As globalization is increasingly affecting individuals, the domestic
influences on international trade become especially important to
understand. A central aspect of any constituent-based political-economy
model of trade policy, explicitly or implicitly, is a theory of
intersectoral factor mobility. The level of factor mobility—or the
degree of factor specificity—defines the ease with which the factors
of production can move among sectors of the economy. As factor mobility
varies, so do the expected constituent coalitions for and against trade
liberalization. The level of factor mobility, then, is an essential tool
in developing theoretical and empirical models of constituent interests
for trade policy. Furthermore, the level of factor mobility also has
significant effects on an individual's income, the depth and breadth
of the economic impact of globalization on society, the ability of the
political parties to represent the vying constituent coalitions, as well
as on the volume of international trade itself. Thus whether factor
mobility is relatively high or low has far-reaching but vastly different
consequences in the economy, in society, and in politics.
The level of factor mobility cannot be measured directly. With a few
notable exceptions, it and the appropriate modeling of constituent
coalitions for and against trade policy are too often assumed or not
considered. Thus it may not be surprising that an examination of the
literature on recent U.S. trade policy and constituent coalitions does not
produce, in the aggregate, consistent conclusions. Although some studies
have shown evidence that factor mobility is consistent with it recently
being relatively low,1
others have reported evidence that is consistent
with its being relatively high.
2 Hiscox provides
probably the most extensive studies of factor mobility to date.
3
He analyzes factor mobility in six countries, including the United States,
over a span of nearly two centuries. He too finds evidence of high and low
factor mobility during his most recent pooled time period for the United
States, 1970–94.
4 In this study, I try to bridge the theoretical and empirical gaps
manifested by the seemingly inconsistent findings among the multiple
studies that have addressed constituent interests and trade politics in
the United States from 1963 to 1992. To do so, I directly test whether
factor mobility is relatively low or high and if it changes.5
Determining the precise level of factor mobility
is nearly impossible. The purpose here is to provide an insight into the
general trend.
Using U.S. House of Representative roll call
voting and an original data set of constituent economic demographics at
the level of House districts, I argue that factor mobility was indeed
relatively low in the 1960s and 1970s. It gradually rose, however, and
reached a pivot point, or critical threshold, near the end of the 1970s
that tipped the balance between relatively low and relatively high factor
mobility. Factor mobility remained relatively high in the 1980s and 1990s.
In other words, I show that factor mobility is a societal characteristic
that takes years to gradually change but has profound effects throughout
the economy and politics.
6 I also demonstrate
that constituent economic coalitions significantly affect legislative
voting behavior in a manner consistent with the predictions of the
economic theories of trade policy. Finally, on an expanded set of
international economic policy votes, I check (1) the robustness of the
findings on the strength of the political parties in supplying these
policies and (2) the influence of the partisan control of the branches of
the federal government. The robustness results confirm the expected
political consequences of a change in the relative level of factor
mobility.
Five sections follow. First I present the two main economic theories
of factor mobility and constituent interests for international trade, and
I examine the aggregate inconsistencies in the current literature. Next I
explain the construction of the data set employed and the study's
hypotheses. Third I report and discuss the findings; and fourth check the
robustness of the findings. I conclude with the theoretical and practical
implications of the study.
Constituent Trade Interests and Factor
Mobility
Constituent trade interests are modeled by one of two well-established
economic theories: the Ricardo-Viner or the Stolper-Samuelson theory. For
the purposes of this study, there are three general but critical
differences between these models. First, the Ricardo-Viner model assumes
that factor mobility is low; the Stolper-Samuelson model assumes that it
is high. Second, the models predict different constituent coalitions in
society, based on the way that gains and losses in factor income are
distributed. The Ricardo-Viner model predicts sectoral constituent
coalitions; the Stolper-Samuelson model predicts factoral constituent
coalitions. Because factor mobility cannot be both high and low at the
same time, the two models are, by definition, mutually exclusive. Over
time, however, factor mobility can gradually change and, hence, can
exhibit characteristics of a continuous variable.7
Hiscox 2002a and 2002b, for instance, makes a similar
argument.
Third, the coalitions predicted by both economic models
vary considerably in form. When factor mobility is high, the resulting
factoral constituent coalitions are, by definition, relatively
homogeneous; when factor mobility is low, the resulting sectoral
constituent coalitions are, by definition, relatively heterogeneous.
8 This conclusion is based on the economics
assumptions of each model.
Specifically, the Stolper-Samuelson model assumes that the factors of
production—labor and capital—are mobile, and thus the model
derives preferences and predicts coalitions based on the relative
abundance or scarcity of the factors.9
Consistent with many recent works on the United States, the
factor of land is omitted.
The more abundant factor will be
advantaged by the comparative gains of international trade and experience
income gains. Therefore, the economic interest of the owners of the
abundant factor is to favor an increasingly liberal trade environment.
Owners of the abundant factor and the firms that intensely employ it will
not only gain absolutely from liberalization but also relative to the
owners of the scarce factor and the industries that intensely use it
because of the redistributive income effects between them.
10 In the United States, given its
economic development and current endowments, capital is the abundant
factor and labor is relatively scarce.
11 Accordingly, capital and
capital-intensive firms derive greater long-term benefits from free trade.
At the same time, labor experiences greater competition and falling wages,
and therefore favors policies that decrease the exposure of the domestic
economy to the international economy.
The Ricardo-Viner model assumes that the factors of production are
“sticky” and cannot be reallocated swiftly to more efficient
sectors of the economy.12
See Alt and Gilligan 1994 for an excellent description of
the full logic of the Stolper-Samuelson and the Ricardo-Viner
models.
As such, labor and capital will not be competing against
each other. In other words, capital and labor in the same sector share the
same fate. Each will either be similarly advantaged or similarly
disadvantaged depending on the sector's competitiveness in the
domestic and international economies. In consequence, the coalitional
cleavages will develop not along factoral but sectoral lines
(export-competitive versus import-competing industries). Specifically,
export-competitive industries will experience income gains from free trade
and so favor trade liberalization; import-competing industries will not.
Thus in the United States, when factor mobility is low, sectoral (that is,
industry) divisions will form; when factor mobility is high, factoral
(that is, class) divisions over trade liberalization will form.
The Ricardo-Viner and the Stolper-Samuelson theories are ideal-type
theories. Using these models is empirically useful, as I discuss below,
even though the actual level of factor mobility is unlikely to conform to
their idealized assumptions. Instead, because the level of factor mobility
is likely to gradually change over time, it is a secular trend.13
As the actual level of factor mobility gradually moves along a continuum
between the two ideal types, there should be a pivot point that
distinguishes the two. On one side of the pivot point, factor mobility is
relatively low; on other side, it is relatively high. As the median
voter's economic interest is tipped from one side to the other, there
should be predictable and observable economic and political implications.
The level of factor mobility, however, is not directly measurable. It is
usually indirectly inferred from these predictable and observable economic
and political manifestations, such as the significance of sectoral or
factoral variables that influence reelection-minded legislators on trade
policy. The aggregate conclusions from such political economy studies of
trade policy, however, offer contradictory arguments or evidence on the
recent balance of the relative level of factor mobility in the United
States. Some have reported evidence of sectoral influence on congressional
lobbying and voting behavior.
14 Others have reported factoral
influences.
15 Still others have found evidence of
both: exporters and importers, as well as labor and business groups, have
had a significant influence.
16 There are good reasons, though, why studies
have varied so much in their conclusions.
First, most of the studies are concerned not with determining the
relative level of factor mobility but with modeling other phenomena. In
such cases, factor mobility is frequently assumed, if it is considered at
all, to be either low or high. For this reason, the tested constituent
economic variables may represent just one of theories, both theories, or
parts of each. Because the two models are theoretically mutually
exclusive, however, methodological concerns arise when the two models are
tested simultaneously. Nonetheless, because factor mobility can vary
continuously between the two ideal-type models, it is possible that both
sectoral and factoral variables may be significant. As this study
emphasizes, however, the balance between relatively low and high factor
mobility has important political and economic implications that are best
informed by evaluating the comparative performance of each model.
Second, in that the relative level of factor mobility varies, it is
possible that at some point during the thirty years that the balance
between the two ideal-type models has changed. For instance, empirical
evidence may show that during one Congress sectoral influences are more
significant and during a subsequent Congress, factoral influences are more
significant. This would be theoretically consistent only if the relative
level of factor mobility has also changed accordingly. Similarly, if roll
call votes are pooled together over a time period in which the balance
between relatively low and relatively high factor mobility changes, the
tests may produce seemingly contradictory results.
Third, the scope of studies on the domestic influences on trade policy
range from analyses of a single roll call vote in one Congress to analyses
of roll call votes over nearly two centuries. Given the limitations of
available data, particularly for the latter, some of the studies have used
state-level data to predict voting patterns in the House. Though this may
provide uniformity over time, it raises concerns of ecological fallacies
and spurious results.
The current literature has provided valuable results, but this study
differs from and contributes to the literature in at least three ways.
First, to accurately model the effects of constituent interests on the
voting behavior of members of the House, I use House-district-level
economic data and analyze each of the fifteen Congresses individually. I
use U.S. Census labor and economic data on each of the twenty two-digit
Standard Industrial Classification (SIC) manufacturing sectors, and
national data on exports, imports, and capital to create a data set of
economic demographic variables for each House district over the thirty
years for each of the factoral and sectoral coalitions for and against
trade liberalization. Testing data across consistent geographic units has
been lacking, largely because of the difficulties in collecting and coding
disaggregated Census data. This data set, however, provides unit
consistency between detailed district-level constituent economic interests
over time and House roll call voting.
In addition, the district-level economic variables that I employ
discretely capture each of the four possible constituent coalitions:
export-competitive versus import-competing sectors or capital versus
labor. These four coalitions can be considered input variables that
influence rational legislators and can be used to reveal their
preferences.17
Others have
used variables that represent the output from trade policies and factor
mobility, such as wage rates and corporate profit data.
18 Under pure neoclassical assumptions, these variables
would bear witness to the pressures of international trade in the same way
that the input variables would conform to the results from revealed
preferences. However, there are reasons to question the reliability of
output variables because they are also directly affected by domestic
policies, such as minimum wage laws and corporate tax policies. Because
government policies may alter the economic equilibrium,
19 it is possible that these data are more likely to
produce a “false positive.” One way to avoid such
difficulties, then, may be to ask individuals directly what their trade
preferences are.
20 This avoids
indirect analyses and endogenous outcomes. Although this is certainly
true, other hurdles arise with public opinion surveys. For instance, the
American National Election Survey, the most reliable public opinion survey
of political preferences over time, has not consistently asked trade
questions in all of the years included in this study. To be sure, there
are other means of estimating the level of factor mobility; however, using
constituent coalitions (input variables) to determine the revealed
preferences of rational legislators is the most reliable and readily
available method in this case.
The second difference between this study and many of the previous
studies is the concentration on determining the balance between relative
low and high factor mobility. Significant exceptions to this include.21
See Hiscox 2002a and
2002b; Scheve and Slaughter
2001a and 2001b. Scheve and Slaughter 2001b analyze the 1993 American
National Election Survey and find evidence of factoral influences. Though
the year 1993 follows the last year of this study, it is included in
Hiscox last time period of 1970 to 1994. Similar to some of the cited
studies, they come to different conclusions.
Identifying the
level of factor mobility may inform previous trade politics studies of
possible omitted variables and guide future studies.
The third difference between this study and many of its forerunners is
its extension of factor mobility's predictable political
consequences: in particular, the ability of political parties to
consistently and coherently represent constituent interests, as well as
the effects of divided government on trade liberalization. Analyzing two
political consequences of changes in the balance of the relative levels of
factor mobility provides a stronger theory and a robustness check on the
initial empirical findings. Furthermore, the inclusion of political
parties and divided government in the analysis aids in accounting for
politically endogenous outcomes.
Factor mobility affects political parties by changing the relative
level of homogeneity of constituent coalitions.22
To see this, a return to the ideal-type models is appropriate. When factor
mobility is high, the disadvantaged factor (labor in the U.S. case) can
change its competitiveness only by altering its relative scarcity—a
process that would mostly likely take decades. Furthermore, the costs and
benefits of international competition extend beyond the tradable
industries to all other industries as well. For instance, because labor in
tradable industries is predicted to experience downward pressure on wages
and greater insecurity, there will be additional negative pressures on
labor in all industries. In this case, the costs and benefits of economic
liberalization become relatively nonexcludable as the number of those
either advantaged or disadvantaged expands across society. Moreover, the
class-based pressures on the redistribution of factor income may align
with other typically partisan class-based interests and provide a potent
political alliance for each party. The effects of high factor mobility,
then, extend beyond the tradable sectors of the economy—and even
trade policy—and can unite broad interests into relatively
homogeneous and polarized coalitions. In this way, as Alt and Gilligan
conclude, “there really seems to be a natural affinity between the
Stolper-Samuelson model and majoritarian politics”
23 such as strong political parties.
Alternatively, when factor mobility is low, an industry that is
disadvantageously positioned in the international economy may quickly find
its competitiveness changed through, for instance, technological advances.
Also, the costs and benefits of international competition are largely
limited to the exporting and importing industries. As the income effects
are distributed across just the tradable sectors and not more broadly
between the classes, there are few income externalities in the nontradable
industries. Hence, the sectoral constituent coalitions produced are
characterized by fragmented and ephemeral alliances: in other words, they
are heterogeneous. Consequently, periods of relatively low factor mobility
produce relatively weak constituent coalitions on issues of trade
liberalization and, consequently, each party will have a relatively weak
constituent foundation on which to base policy alternatives.
In sum, the balance of the relative level of factor mobility affects
the degree of constituent homogeneity on trade policy issues.24
The overall homogeneity of partisan
constituent coalitions and the strength of the political parties have many
contributing influences. This study focuses on just one: factor
mobility.
The overall degree of constituent homogeneity
significantly also affects the degree of homogeneity of the political
parties: the greater the intraparty homogeneity and the interparty
polarization, the stronger the parties will be.
25 This implies that when factor mobility is low,
“partisanship is likely to be irrelevant because each party tends to
represent multiple interests groups with different preferences.”
26 When factor mobility is high, “parties
should be unified internally on trade and divided [between] each
other along class lines.”
27 Ibid., 99.
Hiscox 2002c, 7–8, makes this pair of
arguments as well: low factor mobility facilitates weak parties and high
factor mobility facilitates strong parties.
This conclusion is
further supported if the parties broadly represent class interests across
other issues.
28 If congressional voting for trade policy is along
sectoral lines, then party unity is naturally expected to be less than if
voting for trade policy is along factor lines.
29 I would like to thank the editor for emphasizing this
point.
The introduction of party politics, however, raises the important and
much debated issue of the effects of partisan control of branches of the
federal government; in other words, the effects of divided government on
trade liberalization. For instance, O'Halloran and Lohmann and
O'Halloran argue that legislators in unified governments are more
likely to grant trade-making authority to the president, and that
legislators in divided governments are less likely to do so.30
Karol, on the other hand, stresses the role of
“Presidential liberalism”
31 and
partisan constituent coalitions and preferences. Because of these, Karol
argues, “presidents from the protectionist party gain from divided
government, and those from the liberal party are harmed by it.”
32 A wrinkle in this, though, is that free-trading Democrats of
the 1960s had become protectionists by 1980, while protectionist
Republicans were becoming consistent free traders.
33 These studies on divided government, however, understate or do not
address the relative level of factor mobility and the changing strength of
the political parties. Once the latter are considered, then it is logical
to theorize that the effects of unified/divided government matter most
when the parties are strong and stand for clear and consistent policy
alternatives. Therefore, the effects of unified/divided government
ought to be mitigated when factor mobility is low because the relatively
heterogeneous constituent interests for trade policy weaken the
party-constituency bonds. Conversely, the effects of unified/divided
government ought to be more prominent when factor mobility is high because
of the relatively homogeneous and polarized party-constituency bonds. Even
though the relative free-trading position of the president may also be
affected by partisanship,34
a general degree of
presidential liberalism is assumed. Thus if the parties are strong and the
majority congressional party consists of protectionists, divided
government should be negatively correlated with trade liberalization; if
the majority congressional party consists of free-traders, then divided
government should be positively correlated with trade liberalization.
35 This general theoretical construction depends
on the position of the status quo and does not consider other
possibilities. A more extensive discussion, however, is beyond the scope
of this article.
This construction largely corresponds to
Karol's model (2)
36 and is
similar to the constructions used by Schnietz and Bailey, Goldstein, and
Weingast.
37 If the parties are weak, then both of these effects
should be mitigated.
Data and Hypotheses
The rational decisions of reelection-minded legislators are commonly
used to reveal the preferences of their constituents. In this case, the
influence of constituent coalitions on legislators' trade policy roll
call votes can be employed as a measure of the relative strength of
sectoral and factor coalitions as a proxy for the relative level factor
mobility. Thus I use a dependent variable of roll call votes on trade
policy in the House from 1963 to 1992 for all of the initial
regressions.38
The time period is chosen for
two reasons. First, it begins after the release of the 1960 U.S. census
and the subsequent House reapportionment. The 1960 census provides
substantially better data than any time period before 1960. The economic
data are not available for the entire 1993 to 2002 period, thus they are
not included.
For each of the fifteen Congresses, I chose the
most important trade policy issue as designated by Congressional
Quarterly, such as the Trade Act of 1970, the Trade Reform Act of 1974,
the Omnibus Trade Bill of 1986, and the fast track vote in 1991.
39 To ensure that there was at least one bill for
each Congress, a few commodity specific bills had to be included. This is
not optimal. However, most of the bills, if not all, were debated in the
House in terms of the direction of national trade policy. See the
Congressional Quarterly Almanac various years for the
determination of “important” legislation and the substance of
the debates.
If there were multiple votes on the same trade
legislation, I used the roll call vote that most directly and broadly
invoked trade policy without nongermane amendments or components and that
also had at least a minimal amount of competition.
40 A minimal amount of competition is defined as at least 100
yea and nay votes for each roll call.
The votes selected largely
overlap and expand the roll call votes used in other studies of trade
policy during this time period.
41 For example, in Hiscox's most recent pooled
time period, 1970–94, he analyzes just six House roll call votes to
estimate the level of factor mobility for his twenty-four-year period.
42 Of the six, only four overlap with the slightly different
time period analyzed in this study, 1963–92. Votes on each of his
four trade acts are included in the fifteen roll call votes analyzed in
this article. Each vote was then coded as proliberalization if the bill
formally supported trade liberalization or actually attempted to decreased
trade barriers. All votes are then assigned the value of 1 if the
representative voted in favor of liberalization or a 0 if not.
43 Only votes of yea or nay are recorded. All
other (non)votes are coded as missing.
Appendix 1 lists all the
roll call votes used and how they were coded.
44 At first, only one vote is analyzed in each Congress to
facilitate the specific requirements of the statistical tests used. See
the methods section below.
The two economic models of factor mobility suggest that constituent
divisions will either be along sectoral lines if factor mobility is
relatively low or along factoral lines if it is relatively high. For each
division, there are interests that are advantaged by liberalization and
interests that are disadvantaged. Thus there are four possible economic
coalitions (in two exclusive groupings) for trade policy interests:
export-competitive industries versus import-competing industries (sectoral
division) or capital versus labor (factoral division). The former in both
cases is expected to support trade liberalization. From this I am able to
construct specific assumptions and hypotheses. First I assume that the two
coalitions that are advantageously affected by trade liberalization
(exports or capital) will positively influence legislators' votes for
trade liberalization. Second I assume that the two coalitions
disadvantageously affected (imports or labor) will negatively influence
legislators' votes for trade liberalization. Third I expect that the
sectoral variables will better characterize the 1960s and the 1970s.
Fourth I expect that the factoral variables will better characterize the
1980s and 1990s. In other words, I expect that factor mobility was
relatively low in the 1960s and 1970s, but was rising and reached a pivot
point in the late 1970s. Subsequently, factor mobility has been relatively
high in the 1980s and 1990s.
H1: I expect that the sectoral variables will better characterize
the 1960s and 1970s
H2: I expect that the factoral variables will better characterize
the 1980s and 1990s
The variables for exports, imports, and
capital are all constructed using a similar method that creates a
unique score for each congressional district for each Congress for each
variable. In general, district-level economic demographic data are
weighted by a measure of U.S. exports, imports, or capital stock. Bailey,
Goldstein, and Weingast and Irwin and Kroszner use a similar variable
construction to represent constituent economic interests.45
Specifically, the base district-level economic data
record the number of each of the twenty manufacturing industries (based on
each of the two-digit SIC codes) in each congressional district for each
year from 1963 to 1992.
46 This is the
primary difference from the constructions of Bailey,
Goldstein, and Weingast 1997; and Irwin and
Kroszner 1999. The former also uses the total number of industries
but includes an additional weight of total number of manufacturing
employees; see Bailey, Goldstein, and Weingast
1997, 337. The additional weight is not used here because it is the
labor variable. The latter uses sectoral-based product output but
also tests a measure that approximates Bailey, Goldstein, and
Weingast's; see Irwin and Kroszner 1999,
669. The two measures are highly correlated and produce similar results,
see Irwin and Kroszner 1999, fn. 29. The present
choice of construction was also constrained by the availability of county-
and/or district-level data over the thirty-year time span.
These data are then weighted to account for the national-level economic
impact or importance of exports, imports, or capital investment. The
export and import weights are the product of the annual national value of
the exports or imports, respectively, for each of the twenty manufacturing
sectors divided by the annual total domestic value of production for the
sector.
47 For example,
if in 1968 textile manufacturers produced $1 billion worth of textiles and
exported all of it, then the weight for export competitiveness would be
1.0; if the textiles manufacturers exported only $100 million of the $1
billion worth of production, then the weight for export competitiveness
would be 0.1. For imports, if there were $2 billion in textile imports in
1968 and domestic manufacturers still produced $1 billion worth of
textiles, the weight for import competition would be 2.0. For each
manufacturing sector, the base congressional economic data are then
multiplied by the corresponding annual weight for export competitiveness
or import competition. Each manufacturing sector's scores are
averaged for each two-year period corresponding to each Congress; then the
sectors are aggregated. A single score is produced for each House district
for each Congress that measures the local impact of exports and
imports.
For the capital variable, the weight is the national level of
the current-cost gross stock of fixed private capital for each of the
twenty sectors for each year measured in millions of dollars. The annual
capital stock for each sector is then multiplied by the base congressional
economic data by sector for each year. Same as the construction of the
exports and imports variables, the sectors are averaged
for each Congress and aggregated across sectors to produce a unique score
of capital investment for each congressional district for each Congress.
This measure is an improvement over most other measures because it uses
the same base congressional economic data as exports and imports, thereby
making comparisons more reliable. Moreover, it better captures the level
of capital investment in manufacturing sectors at the
congressional-district level. For the labor variable, the labor
figures for the twenty manufacturing industries for each congressional
district are available. I simply sum the twenty figures and hence a
composite index is not needed. The annual labor data are then averaged
across the two years that correspond to each Congress. For the
regressions, the capital and labor variables are
multiplied by a scalar of 1/1000000 and 1/10000, respectively, so
that the coefficients are easier to display. See Appendix 3 for details
and data sources.
Methods and Results
The first step in testing the hypotheses is to determine which
economic model best fits the data on constituent interests—in other
words, to determine whether factor mobility is relatively low or high and
when the balance between the two may change. Because the ideal types of
the economic models are mutually exclusive, one of the models cannot be
reduced to the other by a set of linear transformations. As such, the two
economic models are considered nonnested and should not be regressed
together.48
For
instance, assume one has two trade models, where H
rv is the
model under the ideal-type assumptions of low factor mobility and
H
ss is the model under the ideal type assumptions of high
factor mobility. Then one would write:
where X1 and X2 are the exports and
imports variables, and X3 and X4 are the
capital and labor variables.49
For simplicity, the models are presented in a generic
functional form and the intercepts are omitted. The same error structure
is assumed across models.
If β
3 and
β
4 are restricted to 0 in H
ss, for example,
H
ss does not equal H
rv because of the fundamentally
different assumptions and variables. As such, the two equations are
defined as nonnested.
Assume, however, that one were to test a third
“supermodel,” Ht, which is a combination of the
Hss and Hrv. Then, one would write:
Here one is assuming that the models are nested because, for example,
if the coefficients of the Hss model, β3 and
β4, are restricted to 0, then Ht is the same as
Hrv. Model Ht, though, makes the contradictory
assumption that factor mobility can be both low and high. This obviously
cannot be the case. Thus the two different and opposing theories
necessitate two different models: Hrv and Hss.
Clarke demonstrates that nonnested models regressed together in one
“supermodel,” such as in Ht, has the potential of
producing unreliable results.50
The difficulty, then, becomes comparing the two models head-to-head.
To do so, I use the same procedure, the J-Test, as did Hiscox.51
This procedure uses an indirect
linear combination of the two models to determine which one (if either)
better fits the data. Specifically, the predicted values from the first
economic model, H
rv, are included in the second economic model,
H
ss, as a new variable and retested; the process is then
repeated with the predicted values from the second model retested in the
first. The coefficients that are produced for the predicted value
variables are essentially the weight given to that model and can be
directly compared. If only one of the two coefficients for the predicted
value variables is significant, then that model outperforms the other
(because the insignificant coefficient indicates that that model's
weight cannot be distinguished from zero). If both coefficients are
significant or neither is significant, then the result is indeterminate
and the appropriate model cannot be assessed.
The J Test, then, compares the ideal-type theories and generally
produces a dichotomous result supporting either Hrv or
Hss. As a result, if the sectoral model outperforms the
factoral model in influencing policy, then it is logical to conclude that
factor mobility must be relatively low. Conversely, if the factoral model
outperforms the sectoral model, then factor mobility must be relatively
high. This solves the problem of comparing nonnested models but raises
another issue. The dichotomous result belies the gradual change that
factor mobility as a secular trend is expected to display. Instead, it
simply determines which side of the balance between relatively low and
relatively high factor mobility the data better fit. Thus a shift in the
balance may appear to be discontinuous even if it is not. Although less
than optimal, this is a methodologically necessary concession. I attempt
to accommodate this, however, through a graphical examination of the
probability of each model's statistical significance.
Because the roll call vote takes only the value of 0
(antiliberalization) or 1 (proliberalization), probit is used to estimate
each model. For each Congress only the economic variables from the models
are regressed on the roll call vote. As discussed above, other factors
such as the legislator's party affiliation and divided government may
also influence voting behavior on trade policy. These additional variables
will be included in the robustness checks, but they are omitted from the J
Tests to maintain consistency with other studies of factor mobility.
Specifically, Hiscox estimates the effects only of constituent economic
variables and argues that he “exclude[s] [controls
for party affiliation] to provide the clearest imaginable test
between the class and the industry-group models.”52
Given
the importance of his work on this subject, it is necessary to maintain
this comparability initially.
Table 1 reports
the results from the J Tests.
J Tests of models of factor mobility, 1963–92
The results shown provide a measure of the general trend of factor
mobility through the applicability of each model. In ten of the fifteen
Congresses, the null hypothesis of only one of the two economic models is
rejected. This allows one to say which model is statistically more
appropriate. In the remaining five Congresses, one cannot decipher the
applicability of the two models because either both models are significant
or neither model is significant. Specifically, from 1963 to 1978, the
Ricardo-Viner model is the only model that is accepted and therefore
indicates that the data are consistent with relatively low levels of
factor mobility. The Ricardo-Viner model, however, is significant in only
half of these Congresses (that is, four of the eight Congresses), which
may indicate some weakness in the sectoral variables. Conversely, from
1979 to 1992, the Stolper-Samuelson model is the only model that is
accepted. This is the case for six of the seven Congresses. Therefore one
is provided with strong evidence consistent with relatively high factor
mobility. This is further supported by examining the ratio between the
pseudo R squares of the Ricardo-Viner model over those of the
Stolper-Samuelson model. From 1963 to 1978, the ratio in all but one case
is above 1.0, which indicates that the Ricardo-Viner model is able to
explain more variance in the data than the Stolper-Samuelson model. The
opposite is true from 1979 to 1992. Both of these measures are consistent
with the balance between relatively low and relatively high factor
mobility reaching a pivot point in the late 1970s, with relatively low
factor mobility beforehand and relatively high factor mobility after.
Factor mobility can change over time, but it is highly unlikely that
it will radically swing from one Congress to another. Accordingly, two
caveats are in order. First, the statistically ambiguous tests may be
because of exogenous forces or idiosyncrasies in the individual pieces of
legislation that are examined. Particularly given the heterogeneous nature
of sectoral coalitions, it is not surprising that aggregate measures of
exports and imports are significant less often than aggregate measures of
capital and labor. (This is also evident in Table
2.) Second, this may also be because of the sectoral coalitions of
relatively low factor mobility gradually weakening and approaching the
pivot point between it and relatively high factor mobility. This gradual
process becomes more apparent if one graphically examines the data over
time; Figure 1 does so by plotting the moving
average of the p-values of the z-scores from each model over time. The
p-values are the probability that each model is significant; the area
below the provided 0.05 line indicates statistical significance.53
Because the moving average is used—in
order to mitigate outlying results—the horizontal line at 0.05 does
not actually represent statistical significance. It is used only
illustratively.
During most of the 1960s and 1970s, the dark line
representing the Ricardo-Viner model is consistently below the 0.05
threshold, while the light line representing the Stolper-Samuelson model
is not. However, from the early 1970s, the moving average for the
Stolper-Samuelson model gradually decreases, overtaking the Ricardo-Viner
model during the 96th Congress (1979–80), and then is consistently
below the significance threshold beginning in the 99th Congress
(1985–86). This evidence better represents a gradually changing
secular trend—in this case, from relatively low factor mobility,
across a pivot point between the two models, to relatively high factor
mobility. In this way, this result is more consistent with expectations
than the dichotomous results provided by the J Tests.
Appropriate economic model tested on trade
liberalization
Changing levels of factor mobility
Using the pivot point that defines the period of relatively low from
relatively high factor mobility, I report in Table
2 the probit results from each of the appropriate economic models
that were used to generate the J Test results and Figure 1. Because the evidence was consistent with
factor mobility being relatively low from 1963 to 1978, the Ricardo-Viner
model (that is, export-competitive versus import-competing industries) is
reported for these years plus the 96th Congress.54
The 96th Congress is reported with both the sectoral and the
factoral variables because the data indicate that it is the tipping point
between relatively low and relatively high factor mobility.
Ten
of the eighteen coefficients are significant. Specifically, the
statistically significant export coefficients are all positive, indicating
support for trade liberalization. In all Congresses but the 94th, the
statistically significant import coefficients are negative, indicating
their opposition to trade liberalization.
55 Imports are significant but positive for the 94th Congress.
This is most likely an artifact of the specific legislation analyzed.
Particularly, one of the amendments to the Bretton Woods agreement allowed
for flexible instead of fixed exchange rates. Given the economic context
and the artificially strong dollar in the early 1970s, importers may have
believed that floating exchange rates would weaken the dollar and, thus,
make imports more expensive, which would, of course, be advantageous to
sectors of the economy that experience high import competition. Regardless
of the direction of this sectoral variable, the Ricardo-Viner model
nonetheless outperforms the Stolper-Samuelson in the 94th
Congress.
The J Tests and these results lend evidence in support
of the assumptions and Hypothesis 1: districts that are relatively
export-competitive support trade liberalization and those that are import
competitive generally do not, and the sectoral variables outperform the
factoral variables during the 1960s and 1970s.
After 1979 the results from the J Tests are indicative of a relatively
high level of factor mobility. Thus the variables from the
Stolper-Samuelson model (that is, capital versus labor) are reported. In
these regressions, eleven of the fourteen coefficients are statistically
significant and all are in the expected directions. Districts with
relatively capital-intensive interests favored trade liberalization; those
with relatively labor-intensive interests did not. This is consistent with
the assumptions and Hypothesis 2. Because the relative level of factor
mobility can be derived from the significance of the two models and their
specific economic coalitions, these data are also indicative of a gradual
trend from relatively low factor mobility to relatively high. Again, there
is naturally a pivot point between the appropriateness of the two
ideal-type models in which the influence of the sectoral variables of
relatively low factor mobility should gradually cease to explain as much
variance as the factor variables of relatively high factor mobility.
Tables 1 and 2 do not
allow one to demonstrate this gradual secular trend well. Instead, they
simply allow one to determine which side of the balance the data best
represent. Despite this methodological concession, the evidence is
consistently indicative of a change in the level of factor mobility from
relatively low to relatively high during the late 1970s.
Robustness Checks
One way to test the robustness of these results is to determine if
other expected effects of low and high factor mobility are also present.
As discussed above, the coalitions predicted by the relative level of
factor mobility vary in their relative homogeneity. The sectoral
coalitions of relatively low factor mobility are relatively heterogeneous;
the factor coalitions of relatively high factor mobility are relatively
homogeneous. If the sectoral variables consist of weaker and more
ephemeral coalitional partners, then their strength in explaining roll
call votes on issues of trade liberalization should also be weaker. I have
already provided some evidence that is descriptively consistent with this
assumption. The Ricardo-Viner model and its corresponding sectoral
variables in Tables 1 and 2 were significant less frequently than the
Stolper-Samuelson model and its corresponding factoral variables. This
evidence is far from conclusive and could be caused by other affects, but
nonetheless, it begins to provide some validity to the results.
To further test the robustness of the results, I expand the analysis
to include political independent variables. According to the conditional
party government literature, relatively homogeneous constituent interests
provide the basis for strong parties overall.56
Alt and Gilligan, Hiscox, and Milner and Judkin
extend this logic to the specific case of factor mobility.
57 They argue that
constituent interests derived from relatively high factor mobility provide
the foundations for the political parties to coalesce and offer clear
policy alternatives on trade issues. Alternatively, the relatively
heterogeneous constituent interests that are present when factor mobility
is relatively low weaken the parties' ability to offer clear policy
alternatives on trade liberalization issues. To evaluate the ability of
the parties to offer clear and consistent trade policy alternatives, I
include a party variable:
republican. It takes the value of 1 if
the House member is a Republican and a 0 if a Democrat.
Brady, Goldstein, and Kessler show that during the late nineteenth and
early twentieth centuries—a period that Hiscox argues is also
characterized by relatively high levels of factor mobility—political
parties had a significant effect on trade policy roll call voting. This
effect is independent of constituent economic interests.58
Consistent with these findings, I expect that during the 1960s and 1970s
when the above evidence indicates relatively low levels of factor
mobility, the variable
republican should not have a significant
impact on trade liberalization votes. This would indicate that the parties
do not statistically differ in their voting patterns, and thus do not
offer clear and consistent policy alternatives. Conversely, during the
1980s and 1990s, when the above evidence is consistent with relatively
high levels of factor mobility, the variable
republican should
have a significant impact that is independent of constituent economic
interests. This would indicate that the parties do offer clear and
consistent policy alternatives on trade liberalization issues.
Furthermore, Keech and Pak show that the partisan positions on trade
liberalization changed during this time period.
59
Specifically, during the 1970s, the Democrats switched from being the
party of free trade to the party of protectionism. Consistent with this, I
expect that the
republican variable will be positively correlated
with trade liberalization during the period of high factor mobility.
The other influence on the legislator's vote choice discussed
above is the partisan composition of the branches of the federal
government. The effects of divided government on trade policy have been
extensively debated; these studies, however, have largely overlooked the
effects of factor mobility on the constituent interests and the political
parties. Extending the logic of factor mobility to the effects of divided
government is rather straightforward. If the parties do not offer clear
and consistent policy alternatives, then the partisan composition of each
branch should be less important. If the parties do offer clear and
consistent policy alternatives and, as Brady, Goldstein, and Kessler
argue, party strength exhibits a significant and independent effect on
legislators' voting behavior, then, divided government should also
have a significant and independent effect.60
To evaluate these effects, I create a
divided
government variable that is coded 0 when the House and the president
are from the same party and 1 when they are from opposing parties. Given
that the Democrats were the majority party during the entire time period
analyzed in this study and that they became the party of protectionism
during the 1970s concomitantly with the parties gaining strength, I expect
that the
divided government coefficient will be insignificant
when factor mobility is low, and significant but negative during the
period of high factor mobility.
61 A study
with a theoretical focus on the effects of divided government would
benefit from a greater range of cases than can be offered
here.
H3: republican is expected to be insignificant
during periods of relatively low factor mobility.
H4: republican is expected to be significant and
positively correlated with trade liberalization during periods of
relatively high factor mobility.
H5: divided government is expected to be
insignificant during period of relatively low factor mobility.
H6: divided government is expected to be
significant and negatively correlated with trade liberalization during
periods of relatively high factor mobility.
Because divided government does not vary within a given
Congress, the data must be pooled to assess its impact. In order to use
the appropriate constituent economic variables, the data are also divided
into periods of relatively low and relatively high factor mobility.
Similar to the J Tests, this construction is not ideal because it may
obfuscate the gradual nature of any change. I try to accommodate this
again by graphically documenting the trends in partisan support for trade
liberalization by Congress.
The dependent is again dichotomous, thus as before, probit is used.
However, members in one Congress may be the same members in another
Congress, and thus the independence assumption is violated. To account for
this, I use Huber and White corrections to produce robust standard errors.
The first two columns of Table 3 show the
results from the regressions of the economic and political variables on
the pooled trade votes used above. In the first column, the two economic
variables for the Ricardo-Viner model are significant and in the expected
direction. However, the coefficient for republican is not
significant—also as expected. In other words, the political parties
did not provide statistically clear and consistent alternatives on trade
policy during the 1960s and 1970s. Given the lack of clear partisan policy
positions on trade liberalization, there is little reason to assume that
the partisan composition of the branches of government should be
significant. As expected, then, divided government is also not
significant during the pooled years 1963–78, that are characterized
by relatively low levels of factor mobility. Thus when constituent
economic interests based on relatively low factor mobility are controlled,
the political variables do not have an independent effect. This is
consistent with the hypothesis that sectoral coalitions produce weak and
ephemeral political alliances and thus do not provide a strong foundation
for the parties consistently to offer clear policy alternatives on trade
liberalization.
Robustness test of factor mobility
In the second column of Table 3, the
factoral variables of relatively high factor mobility are used. Both
capital and labor are again significant and in the
expected directions. In this case, though, both republican and
divided government are also significant, each in its expected
direction. The positive correlation between republican and trade
liberalization indicates that Republicans more clearly and consistently
supported liberalization than did the Democrats. I expected that if this
is the case, then divided government should also have a
significant effect, as it does. The significant and negative coefficient
of divided government indicates that the majority-Democratic
House is less likely to support trade liberalization when the president is
a Republican. These robustness checks are supportive of the primary
findings of a gradual change from relatively low factor mobility to
relatively high factor mobility during the 1970s. Moreover, because
constituent economic interests are controlled, the results also lend
additional evidence to the Brady, Goldstein, and Kessler argument that the
manifestations of party strength extend beyond direct constituent
representation.62
I add the caveat, though,
that the presence of relatively homogeneous constituent interests seems to
be a necessary condition, and, in the case of trade policy, that is
consistent with periods of relatively high levels of factor mobility.
To further check the robustness of the results, I expand the number
and content of the roll call votes used in the dependent variable to
ascertain whether the results were merely an artifact of the trade policy
votes initially chosen. The number of votes is increased from fifteen to
forty-five, a quantity significantly greater than are analyzed in most
other trade policy studies. The expanded dependent variable consists of
additional trade issues as well as such other international economic
policy issues as monetary policy. Changes in monetary policy have a direct
effect on the exchange rate, which, in turn, can cause constituent-level
economic effects similar to those caused by changes in tariff policy.
Bearce and Freiden argue that trade economic models can therefore be used
to examine the constituent effects of monetary policy.63
Thus the inclusion of additional
international economic policies in the dependent variable is consistent
with the theoretical construction presented in this study yet also
provides a broader and more strenuous robustness check on the theory and
results.
I selected additional roll call votes and coded them using the same
guidelines as used in the initial selection of trade policy votes. For
each Congress, I included two additional votes that had at least a minimal
amount of competition and offered the clearest purchase on the policy
issue. I coded each as either proliberalization or antiliberalization. I
coded each legislator, then, 1 if he or she voted in favor of
liberalization and coded a 0 if he or she voted against.64
Consistent with Bearce
2003, the monetary policy votes were coded as proliberalization if
the legislation encourages monetary convergence and antiliberalization if
the legislation encourages monetary autonomy.
(See Appendix 2 for
a list of the additional votes and their coding.) Then the votes for each
legislator were aggregated, creating an index that ranges from 0 to 3.
65 This variable could not be easily used in the
initial J Tests because the predicted values are based on multiple cut
points.
The index allows for greater nuance; in effect, each
legislator could also give one-third or two-thirds support for
liberalization.
66 Only legislators that
voted on all three bills for each Congress are included.
Given
that this dependent variable is ordered and discrete, I use ordered probit
with the Huber and White corrections.
As before, the two models based on the relative level of factor
mobility are used; columns (3) and (4) of Table
3 report the results. As expected, the sectoral variables in column
(3) do not perform well with the broader dependent variable. Still, when
exports is significant, it is positive; imports is not
significant. The latter is descriptively consistent with the assumption
that weaker coalitions of relatively low factor mobility can experience
cross-cutting pressures and hence may be less potent in influencing a
broader set of policy issues. Unlike the test on trade votes in column
(1), republican is significant in column (3). The negative
coefficient, though, is consistent with Keech and Pak's argument that
the Democrats were more supportive of liberalization during the 1960s and
early 1970s than the Republicans.67
Even so, the
partisan effect is relatively small. For instance, holding all of the
other independent variables at their means but allowing
republican to vary from 0 to 1 produces the largest probability
on the third cut point. Specifically, a Democrat with identical
constituent interests as a Republican will be only 4.46 percent less
likely to vote in favor of liberalization on all three pieces of
legislation. Given this rather weak partisan effect, it is not surprising
that the coefficient for
divided government is insignificant.
Column (4) reports the results from the period with relatively high
factor mobility. All of the variables are significant and in the expected
direction. capital is positively and labor is negatively
correlated with trade liberalization. As expected, the republican
coefficient is also significant but changes direction from the 1963 to
1978 analysis in column (3). In this case, the partisan effect is also
much greater. For instance, holding all of the other independent variables
at their means but allowing republican to vary from 0 to 1
increases the probability that a legislator will vote in favor of
liberalization on all three pieces of legislation by 37.48 percent.68
The cut point at 3 is used again for
comparative purposes even though it does not produce the largest
probability this time. Instead, the cut point at 0 produces a probability
that a Democrat with identical constituent economic interests as a
Republican is 50.36 percent more likely to vote against trade
liberalization in all three pieces of legislation.
As expected,
such a strong partisan effect independent of constituent economic
interests creates a circumstance ripe for partisan conflict between the
branches of government. It is no surprise, then, that
divided
government is significant in column (4) and negative: the
majority-Democratic House is less likely to support liberalization when
the president is a Republican. Even though there are certainly numerous
causes of increased partisanship between these two time periods, these
results are consistent with the expectations derived from a change in the
relative level of factor mobility being at least a parallel and possibly a
contributing influence.
Viewing the data graphically for each Congress shows a more gradual
pattern of increased partisanship over issues of liberalization than is
evident in the split pooled cross-sectional analyses of Table 3. In Figure 2, I plot
the moving average of three variables that represent party strength in the
House from 1963 to 1992. The dark line is the Average Party Unity (APU)
scores, the aggregated difference between the percentages of each party
voting yea on each piece of legislation for each Congress. As Rohde
argues, this is the best measure to determine how internally united and
externally divided the parties are—in other words, how strong they
are. The trend displays a significant increase in party resurgence
beginning in the late 1970s and early 1980s.69
This trend has
led many scholars to declare that the political parties are once again
strong.
70
Average party unity, 1963–92
The two light lines in Figure 2 (one with
square and one with triangular data point markers) are the APU scores for
the trade votes and for the international economic policy votes,
respectively. Both exhibit a similar pattern over these issues: trending
from partisan dealignment to partisan realignment. During the latter, the
parties offer clearer and more consistent partisan policy alternatives.
The international economic policy APU scores, however, begin to show the
partisan realignment sooner than the trade policy APU scores. In both
cases, the 1970s is obviously a time of great partisan change. In
addition, the trends correspond to the pattern discussed in the decline
and resurgence of U.S. political parties' literature, the partisan
trade policy literature,71
and the trends displayed in
Table
3. In fact, the correlation between the overall APU and the trade
votes APU is 0.614; the correlation between the overall APU and the
International Economic Policy APU is 0.787. The thirty years analyzed
cover a vast terrain of politics and legislation: the civil rights
movement, the Vietnam War, stagflation and the oil crisis, Reaganomics and
recession, the Iran-Contra affair, and the first Persian Gulf War. So a
strong correlation between a specific set of issues and all the issues
addressed by Congress from 1963 to 1992 is noteworthy at the very
least.
Discussion
The above results both confirm and challenge the existing literature
on factor mobility and constituent trade preferences. The results confirm
the significance and direction of sectoral constituent interests in the
1960s and 1970s and of factoral constituent interests in the 1980s and
1990s on trade policy. They also clarify some of the aggregate
inconsistencies in the literature. For example, Magee, Brock, and Young,
Kaempfer and Marks, and Scheve and Slaughter study U.S. trade preferences
in one particular time period since 1960.72
Magee, Brock, and
Young find evidence that is consistent with low factor mobility in the
early 1970s; the others' evidence is consistent with high factor
mobility in the 1990s. Under an assumption of constant factor mobility
from 1963 to 1992, these conclusions are contradictory. By challenging an
assumption of constant factor mobility, however, the findings presented in
this article help to rectify these differences. Because the general trend
from relatively low to relatively high factor mobility reached a pivot
point during the late 1970s, studies of trade preferences of the 1960s and
1970s should find sectoral influences more significant and studies of the
1980s and 1990s should find factoral influences more significant.
Other studies have provided longitudinal analyses of trade
preferences. Rogowski, for instance, does not test for variability in
factor mobility—and, accordingly, the appropriateness of each
economic model.73
He assumes that factor
mobility is high but finds some contradictory evidence in the United
States from 1948 to the early 1980s.
74 Midford expands
on this point and notes a number of empirical anomalies in the assumption
of high factor mobility: labor in specific sectors, such as “light
industries such as textiles, apparel, and shoe manufacturing,”
supported protectionism in the 1950s; labor in other sectors, such as the
United Auto Workers, continued to support free trade until the end of the
1970s.
75 These instances contradict the
assumptions of high factor mobility but fit the expectations of low factor
mobility; where export-competitive and import-competing industries are the
dominant constituent economic cleavage. Specifically, Midford notes that
the logical extension of his alternative model to
Rogowski's—the multifactoral model—approximates the
Ricardo-Viner model of low factor mobility in the extreme and has some
applicability to the United States for the 1950s through the 1970s.
76 The findings
presented in this study for the 1960s through the 1970s are also
consistent with Hiscox's findings.
77 He argues that factor mobility in the
United States was in decline throughout most of the twentieth century.
Despite the century-long trend, he finds some evidence of high factor
mobility in his last pooled time period, 1970 to 1994.
78 This study and the
other studies that find factoral influences in the 1980s and 1990s suggest
that, at the very least, his pooled time period may be obscuring a change
in the mobility of factors that occurred in the late 1970s.
In addition, I have emphasized two methodological points. First, the
inclusion of both sectoral and factoral variables in the same model may
lead to erroneous results. To test two competing and nonnested models
accurately, alternative statistical procedures are required to
discriminate between them. Second, case-unit consistency between the
geographic dimensions of the economic and political variables, in this
case House districts, produces results that are more reliable. No doubt,
this type of detailed economic data is not consistently available for
studies such as Hiscox's that cover two centuries. Nonetheless,
empirical constructions that are not unit consistent may also obscure
trends that are indeed present.
This study, then, provides the case for a gradual increase in factor
mobility with the balance between relatively low factor mobility and
relatively high factor mobility reaching a pivot point from the former to
the latter in the late 1970s. This claim begs the question: What causes
factor mobility to change? There are multiple determinants of the level of
factor mobility.79
The determinants
discussed represent general national level trends. There could, of course,
be local contradictions such as housing shortages, additional
transportation costs, and the like. The probable presence of such local
anomalies ought to have only diminished the reported results.
The
most obvious is the costs of moving factors across physical space. As
transportation costs decrease so do the costs of mobility, and thus factor
mobility will be increasing. For instance, Kim argues that the diminishing
U.S. regional differences in income disparity, economic structure, and
factor endowments from 1947 to 1990 were caused, in part, by decreasing
costs of transportation and communications.
80 An equally
important determinant is the costs associated with moving factors across
product space. Kim also argues that firms in the United States during this
time period “increased [their] usage of substitutes and
recycled inputs.”
81 It is assumed that factors become increasingly
interchangeable as the costs of introducing new factors across product
space become lower; in other words, higher levels of factor mobility. In
sum, U.S. factors of production seem to have become physically and
practically more mobile during the second half of the twentieth century,
partially because of developments in technology and improvements in
transportation.
The general economic context also affects the relative level of factor
mobility through the decision calculus of individuals.82
Laborers, for example, decide to invest in the optimal type of skill
training, either specific or general, depending on the economic
environment. In an uncertain world, workers vary their training investment
based on expected risk. In an economic environment with low risk (for
instance, high employment stability and retention), a worker has an
incentive to choose to increase personal returns by specializing. In an
economic environment of high risk, however, the same worker foregoes the
increased returns of specialization in favor of the greater employment
security offered by general skill training. The former decreases factor
mobility; the latter increases it. Grossman and Shapiro conclude
optimistically: “factor mobility is more likely to be observed when
it is most important that factors be mobile.”
83 Their conclusion implies that during periods of economic stagnation
and increased job churning, one should witness individuals making
investment decisions that increase their ability to be mobile. In this
vein, Boisjoly, Duncan, and Smeeding demonstrate that from 1968 to 1992,
all labor groups except government workers experienced increasing
incidence of involuntary job loss.84
Despite
the perception of relatively tight labor markets in the 1990s, Farber
shows that involuntary job loss rates were even higher in the 1990s than
in the 1980s.
85 Furthermore, displaced workers that were reemployed
suffered “significant earnings declines relative to what they earned
before they were displaced.”
86 Similarly, the number of union elections and the union voting
rates were rather stable during the 1960s and 1970s but decreased
precipitously beginning in the early 1980s and continued through the
1990s.
87 Union membership
rates correspondingly also decreased significantly during the latter time
period. Farber and Western argue that “increased global
competitiveness and mobility of capital were likely important contributing
factors” to these changes in union strength.
88 These circumstances did not escape the perceptions of workers
either. Using the General Social Survey, Schmidt shows that workers became
increasingly pessimistic about their job security from 1977 to 1996. In
response, they invested “less in firm-specific human capital and are
more concerned about developing general skills.”
89 Dynamic factor mobility is not limited to labor. Grossman also
constructs a theoretical model of changing degrees of capital
mobility.90
The
Stolper-Samuelson model predicts that as the relative level of factor
mobility increases, the fortunes of labor should decrease and the fortunes
of capital should increase. In that vein, Farber and Hallock find that
reductions in the labor force from 1970 to 1997 are positively correlated
with increases in firms' stock prices.
91
It is also not surprising, then, that income disparity between the classes
has also greatly accelerated since about 1980.
92 Although the theories are clear on this point, the empirical
conclusions have been mixed on the extent of trade's influence on
growing income disparity. See Slaughter 2000 for
a survey of the literature.
The increased income disparity also
has been shown to have significantly affected partisanship.
93 One economic trend that seems to be at odds with a gradual increase in
the level of factor mobility is the tremendous concomitant growth in
intraindustry trade. If factor mobility is relatively high, countries that
are endowed with abundant capital, for instance, should gain from trade
with countries endowed with abundant labor. Intraindustry trade, though,
is typically among capital-abundant northern countries. This conventional
wisdom, however, has been theoretically and empirical challenged. When
technological differences among northern countries are considered,
“it is possible to have a strictly conventional [factoral]
model in which the volume of North-North trade exceeds that of
North–South trade.”94
Davis and Weinstein (
2001) further develop the model and test it on ten
countries within the Organization for Economic Cooperation and Development
(OECD) across thirty-four industry types. The results confirm that
intraindustry trade among capital-abundant countries does not violate the
factor endowment theories.
Finally, it is important to note that in analyzing aggregate district
interests and the revealed preferences of legislators through policy
votes, it is difficult to isolate the economic links.95
This study and
all studies similar to it are drawing conclusions about the general trends
of economic phenomena while simultaneously invoking (directly and
indirectly) political institutions and preferences.
96 The inclusion of “political parties” and
“divided government” in this study provides an additional
check on the robustness of the economic findings by controlling for these
political components. The results confirm both the expected economic and
political effects of factor mobility. Economically, the influence of
sectoral and factoral variables remains significant while controlling for
party and divided government. Politically, the significance of a
legislator's partisanship and the effects of divided government are
correlated with relatively high factor mobility. Logically, other periods
of high factor mobility ought to have had a similar potential to generate
strong partisan conflicts over trade policy. For example, Hiscox argues
that factor mobility was high during the late nineteenth century in the
United States.
97 As
expected, this was also a period of strong parties offering clear and
consistent partisan alternatives on trade policy.
98 Given the findings in this article of high factor
mobility in the 1980s and 1990s, it is not surprising, then, that Destler
has argued that “today's debate over the impact of
globalization parallels that over a nationalizing U.S. economy that became
heated a century ago.”
99Conclusion
As economic globalization intensifies, the domestic influences on
international trade policy become ever more important to understand. A
central variable in the formation of domestic trade preferences and
constituent coalitions is the mobility of factors. The analysis presented
in this article is consistent with factor mobility being relatively low
but rising during the 1960s and the 1970s, reaching a pivot point in the
late 1970s and subsequently remaining relatively high during the 1980s and
the 1990s. Furthermore, the constituent coalitions are shown to influence
the voting behavior of legislators in a manner consistent with economic
theory. Constituent coalitions of exporters and capital are positively
correlated with trade liberalization, and coalitions of importers and
labor are negatively correlated with trade liberalization. I have also
shown that the level of factor mobility and the coalitions' predicted
level of homogeneity are significantly correlated with the strength of the
political parties in supplying trade policies and the influence of divided
government.
The relationship among constituent interests, trade policy, and
political institutions is dynamic and continues to deserve much greater
attention. Further study in this area is crucial not only for an improved
understanding of the political system but also as a means of better
isolating economic phenomena. Toward this end, this study sheds
theoretical light on some of the empirical discrepancies currently present
in the literature on trade preferences and policy by using a more detailed
data set of constituent economic interests and challenging the theoretical
assumption of a constant level of factor mobility. Finally, the relative
level of factor mobility is likely to have profound effects on the
distribution of income as well as on numerous other policy areas: for
instance, immigration policy. As such, using factor mobility to explore
political and economic phenomena should continue to be a fertile area of
research.
Appendix 1: Trade Policy Votes
Appendix 2: Additional Roll Call Votes on
International Economic Policy
Appendix 3: Data Sources and Variable
Construction Details
The base economic data are constructed from three sources. Together,
these sources provide data on the number of the twenty manufacturing
industries based on the two-digit SIC classifications (see below) in each
U.S. House district for each year from 1963 to 1992. First, for the years
1963 to 1972 (the first two years of the subsequent decade are included
because redistricting after that decade's census had not yet
occurred), the data are found in the Congressional District Data Book
(Districts of the 88th Congress). Here, the data are presented for
the twenty manufacturing sectors already aggregated into House districts.
Unlike the data for 1973 and after, the Congressional District Data
Book provides data only on firms with twenty or more employees.
Second, for the years 1973 to 1976, the data are aggregated only at the
subcongressional level (that is, county and independent city). The data
for the 3,143 subcongressional units can be found in the Census of
Manufacturers, 1972. Beginning in 1977, a digital copy of annual data
from County Business Patterns, various years, is available at the
I.C.P.S.R. Accordingly, from 1977 to 1992, I use the annual data at the
subcongressional unit level for each of the twenty manufacturing sectors.
Because the annual data are not available before 1977, the base economic
data from 1963 to 1972 and from 1973 to 1976 are held constant.
The subcongressional data (1973 to 1992) are then aggregated into
their appropriate congressional district. If one of the subcongressional
units was entirely contained in one congressional district, it was simply
coded for that House district. If one of the subcongressional units is not
entirely so contained but divided between two or more congressional
districts, the base economic data for the subcongressional unit is
weighted by the percentage of the population of the unit that lives in
each corresponding congressional district. From 1972 to 1982, the
population percentages are calculated using the 1970 Census of the
Population, Characteristics of the Population for each state. From
1983 to 1992, the 1980 Census of Population and Housing: Congressional
Districts of the 98th Congress for each state is used. In both cases,
the base economic data are then aggregated into the appropriate House
districts. Together with the House data from the Congressional
District Data Book, the base of congressional economic data is
created that provides a detailed demographic account of the number and
type of manufacturing industries found in each congressional district for
each year from 1963 to 1992.
Next, the base congressional economic data is weighted by annual
national level economic data for exports, imports, and capital stock,
respectively. The data for the value of the exports and imports can be
found in U.S. Census Bureau, U.S. International Trade in Goods and
Services, series FT-900, various years. The domestic value of
production (GDP) by industry is found in the U.S. Bureau of Economic
Analysis, Survey of Current Business, various years. The gross stock
of fixed private capital is found in the U.S. Bureau of Economic
Analysis, Survey of Current Business, various years. For the years
1963 to 1972, the number of laborers in each congressional district as
found in the Congressional District Data Book is weighted by the
annual percentage of the national rise or decline in manufacturing jobs.
For the years 1973 to 1976, the number of laborers in each of the
subcongressional units (as found in the Census of Manufacturers,
1972) is aggregated into congressional districts using the same
process as described above and is weighted by the annual percentage of
rise or decline in manufacturing jobs in each state.
The following are the manufacturing sectors and their SIC number:
20. Food and kindred products
22. Textile mills products
23. Apparel and other textile products
24. Lumber and wood products
25. Furniture and fixtures
26. Paper and allied products
27. Printing and publishing
28. Chemicals and allied products
29. Petroleum and coal products
30. Rubber and plastic products
31. Leather and leather products
32. Stone, clay and glass products
33. Primary metal industries
34. Fabricated metal product
35. Machinery, except electrical
36. Electrical equipment and supplies
37. Transportation equipment
38. Instruments and related products
39. Miscellaneous manufacturing
