with the interpretation of the NIT results is the distinction between
transitory and permanent responses. The Seattle-Denver experiment was designed to allow investigation
of this potential problem. Hausman (1982b) discusses the issue and finds evidence that it may
well affect the estimated responses.
5XFor a series of papers on the New Jersey experiment, see the Spring 1974 volume of The Journal
of Human Resources.
59These problems are discussed in Hausman (1982b).
Ch. 4 Taxes and Lbhor Supp~v 253
JerryA. Hausman
-H
$
G
Figure 7.1
ditional cost of an NIT program over the existing welfare program is a crucial
consideration. Thus, we are very interested in the overall labor supply response,
rather than just the distortion created by taxation. The second difference is that
for males both the income and substitution effects work to reduce labor supply. In
Figure 7.1 we show how the NIT alters the budget set. Non-labor income y is
replaced by the NIT guarantee G, which will have the effect of reducing labor
supply for an individual who was initially on the first budget segment so long as
leisure is a normal good. At the same time the net wage w,, which was subject
only to FICA contributions, now is lowered to i, which is subject to approximately
a 0.5 tax rate. Thus, labor supply will be reduced since the NIT budget
segment lies uniformly above the first non-NIT budget segment.6 0
For individuals
initially on the second segment, but below breakeven H, the same reasoning
holds. Non-labor income has risen from Y2, to G, and iv is less than w2. Lastly,
many individuals above breakeven hours H will not change their labor supply at
all, but others will shift down below fT because of the income effect of the
guarantee.
In fact, the findings agree with this economic theory. The labor supply
reduction in hours worked for white males in New Jersey was about 4% uncorrected
for attrition. In Gary for black males it was about 6% uncorrected for
attrition and 10% when corrected for attrition, and in Seattle-Denver the
response was 5% uncorrected for attrition. While these overall results are of
interest, they are not sufficient for policy purposes. They are an average response
6 0
So long as the NIT segment lies uniformly above the previous budget segment, the net change in
income must be positive. Thus, the income effect will reinforce the substitution effect and cause a
reduction in labor supply. Thus, the level of the guarantee removes the usual interdeterminancy of the
effect of a change in the net wage.
254
-
over the many NIT plans used in each experiment.6
' To obtain reliable cost
estimates, it is necessary to construct a model which permits determination of
income and substitution effects. Then the cost of different plans can be forecast
from the estimated parameters.6 2
Hausman and Wise (1976) was the first paper which took explicit account of
the form of the NIT budget set in constructing an empirical model. They used an
instrumental variable procedure to predict the net wage and virtual income along
with a budget segment, and estimated a log-linear labor supply specification for
white males in the New Jersey experiment. They found an uncompensated wage
elasticity of 0.14 and an income elasticity of - 0.023. Thus both effects in Figure
7.1 have significant effects in reducing labor supply. The poverty level for a
four-person family in New Jersey was $3,300. Thus, for an individual who
received the poverty limit as the guarantee and faced a 50% marginal tax rate, the
uncompensated wage effect would lead to an expected labor supply reduction of
about 8%, while the income response would lead to an expected reduction of
between 10-16% if the person had initially been on the first budget segment.
Taking midpoints we would have an overall expected response of 21% in labor
supply. An individual initially on the second segment might have no response to
an NIT at all. For those initially below breakeven hours H on the second
segment, the wage effect is 6% with the income effect leading to a reduction of 1%,
so that the overall response is about 7%. Taking weighted averages of the two
responses leads to an expected labor supply reduction for those individuals below
breakeven hours of 16.5%. It is very important to note that the model predicts
only 17.6% of the population will fall below breakeven, so that the overall
population response is about 4%.63 Some confusion has arisen over the response
conditioned on being below breakeven hours and the overall population response.
The latter response is appropriate for cost estimates of an NIT.
Burtless and Hausman (1978) analyze the labor supply response among black
males in the Gary NIT experiment. They use a procedure to treat taxes very
similar to the technique used in Section 5 except for the choice of a log-linear
labor supply function. In particular, they treat the budget set as exogenous rather
61Unfortunately, insufficient subjects were included in each cell of the experiment to use classical
analysis of variance techniques to compute an accurate estimate of the response to each NIT plan.
Statistical problems which arose during the design and duration of the experiment may preclude use of
these techniques anyway. See Hausman and Wise (1977, 1979, 1981).
62 Two potential problems arise in using the experimental results to produce cost estimates. First,
the demand side of the market could change significantly for a nationwide NIT. In particular,
individuals could choose work patterns to convexify their budget sets by working and not working in
alternative accounting periods. Also, the limited duration of the experiment may miss important
long-range effects on both the supply and demand sides of the labor market.
6The low proportion below breakeven is due to the study of white males who were relatively well
off in New Jersey and Pennsylvania. Ashenfelter (1983) estimates a probit model which focuses on the
probability of an individual being below the breakeven point.
C& 4 Txes arnd Labor Supply 255
Table 7.1
Location of fifteen budget lines in the Gary experiment.
Change Probability
Financial plan Hours Ex- from 95% confidence of below
and gross w w2 Yl Y2 at kink pected control range of break-even
wage/hr ($) ($) ($) ($) ($) point hours (%) expected hours point
Control:
2.25 2.07 1.67 2.72 27.82 43.16 43.55 - 36.8, 45.38 -
4.25 3.92 3.15 2.72 27.82 22.85 40.37 36.8, 45.38 -
6.25 5.76 4.63 2.72 27.82 27.82 40.34 - 36.8, 45.38 -
40% tax/low
guarantee:
2.25 1.35 1.67 78.63 27.82 159.59 38.68 11.8 34.15, 44.95 1.00
4.25 2.53 3.15 78.63 27.82 81.77 38.68 - 4.3 34.15, 44.95 1.00
6.25 3.75 4.63 78.63 27.82 57.45 38.68 4.2 34.15, 44.95 1.00
60% tax/low
guarantee:
2.25 0.90 1.67 78.09 27.82 65.42 38.69 -11.8 34.17, 44.96 1.00
4.25 1.70 3.15 78.09 27.82 34.63 39.62 1.9 34.16, 43.48 0.21
6.25 2.50 4.63 78.09 27.82 23.55 40.23 - 0.3 34.16, 45.38 0.02
40% tax/high
guarantee:
2.25 1.35 1.67 102.63 27.82 234.97 38.27 -12.9 33.50, 44.85 1.00
4.25 2.53 3.15 102.63 27.82 120.39 38.27 - 5.3 33.50, 44.85 1.00
6.25 3.75 4.63 102.63 27.82 84.59 38.27 - 5.3 33.50, 44.85 1.00
60% tax/high
guarantee:
2.25 0.90 1.67 101.09 27.82 96.66 38.29 -12.9 33.50, 44.84 1.00
4.25 1.70 3.15 102.09 27.82 51.17 38.29 - 5.3 33.50, 44.84 1.00
6.25 2.50 4.63 102.09 27.82 34.86 39.38 2.4 33.50, 45.38 0.23
than using an ad hoc instrumental variable procedure, and they also allow for a
distribution of tastes in the population. Here in Table 7.1 we present their results
for both control individuals and for experimental individuals on a weekly basis
for the mean individual in the sample. We first note that breakeven hours are
quite high for some of the plans, so that the individual will almost certainly be
below breakeven. Also note that a significant dispersion exists in the expected
response - it is about 13% for low-wage groups. Perhaps even more importantly
the distribution of tastes parameter indicated that most of the response takes
place via the income effect for a small number of individuals. The great majority
of individuals do not significantly alter their work response, so that the effect of
the NIT leads to a very skewed response in the population. On the other hand,
the uncompensated wage change has very little effect. We can see the income and
substitution effects by comparing the rows which correspond to a $2.25 wage
since the individuals will always be on the first budget segment. No difference in
response at all is found for the 0.4 or 0.6 tax rate, while the high guarantee leads
to a 9% greater response than does the low guarantee. At higher wage rates the
Jerry A. Hausman256
amount of the tax does play a role, but only because it changes the amount of
breakeven hours and thus the probability of being above breakeven. The finding
that the income effect is the major determinant of labor supply reduction among
males was also found by Moffitt (1978), who used a quite different probit type of
model. The results differ markedly from the Hausman-Wise findings for New
Jersey, where the income effect explained about 68% of the change in hours. It
would be interesting to determine if this result occurs because of different model
specification or because of a fundamentally different response pattern among the
two populations.6 4
Other labor supply results for males that we briefly review are the findings of
Keeley et al. (1978) for the Seattle-Denver experiment. While the SeattleDenver
experiment is superior in certain respects to the other urban experiments,
the ad hoc method used by the authors to treat the budget set is not entirely
satisfactory. They use a first difference specification, where the change in income
is done at pre-experimental hours of work for the individual as is the change in
the net wage rate. Since pre-experimental hours are an endogenous variable, an
important simultaneous equation bias may be introduced.65
However, the magnitude
of the bias is difficult to estimate. At the mean of the sample, Keeley et al.
found the income effect to explain 46% of the reduction in hours, while the
change in the wage explained the other 54%. These results differ markedly from
the results in the New Jersey and Gary sample, where the change in non-labor
income is the more important determinant of the reduction in labor supply.
Again, it would be interesting to ascertain whether the different results arise
because of the model used.
The last study of male labor supply response in the NIT experiments that we
review is MaCurdy (1983), who estimates a marginal rate of substitution function
between consumption and labor supply from the Denver NIT. As we discussed
previously, MaCurdy estimates his model within a life cycle context. MaCurdy's
results are extreme in relation to other research on labor supply response. His
estimated compensated substitution elasticity lies in the range of 0.74 to 1.43,
which is on the order of 10-20 times higher than other estimates. Similarly, his
estimated income elasticities range from - 0.74 to - 0.43, which are much greater
in magnitude than other studies. It is possible that these results arise from the
incorporation of the previously neglected life cycle considerations; but it seems
more likely that they arise because of econometric problems associated with his
marginal rate of substitution specification.
64
The finding that it is the income effect which creates almost the entire labor supply response, is
corroborated further by the results of Hausman and Wise (1979) who consider a model which corrects
for attrition.
65
A further problem exists since people initially above breakeven hours will not have their net wage
or income affected by the experiment. The authors attempt to treat this problem by including a
dummy variable which again would create simultaneous equation bias.
Ch. 4 Taxes ad Lor Supply 257
The other group whose labor supply might be markedly affected by introduction
of an NIT, is working wives. Neither the New Jersey nor the Gary
experiments had sufficient numbers of working wives to allow model estimation.
Keeley et al. found an average response elasticity about four times as large for
wives as for husbands. The mean labor force reduction is 22%. Here the change in
income accounted for 75% of the total effect. Since most of these women
presumably had working husbands, such a large-scale withdrawal from the labor
force could be an important effect of an NIT.
The last group to be considered is female headed households. Most of the
affected population qualifies for AFDC, so that introduction of an NIT leads to a
substantially higher guarantee and somewhat lower tax rate under most NIT
plans. Keeley et al. found the female heads response to be about twice as large as
the male response. The mean labor force reduction is 11%. Here they found the
income effect to explain about 66% of the total response. Hausman (1980), in a
study of labor force participation among black females who headed households in
Gary, again found the level of the guarantee to be much more important than the
NIT tax rate. For instance he finds that the change from 0.4 to 0.6 NIT tax rate
reduces the probability of participation by about 2.5%, while a change in the NIT
guarantee from 0.75 of the poverty limit to the poverty limit reduces the
probability of participation by 6.5%. In terms of comparing the expected effort to
that of AFDC, it seems likely that a reduction in labor supply would result. Even
if the marginal tax rate fell from the AFDC level of 0.67 to an NIT level of 0.4,
the accompanying higher benefits would create a net disincentive effect. The net
result would be a significant increase in the cost of family support for female
headed households. At the same time the extra income, which would go to the
lowest-income group in the economy, might well lead to a net gain in social
welfare.
The other literature which we review considers the effect of the social security
earnings test on retirement behavior and labor supply. We discussed the social
security beneficiary budget set in Section 4, where we emphasized the intertemporal
aspects of the model. An important empirical fact does appear with respect
to social security. Labor force participation has decreased among the elderly over
the post-war period in the United States. From 1960 to 1975, labor force
participation for males over 65 fell from 33% to 22%. Over the same period for
men aged 62-64, it fell from 81% to 60%. 1961 is the year in which social security
eligibility for men 62-64 was introduced. An important policy question is
whether the decline in male labor force participation is almost wholly a result of
the early retirement provision of social security and the rising real benefit level.
Rising real income for potential retirees during the period offers an alternative
explanation for part of the observed behavior. Given recent policy proposals to
extend the age of early retirement from 62 to 65 years of age, the causes of early
retirement assume an important role in financial projections for the social security
258 JerrvA. Hausman
Ch. 4: Taxes and Labor Supply
$
y+B
Figure 7.2
system. Three recent papers consider the causes of retirement over the 1965-1975
period. Boskin and Hurd (1982) ascribe almost all the decrease in labor force
participation to social security. Diamond and Hausman (1984a, 1984b) find that
social security is the most important factor, yet if early retirement between ages
62-65 were stopped, the retirement probability would decrease by about 50%, so
that a significant number of men would still retire during that age period.66
Further research is required here because of the complex interaction of non-retirement
labor supply and its effect on future social security benefits; cf. Blinder et al.
(1980).
The other dimension of the effect of social security is the earnings test for social
security recipients. In 1982 earnings beyond $6,000 are subject to a 50% tax rate
until all social security benefits are recovered, i.e., the breakeven point H is
reached in Figure 7.2. Burtless and Moffitt (1982) in a recent study find that the
earnings test has a major effect on retired males labor supply decisions.6 7
Among
retired men who are working the frequency distribution of hours worked has a
pronounced spike at the kink point H in Figure 7.2 which provides strong
evidence of the incentive effect of the earnings test. This effect is to be expected
66
These studies are in stark contrast to Gordon and Blinder (1980) who in their study of retirement
decisions "assume(s) that social security is irrelevant to retirement decisions". No empirical study, to
the best of my knowledge, has come close to verifying this assumption.
67
Burtless and Moffitt go a long way towards a complete model of the lifetime budget set. However,
since their model is basically cross-sectional, they do not account for increases in future social security
benefits from extra years of work in an entirely satisfactory manner.
259
given the pronounced kink at HI hours, where the net wage is reduced from w to
0.5w. About 50% of working males were located at the kink point and 90%
worked hours either at or below the kink point H1 . However, to analyze the
overall effect of the earnings test, non-participation must be accounted for since
upwards of 80% of the men in the Burtless-Moffitt sample worked zero hours.
Thus, overall the earnings test leads to a reduction in expected hours of about 50
hours per year. However, for those men who are working, the removal of the
earnings test would increase expected labor supply by about 400 hours per year.
Thus, the situation is very similar to the case of wives discussed in Section 5,
where tax changes have only a small expected effect on non-labor force participants.
But considering the problem in this context, Burtless and Moffitt may
have overestimated the wage elasticity by their neglect of fixed costs to working.6 8
Still, they have provided strong evidence of the effect of the earnings test on the
labor supply behavior of social security recipients.
In this section we have considered the empirical evidence from the NIT
experiments. Although numerous statistical and econometric problems arise, I feel
we have learned much about labor supply behavior of low-income workers. We
now return to our question of the last section. There, we decided that labor
supply behavior of high-income persons was not too different from that of
middle-income individuals. What about low-income people? From the experimental
results, I conclude that the income effect is probably larger than we previously
had thought. Especially for male heads of households I feel that introduction of
an NIT would lead to a significant labor force supply reduction by a small
proportion of the population. I doubt that the NIT tax rate is nearly as important
as the level of the NIT guarantee. Thus, low-income males do have low wage
elasticities as does the rest of the population; but their income elasticities may
have an important effect on labor supply behavior given the size of the NIT
guarantee. Similar results were found for female headed households, although
they presently have AFDC so that the change might not be as large. Lastly, the
NIT results for wives seem quite different than the usual results. Their wages
elasticities are much lower and their income elasticities are much higher than had
been found for middle-income wives. But, the evidence on wives is based on only
one sample and one estimation technique. More research needs to be done on
wives behavior under an NIT before we can be confident about the results.
Research of the effects of social security is still in an early stage of development.
Little doubt would seem to exist that social security benefits are an
important determinant of retirement decisions. Furthermore, the earnings test
does have an important effect on labor supply behavior of retirees. Further
research that accounts for the intertemporal aspects of the problem and the form
of the lifetime budget set is still needed.
68
Hausman (1981a) found a lower wage elasticity for wives when fixed costs to working here
accounted for. Cogan (1981) made a similar finding in a model without taxes.
260 JerryA. Hausman
Ch. 4: Taxes and Labor Supply
References
Ashenfelter, O., 1983, Discrete choice in labor supply: The determinants of participation in the
Seattle-Denver income maintenance experiment, Journal of the American Statistical Association.
Ashenfelter, 0. and J. Heckman, 1974, The estimation of income and substitution effects in a model of
family labor supply, Econometrica 42.
Ashworth, J.A. and D.T. Ulph, 1981a, Estimating labour supply with piecewise linear budget
constraints, in: C. Brown, ed., Taxation and labor supply (London).
Ashworth, J.A. and D.T. Ulph, 1981b, Household models, in: C. Brown, ed., Taxation and labour
supply (London).
Atkinson, A. and N. Stern, 1980, On labour supply and commodity demand, Journal of Public
Economics 14.
Atkinson, A. and J. Stiglitz, 1976, The design of tax structure: Direct versus indirect taxation, Journal
of Public Economics 6.
Atkinson, A. and J. Stiglitz, 1980, Lectures on public finance (New York).
Atkinson, A. et al., 1980, On the switch from direct to indirect taxation, Journal of Public Economics
14.
Auerbach, A., 1984, Measurement of excess burden and optimal taxation, this volume.
Barlow, R., H. Brazer and J. Morgan, 1966, Economic behavior of the affluent (Washington, DC).
Blinder, A. et al., 1980, Reconsidering the work disincentive effects of social security, National Tax
Journal 33.
Blomquist, S., 1983, The effect of income taxation on male labor supply in Sweden, Journal of Public
Economics 22.
Blundell, R. and C. Walker, 1982, Modelling the joint determination of household labor supply and
commodity demands, Economic Journal 92.
Blundell, R. and C. Walker, 1984, A life cycle consistent empirical model of family labor supply using
cross section data, Mimeo.
Borjas, G. and J. Heckman, 1978, Labor supply estimates for public policy evaluation, Proceedings of
the IRRA.
Boskin, M. and M. Hurd, 1982, The effect of social security on retirement in the early 1970's, Mimeo.
Break, GC.,1957, Taxes and incentives to work: An empirical study, American Economic Review 48.
Brown, C. et al., 1976, Estimate of labour hours supplied by married male workers in Great Britain,
Scottish Journal of Political Economy 23.
Brown, C. et al., 1982, Preliminary family labour supply estimates, Mimeo.
Burtless, G. and J. Hausman, 1978, The effect of taxes on labor supply: Evaluating the Gary NIT
experiment, Journal of Political Economy 86.
Burtless, G. and R. Moffitt, 1982, The effect of social security on labor supply of the aged, Mimeo.
Cain, G. and H. Watts, 1973, Income maintenance and labor supply (Chicago, IL).
Cogan, J., 1980, Married women's labour supply, in: J. Smith, ed., Female labor supply (Princeton,
NJ).
Cogan, J., 1981, Fixed costs and labor supply, Econometrica 49.
Deaton, A., 1981, Theoretical and empirical approaches to consumer demand under rationing, in: A.
Deaton, ed., Essays in the theory and measurement of consumer behavior (Cambridge, MA).
Dcaton, A. and J. Muellbauer, 1980, Economics and consumer behavior (Cambridge, MA).
Deaton, A. and J. Muellbauer, 1981, Functional forms for labor supply and commodity demand with
and without quantity rationing, Econometrica 49.
Diamond, P. and J. Hausman, 1984a, Individual retirement and savings behavior, Journal of Public
Economics 23.
Diamond, P. and J. Hausman, 1984b, Retirement and unemployment behavior of older men, in: H.
Aaron and G. Burtless, eds., Retirement and economic behavior (Washington, DC).
Diamond, P. and D. McFadden, 1974, Some uses of the expenditure function in public finance,
Journal of Public Economics 3.
Diewert, E., 1971, Choice on labor markets and the theory of the allocation of time, Mimeo.
Diewert, E., 1982, Duality approaches to consumer theory, in: K. Arrow and M. Intriligator, eds.,
Handbook of mathematical economics (Amsterdam).
261
Feenberg, D. and H. Rosen, 1983, Alternative tax treatment of the family: Simulation methodology
and results, in: M. Feldstein, ed., Behavioral simulation methods in tax policy analysis (Chicago,
IL).
Fields, D. and W. Stanbury, 1970, Disincentives and the income tax: Further empirical evidence,
Public Finance 25.
Fields, D. and W. Stanbury, 1971, Income taxes and incentives to work: Some additional empirical
evidence, American Economic Review 60.
Gordon, R. and A. Blinder, 1980, Market wages, reservation wages, and retirement decisions, Journal
of Public Economics 14.
Greenburg, D. and M. Kosters, 1973, Income guarantees and the working poor, in: G. Cain and H.
Watts, eds., Income maintenance and labor supply (Chicago, IL).
Hall, R., 1973, Wages, income and hours of work in the U.S. labor force, in: G. Cain and H. Watts,
eds., Income maintenance and labor supply (Chicago, IL).
Hall, R., 1982, The importance of lifetime jobs in the U.S. economy, American Economic Review 72.
Ham, J., 1982, Estimation of a labour supply model with censoring due to unemployment and
underemployment, Review of Economic Studies 49.
Hanoch, G., 1980, A multivariate model of labor supply, in: J. Smith, ed., Female labor supply
(Princeton, NJ).
Hanoch, G. and M. Honig, 1978, The labor supply curve under income maintenance programs,
Journal of Public Economics 9.
Harberger, A., 1964, Taxation, resource allocation and welfare, in: The role of direct and indirect
taxes in the Federal Revenue System (NBER).
Hausman, J., 1979, The econometrics of labor supply on convex budget sets, Economics Letters 3.
Hausman, J., 1980, The effect of wages, taxes, and fixed costs on women's labor force participation,
Journal of Public Economics 14.
Hausman, J., 1981a, Labor supply, in: H. Aaron and J. Pechman, eds., How taxes effect economic
activity (Washington, DC).
Hausman, J., 1981b, Exact consumers' surplus and deadweight loss, American Economic Review 71.
Hausman, J., 1981c, Income and payroll tax policy and labor supply, in: L.H. Meyer, ed., The
supply-side effects of economics policy (St. Louis, MO).
Hausman, J., 1982a, The econometrics of non-linear budget sets, Mimeo.
Hausman, J., 1982b, The effect of time in economic experiments, in: W. Hildenbrand, ed., Advances in
econometrics (Cambridge).
Hausman, J., 1983, Stochastic problems in the simulation of labor supply, in: M. Feldstein, ed.,
Behavioral simulation methods in tax policy analysis (Chicago, IL).
Hausman, J. and P. Ruud, 1984, Family labor supply with taxes, American Economic Review 74.
Hausman, J. and D. Wise, 1976, The evaluation of results from truncated samples, Annals of
Economic and Social Measurement 5.
Hausman, J. and D. Wise, 1977, Social experimentation, truncated distributions, and efficient
estimation, Econometrica 45.
Hausman, J. and D. Wise, 1978, A conditional probit model for qualitative choice, Econometrica 46.
Hausman, J. and D. Wise, 1979, Attrition bias in experimental and panel data, Econometrica 47.
Hausman, J. and D. Wise, 1981, Stratification on endogenous variables and estimation, in: C. Manski
and D. McFadden, eds., The econometrics of discrete data (Cambridge, MA).
Holland, D., 1969, The effect of taxation on effort: The results for business executives, in: S.J. Bowers,
ed., Proceedings of National Tax Association.
Holland, D., 1976, The effect of taxation on incentives of higher income groups, in: IFS, Fiscal policy
and labour supply, No. 4.
Kay, J., 1980, The deadweight loss from tax system, Journal of Public Economics 10.
Keeley, M. et al., 1978, The estimation of labor supply models using experimental data, American
Economic Review 68.
Killingsworth, M., 1983, Labor supply (Cambridge).
Kosters, M., 1969, Effects of an income tax on labor supply, in: A. Harberger and M. Bailey, eds., The
taxation of income capital (Washington, DC).
Lewis, G., 1956, Hours of work and hours of leisure, IRRA Proceedings 9.
MaCurdy, T., 1981, An empirical model of labor supply in a life-cycle setting, Journal of Political
Economy 89.
262 JerryA. Hausman
MaCurdy, T., 1983, A simple scheme for estimating an intertemporal model of labor supply and
consumption in the presence of taxes and uncertainty, International Economic Review 24.
Mirrlees, J., 1971, An exploration in the theory of optimal income taxation, Review of Economic
Studies 38.
Mirrlees, J., 1982, The theory of optimal taxation, in: K. Arrow and M. Intriligator, eds., Handbook of
mathematical economics (Amsterdam).
Moffitt, R., 1979, A note on the effect of taxes and transfers on labor supply, Southern Economic
Journal.
Moffitt, R. and W. Nicholson, 1982, The effect of unemployment insurance on unemployment, Review
of Economics and Statistics 64.
Nakamura, A. and M. Nakamura, 1981, A comparison of the labor force behavior of married women
in the U.S. and Canada, with special attention to the impact of income taxes, Econometrica 49.
Neary, J. and K. Roberts, 1980, The theory of household behavior under rationing, European
Economic Review 13.
Palmer, J. and J. Pechman, 1978, Welfare in rural areas: The North Carolina-Iowa income maintenance
experiment (Washington, DC).
Pechman, J., 1976, Federal tax policy, 3rd ed. (Washington, DC).
Rosen, H., 1976, Taxes in a labor supply model with joint wage-hours determination, Econometrica
44.
Ruffell, R.J., 1981, Direct estimation of labor supply functions with piecewise linear budget constraints,
in: C. Brown, ed., Taxation and labor supply (London).
Sanders, T., 1951, Effects of taxation on executives (Boston, MA).
Seater, J., 1982, Marginal federal personal and corporate income tax rates in the U.S., 1909-1975,
Journal of Monetary Economics 10.
Smith, J., 1980, Female labor supply (Princeton, NJ).
Steverle, E. and M. Hartzmark, 1981, Individual income taxation, 1947-79, National Tax Journal 34.
Stiglitz, J., 1982, Utilitarianism and horizontal equity, Journal of Public Economics 18.
Tax Foundation, 1981, Facts and figures on government finance (Washington, DC).
Varian, H., 1978, Microeconomic theory (New York).
Wales, T., 1973, Estimation of a labor supply curve for self-employed business proprietors, International
Economic Review 14.
Wales, T. and A. Woodland, 1979, Labor supply and progressive taxes, Review of Economic Studies
46.
Zabalza, A., 1983, The CES utility function, nonlinear budget constraints, and labor supply, Economic
Journal.
Zabalza, A. et al., 1980, Social security and the choice between full-time work, part-time work and
retirement, Journal of Public Economics 14.
Ch. 4: Taxes and Labor Supply 263