Republican Elites, Employer Mobilization, and the Politics of State Fair Employment Practices Legislation in the North, 1945-19641
Anthony S. Chen University of Michigan
E-mail: [email protected]
July 2004
[APPROXIMATELY 14,000 WORDS, INCLUDING TABLES AND REFERENCES]
1 Working Paper No. 04-004, Gerald R. Ford School of Public Policy, University of Michigan, Ann Arbor, MI 48109. This research was supported by grants from the National Science Foundation (SES-0000244), Rackham Graduate School at the University of Michigan, Graduate Division at the University of California, Berkeley, and a Soros fellowship. The author gratefully acknowledges the encouragement and feedback of Becky Blank, Jake Bowers, Nancy Burns, Ken Chay, Sandy Danziger, John DiNardo, Ben Hansen, Mike Hout, Greg Huber, Jerome Karabel, Michael Katz, Don Kinder, Sam Lucas, Justin McCrary, Isaac Martin, Robert Mickey, Mark Mizruchi, Andrew Noymer, Joseph Palacios, Trond Peterson, Tom Sugrue, Arland Thornton, Rob Van Houweling, Margaret Weir, Yu Xie, and Dean Yang. Thanks to Pooja Patel and Robin Phinney for research assistance. Special thanks to Sheldon Danziger for his extraordinarily detailed set of comments. All errors and infelicities are mine.
2
Manuscript Title: Republican Elites, Employer Mobilization, and the Politics of State Fair
Employment Practices Legislation in the North, 1945-1964
Abstract: From 1945 to 1964, more than a score of “northern” states passed laws mandating non-
discrimination in employment. Why did some state legislatures pass such fair employment
practice (FEP) laws sooner than other states? Through discrete-time, event-history analysis of
new data, I find that Republican control of veto points—and to a lesser extent the political
mobilization of organized business—reduced the likelihood of passage. This finding casts doubt
on the leading theory of the electoral realignment that began in the mid-1960s. Well before the
advent of affirmative action, Republican elites exhibited racial conservatism, using their
institutional control over the legislative process in the states to frustrate the spread of color-blind
anti-discrimination policies.
3
The history of civil rights in the United States is often seen as reaching a luminous
summit with the enactment of the Civil Rights Act of 1964, the Voting Rights Act of 1965, and
the Fair Housing Act of 1968. This triumvirate of legislation is undoubtedly a historic
achievement. While the deepest hopes of their backers went unfulfilled and the darkest fears of
their critics failed to materialize, civil rights bills of comparable breadth and import had not
cleared Congress since Reconstruction (Klinkner and Smith 1999).
The legislative successes of the period are worthy of the prominence given them, but they
have had the effect of overshadowing the conflict over civil rights that began sweeping northern
states at the close of World War II. It is only vaguely remembered that more than fifty pieces of
civil rights legislation mandating fair employment, open accommodations, and fair housing were
passed by “northern” state legislatures from the 1940s to the 1960s (Lockard 1968).2 Of these,
state fair employment practice (FEP) laws were the most significant, politically and
economically.3 State FEP laws prohibited racial, religious, and national origin discrimination in
public and private employment—that is, they mandated non-discrimination in employment. For
enforcement, most laws also created a new state agency, usually called a Fair Employment
Practice Commission (FEPC). The first FEP law was passed in New York State in 1945. By
1964, when the operation of state FEP agencies was incorporated into Title VII of the Civil
Rights Act, twenty-nine states had passed FEP bills of one kind or another.
This twenty-year burst of legislation gives credence to Justice Brandeis’s metaphor that
the states are the laboratories of democracy, but if his metaphor remains felicitous, it is also clear
that the laboratories of democracy were not all created equal. States varied considerably in the
2 I use the term “northern” to loosely designate states outside the South, which I define, following
V.O. Key (1949), as the eleven states belonging to the former Confederacy.
3 The impact of FEP laws on black labor market outcomes is the subject of ongoing research. See
Collins (2001), Neumark and Stock (2001) and Landes (1967, 1968).
4
timing of their legislation. New Jersey passed a FEP law in 1945, following the lead of New
York. By contrast, Ohio and California did not pass such legislation until 1959. Why were some
states clearly more willing to experiment with civil rights than other states?
It is tempting to think that partisanship played a critical role. However, according to one
theory of the electoral realignment that began unfolding in the 1960s, civil rights was simply not
a partisan issue in the 1940s and 1950s (Carmines and Stimson 1989; Edsall and Edsall 1991;
Thernstrom and Thernstrom 1997). Before 1964, Republicans and northern Democrats had both
professed and taken highly liberal positions on civil rights. Civil rights became a defining axis of
partisan conflict only after 1964, when the rise of color-conscious policies like affirmative action
and busing precipitated a “backlash” that contributed to the eventual breakup of the New Deal
coalition and catalyzed the emergence of racial conservatism among Republicans and racial
liberalism among Democrats.4 Hence, on this view, the legislative timing of state FEP laws would
have been largely unrelated to the relative strength of the parties.
But there are good reasons to believe that the passage of state FEP legislation was in fact
a partisan issue—and that Republican elites were far less supportive of it than northern
Democratic elites. Recent evidence of racial liberalism among Republican elites during the 1940s
and 1950s is not conclusive, and older evidence indicates that party control was relevant (Lockard
1968; Erikson 1971). Most pertinently, case studies of FEP campaigns in Michigan (Sugrue 1996;
Fine 2000), Pennsylvania (Siskind 1997), New York (Chen 2002), and California (Chen 2002),
provide evidence that Republican officeholders actively sought to block state FEP legislation.
4 Sociologists have employed the terms “racial liberalism” and “racial conservatism” to signify a
variety of political ideologies (for one recent usage, see Brooks 2000). I use them to describe
different ideas about the proper role and scope of government intervention in facilitating social
change, particularly as it pertained to race. Racial liberals were less reluctant than racial
conservatives to invoke governmental authority to combat discrimination and promote equality.
5
Largely at the behest of organized business, Republicans used their control of key legislative
committees to delay the passage of FEP laws.
What, then, was the precise role of Republican elites as well as business interests in the
politics of state FEP legislation? This question has unmistakable theoretical significance. If state
FEP legislation was a partisan issue in the 1940s and 1950s, then there would be reason to
reconsider the causes of electoral realignment. Republican resistance to “color-blind” laws like
FEP would demonstrate that their racial conservatism antedated the “color-conscious” turn in
public policy. It would suggest that if affirmative action had not risen to prominence in the late-
1960s, Republican elites would have sought out another racial wedge issue with which to split the
New Deal coalition and build a new electoral majority. The sources of realignment would have to
be sought elsewhere in the political economy of the postwar United States.
This article addresses these theoretical concerns by presenting a new empirical analysis
of the effect of party control and employer strength on the passage of FEP legislation. I apply
discrete-time, event-history methods to a new, state-level data set containing a rich mix of time-
varying and time-constant covariates. I find evidence that the likelihood of passage varies
inversely with Republican control of legislative veto points, and to a lesser extent, the political
mobilization of the business lobby—even when controlling for other variables commonly
associated with policy innovation. Unobserved heterogeneity cannot be conclusively ruled out,
but the results for Republican control in particular are robust to a variety of fairly stringent
assumptions.
This article has four sections. The first section provides background on the origins and
operation of state FEP laws. A second section formulates the empirical question and provides a
theoretical motivation. The third section reports the results of the empirical analysis. A final
section draws conclusions about the theoretical significance of the empirical results and suggests
new directions for future research.
6
A BRIEF SKETCH OF STATE FEP LAWS
The history of state FEP laws begins 1941, when President Franklin D. Roosevelt issued
Executive Order 8802, prohibiting racial and religious discrimination in war-related employment
and creating a new Fair Employment Practice Committee (FEPC) to encourage compliance. The
president issued his order not so much out of moral conviction, but rather out of a desire to halt
the growing threat posed by the March on Washington Movement (MOWM). Led by black
unionist A. Philip Randolph, MOWM had been mobilizing a march on Washington to demand the
integration of the national defense program. Roosevelt refused to integrate the military on the eve
of war, deferring to the concerns of his defense chiefs, but he was less wary of integrating war
jobs—and he sought to defuse Randolph’s threat by creating the FEPC (Ruchames, 1952;
Kesselman 1948; Reed 1991: 13; Garfinkel 1959: 37-61; Kryder 2000).
Although the wartime committee possessed scant enforcement authority, it never escaped
the cloud of controversy under which it had been established. The most venomous opprobrium
surfaced in the rhetoric of southern demagogues like Theodore Bilbo (D-MI). But the committee
also attracted the reproach of conservative Republicans like Robert A. Taft (R-OH), who
considered it a dangerous aggrandizement of federal authority over private economic activity
(Chen 2002). In 1944, due to southern hostility, FDR reconstituted the FEPC through a second
executive order. By 1946, however, even the second FEPC had fallen. Congress had rejected
legislation that would have given the FEPC a statutory basis, ordering it instead to close down
operations (Kesselman 1948; Ruchames 1952; Reed 1991).
The short-lived existence of the FEPC nonetheless had a lasting influence on the politics
of civil rights. Its most direct consequence was to kindle a new interracial, interfaith coalition that
sought to use normal electoral channels to establish a new FEPC in the mold of the National
Labor Relations Board (Chen 2002). That coalition, beginning in 1944, introduced scores of FEP
bills into Congress. Leading the campaign were northern Democrats like Hubert Humphrey (D-
MN) and liberal Republicans like Irving M. Ives (R-NY). Although they eventually achieved
7
limited success with the Civil Rights Act of 1964, aspirations for FEP legislation—particularly in
the Truman years—were continually frustrated by a conservative coalition of Republicans and
southern Democrats (Chen 2002; Katznelson, Geiger, and Kryder 1993; Burstein 1985; Santoro
2002).
In response, liberals turned to state legislatures, where they thought prospects for success
were better. They assessed their chances correctly. In 1945, more than a dozen FEP proposals
were introduced into state legislatures (American Council on Race Relations 1945). That year,
New York became the first state to adopt FEP legislation when Governor Thomas E. Dewey
signed the Ives-Quinn bill (Chen 2002). By 1964, when Congress began to seriously consider
omnibus civil rights legislation, twenty-nine states had already passed FEP legislation of one type
or another (Bureau of National Affairs 1964; Lockard 1968).
The regulatory scope of such legislation was comparable. Most laws followed the New
York model by declaring it unlawful for employers, employment agencies, or labor organizations
to discriminate against a person in almost every phase of the employment relationship, from
hiring and promotion to termination. Most laws prohibited discrimination on account of race,
color, religion, national origin, or ancestry. With the exception of Oklahoma, state FEP laws
applied to discrimination in both public and private sectors of employment.
Of the twenty-nine states passing a fair employment law, only three passed voluntary
laws that lacked enforcement provisions. The other twenty-six laws were at least nominally
enforceable through civil proceedings or criminal penalties. Most laws relied on the NLRB-style
framework of administrative regulation that Congress was repeatedly rejecting. This framework
required individuals to take complaints of discrimination not to the courts but rather a state
commission. Such commissions were typically given the authority to hold public hearings;
subpoena witnesses and compel testimony from them; and initiate conciliation proceedings if it
made a determination that discrimination had occurred. If the commission could not elicit
voluntary compliance from companies or unions involved, it had the authority to the offending
8
parties to cease-and-desist from their discriminatory practices and take “affirmative action” to
compensate the victims of discrimination for the harms they had suffered. Any such orders were
typically subject to judicial review (Bureau of National Affairs 1964).
[Table 1 about here]
If the design and application of state FEP laws shared many similarities, however, they
varied widely in the timing of their passage. It is clear from Table 1 that the laws did not pass in a
steady progression. In the mid-1940s, northeastern states like New York, New Jersey,
Massachusetts, and Connecticut were clear pioneers. Three western states—Oregon, Washington,
and New Mexico—adopted FEP laws at the end of the 1940s. There followed a temporary lull,
but the mid-1950s saw another acceleration of adoption, led by Michigan and Minnesota. A group
of laggards, including California, Illinois, Ohio, and Missouri, finally passed FEP laws in the late-
1950s and early-1960s. By 1964, sixty-five percent of states outside the South had passed
enforceable FEP laws. This was a major historical achievement. But why had some states taken
longer to pass FEP legislation than other states?
RACIAL POLITICS AND ELECTORAL REALIGNMENT
Today, few issues define the difference between the two major parties more clearly than
civil rights. It is hence natural to think that the balance of partisan forces in state politics was
partly responsible for variation in the legislative timing of state FEP legislation.
Yet one of the most influential theories of race and the party system, which was
formulated to explain the electoral realignment that began in the 1960s, predicts that partisanship
was largely unrelated to the passage of state FEP legislation. This expectation rests on the claim
that “issues of race were not partisan issues” before 1964 (Carmines and Stimson 1989: 184).
Before 1964, Republicans and northern Democrats were both racially liberal, committed to the
pursuing the dream of a color-blind society in which individuals would be judged by the content
of their character and not their color of their skin. In fact, Republicans might have been more
9
liberal than northern Democrats, since the latter were sometimes forced to downplay their racial
liberalism in order to avoid antagonizing their segregationist co-partisans from the South.
Carmines and Stimson (1989: 184) clearly articulate a key factual premise of the dominant
theory: “Advocates of racial liberalism were to be found equally among northern Democrats and
Republicans. Hostility to the aspirations of black Americans was almost exclusively the province
of the southern wing of the Democratic party.”5
Before 1964, Republicans themselves frequently accentuated their racial liberalism,
hoping that it would make them more competitive for black ballots. The strategy reached a zenith
in 1963, when it became clear that Kennedy’s civil rights proposal would receive serious
consideration in Congress. Policy analysts at the Research Division of the Republican National
Committee drafted and circulated a memo of talking points about the Republican record on civil
rights. The memo mentioned Lincoln’s emancipation of the slaves, Eisenhower’s role in the
passage of the Civil Rights Act of 1957 and 1960, and Republican votes for other civil rights
legislation in Congress. It also prominently cited Republican support of state FEP legislation,
pointing to patterns of party control at the time of passage: More of such laws had passed under
Republican than Democratic majorities, and states that had “pioneered” FEP in the late-1940s
were more likely to have been governed by Republicans (Republican National Committee 1963).
Subsequent studies of racial liberalism and the party system seem to confirm Republican
professions at the time. A roll-call analysis of Congressional votes on civil rights legislation
before 1964 reveals that Republicans consistently scored higher on a scale of racial liberalism
than Democrats (Carmines and Stimson 1989: 63-4). Moreover, analysis of national party
5 Although he is primary concerned with elite models of public opinion, Lee (2002: 5-6) also
notes that Carmines and Stimson (1989) invoke the idea that the 1960s was a turning point in
racial politics.
10
platforms for this period shows that GOP platforms contained more paragraphs on race and
placed a greater priority on civil rights than Democratic ones (Carmines and Stimson 1989: 55-6).
Indeed, most students of the period believe that civil rights developed into a touchstone
of partisan identity only after 1964, with Republicans becoming the party of racial conservatism
and Democrats the party of racial liberalism. The realignment had numerous sources, but it is said
to have been greatly “reinforced by a change in the civil rights agenda itself,” which “shifted
away from an initial, pre-1964 focus on government guarantees of fundamental citizenship rights
(such as the right to vote and the right to equal opportunity), and shifted toward a post-1964 focus
on broader goals emphasizing equal outcomes or results for blacks, often achieved through racial
preferences” (Edsall and Edsall 1991: 7). This color-conscious turn in public policy—epitomized
by affirmative action and busing—fueled white backlash and led decisively to the breakup of the
“New Deal Democratic bottom-up coalition” of northern workers, African Americans, and white
southerners (Edsall and Edsall 1991: 7).6
Color-conscious policies, it is thought, gave Republicans an unparalleled political
opportunity to split the faltering New Deal coalition and assemble a new electoral majority.
Championing programs like the Nixon administration’s Philadelphia Plan, Republican elites
sought to divide working-class whites from African Americans. At the same time, however, they
also began to embrace racial conservatism, hoping eventually to broaden the GOP’s electoral
appeal among resentful whites in both the North and South (Skrentny 1996; Kotlowski 1998).
6 Edsall and Edsall (1989) are not alone in arguing that the mid-1960s marked a shift in the
orientation of the civil rights movement and public policy. Authors of varying political
commitments have made similar observations, including Sleeper (1990), Thernstrom and
Thernstrom (1997: 179-80), and Matusow (1984). For critical perspectives on extent and nature
of the discontinuity exhibited during the period, see Sugrue (1996, 2004), Lee (2002), and Chen
(2002).
11
This elite-driven strategy succeeded brilliantly and came to fruition with Reagan’s election in
1980. The “New Deal order” (Gerstle and Fraser 1989) had decisively fallen, and a new racially
conservative Republican majority—forged out of the shards of the New Deal coalition—had
emerged to deliver the White House into GOP hands “How did Mr. Reagan manage to pull off
what many people regarded as impossible?” asked The Economist in a recent essay tracing the
history of Reagan’s electoral coalition. “The underlying reason was the implosion of liberal
America,” it wrote. Much of the blame for the implosion, it went on to imply, could be pinned on
one policy in particular: “The Democratic Party's embrace of affirmative action…[which] stirred
up a mighty backlash among whites” (The Economist, June 10, 2004).
There are nevertheless a number of reasons to ask whether Republican elites were as
uniformly racially liberal before 1964 as the dominant theory claims. The most obvious one is
that professing and practicing racial liberalism, especially at the national level, was smart politics
for the GOP. By endorsing racial equality and voting for sectional civil rights proposals (e.g.,
anti-lynching bills), Republican elites could drive a wedge between the northern and southern
wings of the Democratic party at little electoral cost to themselves. Second, measuring party
control at the time of legislative passage is a poor method of gauging party support for a
particular policy. A better metric is the number of times a party passes legislation relative to the
number of opportunities they had to do so. This is the metric Erikson uses in an older study of
party control and state civil rights legislation, and he finds evidence that Democratic control of
state legislatures was favorable for passage (Erikson 1971: 179). Lastly, and even more to the
point, recent case studies (e.g., Sugrue 1996, Siskind 1998; Fine 2000) of various states have
unearthed evidence that Republican elites actively resisted the passage of FEP laws.
This partisan dynamic is nowhere more clearly illustrated than in the case of California,
where a FEP bill was first introduced the legislature in 1945. The proposal failed to pass, but
comparable bills were reintroduced in every subsequent legislative session until one was finally
enacted in 1959, when Democrats gained control of the state government for the first time in the
12
twentieth century. The fifteen-year struggle in the legislature was led by two black Democrats,
Assemblymen August F. Hawkins (South-Central Los Angeles) and Assemblyman W. Byron
Rumford (Berkeley-Oakland). Hawkins and Rumford took turns serving as main sponsor of FEP
proposals, and they received support from an interfaith, interracial coalition of liberal groups that
Chen (2002) has called the “other” civil rights movement—to distinguish it from the Southern-
based, direct action movement to dismantle legal apartheid. The “other” civil rights movement,
which sought to eradicate racial and religious discrimination in the North through electoral
politics, included groups like the National Association for the Advancement of Colored People
(NAACP), the Brotherhood of Sleeping Car Porters (BSCP), the Jewish Community Relations
Council, the American Friends Service Committee as well as internationals and locals affiliated
with the Congress of Industrial Organizations (CIO) (Chen 2002).
It took Hawkins, Rumford, and the “other” civil rights movement so long to realize their
aims primarily because of Republican obstruction, which was facilitated by their control of
legislative institutions. In a striking reprise of the tactics that southern Democrats were deploying
in Congress against civil rights proposals, Republicans used their control of key committees to
block the passage of FEP bills, claiming that such proposals “must discriminate in favor of
members of such minority races” (Los Angeles Examiner, May 23, 1955) and that FEP was
“intended to take care of people who are not qualified” (California Voice, April 19, 1957). From
the mid-1940s to the early-1950s, GOP dominance of the California Assembly’s Government
Efficiency and Economy Committee bottled up successive FEP proposals. The battle shifted to
the California Senate by the mid-1950s, when Republicans in the Labor Committee kept FEP
bills from reaching the Senate floor (Chen 2002).
In their efforts, Republican elites had the backing of organized business. Nearly all
segments of the California business lobby—which saw FEP legislation as a violation of the
sacrosanct prerogative of management to hire, promote, and fire whomever it pleased—opposed
FEP proposals in the postwar period. The most influential among them included the Associated
13
Farmers (AF) and the California Chamber of Commerce (CCC), but many smaller and less
prominent employers shared similar views (AF 1945; California Chamber of Commerce, various
years; “Old Employer” to W. Byron Rumford 1955; Chen 2002).
Outside of California, similar coalitions of Republican elites and organized business led
the charge against FEP. Michigan and Pennsylvania both passed FEP laws only after Democrats
and urban Republicans had successfully pried FEP bills away from Republican-controlled
committees (Siskind 1997; Fine 2000). Even in New York, birthplace of Rockefeller
Republicanism, Republican elites mounted a fierce resistance. In coalition with the Associated
Industries of New York State, upstate Republicans led a revolt of rank-and-file GOP legislators
against party leaders. Only the belated support of Governor Dewey (R) and unprecedented
mobilization by the civil rights movement averted defeat (Chen 2002).
These new case studies challenge raise a testable empirical question. What was the effect
of Republican control and employer mobilization on the likelihood that a northern state would
pass a FEP law? The question has clear theoretical relevance. If it appears that Republicans elites
supported state FEP legislation with equal vigor as Democrats, then the dominant explanation
would enjoy further empirical support. It would strengthen the argument that policies like
affirmative action and busing gave Republican elites the ideal wedge issue with which to
strategically split the Democratic coalition.
But if the GOP resisted “color-blind” state laws like FEP well before 1964, then it would
appear that the racial conservatism among Republican elites antedated the “color-conscious” turn
in public policy. This early conservatism, in turn, suggests that Republicans would have quickly
found their way toward the same political and electoral strategy whether or not the regulatory
framework governing job bias eventually came to include race-attentive policies such affirmative
action. In lieu of the chance to exploit affirmative action, Republicans would have fanned the
flames of white resentment against civil rights using a different racial wedge issue—whether it
involved a “color-conscious” like affirmative action, or “color-blind” policy like FEP. Theorists
14
of electoral realignment should refrain from ascribing too much causal significance to the “color-
conscious turn” in public policy.
At least one previous researcher has sought to systematically analyze the politics behind
the passage of state FEP laws. In a study using continuous-time, event-history methods, Collins
(2003b) finds that the mobilization of Jewish organizations, civil rights groups, and labor unions
are the strongest predictors of passage, while unemployment, Catholic population, electoral
competition, and Democratic governors are less important. However, his central measure of party
control is conflated with a measure of electoral competition, making it impossible to separate the
effect of one variable from the other. Nor does he explicitly estimate the effects of employer
mobilization. Hence the question remains largely unanswered. This article uses a clearer, time-
varying measure of party control (one that is separate from electoral competition) as well as a
new measure of employer mobilization to test the following hypothesis: Republican control of
state government and the political mobilization of employers reduced the likelihood that a
northern state would pass a FEP law.
MODEL, DATA, AND VARIABLES
Model
Based on the foregoing theoretical discussion, I estimate the impact of Republican
control and employer strength on the passage of FEP legislation, using discrete-time, event-
history methods (Allison 1982; Peterson 1991; Yamaguchi 1991; Box-Steffensmeier and Jones
2004). Although others have employed continuous-time methods, I prefer discrete-time methods
for the question at hand because they handle ties more easily and because the passage of
legislation is fundamentally a discrete-time process. I specify the model as a logistic regression of
the functional form:
log(Pit/(1-Pit)) = α + β1xi + β2zit
15
in which Pit is the probability that state i passes a fair employment law at time t provided that it
has not yet done so, α is a constant, xi is a time-constant vector of covariates for state i, zit is a
vector of time-varying covariates for state i that varies according to time t, β1 and β2 are vectors
of effects associated with xi and zit, respectively. Time is modeled as a linear trend that is included
in the vector of covariates zit.7
Data
Conducting a discrete-time event-history analysis of FEP legislation requires an annual
data set on the social, political, economic, and institutional characteristics of thirty-seven
“northern” states during the period 1941-1964.8 I constructed such a data set from cross-sectional
data on the states that I collected from a wide range of published and unpublished sources,
including government reports, private publications, and archival records. Whenever possible, I
sought annual data, but in the instances where they were not available, I collected as much data as
possible and then generated annual data through linear interpolation. (Appendix A presents
descriptive statistics for, and identifies the sources of, the variables used in the analysis.)
My data set is organized in the standard unit-time format required by discrete-time,
event-history models—state-year observations. The first year for which I record observations is
1941. I continue to record observations on all thirty-seven states for each subsequent year in
which their legislatures met in regular or special session, as reported by Book of the States
7 I do not report robust standard errors clustered on the state level because states are clearly not
independent of one another. See Berry and Berry (1992, 1990). The reported results in Table 3
and Table 6 are highly comparable to those obtained with robust (but non-clustered) standard
errors.
8 I exclude thirteen states altogether, eleven states from the South as well as Alaska and Hawaii,
following the convention in studies of state economic performance (e.g., Brace 1993).
16
(Council of State Governments, various years).9 Once a state passes FEP legislation, it is
excluded from the data set. The last year for which I observe a state that has not yet passed a FEP
law is 1964. This procedure translates into a data set or “risk set” of 502 state-year observations.
My periodization of the risk set rests on a straightforward rationale. I define 1941 as the
first year in the risk set because states initially became at risk for passing FEP legislation as a
result of Roosevelt’s wartime FEPC in 1941. Its establishment touched off a cascade of state-
level political developments that culminated in formal campaigns for state FEP laws. The
political landscape changed in 1964, when Congress passed the Civil Rights Act, which stipulated
that all states without a FEP agency would effectively relinquish their right to first investigate
complaints through their own commission. This gave states that had not yet passed FEP
legislation a strong incentive to do so. Thus I define 1964 as the final year in the risk set.
Variables
My dependent variable is the passage of a state FEP law. This time-varying, indicator
variable is set to 1 if a state adopted a nominally enforceable FEP law in a given year, and set to 0
if it did not.10 Twenty-four such laws, not including Alaska and Hawaii, passed in the period from
1945 to 64.
9 The decision to include only years in which state legislatures met in regular or special session
represents a compromise between two approaches of contrasting rigor. The least exacting
approach, which would severely understate the hazard rate, would be to include all of the years in
the period 1945-1964. The most exacting (but prohibitively time-consuming) approach would be
to include all of the years in which it is known that a legislator introduced a fair employment bill.
10 Indiana and Wisconsin passed non-enforceable laws in 1945 and 1961, and enforceable laws in
1957 and 1963, respectively. I consider only the passage of enforceable laws; hence, I code
Indiana and Wisconsin as passing having passed FEP laws in 1957 and 1963, respectively. To see
17
My two key independent variables are Republican control and employer political
strength. Let me discuss each measure in turn.
I measure Republican control separately from electoral competition, unlike previous
researchers who use a combined measure of party control and electoral competition. For instance,
Collins (2003b: 36) employs the Ranney index (1965) of inter-electoral competition, which
assigns each state a score from 0 to 100. A score of 0 indicates complete Republican control and
little electoral competition, while a score of 100 indicates complete Democratic control and little
electoral competition. Hence a score of 50 indicates a highly competitive political system in
which neither Republicans nor Democrats held the upper hand. To construct his index, Ranney
gathered data on state politics for the 1946-1963 period and averaged four components for each
state: the average percentage vote for Democratic gubernatorial candidates, the average
percentage of seats held by Democrats in the upper house, the average percentage of seats held by
Democrats in the lower house, and the percent of all legislative sessions during the period in
which Democrats held control over all three institutions. To capture any potential non-linearities,
Collins (2003b) includes a quadratic transformation of the index.
Using the Ranney index to identify the impact of Republican control has several
problems. One concern is that it is time-constant for each state. This is problematic because there
were significant changes in inter-electoral competition in the postwar period. A state that
gradually shifts from total Republican dominance to marginal Democratic control would have a
similar value on the index as a highly competitive state that consistently remains under
Republican control. Under these circumstances, it is not possible to identify the effect of party
control. A more serious problem is that the Ranney index does not distinguish the effect of party
if this affects the results, I estimate the final trimmed model (Table 3, Model 4) excluding all
observations from both Indiana and Wisconsin. The results, which are highly comparable, are
available upon request.
18
control from electoral competition. A government in which Democrats hold the governorship and
both houses of the legislature is obviously politically different than one in which a Democratic
governor faces a divided legislature. Such differences might not be properly reflected in the
Ranney index. Consider, for instance, two hypothetical states. In one, Republicans narrowly win
the gubernatorial race and barely win control of the legislature. In the other, Republicans win the
gubernatorial race by a landslide, narrowly win the Assembly, and marginally lose the Senate.
Using the Ranney index, both states are observationally equivalent, even though one would exist
under unified Republican control, while the other would exist under divided government.
For these reasons, I measure Republican control separately from electoral competition.
Since case studies of individual states suggest that Republicans typically used their control to
block FEP legislation, primarily through their control of key committees, I operationalize
Republican control as a binary variable indicating whether Republicans held control over a “veto
point” in the legislative process. This operationalization is highly consistent with the
“institutional politics” theory of policy-making, which predicts that the impact of party
organizations is mediated by the institutional structure of political authority (Amenta and
Halfman 2000). The variable is time-varying, and it is set to 1 if Republicans hold a majority of
seats in the lower house, a majority of seats in the upper house, the governorship, or any
combination of the three. The variable is set to 0 when Democrats hold unified control (Council
of State Governments, various years). 11 A negative, statistically significant coefficient indicates
that Republican control depressed the likelihood of passage.
11 My coding scheme raises several potential concerns. First, it is possible to classify only state
legislatures controlled by a Democratic supermajority (i.e., a veto-proof majority) as “Democratic
control” (that is, set to 0). This circumstance was extremely rare outside the South, however.
Moreover, I am not aware of any case in which a Democratic or Republican governor actually
vetoed a FEP bill. Republican governors very occasionally threatened a veto, but these cases are
19
My other independent variable is the political strength of employers. It is obvious that
their political strength can vary considerably, depending on the composition, cohesion, and
competitiveness of the firms and industries in a state. But since the records of business lobbies are
not generally available, it is difficult to measure their strength by tallying up their membership,
correctly coded as Republican control. Second, using a dummy variable discards information
about the magnitude of party control. This is true, but I am less interested in whether the passage
of FEP laws is a continuous function of incremental changes in Republican control and more
interested in whether the passage of FEP legislation is a step function of Republican control—that
is, I am more concerned about identifying the difference between a Republican-controlled
government and a Democratic-controlled government than identifying the difference between a
weakly Republican-controlled government and a strongly Republican-controlled government. To
gauge the impact of alternate coding, I estimate the final trimmed model (Table 3, Model 4) with
a more differentiated measure of party control; namely, six dummies indicating unified
Republican control, Republican governor and divided legislature, Republican governor and
Democratic legislature, Democratic governor and Republican legislature, Democratic governor
and divided legislature, and unified Democratic control. I use unified Democratic control as the
reference category. I also estimate the final trimmed model with a dummy variable that is set to 1
for a state under unified Democratic control, and 0 otherwise. All of the results are substantially
similar. Minnesota and Nebraska hold non-partisan elections for the legislature. For these states, I
code Republican control based on the party of the governor. To determine if this coding decision
drives the results, I re-estimate the final trimmed model excluding all observations from
Minnesota and Nebraska. The coefficient for Republican control is robust, but the coefficient for
employer strength exhibits some sensitivity.
20
calculating their annual expenditure on political activities, or considering other direct indicators.12
It is necessary to gauge their strength indirectly. One strategy is to use the passage of employer-
friendly legislation as a proxy measure on the assumption that states with employer-friendly laws
are also states with politically powerful and successful employers. This warrants caution. Laws
beneficial to employers may have passed for reasons that were unrelated to their political
strength. Also, employers may view certain laws as serving (or not harming) their interests only
ex post. Before such laws passed, employers may have opposed them. Thus, it is crucial to select
laws that enjoyed clear, ex ante support from employers and whose passage appears to have been
a consequence of employer mobilization.
Of the many possibilities, state “right-to-work” (RTW) laws, which banned union shops,
seem most attractive.13 RTW laws began spreading quickly across the states after 1947, when
Congress enacted the Taft-Hartley Act over President Truman’s veto. Taft-Hartley outlawed the
closed shop and gave states the authority to decide whether to outlaw the union shop (Labor-
Management Relations Act 61 Stat. 136, 29 U.S.C. 141 [1947]). Many states chose to do so. By
1964, nearly two-thirds of the states had passed RTW laws prohibiting union shops. Few types of
state legislation addressed employer interests more squarely or inspired their political
12 The records of the Minnesota Manufacturers Association and the California Chamber of
Commerce are informative about their operation in a limited number of years. Few other business
lobbies, however, have made their historical records publicly available. The records of the
National Association of Manufacturers and U.S. Chamber of Commerce contain only limited
amounts of relevant information on their respective state and local affiliates.
13 A “closed shop” is a company that will employ only union workers. A “union shop” is a
company that does not require union membership as a condition of hiring, but requires it for
continued employment after a specific period of time. A so-called open shop is a company that
does not require union membership for either hiring or continued employment.
21
involvement more effectively. After the Second World War, corporate and business leaders
launched a counter-mobilization against liberalism, labor, and the New Deal (Fones-Wolf 1994),
and securing right-to-work legislation was a major element of their political agenda in the states.
For these reasons, state RTW laws are a useful, if limited, proxy for employer strength. I
code the measure as a time-constant dummy variable coded 1 if a state passed a right-to-work law
by 1964, and 0 if it did not.14 This measure is extremely coarse, and measures employer strength
with error, but it is broadly consistent with the assumption that states that had passed RTW laws
by 1964 were states in which employers were politically powerful enough to secure and defend
their passage. The validity of the measure naturally hinges on whether one finds this assumption
plausible. If one does, then a negative, statistically significant coefficient would indicate that the
political strength of employers is inversely related to the likelihood of passage.
Research on the impact of party organizations, interest groups, social movements on
public policy is extensive, and I control for some of the most important variables that have been
shown to influence the adoption of state civil rights legislation or the pace of policy innovation
more generally (see Besley and Case 2003).15
14 According to Lumsden and Peterson (1975: 1242), states that had passed a RTW law by 1964
include Arizona (1946), Nebraska (1946), South Dakota (1946), Iowa (1947), North Dakota
(1947), Nevada (1951), Utah (1955), Indiana (1957), Kansas (1958), Wyoming (1963). Indiana
repealed its law in 1965.
15 I choose not control for innovative propensity using Walker’s score, as other researchers have
done for reasons that are sound to their purposes (e.g., Soule and Zylan 1997; Zylan and Soule
2000). Ranging from 0 to 1, Walker’s score is constructed from data on 88 different programs
that had been enacted in at least 20 states by 1965 (Walker 1969: 882). The first state to enact a
particular program is given a score of 0, while the last state to enact a program is given a 1. States
enacting programs in the interim are given a score that corresponds to the proportion of time
22
The oldest studies find that economic modernization is very strongly correlated with
policy innovation (Dye 1969; Walker 1969; Gray 1973). I control for economic modernization
using three variables. The first is a time-varying variable for income, measured by a state’s
personal income per capita (U.S. Bureau of the Census, SA, various years).16 The second is a
time-varying variable for industrialization, measured by the value-added in manufacturing per
capita (U.S. Bureau of the Census, SA, various years). Both amounts are adjusted for inflation
elapsed between the enactment of the first and last program. Scores for each program are then
averaged by state. This score is the dependent variable in Walker’s study, but it seems
inappropriate for use as an independent variable. This is mainly because it is unclear what the
score measures. Using it would only tell us that innovative states tend to innovate—that they have
a propensity to innovate—but not necessarily why they tend to innovate. In results available upon
request, I re-estimated the final trimmed model (Table 4, Model 4) including Walker’s score as an
additional time-constant covariate. The coefficient for Walker’s score is large, positive, and
statistically significant, but the coefficient for Republican control remains negative, large, and
statistically significant. The coefficient for employer strength remains negative and large, but it
becomes indistinguishable from zero. If Walker’s score is substituted in the final trimmed model
for employer strength, the coefficients for Republican control and Walker’s score remain
significant. Since my measure of employer strength is plausibly more precise about the sources of
innovation than Walker’s score, I choose to retain in it in the final trimmed model.
16 It is worth pointing out that Erikson, Wright, and McIver (1993: 86-7) find evidence that
income and other demographic variables influence policy primarily “because income is correlated
with the degree of liberal sentiment of state public opinion.” Hence controlling for income
partially controls for the ideological character of the state electorate.
23
using the Consumer Price Index for urban consumers (CPI-U).17 Urbanization is a third time-
varying variable, which I measure as the percentage of individuals living in urban areas of the
state (U.S. Bureau of the Census, SA, various years).
But various aspects of electoral politics also matter. Among them, electoral competition
is one of the most relevant (Walker 1969; Skocpol et al. 1993; Holbrook and Van Dunk 1993;
Barrileaux, Holbrook, and Langer 2002). In an electorally competitive environment, where the
electoral strength of the parties is comparable, partisan legislators make broader appeals than they
would otherwise, thereby improving the chances of policy innovation. I control for electoral
competition through a modified version of a time-varying measure initially developed by Skocpol
et al. (1993: 699). This measure is constructed by averaging three percentages: the percentage
margin of victory for the sitting governor in the previous election, the seat margin of the majority
party in the upper house expressed as a percentage of the total number of seats in the upper house,
and the seat margin of the majority party in the lower house expressed as a percentage of the total
number of seats in the lower house (Council of State Governments, various years; Congressional
Quarterly 1994). I then subtract the average from 100. This yields a variable for electoral
competition that is measured independently of party control. A score of 100 indicates a highly
competitive state, while a score of 0 indicates a grossly non-competitive state.
In order to plausibly identify the effect of Republican control, it is essential to control for
public opinion and the policy preferences of the electorate. To be sure, the link between public
opinion, political parties, and policy outcomes remains a much-debated area of research (Manza,
Cook, and Page 2002). But research on the states has consistently demonstrated a consistent link
between “general mass political attitudes and the general choices of state policy makers” (Brace
et al 2002: 173). At a general level, “party control is not a particularly good indicator of state
17 U.S. Bureau of the Census, Statistical Abstract (Washington, D.C.: U.S. Government Printing
Office, various years). For CPI-U figures, see ftp://ftp.bls.gov/pub/special.requests/cpi/cpiai.txt.
24
policy” (Erikson, Wright, and McIver 1989: 743)—primarily because the responsiveness of state
parties to state opinion (which is more moderate that party positions) leads Democratic and
Republican legislators to moderate their policy positions. Hence, public opinion can strongly
shape policy outcomes in the states (Wright, Erikson, McIver 1987; Erikson, Right, and McIver
1993). Failing to control for the ideological character of public opinion can greatly exaggerate the
role of party organizations in policy-making (Burstein and Linton 2002; Burstein 1998).
While public opinion is significant generally, it is important to focus on the specific issue
of state FEP laws because “mass belief systems show little internal consistency” (Brace et al
2002, citing Converse 1964). This specific focus is all the more important because the most
extensive studies of Congressional action on equal employment opportunity legislation suggest
the importance of public attitudes regarding civil rights (Burstein 1985; Santoro 2002). I control
for across-state differences in public opinion on state FEP laws by using data from a Gallup Poll
(N=1,581) taken in 1945.18 From the raw Gallup poll data I calculate the percentage of
18 I also control for public opinion using additional variables. In results available upon
request, I use two different measures to control for the ideological character of mass opinion in
the states. The first is a score of citizen ideology developed by Berry and his collaborators (Berry
et al 1998) from data on Congressional roll call votes and other sources. Although scores are now
available annually for the period 1960-2002, I use only the average of the scores from 1960 to
1964 as a time-constant covariate. The averaged score ranges from a low (conservative) of
28.83112 for Nebraska and a high (liberal) of 78.90062 for Rhode Island. When substituted for
the main measure of public opinion, it does not yield a statistically significant coefficient in any
specification reported in Table 3. The second variable is a survey-based, time-constant measure
of state opinion developed by Wright, Erikson, and McIver (1985) from pooled (1974-1982)
CBS/New York Times surveys. The measure ranges from a low (liberal) of -.053 for Nevada to a
high (conservative) of .333 for Utah. When substituted for the main measure of public opinion, it
25
respondents in each state answering yes to the following question: “Would you favor or oppose a
state law which would require employers to hire a person if he is qualified for the job, regardless
of his race or color?” (Gallup Organization 1945). This variable is measured with error since it
disaggregates data from what is meant to be a nationally representative sample, but the results do
have face validity, according with generally held conceptions of racial liberalism in the states. For
instance, New York (75%) was among the most supportive of state FEP legislation; Michigan
(51%) was moderately supportive; while Missouri (21%) was least supportive. I retrieve this
information by recoding it as a binary variable, which is coarse enough to ensure that a
“favorable” state is not misclassified as “opposed,” and vice versa. I specify public opinion on
FEP as a time-constant dummy variable set to 1 if the percentage of residents in a state expressing
support for a FEP law is higher than the mean level of support for all thirty-seven states in the
risk set; and 0 if it is lower than the mean.19
does not yield a statistically significant coefficient in any specification reported in Table 3. Using
a third time-constant variable developed by Brace et al (2002) from pooled data (1974-1998) in
the General Social Survey, I control for mass opinion about racial integration. The variable ranges
from a low of .5 for West Virginia to a high of .88 for Rhode Island. I used the sample mean (.75)
to replace missing data for five states. When substituted for the main measure of public opinion, it
does not yield a statistically significant coefficient in any specification reported in Table 3.
19 The distribution of the variable closely approximates the normal standard distribution. I used
the sample mean of the thirty-seven states to replace missing data that were not available for
Oregon, New Mexico, Delaware, and North Dakota. This is admittedly a crude control, but it is
the best one presently available, and it partially addresses the total absence of public opinion from
prior models. I also tried using the raw state percentages from the Gallup poll, but they do not
come out statistically significant in any specification.
26
The malapportionment of state legislatures—whereby rural areas enjoyed
disproportionate representative relative to urban areas—has been shown to shape certain policy
outcomes. Most recently, Ansolabehere, Gerber, and Synder (2000) find that malapportionment
influences the distribution of public expenditures by state governments. I control for
malapportionment using the Right-To-Vote (RTV) index developed by Ansolabehere, Gerber,
and Synder (2000: 30-1). The index varies from 1.07 (NH) to 3.54 (CA), where a score of 1
indicates a well-apportioned legislature in 1960 under the one-person, one-vote rule and higher
scores indicate overrepresentation. The RTV index is time-constant.
Organized business was not the only interest group with a stake in fair employment
legislation. The “other” civil rights movement aggressively lobbied for it.20 Interest groups like
the NAACP, American Jewish Congress (AJ Congress), and Catholic Interracial Council (CIC)
promoted FEP legislation because it offered protection against discrimination to their members.
Many unions also supported FEP legislation out of perceived self-interest. In the early postwar
years, unions affiliated with the Congress of Industrial Organizations (CIO), such as the United
Automobile Workers, backed FEP more strongly and consistently than craft unions in the
American Federation of Labor (AFL). By 1964, however, nearly all local and international unions
as well as the AFL-CIO itself publicly supported FEP legislation—even if their actual practices
fell short of their articulated ideals (Frymer 2003). Collins (2003b) finds that the mobilization of
Jewish groups, civil rights organizations, and unions—as well as the size of the Catholic
population—were positively related to passage. I control for the electoral and political
significance of these groups by including measures of the percentage black, percentage Jewish,
and percentage Catholic for each state-year (U.S. Census Bureau, 1975; American Jewish
20 For empirical studies of how the civil rights movement shaped public policy, see Andrews
(2001) and Button (1997).
27
Committee, various years; Official Catholic Directory, various years).21 I control for union
strength by including a measure of the percentage of the non-agricultural workforce in a union in
each state-year (Troy 1985). To control for the actual mobilization and not merely the potential
mobilization of the African American community (Andrews 2002), I include a measure of the
percentage of African Americans with NAACP membership for each state year (NAACP, various
years).22
A different theory predicts that the electoral importance of social groups standing to
benefit from legislation might actually depress the likelihood of passage. Specifically, “racial
competition” or “racial threat” theory (Olzak 1992; Behrens, Uggen, and Manza 2003) predicts
that passage varies inversely with the size of the black, Jewish, and Catholic population. This is
because racial, ethnic, and religious groups are thought to compete against one another for scarce
economic resources and white Protestants viewed blacks, Jewish, and Catholics as a threat to their
21 Data on Catholic residents by state was graciously provided by Mary Gautier at the Center for
Applied Research in the Apostolate at Georgetown University.
22 To identify potential non-linear relationships, I substituted logged measures of the black,
Jewish, and Catholic population, as well as NAACP membership in the full specification. The
results differ slightly for the control variables. While the coefficients for Republican control and
employer strength remain similar, the coefficients for Jewish population (ln), Catholic population
(ln), and NAACP membership (ln) become statistically significant, and the coefficient for black
population (ln) becomes statistically insignificant. They are all highly multicollinear, however.
This is clear when the full specification with logged measures is estimated using OLS regression,
and a Variance Inflation Factor (VIF) is calculated for each coefficient. Jewish population (ln),
black population (ln), Catholic population (ln), and NAACP membership (ln), all exhibit high
VIF scores. Since the main results do not change, and since the interpretation of the percentage
measures is more straightforward, I retain the results reported in Table 3.
28
dominant position. Since theory gives contradictory predictions regarding the directionality of
these three variables, I treat it as a strictly empirical question, following Collins (2003b).
Previous studies find a complex series of diffusion effects associated with policy
innovation (Berry and Berry 1990; Berry and Berry 1992; Strang and Tuma 1993; Zylan and
Soule 2000). Analysts have given different explanations of the effect, but the most robust and
consistent finding is that the adoption of legislation in a neighboring state raises the likelihood
that a non-adopter will pass similar legislation. I control for diffusion by including a variable that
measures the percentage of neighboring states that have adopted FEP legislation.
EMPIRICAL ANALYSIS
Descriptive Findings
Table 2 presents the FEP passage rate across all of the explanatory variables. For the
bivariate analysis, I convert all continuous measures into quartiles. The results offer preliminary
support for the hypothesis that Republican control and employer strength are both inversely
related to passage. Only 4.2% of the states (i.e., state-years) under Republican control saw the
passage of FEP legislation, compared to 7% of the states under unified Democratic control. Only
2.6% of RTW states passed a FEP law, while 5.8% of non-RTW states did.
[Table 2 about here.]
The control variables also exhibit strong associations with legislative passage, raising the
possibility that the bivariate results are spurious. Income, industrialization, urbanization, the
percentage of Jewish residents, percentage of Catholic residents, and union density, all exhibit a
positive (though not necessarily monotonic) relationship with the likelihood of adoption. The
passage of FEP legislation in a neighboring state could also raise the chances of passage in a non-
adopter. FEP laws passed in ten percent of states in which a neighboring state had passed a FEP
law, compared to only one percent in which no neighboring state had passed a FEP law.
Malapportionment and percent black similarly display strong relationships with the likelihood of
29
passage, although the effects appear non-linear. Only public opinion and NAACP membership
(relative to the African American population) appear unrelated to passage. These associations call
for multivariate analysis. If Republican control and employer strength remain negatively and
significantly associated with the likelihood of passage, even after including the relevant control
variables, then there would be stronger evidence supporting the hypothesis.
Multivariate Findings
Table 3 reports the parameter estimates from the multivariate analysis. Model 1 is the full
specification, which includes Republican control, employer strength, and all of the control
variables. The results offer mixed support for the hypothesis. The model as a whole is statistically
significant (χ2=75.88, df=15, p<.00), and it results in a proportional reduction-in-error of .39. The
coefficient for Republican control is large, negative, and statistically significant, but the
coefficient for employer strength—while large and negative—does not reach statistical
significance at conventional levels. Several control variables are statistically significant. Income,
electoral competition, union density, and adjacency are positively related to passage, as expected.
Percentage black is negatively related to passage. The remaining control variables are not
significant, including industrialization, urbanization, malapportionment, percent Jewish, percent
Catholic, percent NAACP membership, and public opinion. The results of Model 1 provide
preliminary evidence that the bivariate result for Republican control is not spurious to the
multiple controls included in the specification.
[Table 3 about here]
But the results for employer strength and public opinion should be subjected to further
scrutiny. While the estimated coefficients for these variables are not statistically significant at the
conventional threshold, they are each reasonably close (p≈.10). This requires special attention
because the event-per-variable (EPV) ratio is extremely small (24 events/15 variables = EPV 1.6).
In logistic regression models with small EPV ratios, estimated coefficients can be severely biased
30
and significance tests can be too conservative (Peduzzi et all 1996). It is hence imperative to trim
the model to include only the best-fitting controls. While it is normally preferable to incorporate
all theoretically relevant variables in the model specification, this approach is not advisable in this
instance since there are only twenty-four events in the data set.23
I estimate a trimmed specification by removing variables from Model 1 that are not
significant at the p<.15 level. The results offer more consistent support for the hypothesis. The
resulting specification, Model 2, is statistically significant as a whole (χ2=70.44, df=9, p<.00),
and it results in a pseudo-R2 comparable to Model 1. The coefficient for Republican control
remains large, negative, and statistically significant, and the coefficient for employer strength,
which remains large and negative, becomes statistically significant at the p<.05 level. The
direction, magnitude, and significance of the coefficients for the controls retained from Model 1
remain similar, and the coefficient for public opinion, which remains large and positive, becomes
significant as well. A likelihood ratio test between Model 1 and Model 2 (χ2=5.44, df=6, p<.49)
clearly indicates that the removed variables are not jointly significant.
How robust are the results of Model 2? A “jackknife” diagnostic reveals that the main
results are robust, but the results for the control variables are highly sensitive.24 In particular, the
23 Thanks to Yu Xie for pointing this out to me. For a general discussion, see Hosmer and
Lemeshow (2000: 345-6). Using Monte Carlo simulations, one study of the EPV ratio in logistic
regression finds that more than one-third of the estimated coefficients are severely biased (twice
as large or half as small as the true value) at an EPV of 2 (Peduzzi et al 1996).
24 This procedure entails successively re-estimating a model and systematically excluding a
different state and year from the risk set each time. Hence as many as 37 observations are
excluded in some instances. This is obviously a more rigorous test than simply excluding only
one state-year observation at a time. The coefficient for public opinion is sensitive to the
31
coefficient for public opinion is driven by outlying states and years. This likely reflects the
coarseness of the measure. When public opinion from is removed the specification, union density
becomes insignificant, although the coefficients associated with Republican control and employer
strength remain large, negative, and statistically significant. Further analysis reveals that union
density and public opinion are significant only when they are jointly included in the model.
I estimate another trimmed specification by removing both union density and public
opinion from Model 2. I report the results as Model 3. The model as a whole is significant
(χ2=64.06, df=7, p<.00), and it results in a proportional reduction-in-error of .33, making it
comparable to Model 1. The results from Model 3 offer consistent support for the hypothesis as
well. The coefficients for Republican control and employer strength are large, negative, and
statistically significant, though the size of the coefficient for Republican control is slightly
smaller and the coefficient for employer strength is slightly larger than in previous models. The
directionality, magnitude, and significance of the retained control variables in Model 3 are highly
similar to earlier models. As expected, a likelihood-ratio test between Model 2 and Model 3
indicates that the removed variables are jointly significant (χ2=6.38, df=2, p<.05). A jackknife
diagnostic indicates that the main results are highly robust to the exclusion of outlying states and
years. Of the control variables, only percentage black exhibits any sensitivity.25
Model 4 is the final trimmed specification, which excludes percentage black. It yields
highly similar results for Republican control as well as the control variables, but employer
exclusion of 1945, 1949, 1955, 1959, California, Colorado, Kentucky, Michigan, Montana, Ohio,
and Oregon. When these observations are excluded, it becomes statistically insignificant.
25 Further analysis indicates that the coefficient for percentage black is sensitive primarily to the
exclusion of Maryland, indicating that it could be driven by the large black populations in the
border states. When Maryland and Delaware are both excluded, the size and significance of the
coefficient falls substantially.
32
strength exhibits some sensitivity to outlying observations, notably the exclusion of Nebraska and
Nevada. But the coefficient for Republican control in Model 4 is fairly robust to different
functional forms (Buckley and Westerland 2004) and different assumptions about the
parameterization of the hazard rate. Probit, linear probability, and complementary log-log models
yield very similar results in the discrete-time framework, as do exponential, Gompertz, Weibull,
and Cox semi-parametric (continuous-time) models. Employer strength is robust as well, except
in the linear probability framework. 26
In all four specifications so far reported, Republican control—and to a lesser extent,
employer strength—shows robustness to the exclusion of theoretically and methodologically
relevant observations. But how robust are the results to the possibility that there are significant
unobserved differences across states or time? This is known as the problem of unobserved
heterogeneity, or frailty, which can inflate negative duration dependence or deflate positive
duration dependence (Peterson 1991; Powers and Xie 2000). More conventionally, unobserved
heterogeneity can work as a form of omitted variable bias. For instance, if more “conservative”
states are less likely to pass FEP legislation and tend to be more Republican than other states, it is
possible that the negative effect of Republican control is smaller than actually estimated, since
conservatism is not directly observed.
One increasingly popular way of addressing unobserved heterogeneity is the estimation
of random-effects models, but Peterson and Koput (1991: 408) as well as Powers and Xie (2000:
190) advise caution in their use.27 This advice is even more pertinent given the sparseness of the
26 All results are available upon request.
27 Parametric random-effects models can be sensitive to assumptions about the distribution of
unobserved heterogeneity (Heckman and Singer 1984), while estimates from Heckman-Singer,
non-parametric models can be unstable (Hoem 1989, cited in Peterson and Koput 1991). Using
the final trimmed specification (Table 3, Model 4), I estimated a state random-effects model with
33
data. A better option is perhaps a state-fixed effects model, which would identify only within-
state differences.28 This obviously disallows the inclusion of time-constant covariates such as
employer strength, but time-varying covariates such as Republican control can be included. State
fixed-effects are not possible in a logistic-regression framework because the number of variables
would exceed the number of events, but a linear probability model (OLS) does not impose similar
constraints in this regard. The estimation of standard errors in a linear probability model can be
inconsistent, but it can serve as a useful robustness check.
[Table 4 about here.]
To check the robustness of the Republican control coefficient, I estimate several fixed-
effects models using OLS regression. The results are displayed in Table 4. Model 1 is a minimal
specification that includes state fixed-effects, Republican control, and a linear time trend. The
coefficient for Republican control is consistent with the previous results. Its magnitude is not
readily interpretable, but it is negative and statistically significant. Model 2 is a full specification
that includes all time-varying covariates; namely, income, industrialization, urbanization,
electoral competition, union density, percentage black, percentage Jewish, percentage Catholic,
NAACP membership, and percent of adjacent states adopting. The results differ slightly for the
control variables (though the ones for electoral competition and percent adjacent remain the
same), but the coefficient for Republican control is highly comparable. Both a minimal
specification (Model 3) and a full specification (Model 4) that add time fixed-effects consistently
show the same results for Republican control.
Gaussian frailty. The results indicate that unobserved heterogeneity is not present, but the model
is incapable of detecting unobserved heterogeneity when covariates of known importance are
excluded.
28 Thanks to Ken Chay for the suggestion.
34
The problem of unobserved heterogeneity is more or less severe depending on the
magnitude of the observed effect. Since the size of the effect associated with logit coefficients is
not directly interpretable, Table 5 presents the predicted probabilities associated with the
independent variables as calculated using the SPost suite (Long and Freese 2001). Republican
control strongly reduces the probability of passage across specifications. The baseline probability
of all models (where the covariates are set to their sample means) is 1 percent. But if a state goes
from Democratic to Republican control, with all other variables held at their sample means, then
it is anywhere from 4 to 6 percent less likely to pass a FEP law. The effect is slightly smaller for
employer strength. If a state has a powerful business lobby, it is 1 to 2 percent less likely to pass a
FEP law. In both cases, however, it seems that the size of the effect is considerable relative to the
baseline probability.
[Table 5 about here]
How should these multivariate results be interpreted overall? It would be overreaching to
claim that they are evidence of a “causal effect” in a strict sense. This would require random
assignment of Republican control or employer strength, which is clearly impossible; or it would
require some kind of instrument or exogenous shock whose existence is unlikely. The results of
the fixed-effects models provide some reassurance against unobserved heterogeneity or omitted
variables bias, but if one finds the use of such models unconvincing on technical grounds, then
the results of the discrete-time models could be spurious. The possibility cannot be ruled out that
there is some unmeasured characteristic among Republican-controlled or employer-dominated
states responsible for depressing the likelihood of passage.
Of special concern is whether the coefficient for Republican control identifies a “party
control” effect or whether it reflects the underlying preferences of the electorate in a broadly
Downsian sense (Erikson, Wright, and McIver 1993; Burstein 1998). The latter idea lends itself
to a straightforward and compelling interpretation of the empirical results. When and where
public opinion was conservative, running against FEP legislation, voters tended to elect
35
Republicans to office. When and where public opinion was liberal, favoring FEP legislation,
voters tended to elect Democrats to office. This is essentially a story about selection bias. States
with conservative electorates selected Republicans to derail FEP proposals, while states with
liberal electorates selected Democrats to pass FEP proposals.
But is such a view consistent with further empirical analysis of the data? This question
can be assessed in two ways if one assumes (as one must under the “public opinion”
interpretation) that partisan representation in elective office is a reasonable proxy for public
sentiment. The first approach involves estimating a model that controls for public sentiment
through a measure of Republican electoral strength. If the coefficient for Republican control
remains large, negative, and significant across all levels of Republican strength—that is, when the
public favors and disfavors FEP laws—then it strengthens the case for a “party control”
interpretation. To operationalize Republican strength, I average the vote share of the Republican
candidate in the previous gubernatorial election, the Republican share of the upper house, and the
Republican share of the lower house. This yields a number that, when multiplied by 100, varies
between 0 and 100—where 0 indicates extreme Republican weakness and 100 indicates extreme
Republican strength.
A second empirical strategy involves comparing states that are barely under Republican
control with states that are barely under Democratic control. This “controls” for public opinion
because the “public opinion” interpretation implies that states under the marginal control of either
party are very similar in their underlying policy preferences.29 The strategy can be implemented
by interacting Republican control with the existing measure of electoral competition. Recall that
the measure for electoral competition does not distinguish which of the two parties is dominant; it
29 This empirical strategy borrows the intuition of DiNardo and Lee (2004), who compare unions
that barely won a certification election to unions that barely lost a certification election in order to
identify the effects of unionization on various labor market outcomes.
36
merely measures how close the parties are to one another in terms of their representation in
elective office. If the effect of Republican control vanishes at the highest levels of electoral
competition—that is to say, if Republican control has a negative effect only at low levels of
electoral competition—it would be consistent with a “public opinion” interpretation. But if it
remains negative, large, and significant (or perhaps even grows larger) at the highest levels of
electoral competition, where either party is barely in control, then it would strengthen the case for
a “party control” interpretation.
[Table 6 about here]
Table 6 presents the estimated logit coefficients from several additional event-history
models of state FEP legislation. Model 1 presents the results of a specification identical to the
final trimmed specification (Table 3, Model 4) except that the measure of Republican strength is
substituted for the electoral competition. The results support the “party control” interpretation.
The electoral strength of Republicans is positively and significantly related to passage, but the
coefficient for Republican control nonetheless remains negative, large, and significant. A plot of
predicted probabilities against Republican strength indicates the possibility that a quadratic
transformation of the variable might provide a better fit. Model 2 presents the results of such a
specification. Adding a quadratic term does improve the fit of the model, as indicated by the
increase in the proportional reduction-in-error. The results continue to support the “party control”
interpretation. Even when controlling for the electoral strength of the GOP (as a proxy of public
sentiment), Republican control of veto points strongly reduces the likelihood of passage. In both
Model 1 and Model 2, Republican control lowers the probability of passage by 10%, holding all
other variables at their sample means.
Model 3 of Table 6 reports the coefficients from a specification identical to the final
trimmed specification (Table 3, Model 4), except that electoral competition is entered as a
categorical variable by quartile. Here the probability of passage falls by 4% under Republican
control, holding all other variables by the sample means. Model 4 includes an interaction term
37
between Republican control and electoral competition. These results also strongly support a
“party control” interpretation. The interaction term is positive and statistically significant,
indicating that Republican control has a greater negative effect on passage at high levels of
electoral competition than it does at lower levels of electoral competition. At the highest level of
electoral competition (top quartile), Republican control lowers the chances of passage by at
massive 23%, with all other variables set to their sample means. Quite in contrast to the “public
opinion” interpretation, when states are very similar in their public sentiment about FEP,
Republican control of veto points has the greatest effect.
These results strengthen the case that Republican control lowers the likelihood of passage
independently of the underlying preferences of the electorate. But even if the quasi-Downsian
interpretation were valid, it would necessarily imply that the Republican masses in the North
were more reluctant to support FEP laws compared to the Democratic masses. It would also
necessarily imply that the parties were already aligned on opposite sides of the FEP issue. Voters
wary of FEP laws were putting Republicans in office only because it was clear to them that GOP
control of policy-making would reduce the chance of getting a FEP law. While not fundamentally
elite-driven, these conclusions would pose equally important challenge to existing accounts of
realignment.
Ultimately, it is untenable to maintain that public opinion played no part in the passage of
FEP laws. But it seems equally untenable to regard the effect of Republican control as totally
epiphenomenal. The foregoing empirical analysis has yielded fairly robust evidence that the
effect of Republican control is consistently large and non-zero. Ceretis paribus, whenever
Republicans controlled a veto point in the legislative process, FEP legislation was less likely to
pass. The direction, magnitude significance of the coefficient for Republican control persists
across a broad range of model specifications, including models where mass opinion about state
FEP laws is directly controlled. The coefficient would still constitute a large effect, even if it were
half the size estimated. It is not driven by outlying observations, and it remains even when
38
sensitive controls are removed. In fact, it appears to grow larger and more significant as
additional controls are added to the specification. I cannot conclusively rule out unobserved
heterogeneity or omitted variables bias, but given the evidence amassed, it seems increasingly
difficult to conclude that the effect of Republican control is artifactual or spurious.
The empirical analysis provides more modest evidence about the effect of employer
strength. The estimated coefficient is somewhat sensitive to a variety of tests, and the measure
itself is extremely coarse, raising doubts about identification. But the results are broadly
consistent with the findings of the case-study literature and suggest that the political mobilization
of business interests retarded the passage of FEP legislation in the states.
DISCUSSION AND CONCLUSION
The electoral realignment that began in the mid-1960s is undoubtedly one of the most
significant developments in modern U.S. history. Many accounts of realignment acknowledge its
complex origins, but the most influential among them maintain that race and civil rights were
simply not partisan issues among political elites before 1964 (Carmines and Stimson 1989;
Sleeper 1990; Edsall and Edsall 1991; Thernstrom and Thernstrom 1997). These accounts argue
that racial liberalism was prevalent among Republicans and northern Democrats alike, who
struggled to forge color-blind policies aimed at dismantling racial apartheid and eliminating racial
discrimination. If their efforts proved largely unfruitful, it was primarily due to southern
Democrats, who obstructed the passage of civil rights legislation and stoked the fires of massive
resistance across the South.
Things are said to have been different after 1964. Both the civil rights movement and
public policy took a decidedly color-conscious turn, fueling backlash. This backlash, in turn,
contributed breakup of the New Deal coalition, setting organized labor and northern workers
against the civil rights movement and African Americans. Moreover, it gave Republican elites a
gilded opportunity to execute a “southern strategy” that would help them build a new electoral
39
majority out of the discontent of white voters in the North and South alike. Post hoc, ergo propter
hoc: Democratic fortunes fell and Republican fortunes rose after affirmative action; therefore,
Democratic fortunes fell and Republican fortunes rose because of affirmative action.
This article presents empirical evidence that contradicts the dominant account of electoral
realignment. Analyzing a new data set using event-history methods, I confirm the suggestive
findings of earlier research (Erikson 1971; Sugrue 1997; Siskind 1998; Fine 2000; Chen 2002)
that highlights the key role of Republican officeholders in the politics of fair employment and
civil rights in the North. In contrast to existing research on public opinion, political parties, and
public policy (Wright, Erikson, and McIver 1985; Erikson, Wright, and McIver 1989; Burstein
and Linton 1998), I find fairly robust statistical evidence that party control mattered. Consistent
with the “institutional politics” theory of policy-making (Amenta and Halfman 2000), Republican
elites leveraged their control over legislative institutions—however limited it might have been in
some cases—to obstruct and delay passage of FEP laws, which merely mandated non-
discrimination in employment. Even as Republicans in Congress professed racial liberalism,
partially to aggravate sectional divisions in the Democratic Party, Republican elites in northern
states firmly opposed color-blind legislation like FEP. Ironically, in the politics of state FEP
legislation, Republicans played the same role in northern state legislatures as southern Democrats
did in Congress. There is less statistically robust evidence that the political power of organized
business slowed the passage of FEP laws, but the evidence is highly suggestive nonetheless.
The broader implications of these findings for accounts of electoral realignment are
evident. GOP resistance to “color-blind” state laws suggests that claims about the racial
liberalism of the party of Lincoln are somewhat overstated. It also raises doubts, with Sugrue
(1996, 2004), that affirmative action and other “color-conscious” policies were uniquely
responsible for fanning the flames of backlash and providing Republican elites with the ideal
racial wedge issue. To be sure, strategists like Kevin Phillips correctly recognized the extent to
which backlash had become a substantial force in national politics by the late-1960s. But GOP
40
resistance to FEP laws in the North suggests that affirmative action bears no special blame for the
invention of the “southern strategy” or its eventual success. If Republican elites had not
recognized the political utility of affirmative action in the late-1960s, they would have probably
tried to use a different civil rights policy—whether color-conscious or color-blind—to
accomplish the same ends. Hence the sources of realignment should be sought elsewhere.
The foregoing findings raise a host of new questions worthy of further investigation. The
field of scholarship on civil rights remains replete with studies of heroism and massive resistance
in the South during the 1950s and 1960s. It is also well occupied with studies of white backlash in
North and South during the 1970s. What still remain rare are studies of race, politics, and civil
rights in the urban and suburban North before 1964. It would be valuable for future researchers to
follow the pioneering work of Hirsch (1983), Sugrue (1996), and Self (2003). Additional studies
could provide a further basis for reassessing the dominant account of electoral realignment.
A focus on legislative battles in the North seems particularly fruitful. FEP legislation was
only one of several types of civil rights laws passed by northern states; it remains to be seen
whether Republican elites opposed fair housing legislation with equal skill and fervor.30 Northern
cities like Chicago, Cleveland, Detroit, and San Francisco were the focal point of significant
campaigns to pass civil rights ordinances, fair employment practices among them. What was the
role of Republican elites in these battles? Equally importantly, future studies should follow the
lead of Lee (2002) and Brooks (2000) in parsing the relationship between mass opinion, racial
liberalism, and partisan identification. 31 Does racial liberalism appears weaker among
Republican masses as well? When faced with the possible extension of civil rights in their own
30 See Chen and Phinney (2004), for an analysis of state fair housing legislation.
31 Chen, Mickey, and Van Houweling (2003) analyze the attitudes and behavior of Republican
voters on the question of fair employment practices using precinct-level election returns from a
1946 referendum on FEP in California, as well as NES data for the period 1956-1960.
41
villages, towns, and cities, what were the political attitudes and voting behavior of ordinary
Republicans? Did they consistently support civil rights, or did they demonstrate as little fidelity to
Lincoln’s legacy as their party leaders? Answering these and other questions will make it possible
to clarify the true sources of electoral realignment and identify the deep roots of contemporary
racial politics.
42
REFERENCES
Allison, Paul. 1982. Event History Analysis. Newbury Park: SAGE Publications.
Amenta, Edwin and Drew Halfman. 2000. “Wage Wars: Institutional Politics, WPA Wages, and
the Struggle for U.S. Social Policy.” American Sociological Review: 506-528.
American Council on Race Relations. 1945. “State FEPC—What the People Say.” Chicago:
American Council on Race Relations.
American Jewish Committee. Various years. American Jewish Yearbook. New York: American
Jewish Committee.
Andrews, Kenneth T. 2001. “Social Movements and Policy Implementation,” American
Sociological Review 66: 71-95.
----------. 2002. “Movement-Countermovement Dynamics and the Emergence of New
Institutions.” Social Forces 80: 911-936.
Ansolabehere, Stephen, Alan Gerber, and James Snyder, Jr., 2000. ”Equal Votes, Equal Money,”
published subsequently in American Political Science Review 96: 767-777.
Associated Farmers of California. 1945. The Associated Farmer 5:12 (June 28, 1945), Folder:
Legislature 1945, Box 15, Cleary Collection, Bancroft Library, UC Berkeley.
Barrileaux, Charles, Thomas Holbrook, and Laura Langer. 2002. “Electoral Competition,
Legislative Balance, and American State Welfare Policy.” American Journal of Political
Science 46: 415-427.
Bartley, Numan V. 1969. The Rise of Massive Resistance: Race and Politics in the South during
the 1950s. Baton Rouge: Louisiana State University Press.
Behrens, Angela, Christopher Uggen, and Jeff Manza. 2003. “Ballot Manipulation and the
‘Menace of Negro Domination’.” American Journal of Sociology 109: 559-605.
Berry, Frances Stokes and William D. Berry. 1990. “State Lottery Adoptions as Policy
Innovations.” American Political Science Review 84: 395-415.
43
----------. 1992. “Tax Innovation in the States: Capitalizing on Political Opportunity.” American
Journal of Political Science 36: 715-742.
Berry, William D., Evan J. Ringquist, Richard C. Fording, and Russell L. Hanson. 1998.
“Measuring Citizen and Government Ideology in the American States, 1960-93,”
American Journal of Political Science 42: 327-48.
Besley, Timothy and Anne Case. 2003. “Political Institutions and Policy Choices: Evidence from
the United States.” Journal of Economic Literature 41: 7-43.
Box-Steffensmeier, Janet M. and Bradford S. Jones. 2004. Timing and Political Change: Event-
History Modeling in Political Science. New York: Cambridge University Press.
Brace, Paul. 1993. State Government and Economic Performance. Baltimore: Johns Hopkins
University.
Brace, Paul, Kellie Sims-Butler, Kevin Arceneaux, and Martin Johnson. 2002. “Public Opinion in
the American States: New Perspectives Using National Survey Data.” American Journal
of Political Science 46: 173-189.
Brooks, Clem. 2000. “Civil Rights Liberalism and the Suppression of a Republican Political
Realignment in the United States.” American Sociological Review 65: 483-505.
Buckley, Jack and Chad Westerland. 2004. “Duration Dependence, Functional Form, and Correct
Standard Errors.” State Politics and Policy Quarterly. 4(1): 94-114.
Bureau of National Affairs. 1964. State Fair Employment Laws and Their Administration.
Washington, DC: Bureau of National Affairs.
Burstein, Paul and April Linton. 2002. “The Impact of Political Parties, Interest Groups, and
Social Movement Organizations on Public Policy.” Social Forces 81: 381-408.
----------. 1998. “Should Sociologists Consider the Impact of Public Opinion on Public Policy?”
Social Forces 77: 27-62.
----------. 1985. Discrimination, Jobs, and Politics. Chicago: University of Chicago Press.
Button, James W. 1989. Blacks and Social Change. Princeton: Princeton University Press.
44
California Chamber of Commerce. Various years. Minutes, Board of Directors Meetings, Boxes
2243-4, Records of the California Chamber of Commerce, California State Library.
Carmines, Edward G. and James A. Stimson. 1989. Issue Evolution. Princeton: Princeton
University Press.
Chen, Anthony S. 2002. “From Fair Employment to Equal Employment Opportunity and
Beyond: Affirmative Action and the Politics of Civil Rights in the New Deal Order,
1941-1972," Ph.D. diss., Department of Sociology, University of California, Berkeley.
Chen, Anthony S. Robert Mickey, and Robert P. Van Houweling. 2003. “Backlash
Reconsidered.” Paper presented at the SSHA Meetings, Baltimore, MD, Nov. 13.
Chen, Anthony S. and Robin Phinney. 2004 “Did the Civil Rights Movement Really Affect
Public Policy?” Paper presented at the ASA Meetings, San Francisco, CA, August 14-7.
Collins, William J. 2003a. “The Labor Market Impact of State-Level Anti-Discrimination Laws,
1940-1960.” Industrial and Labor Relations Review 56: 244-272.
----------. 2003b. “The Political Economic of State-Level Fair Employment Laws, 1940-1964.”
Explorations in Economic History 40: 24-51.
Congressional Quarterly. 1994. Congressional Quarterly’s Guide to U.S. Elections, 3rd ed.
Washington, DC: Congressional Quarterly.
Council of State Governments. Various years. Book of the States. Chicago: Council of State
Governments.
DiNardo, John and David S. Lee. 2004. “The Economic Impact of Unionization on Private Sector
Employers.” NBER Working Paper No. 10598. Cambridge, MA: National Bureau of
Economic Research.
Dye, Thomas R. 1969. “Inequality and Civil-Rights Policy in the States.” Journal of Politics 31:
1080-1097.
The Economist, June 10, 2004
Edsall, Thomas B. and Mary Edsall. Chain Reaction. New York: Norton.
45
Erikson, Robert S. 1971. “The Relationship between Party Control and Civil Rights Legislation
in the American States.” Western Political Quarterly 24: 178-182.
Esping-Anderson, Gösta. 1985. Politics against Markets. Princeton: Princeton University Press.
Executive Order 8802. 1941. Federal Register 6: 3109.
Fine, Sidney. 2000. “Expanding the Frontiers of Civil Rights”: Detroit: Wayne State Press.
Fones-Wolf, Elizabeth. 1994. Selling Free Enterprise. Urbana: University of Illinois Press.
Frymer, Paul. 2003. “Acting When Elected Officials Won’t.”” American Political Science
Review 15: 1-17.
Gallup Organization. 1945. Gallup Poll #349. Storrs, CT: Roper Center.
Garfinkel, Herbert. 1959. When Negroes March. Glencoe, IL: Free Press.
Gerstle, Gary and Steven Fraser. 1989. The Rise and Fall of the New Deal Order, 1930-1980.
Princeton: Princeton University Press.
Gray, Virginia. 1973. “Innovation in the States.” American Political Science Review 67: 1174-
1185.
Heckman, James and Burton Singer. 1984. “A Method for Minimizing the Impact of
Distributional Assumptions in Econometric Models of Duration Data.” Econometrica 52:
271-320.
Hirsch, Arnold. 1983. Making the Second Ghetto. New York: Cambridge University Press.
Hoem, Jan M. 1989. “Limitations of a Heterogeneity Technique” (Stockholm Research Reports
in Demography No. 56). University of Stockholm, Sweden.
Holbrook, Thomas M. and Emily Van Dunk. 1993. “Electoral competition in the American
States.” American Political Science Review 87 (December 1993): 955-962.
Hosmer, David W. and S. Lemeshow. 2000. Applied Logistic Regression. New York: Wiley.
Katznelson, Ira, Kim Geiger, and Daniel Kryder. 1993. “Limiting Liberalism: The Southern Veto
in Congress, 1933-1950.” Political Science Quarterly 108: 283-305.
46
Kesselman, Louis C. 1948. The Social Politics of FEPC: A Study in Reform Pressure
Movements. Chapel Hill: University of North Carolina Press, 1948.
Key, V. O. 1949. Southern Politics in State and Nation. New York: A. A. Knopf.
Klinkner, Philip and Rogers M. Smith 1999. The Unsteady March. Chicago: University of
Chicago.
Kryder, Daniel. 2000. Divided Arsenal.. New York: Cambridge University Press.
Kotlowski, Dean J. 1998. “Richard Nixon and the Origins of Affirmative Action,” Historian 60:
523-41.
Landes, William M. 1967. “The Effect of State Fair Employment Laws on the Economic Position
of Nonwhites.” American Economic Review 56: 578-90.
----------, ----------. 1968. “The Economics of Fair Employment Laws.” Journal of Political
Economy 76: 507-552.
Lee, Taeku. 2002. Mobilizing Public Opinion. Chicago: University of Chicago Press.
Lockard, Duane. 1968. Toward Equal Opportunity. New York: Macmillan.
Long, J. Scott and Jeremy Freese. 2001. Regression Models for Categorical Dependent Variables
Using Stata. College Station, TX: Stata Press.
Lumsden, Keith and Craig Peterson. 1975. “The Effect of Right-to-Work Laws on Unionization
in the United States.” Journal of Political Economy 83: 1237-1248.
Manza, Jeff, Fay Lomax Cook, and Benjamin I. Page. 2002. Navigating Public Opinion: Polls,
Policy, and the Future of American Democracy. New York: Oxford University Press.
Matusow, Allen J. 1984. The Unraveling of America: A History of Liberalism in the 1960s. New
York: Harper and Row.
National Association for the Advancement of Colored People (NAACP). Various years. Records
of the NAACP, Part II, 1910-1955. Washington, DC: Library of Congress.
National Association for the Advancement of Colored People (NAACP). Various years. Records
of the NAACP, Part III, 1909-1965. Washington, DC: Library of Congress.
47
Neumark, David and Wendy A. Stock. 2001. “The Effects of Race and Sex Discrimination
Laws.” (Working Paper 8215) National Bureau of Economic Research, Cambridge, MA.
Official Catholic Directory. Various years. New York: P.J. Kenedy.
“Old Employer” to W. Byron Rumford, May 4, 1955. Folder: FEPC—A.B. 971—1955, Box 3,
Rumford Papers, Bancroft Library, University of California, Berkeley.
Olzak, Susan. 1992. The Dynamics of Ethnic Competition and Conflict. Palo Alto: Stanford
University Press.
Peduzzi, Peter, John Concato, Elizabeth Kemper, Theodore R. Holford and Alvan R. Feinstein.
1996. “A Simulation Study of the Number of Events Per Variable in Logistic Regression
Analysis.” Journal of Clinical Epidemiology 49: 1373-1379.
Peterson, Trond. 1991. “The Statistical Analysis of Event Histories.” Sociological Methods and
Research 19: 270-323.
----------, ----------. 1995. “Analysis of Event Histories.” In Handbook of Statistical Modeling for
the Social and Behavioral Sciences, edited by Gerhard Arminger, Clifford C. Clogg, and
Michael E. Sobel (New York: Plenum Press).
Phillips, Kevin P. 1969. The Emerging Republican Majority. New Rochelle, NY: Arlington
House.
Powers, Daniel A. and Yu Xie. 1999. Statistical Methods for Categorical Data Analysis. San
Diego: Academic Press.
Ranney, Austin. “Parties in State Politics.” Pp. 61-99 in Politics in the American States:, edited
by Herbert Jacob and Kenneth N. Vines. Boston: Little, Brown and Company.
Reed, Merl E. 1991. Seedtime for the Modern Civil Rights Movement. Baton Rouge: Louisiana
State University Press.
Republican National Committee, Research Division. 1963. “Some Facts on Civil Rights
Legislation” (August): Part 1, Box 465, Folder: “H.R. 7152 (88),” Emanuel Cellar
Papers, Library of Congress.
48
Ruchames, Louis. Race, Jobs, and Politics. New York: Columbia University Press.
Santoro, Wayne A. “The Civil Rights Movement’s Struggle for Fair Employment: A “Dramatic
Events—Conventional Politics Model.” Social Forces 81: 177-206.
Schubert, Glendon and Charles Press. 1964. “Measuring Malapportionment.” American Political
Science Review 58: 302-327.
Self, Robert O. 2003. American Babylon. Princeton: Princeton University Press.
Sharp, Elaine B. 1999. The Sometime Connection. Albany: SUNY Press.
Siskind, Peter. 1997. “Struggling for Fair Employment.” Department of History, University of
Pennsylvania. Unpublished manuscript.
Skocpol, Theda, Marjorie Abend-Wein, Christopher Howard, and Susan Goodrich Lehmann.
1993. “Women’s Associations and the Enactment of Mothers’ Pensions in the United
States.” American Political Science Review 87: 686-701.
Skrentny, John D. 1996. The Ironies of Affirmative Action. Chicago: University of Chicago.
Sleeper, Jim. 1990. The Closest of Strangers. New York: W.W. Norton.
Soule, Sarah A. and Yvonne Zylan. 1997. “Runaway Train?” American Journal of Sociology
103: 733-762.
Strang, David, and Sarah A. Soule. 1998. “Diffusion in Organizations and Social Movements:
From Hybrid Corn to Poison Pills.” Annual Review of Sociology 24: 265-90.
Strang, David and Nancy Brandon Tuma. 1993. “Spatial and Temporal Heterogeneity in
Diffusion.” American Journal of Sociology 99: 614-639.
Sugrue, Thomas J. 2004. “Affirmative Action from Below.” JAH 92: 145-173.
----------. 1996. The Origins of the Urban Crisis. Princeton: Princeton University Press.
Thernstrom, Stephan and Abigail Thernstrom. 1997. America in Black and White. New York:
Simon and Schuster.
Troy, Leo. 1957. “Distribution of Union Membership among the States, 1939 and 1953”
(Occasional Paper No. 56) Cambridge, MA: National Bureau of Economic Research.
49
U.S. Bureau of the Census. Various years. Statistical Abstract of the United States, 1950.
Washington, DC: U.S. Government Printing Office.
----------, ----------. 1975. Historical Statistics of the United States. Washington, DC: US
Government Printing Office.
Walker, Jack L. 1969. “The Diffusion of Innovations among the American States.” American
Political Science Review 63: 880-899.
Wright, Gerald C., Robert S. Erikson, and John P. McIver. 1985. “Measuring State Partisanship
and Ideology with Survey Data,” Journal of Politics 47: 469-89.
Yamaguchi, Kazuo. 1991. Event History Analysis. Newbury Park, CA: Sage Publications.
Zylan, Yvonne and Sarah A. Soule. 2002. “Ending Welfare As We Know It (Again): Welfare
State Retrenchment, 1989-1995.” Social Forces 79: 623-652.
50
TABLE 1. Passage of Enforceable State Fair Employment Practice Laws in Thirty-Nine, Non-Southern States, 1945-1964
Year
State(s)
Annual
frequency
Cumulative frequency
Cumulative percentage
1945
New York, New Jersey
2
2
5%
1946 Massachusetts 1 3 8 1947 Connecticut 1 4 10 1948 1949 New Mexico, Oregon, Rhode Island,
Washington 4 8 21
1950 1951 1952 1953 Alaska 1 9 23 1954 1955 Michigan, Minnesota, Pennsylvania 3 12 31 1956 1957 Wisconsina, Colorado 2 14 36 1958 1959 California, Ohio 2 16 41 1960 Delaware 1 17 44 1961 Idahob, Illinois, Kansas, Missouri 4 21 54 1962 1963 Vermont b, Indianaa, Iowa b, Nebraska,
Hawaii 5 26 67
1964 Total 26 26 67%
Source: (Bureau of National Affairs 1964). Note: Following V.O. Key (1949), I define the South as the eleven states once comprising the
former Confederacy. States altogether failing to pass fair employment laws before Congressional passage of the 1964 Civil Rights Act include Arizona, Kentucky, Louisiana, Maine, Maryland, Montana, New Hampshire, North Dakota, South Dakota, Utah, and Wyoming. Of course, no southern state (Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, and Virginia) passed a FEP law. States passing non-enforceable FEP laws include Wisconsin in 1945, Indiana in 1961, Nevada in 1961, West Virginia in 1961, and Oklahoma in 1963.
a pre-existing commission given administrative enforcement powers in the form of cease-
and-desist authority b civil or penal enforcement
51
TABLE 2. FEP Passage Rate by Explanatory Variables for Thirty-Seven, Non-Southern States, 1945-1964
Variable % Pass Variable % Pass Republican Control Employer Strength (Right-to-Work Law) Rep. Control 4.2% RTW 2.6 Unified Dem. Control 7.0 Non-RTW 5.8 Income Jewish Residents (%) First Quartile 10.4 First Quartile 7.9 Second Quartile 6.4 Second Quartile 4.0 Third Quartile 1.6 Third Quartile 4.0 Fourth Quartile 0.8 Fourth Quartile 3.2 Industrialization Catholic Residents (%) First Quartile 10.3 First Quartile 7.9 Second Quartile 3.2 Second Quartile 5.6 Third Quartile 4.0 Third Quartile 3.2 Fourth Quartile 1.6 Fourth Quartile 2.3 Urbanization NAACP Membership (%) First Quartile 9.5 First Quartile 6.3 Second Quartile 4.8 Second Quartile 6.5 Third Quartile 4.0 Third Quartile 6.4 Fourth Quartile 0.8 Fourth Quartile 0.0 Electoral competition Union Density First Quartile 7.1 First Quartile 7.9 Second Quartile 6.2 Second Quartile 4.0 Third Quartile 5.0 Third Quartile 4.8 Fourth Quartile 0.8 Fourth Quartile 2.4 Malapportionment (RTV) Public Opinion First Quartile 2.6 Favorable 5.0 Second Quartile 6.3 Unfavorable 4.5 Third Quartile 7.3 Fourth Quartile 2.9 Adjacent States with FEP At Least One 10.7 Black Residents (%) None 1.0 First Quartile 5.6 Second Quartile 3.2 Third Quartile 8.8 Fourth Quartile 1.6
Notes: Figures are to be interpreted as the percentage of state-years in which a FEP law passed. The raw passage rate is 5% (24 events/502 state-years). All monetary variables expressed in 1964 dollars or cents.
52
TABLE 3. Logit Coefficients and Standard Errors from a Discrete-Time, Event-History Analysis of the Passage of State FEP Legislation in Thirty-Seven, Non-Southern States, 1945-1964
Model 1 Model 2 Model 3 Model 4 Republican Control -2.972**
(.871) -2.968**
(.822) -2.383**
(.733) -1.817**
(.667) Employer Strength (RTW Law = 1) -1.581
(1.023) -1.669*
(.796) -2.191**
(.774) -1.502*
(.685) Income (1964 dollars) .002*
(.001) .003**
(.001) .003**
(.001) .002*
(.001) Industrialization (1964 cents) .008
(.011) ------ ------ ------
Urbanization (%) -.007 (.038)
------ ------ ------
Electoral competition .049* (.024)
.045* (.020)
.053** (.019)
.063** (.019)
Public Opinion (Favorable = 1) 1.268 (.813)
1.329* (.691)
----- ------
Malapportionment: RTV Index -.679 (.596)
------ ----- ------
Black (%) -.264* (.127)
-.229* (.099)
-.176* (.084)
------
Jewish (%) .167 (.131)
------ ----- ------
Catholic (%) -.033 (.046)
------ ----- ------
NAACP Membership (%)
.106
.096 ------ ----- ------
Union Density .097* (.044)
.090* (.038)
------ ------
Adjacent States with FEP Law (%) .056** (.014)
.055** (.012)
.048** (.011)
.054** (.012)
Time .079 (.079)
.002 (.062)
.015 (.060)
-.002 (.061)
Constant -12.903** (2.749)
-13.553** (2.425)
-11.135** (2.041)
-10.632** (1.921)
Pseudo-R2 .39 .37 .33 .31 Model χ2 75.88 70.44 64.06 59.13 Degrees of freedom 15 9 7 6
Note: Standard errors are in parentheses. N = 502. * p<.05 ** p<.01 (two-tailed test)
53
TABLE 4. OLS Coefficients and Standard Errors from Fixed-Effects Linear Probability Models of FEP Legislation in Thirty-Seven Northern States, 1945-64
Model 1 Model 2 Model 3 Model 4 Republican Control -.151**
(.050) -.154** (.053)
-.128** (.051)
-.114* (.057)
Income (1964 dollars) ----- .000 (.000)
----- .000 (.000)
Industrialization (1964 cents) ----- -.002 (.002)
----- -.002 (.001)
Urbanization (%) ----- -.003 (.003)
----- -.005 (.003)
Electoral competition ----- .002* (.001)
----- .002* (.001)
Union Density ----- -.005 (.004)
----- -.001 (.005)
Black (%) ----- .046 (.033)
----- .035 (.032)
Jewish (%) ----- -.177** (.057)
----- -.211** (.062)
Catholic (%) ----- .004 (.007)
----- .004 (.008)
NAACP Membership (%) ----- .001 (.004)
----- -.001 (.005)
Adjacent States (%) ----- .004** (.001)
----- .004** (.001)
Time .012** (.002)
.003 (.006)
----- -----
Constant -.078** (.028)
.005 (.194)
-.076* (.034)
.029 (.207)
R2 .16 .27 .20 .30 Degrees of freedom 38 48 58 68
Note: N=502. Time-constant covariates are excluded from all specifications. Model 1 and 2 includes state fixed-effects. Model 2 includes state and time fixed-effects, but it excludes a linear time counter variable. Robust standard errors calculated to adjust for heteroskedasticity. * p<.05 ** p<.01 (two-tailed test)
54
TABLE 5. Predicted Probabilities for Key Independent Variables Model 1 Model 2 Model 3 Model 4 Employer Strength (RTW Law = 1) -.01 -.01 -.02 -.01 Republican Control -.04 -.07 -.06 -.04 Baseline Probability .01 .01 .01 .01
Note: Predicted probabilities based on specifications reported in Table 3. Predicted probabilities are calculated using the SPost Suite (Long and Freese 2001). Baseline probabilities are calculated by setting all of the right-hand side variables to their sample means. Predicted probabilities for control variables are not reported for simplicity of presentation.
55
TABLE 6. Logit Coefficients from Additional Discrete-Time Models of State FEP Legislation, 1945-1964
Model 1 Model 2 Model 3 Model 4 Republican Control -2.454**
(.867) -3.330**
(.900) -1.565*
(.643) 2.272
(1.893) Republican Strength .046*
(.022) .551**
(.177) ------ ------
Republican Strength - Squared ------ -.005** (.002)
------ ------
Employer Strength (RTW Law = 1)
-1.197* (.620)
-1.830* (.730)
-1.273* (.642)
-1.537* (.684)
Income (1964 dollars) .002** (.001)
.002** (.001)
.002* (.001)
.002** (.001)
Adjacent States with FEP Law (%)
.046** (.010)
.054** (.011)
.051** (.011)
.052** (.011)
Electoral competition (Quartiles, 4=Highest)
------ ------ .663** (.250)
1.520** (.491)
Republican Control * Electoral competition
------ ------ ------ -1.312* (.570)
Time -.008 (.059)
.009 (.062)
-.004 (.061)
-.007 (.061)
Constant -9.550** (2.004)
-20.868** (4.858)
-7.961** (1.425)
-11.051** (2.238)
Pseudo-R2 .26 .34 .27 .31 Model χ2 50.10 64.90 52.98 59.25 Degrees of freedom 6 7 6 7
Note: N=502. Standard errors in parentheses. * p<.05 ** p<.01 (two-tailed test)
56
APPENDIX A Descriptive Statistics and Sources for the Variables in the Analysis Variable
Mean SD Time-
Varying Inter-polated
Source
Pass .05 .21 yes no Lockard 1968 Republican Control (Republican Control = 1)
.80 .40 yes no Council of State Gov., various; Congressional Quarterly
Employer Strength (RTW = 1)
.31 .46 no no Lumsden and Peterson (1975)
Income (1964 Dollars) 1987 415 yes yes U.S. Bureau of the Census, various years
Industrialization (1964 Cents) 648 38 yes yes U.S. Bureau of the Census, various years
Urbanization (%) 55 16 yes yes U.S. Bureau of the Census, various years
Electoral competition (%) 68 19 yes yes Council of State Governments, various
Malapportionment (Right-To-Vote Index)
2 1 no no Ansolabehere, Gerber, and Synder (2000)
Public Opinion (Favorable = 1) .55 .50 no no Gallup 1945 Black residents (%) 4 4 yes yes U.S. Bureau of the
Census, various years Jewish residents (%) 1 2 yes no American Jewish
Committee Catholic residents (%) 18 10 yes yes Official Catholic
Directory NAACP Membership (%) 4 3 yes no NAACP, various years Union Density 74 9 yes yes Troy 1957 Adjacent States with FEP (%) 14 20 yes no Lockard 1968 Republican Strength 54 17 yes no Council of State Gov.,
various; Congressional Quarterly
Republican Strength - Squared 3219 1790 yes no Council of State Gov., various; Congressional Quarterly
Electoral Competition (Quartile)
3 1 yes no Council of State Gov., various; Congressional Quarterly
Electoral Competition * Republican Control
2 1 yes no Council of State Gov., various; Congressional Quarterly
Note: N = 502.