Estimating Racially Polarized Voting: A View From the States

• Author: Daron R. Shaw
• Published date: 01 March 1997
• Date: 01 March 1997
• DOI: http://doi.org/10.1177/106591299705000103
• Subject Matter: Articles

Estimating Racially Polarized
Voting: A View From the States
DARON R. SHAW, UNIVERSITY OF TEXAS AT AUSTIN
In the aftermath of the 1990 United States Census, much attention is
being paid to the drawing of electoral districts. Leaders of the black and
Latino communities are among the most interested in these enterprises
because they see districting as a way for white majorities to minimize the
prospects for electing minority representatives. Decidedly less attention is
being paid to the use of ecological regression in evaluating districting ar-
rangements. Ecological regression, long viewed with caution by social sci-
entists, is especially important in ruling on racial polarization in voting
rights cases. The courts’ reliance on ecological regression is a product of
the lack of reliable district-level survey data on individual-level preferences.
This study focuses on the potential pitfalls of using aggregate-level data to
infer intergroup voting differences. The design compares ecological regres-
sion estimates of group voting behavior in each of the states with estimates
gleaned from national tracking polls for the 1992 presidential election.
The data show that the technique detects racial polarization in states, though
the accuracy of ecological regression’s point estimates is variable.
The past three decades have seen much progress toward political equality
for racial minorities in the United States. Poll taxes, literacy tests and other
veiled (and not so veiled) attempts to deny minority citizens the vote have
been done away with. Black voting rates in the south, where discrimination
was most prominent, now approach those of black communities in the north.
In ways, however, the magnitude of this progress is less clear. A case could be
made that minorities’ interests are inadequately represented in our elected
institutions. In particular, there is evidence that districting arrangements di-
lute minority influence in elections today in much the same way that
NOTE: I wish to thank Stephen Ansolabehere, Robert Luskin, Tse-min Lin, Franklin Gilliam,
John Petrocik, Lawrence Bobo, Gary King, Rudolfo De La Garza and several anony-
mous reviewers for their helpful comments.
49

---

statutory impediments used to do. Disputes over districting have thus
become the cutting edge engagements of the civil rights movement.
Since 1962 the courts have adjudicated the legality of districting arrange-
ments. Indeed, the courts’ importance for settling voting rights issues rivals
their importance on school desegregation in the 1950s. Given their role in
handling districting cases, and given the implications of these cases to civil
rights, it is puzzling that the courts accept ecological regression estimates of
group voting behavior. Social scientists have long noted the pitfalls of infer-
ring individual behavior from ecological (or aggregate) data (Robinson 1950;
Goodman 1953, 1959; Cramer 1964; Hanushek, Jackson, and Kain 1974;
Kmenta 1986; Erbring 1990). Yet the courts have consistently used ecological
regression to gauge voting differences between blacks, whites,and Latinos.
One might reasonably ask whether districting arrangements have been de-
cided on the basis of an inappropriate estimation technique.
This article considers the use of ecological data for voting rights cases. In
particular, I explore the legal system’s use of ecological regression techniques
to estimate group voting tendencies. The 1992 presidential election is treated
as a case study for analysis. I use state-level data to show that although the
magnitude of group differences is sometimes mispredicted, the direction of
the differences predicted by ecological regression is almost always correct.
ECOLOGICAL ESTIMATION IN VOTING RIGHTS CASES
A Legal History
In the voting rights cases of the 1960’s, a tension developed between the
anonymity of the ballot box and the need to examine individual-level behav-
ior. Advocates of minority rights believed that black candidates were shut out
because they ran in districts where whites outnumbered blacks. It was impos-
sible to tell, however, for whom individuals voted. White candidates could
have been getting equally large majorities of both black and white votes. What
the plaintiffs in these voting rights cases could see was that votes for minority
candidates increased with the percentage of minorities in the district. They
estimated this relationship with correlation or regression techniques and used
the estimates to infer group voting patterns.
The legitimacy of this application of aggregate data has been consistently
acknowledged by the courts. In the critical case, Baker v. Carr (1962), the
claim was that a 1901 Tennessee statute arbitrarily and capriciously appor-
tioned seats in the General Assembly among the state’s ninety-five counties.
The plaintiffs brought suit under the Civil Rights Act of 1875 in United States
District Court, arguing that the districting arrangements debased their votes
and thus denied them equal protection of the laws guaranteed by the Four-
50

---

teenth Amendment. The Supreme Court ruled that the plaintiffs were entitled
to judicial protection under the Equal Protection Clause. In the process, the
Court ruled that ecological estimates could be used to demonstrate a dis-
criminatory effect in districting.
Cases immediately following Baker built upon both of these rulings. For
example, in Reynolds v. Sims (1964), the Supreme Court invalidated the legis-
lative apportionments of Alabama, Colorado, Delaware, Maryland, New York,
and Virginia, again admitting ecological estimates as evidence for the exist-
ence of a discriminatory effect in districting.
The options available to those seeking to restructure the electoral system
were expanded soon after Reynolds. In 1965, Congress passed the Voting Rights
Act, outlawing discriminatory practices such as the poll tax and literacy tests,
which had been common in southern states. The act also created a federal
bureaucracy charged with reviewing electoral arrangements. Federal reviews
of voting districts have occurred since 1965 and have resulted in an increased
number of court cases.
The Voting Rights Act, of course, does not speak directly to the issue of
proof and is entirely mute on the question of inferring group behavior from
summary statistics. Moreover, the most detailed ruling on how discriminatory
effect could be proven, in the Baker decision, left many questions unanswered.
How
does one establish that racial groups vote differently? Aggregate data are
admissible, but what is the preferred form and when are variations justified?
How much weight should be given to survey data? Who counts as a person
for issues of turnout and vote choice estimation? As legal challenges contin-
ued to be heard, the need for guidelines became acute.
It wasn’t until Thornburg v. Gingles (1986) that the Supreme Court estab-
lished a binding set of rules for redistricting cases. In Thornburg, the Court
stated that plaintiffs need to establish three things in order to prove discrimi-
natory effect:
1. the minority group must be capable of constituting a majority in at
least one district;
2. the minority group must be politically cohesive, and;
3. the majority group must usually defeat the candidate preferred by the
minority
According to Thornburg, ecological regression, exit polls, or analyses of
racially homogenous districts could be used to demonstrate racially polarized
voting. Since Thornburg, and especially since the 1990 Census, cases in which
1
Though Thornburg v. Gingles involved multimember districting and not district lines,
the same basic adjudication framework was adopted for districting cases in Growe v.
Emisten (p.3).
51

---

minorities attempted to compel the redrawing of political lines have prolifer-
ated. Courts have recently heard challenges to city districts in California, Illi-
nois, Pennsylvania and Texas, to county districts in California and Tennessee,
and to congressional districts in North Carolina and Alabama.’
The problem is this: it is not clear that techniques used by courts provide
reliable indicators of racially polarized voting. Ansolabehere and Rivers (1990)
point out that plaintiffs usually rely on ecological estimates of polarization
while defendants usually oppose them, suggesting that ecological estimates tend
to find racial differences. Furthermore, when ecological regression, exit polls,
and results from racially homogenous districts are all available, they can pro-
duce different estimates (Grofman 1991; Lichtman 1991). This might not be
so troubling if all these approaches were always feasible. Then we could at
least compare them. But in many legal proceedings to determine districting
arrangements, ecological regression is all we have. It influences the character
of political representation but we are unsure about the nature of this influence.
CRITICISMS OF ECOLOGICAL REGRESSION
Although the precedent established in Baker and clarified in Thornburg
stands, the reliance on ecological regression estimates of racially polarized
voting has been an issue in all of the cases previously mentioned. The contro-
versy was exhaustively revisited in the recent Los Angeles Voting Rights case
of Garza v. County of Los Angeles (1990).
In Garza, the Mexican American Legal Defense and Educational Fund
(MALDEF), the American Civil Liberties Union (ACLU), and the Department of
Justice brought suit against the County of Los Angeles. They claimed that the
districts electing the County’s five supervisors (as well as a...