\documentstyle[12pt,widetext,titlepage,hss,doublespace]{article}
\renewcommand{\baselinestretch}{1.5}
\begin{document}
\title{The Russian Electorate from 1991 to 1995\thanks{The research
reported in this paper was supported by a grant from the National
Council for Soviet and East European Research, which, among other
things, allowed Dr. Sobyanin to visit Caltech to pursue this
collaborative effort.}}
\author{Mikhail Myagkov, Peter Ordeshook and
Alexander Sobyanin \\ California Institute of Technology}

%\today
%\vglue 1in
%\noindent
\maketitle
%\leavevmode
%\epsfxsize=8in

\begin{abstract}

\begin{singlespace}
The results of Russia's two parliamentary elections prompted much
speculation about the stability of the Russian electorate, its
commitment to reform, and the viability of Russia's democratic
institutions. The concern is that electoral trends signal a
fundamental shift in attitudes away from the euphoria of the 
late 1980's and early 1990's so that, in the likely event that the
Russian economy experiences only a gradual and painful ascent that
leaves much of the population impoverished for the foreseeable future,
any optimism about the future of Russian democracy and reform 
seems wholly inappropriate. However, most commentary on these
elections, journalistic or otherwise, focus on aggregate national
election returns. Here, in contrast, we reexamine trends in voting
using data from all four of Russia's democratic elections beginning
with the 1991 presidential contest through December 1995, that are
aggregated only up to the level of individual rayons. Conducting our
analysis so that we can also trace the impact of fluctuations in
turnout as well as in the percentage of invalid ballots, our analysis
paints a remarkably consistent picture of the electorate. That picture
reveals an electorate that consists of three basic parts: a
hard-and-fast opposition that changes little over four years; a part
that moves from supporting reform (by voting for Yeltsin in 1991) to
opposing it in 1995 after first becoming indifferent to all candidates
and abstaining in 1993; and a reform core that succeeded in both
parliamentary elections in dividing itself up among a plethora of
candidates and parties. It remains to be seen, of course, whether
reform candidates in the future can recapture some of their lost
support or whether the erosion witnessed over the last four years will
continue. Our analysis does not allow us to make such predictions. But
it does permit us to reject the idea that Russian voters are unlike 
voters elsewhere or that they are an unstable lot that moves easily
between reform and anti-reform positions.

\end{singlespace}
\end{abstract}



\clearpage

\section{Introduction}

The results of Russia's December 1993 parliamentary election, judged
by most observers as an upset victory for the ultra-nationalist party of
Vladimir Zhirinovsky, and of the December 1995 contest in which
Communists gained the upper hand in the State Duma, prompted a
good deal of speculation about the stability of the Russian electorate,
its commitment to reform, and the viability of Russia's infant
democratic institutions. Indeed, it seemed inconceivable that four
years or so of declining economic output and life expectancy, rising
crime, and a civil war in the Caucasus would leave people as
hospitable to reform as they appeared to be in 1991 when Boris Yeltsin
first ascended to the presidency. The chief concern is that electoral
trends signal a fundamental shift in attitudes away from the euphoria
of the late 1980's and early 1990's so that, in the likely event that the
Russian economy experiences only a gradual and painful ascent that
leaves much of the population impoverished for the foreseeable future
and that the war in the Caucasus must necessarily entail increased
sacrifice and cost, any optimism about the future of Russian
democracy and reform seems wholly inappropriate.

Of course, we cannot discount the possibility that vote counts are
seriously contaminated as a rule by widespread fraud (Myagkov and
Sobyanin 1995). After all, Russian democracy is a mere five years old,
and Stalinist rules for administering elections still apply in many
regions; how else do we explain turnout rates in some conservative and
remote voting districts that remain above 95 percent? Nor should we
discount the possibility that Zhirinovsky and Zyuganov's successes
derived as much from superior campaign tactics and an ability to
appeal to voter aspirations as from any disenchantment with reform
and democracy. We suspect, in fact, that Gaidar as poster boy of a
party with the unintelligible platform "macro-economic stability"
would hardly attract voters in the West any more than he captured the
imagination of Russia's electorate; nor should we suppose that the
image of Viktor Chernomyrdin promising, in effect, more of the
same, would inspire voters in any electorate. However, it is folly to
assume that those returns contain no information about preferences,
that they are wholly contaminated by fraud, or that they are the mere
consequence of campaign rhetoric. They are, after all, largely
consistent with objective economic circumstances. Here, then, abiding
by the same assumption made implicitly by those who offer their
assessments of the ebb and flow of Russian democracy, we treat
election statistics at face value.

The questions we address here, moreover, are not much different from
the ones that concern earlier analyses. We ask whether a significant part
of the Russian electorate is growing weary of Yeltsin, democracy and
reform and whether there is evidence that the electorate is becoming
increasingly polarized between reform and anti-reform positions. We want
to know what share of those who stayed the course of reform through
December 1993 (by voting either for, say, Gaidar or Yavlinski), changed
their opinions and voted Communist in 1995. We address the issue of the
extent to which reformers damaged their electoral prospects more than
Communists in 1995 by failing to unify under the banner of a single
candidate. And we ask whether Yeltsin's `party of power,' Our Home Is
Russia, succeed in attracting voters from the center or whether it secured
its eleven percent of the vote in 1995 primarily at the expense of those
parties that championed reform in 1993? 

We understand that answers to such questions are, in principle at least,
best learned from public opinion polls and in-depth surveys of
people's attitudes and perceptions. Unfortunately, the usual problems
associated with polling -- sampling error, respondents unwilling to
admit they didn't vote or that they voted for a looser, and an
electorate uncertain of its preferences -- are multiplied in Russia by
the inaccessibility of significant parts of the electorate (those in
remote rural regions), the non-existence of panel surveys extending
across two or more elections, the uncertainties inherent in an unstable
multiparty system in which few voters are familiar with all but a small
handful of parties, and by respondents who often seem more inclined
to answer "don't know" than to give any other response.
Unsurprisingly, then, most analyses of electoral trends as well as
nearly all journalistic interpretations of events are based on official
aggregate election returns and, with respect to the parliamentary
contests, on the share of Duma seats won by one party's list or
another. In this respect, perhaps the most thoughtful assessment of
electoral trends is McFaul's (1996) recent analysis in which he
concludes that although there are no radical shifts from left to right
or vise-versa within the electorate, Russian voters are becoming
increasingly and dangerously polarized between reform and anti-
reform positions.

McFaul's analysis is in fact the starting point for our own, and so we
summarize it here in the form of a set of testable hypotheses, the most
important of which are the following:

\begin{description}
\item
H1:
``despite the presence of 43 parties [in December 1995] ...
Russian party politics is still essentially bipolar" (p. 94).

\item
H2: \
``there has been little change in the balance of support
between [reform and anti-reform] camps" (p. 94).

\item
H3:\
``the 1995 results do not signal a radical shift away from
Zhirinovsky-style `nationalism' and a strong move toward
`communism'" (p. 94).
\item
H4:\
``many centrist voters from 1993 voted for the opposition
in 1995 ... [but] "the core opposition parties from 1993 (Communists,
LDPR, and Agrarians) did not benefit from this centrist migration ...
'new' votes for the opposition parties went to ... radical communists
[Anpilov] and the new nationalist parties like Alexandr Lebed's CRC,
Alexandr Rutskoi's Derzhava, and Power to the People" (p. 95).

\item
H5:\
``the opposition received a big boost from the return of the
three million radical communist voters who had mostly boycotted the
1993 elections ... [and] voted primarily for Anpilov's Working Russia"
(p. 95).

\item
H6:\
The success of Our Home Is Russia ``came at the expense of [Russia's
Choice]" (p. 98).

\item
H7:\
``Party proliferation within the reformist camp ...
dramatically weakened the representation of reformists within the
Duma" (p. 98).
\end{description}

Like previous assessments, McFaul's analysis and these seven
hypotheses are based primarily on an informed assessment of national
aggregate returns and, with the exception of hypothesis H5, on the
breakdown of support among those who voted. In a sense, then, these
seven hypotheses are based on but two observations -- the national
aggregate results of the 1993 and 1995 parliamentary elections.
Unfortunately, elections returns at this level of aggregation cannot
provide sufficient data for definitive conclusions: there are too many
variables (possible shifts in voter loyalties) and too few observations
(percentages of the vote received by the different parties). Put simply,
there is an inherent mathematical indeterminacy in the results. This is
not to say that McFaul's scenario is incorrect, but only that it is not
the only possibility consistent with aggregate national returns. In this
essay, then, we reconsider these hypotheses using data aggregated only
to the level of individual rayons drawn from all Russian elections
since 1991. Since we are thereby able to operate with as many as
approximately 2,000 observations per election, we can reduce
considerably the number of logically admissible scenarios.


An additional problem with most interpretations of official returns is
that conclusions typically rely on a classification of parties into
reform, anti-reform, and centrist blocks. Unfortunately, parties and
candidates do not come with ready-made reformist and anti-reformist
labels, and the details of any such classification can be manipulated to
admit a variety of alternative conclusions. For example, how do we
classify, say, Women of Russia in 1993 and 1995? If we look at the
legislative voting records of Duma deputies elected under their list and
classify the party as centrist, we still do not know if this is the view
voters held of them, nor will national aggregate returns tell us who
secured the votes they lost in 1995. Should we classify Shakrai's Party
of Unity and Accord as reformist or centrist? Although Shakrai
himself is associated with reform, several deputies elected under his
party's label in 1993 ran unsuccessfully on Rybkin's stillborn party list
in 1995, which was originally intended to appeal to right-of-center
voters. Although we might agree that Our Home Is Russia belongs in
the same category as Russia's Choice since each, in their time, drew
largely upon Yeltsin's presidential resources, can we safely assume that
each secured their votes from the same pool of voters or did Our
Home Is Russia win a significant share of its votes from parties we
might classify as centrist in 1993? And what do we do with the
LDPR? Is it anti-reform or should we, on the basis of Zhirinovsky's
legislative record of compromise, place it in its own category --
pro-reform but anti-government? Of course, classification is required
whenever we try to say whether the electorate is becoming more or
less conservative. But here we focus on the source of support of the
various parties, candidates and referenda options that appeared on
ballots so that any classifications we intended only to help interpret
our more formal statistical analysis.


Finally, answering questions about electoral stability and preference
is complicated also by the fact that turnout and the percentage of
invalid ballots varies greatly from one election to the next. Officially,
turnout declined steadily from 1991 to December 1993 (75.4 percent
in March 1991, 74.7 percent in June 1991, 64.3 percent in April 1993,
and 54.8 percent in December 1993) but rose sharply again in
December 1995 to 63.2 percent. Similarly, the percentage of the
electorate casting invalid or blank ballots when voting for party lists
rose to 7.62 percent in December 1993, but declined to an historically
more normal level (less than two percent) in 1995. Who cast invalid
ballots in 1993, who lost votes as turnout declined, and who gained
them in 1995 when turnout reversed its trend? More importantly, is
there evidence that parts of the electorate are becoming increasingly
disenchanted with reform and Yeltsin but, instead of shifting directly
from pro-reform to anti-reform positions, the disenchanted first
become indifferent to all parties and positions and register their
dissatisfaction by staying home? Thus, rather than ignore variations
in turnout and the number of invalid ballots, here we examine closely
who failed to vote or voted improperly in 1993 and who returned to
the polls in 1995.


Thus, using more disaggregated than was available to earlier analysts,
imposing no specific classification of parties, and accommodating
changes in turnout, this essay offers a reconsideration of McFaul's
seven hypotheses and the general scenario he offers as a description of
the Russian electorate. Briefly, we sustain the first hypothesis
concerning the stability of that electorate and conclude that if there is
any instability it lies in two places: in the vote Zhirinovsky received
in 1993, a significant share of which supported reformist positions in
1991 and April 1993, and which (in accordance with hypothesis H4)
moved to support communist parties in 1995; and in the changing
patterns of non-voting and invalid ballots. Although the data supports
Hypothesis H5 that Anpilov's party gained from increased turnout in
1995, it also reveals that the Communist Party was the biggest net
beneficiary. And although we agree in part with hypothesis H6 that
many of the lost votes of Russia's Choice went to Our Home Is Russia
and, in accordance with H7, to other unsuccessful reformist parties,
Our Home Is Russia gained as well from increased turnout, although
not as much as the Communists. Trends in turnout also cause us to
question hypothesis H2, in that between 1991 and December 1993,
much of the votes lost to nonvoting came from reformist positions.
The most curious and, perhaps, suspicious finding (at least with
respect to allegations of fraud in December 1993) is that, rather than
being spread uniformly across party lists, virtually all voters who cast
invalid ballots in 1993 voted in 1995 for the Communist Party, the
LDPR, and Our Home Is Russia.


In reaching these conclusions and in offering other refinements of
hypotheses H1 through H7, this paper is organized as follows: Section
2 gives a brief history of the elections that concern us, describes our
data, and discusses our methodology. Section 3 discusses our primary
results about the movement of voters for three pairs of elections: June
1991 (the first presidential contest) versus the April 1993 referendum,
April 1993 versus the December 1993 parliamentary election, and
December 1993 versus December 1995. To gain a better overview of the
electorate across Yeltsin's tenure in office, Section 4 considers two
non-adjacent pairs of elections -- the flow of votes between 1991 and
December 1993 and between 1991 and December 1995 -- and concludes
with an examination of potential differences between urban and rural
voters.

\section{History, Data, and Methods}

Before describing our data it is useful to first summarize the basic
changes that occurred across the four elections that concern us -- the
June 1991 presidential election, the April 1993 referendum, the
December 1993 parliamentary election and constitutional referendum,
and the December 1995 parliamentary election. Briefly, the June 1991
election was the first popular balloting for president in Russian
history, at which time the unique personality associated with reform
and opposition to the existing regime was Boris Yeltsin, then Speaker
of the Russian Congress of People's Deputies. His opposite number
was Nikolai Ryzhkov, the former USSR prime minister, who, as a
nominee of the Communist Party, was widely viewed as a
representative of the status quo. Zhirinovsky, in contrast, although
positioning himself as an anti-western ultra-nationalist, was not
presumed to be opposed to reform or portrayed as a stand in for any
status quo. The remaining candidates were General Albert Makashov,
USSR Minister of the Interior and future KGB director Vladimir
Bakatin, and the communist governor of Kemerovo oblast, Aman
Tuleev. Despite Bakatin's subsequent attempt to liberalize the KGB,
all three of these candidates can be put into the anti-reformist camp,
or at least the camp opposed to Yeltsin's then widely publicized
opposition to the existing regime. Table 1 shows the results of this
election and, with 45.6 million voters supporting Yeltsin, establishes
the high water mark of his support, and, presumably, of enthusiasm
for democratic market reform.


The April 1993 referendum was called jointly by the President and the
Congress of People's Deputies as part of their ongoing power struggle.
Voters were asked four questions, three of which concerned
confidence in Yeltsin and his economic policy, and the fourth asking
about the advisability of holding early elections for a new Congress.
Contrasting the outcome of the referendum (see Table 1), note first
the sharp drop in turnout, from 74.7 percent to 64.2 percent. Second,
although a clear majority answers `yes' to the first question on the
ballot -- the question most clearly directed at gauging confidence in
Yeltsin -- the absolute number of Yeltsin supporters declines from
45.6 million in 1991 to 40.4 million in 1993. Thus we can ask: did
Yeltsin's `missing' 5.2 million voters defect to other, explicitly anti-
reformist positions, or did they merely stay home and contribute to a
growing pool of non-voters?


The third measure of the electorate's mood occurred in December
1993. This election included a referendum on a new constitution as
well as the choice of deputies to Russia's upper legislative chamber,
the Federation Council. But because not only Yeltsin but Zhirinovsky
urged voters to approve the constitution and because most candidates
for the Council ran without party labels, perhaps the most
interpretable assessment of voter preferences occurred with respect to
the party-list voting to fill seats in the new State Duma. Table 1
summarizes the election outcome with respect to the thirteen parties
on the ballot as well as the number of votes cast against all parties, the
number of invalid ballots, turnout, and the `for' and `against' vote on
the constitution. Notice first that turnout continues its decline, from
64.3 percent in April to 54.8 percent. Second, even if we classify only
the Communist, Agrarian, and Zhirinovsky parties as explicitly
`anti-Yeltsin,' Yeltsin's support, like turnout, declines as well -- from
40.4 million voters in April to 34.1 million in December. However,
confounding any simple assessment of who gained what from where
is the fact that the number of voters who voted against Yeltsin in
April and in support of the old Congress (the estimated 21.2 million
who voted No on the first question and No on the fourth) far exceeds
the number who supported the Communist or Agrarian parties (11.0
million) in December. Finally, we should also take note of the sharp
increase in the number of invalid ballots -- from 1.5 million to 4.4
million. Since turnout is declining as well, the question naturally arises
about the political sympathies of these ostensibly incompetent or
confused voters.

Russia's most recent parliamentary election, in December 1995,
offered voters a choice of forty three parties, and, with a significant
cluster of ostensibly pro and anti-reform parties garnering votes in the
3 to 5 percent range, only four parties surpassed the five percent
threshold and won seats on the party-list half of the contest (see Table
1).\footnote{Table 1 as well as our subsequent analyses of the 1995
election groups party lists as follows: the row denoted `other
democrats' consists of B. Federov, Pamfivoliva, and Borovoy's lists, as
well as `89', Christian Democrats, and FDD; the row denoted Rybkin
and Shakrai (Unity and Accord) includes as well Stable Russia, Block
of Independents, and Transition of the Homeland; and the row
denoted Derzhava also includes For the Homeland and Power to the
People.} The most notable `success' was scored by the Communist
party, which more than doubled the vote it and its fellow traveller, the
Agrarians, received in 1993. In contrast, Russia's Choice virtually
disappeared from view and Zhirinovsky lost half of his support. But
aside from these well publicized outcomes, we can find other patterns
that warrant attention. First, turnout increases substantially, from 54.8
percent to 64.4 percent, while the share of invalid ballots decreases to
`more normal' levels -- to 1.3 million ballots from 4.4 million in 1993.
Second, although a new `reform' or pro-government party appeared on
the scene, Our Home Is Russia, and was credited with taking votes
away from Russia's Choice, its vote share cannot account fully for the
losses incurred by that party and by those others we might classify as
centrist in 1993, such as Women of Russia, Travkin's Democratic
Russia, and Shakrai's Party of Unity and Accord.

Finding meaningful trends in this electoral sequence is not difficult,
but as we note earlier, the trends we find will depend on which parties
we label pro-reform, anti-reform, and centrist. Thus, to better assess
the mood of the electorate, we look instead at each party or
candidate's source of support. Saying, for instance, that Russians are
more conservative today than two or four years ago or that significant
numbers are tiring of reform requires finding a measurable part of the
electorate moving across the ideological spectrum from left to right.
Similarly, the argument that Zhirinovsky's vote in 1993 was an
anti-reform protest vote that Yeltsin is unlikely to capture in the June
1996 presidential election requires explicit confirmation of the
hypothesis that a predominant share of the votes he lost in 1995 went
to, say, the Communists. However, mere `declassification' of parties
cannot resolve matters; as we also indicate earlier, we must
disaggregate our data. Here, then, we consider data aggregated only to
the level of individual rayons.

Briefly, the Russian Federation consists of 89 semi-autonomous
regions (oblasts, krays, republics, autonomous regions, and the federal
cities of Moscow and St. Petersburg), and each region, in turn, consists
of a number of ``rayons". The average region has a population of about
two million (exceptions are the city of Moscow and some of the
sparsely populated northern oblasts), whereas the average rayon has a
population of less than a hundred thousand. Typically, each region's
capital is a large city with a population that ranges between several
hundred thousands to several million and may itself consist of several
rayons. The administrative capital of each rayon (excepting those that
are part of a regional capital) is generally a small town with a
population ranging from several thousand to several hundred
thousand. There are, then, several alternative levels of aggregation of
the data -- rayon, region, Duma district (after December 1993) or the
whole country -- but until December 1995, the only official results
were aggregated at the national level, whereas the Central Election
Commission, following the December 1995 vote, published data at the
rayon level. However, through a variety of sources (generally
representatives of the Russia's Choice, who in most cases worked
directly with local administration officials), we have obtained
`unofficial' returns aggregated at the rayon level for the first three
elections in our study. Although not covering all of Russia, our data
appears nevertheless to be representative of the whole in that no
category of return varies from official numbers by more than two
percent. Hence, it is this data plus the official rayon-level
returns for December 1995 that is the basis of this study.

The data set for the June 1991 presidential elections is, with
the exception of 1995, the most comprehensive one in our study. It
consists of 2551 rayon-level observations, covers 87 of 89 regions, and
accounts for 105.7 million of the 107 million people who formed the
eligible electorate. The two missing regions are the autonomous
republics located on the far north of Russia. The data set includes
voting totals for all of the six candidates as well as turnout rates, and
the number of votes against all of the candidates and the number of
invalid and unused ballots. Data for the April 1993 referendum take
the form of 2125 observations covering 68 of 89 regions and about
91.1 million eligible voters (about 90 percent of the total). This data
set includes vote counts for all but one oblast and none of the
republics. The December 1993 election is the most difficult to
document, since the Central Election Commission to this day refuses
to publish official rayon-level returns. Inexplicably, the bottom line,
as asserted by the Commission's chair, is that such data does not exist.
Our data set here, then, comes from a variety of different sources and
includes 1298 rayon-level observations for voting on the constitution
and 1167 rayon-level observations for voting on party-lists. The
constitutional data set covers 41 regions and about 52.9 million eligible
voters, whereas the party-list data set includes 36 regions and about
48.2 million eligible voters. Most of the missing regions are the
autonomous republics, the southern oblasts and the cities of Moscow
and St. Petersburg. Despite these gaps, this data set appears to be fairly
representative of the country as a whole -- the share of votes given to
the constitution and to the several major parties does not differ from
the officially reported national average by more than two percent.
Finally, the December 1995 election returns have been published by
the Central Election Commission, and include all rayons and
regions.\footnote{Notice that the data from one election to the next
are not strictly comparable, and that even within the December 1993
election, we have data covering a slightly different set of rayons when
comparing the Duma party list elections and the constitutional
referendum. Thus, when analyzing pairs of elections, each data set
must be readjusted so that it concerns only those rayons for which we
have data on both elections.}


Turning now to our method of analysis, suppose by way of example
that two parties, $A$ and $B$, compete against each other in two
consecutive elections, $e$ = 1 and 2. Let the percentage of the eligible
electorate voting for the two parties in each election be denoted by
$x^{A}_{e}$ and $x^{B}_{e}$, and let $x^{0}_{e}$ denote the share
of the electorate that failed to vote. Suppose now that party $A$ lost
votes, that $B$ gained votes, and that the number of nonvoters
decreased so that $B$'s gain actually exceeds $A$'s loss. Since there
are only two parties, it is not too difficult to interpret these numbers,
but even still such aggregate numbers allow for an infinity of
possibilities when it comes to describing the flows of voters and
nonvoters. For example, $B$'s increase might have come directly from
$A$'s voters in the previous election, as well as from the ranks of
nonvoters; alternatively, $B$'s new support might have come
exclusively from old nonvoters, with a larger share of $A$'s lost vote
going into the ranks of nonvoters. Although both possibilities yield the
same outcome, they present different interpretations of the outcome.
If, in the first instance, the espoused ideologies of $A$ and $B$ differ
radically, we would infer a somewhat unstable electorate in which
voters switch from one extreme to the other, whereas in the second
case we see fewer voters jumping across the ideological spectrum and
more of them, when switching loyalties, first becoming indifferent
between the parties.


To see, then, how we might disentangle these possibilities using
aggregate data, let us focus first on $A$'s supporters in the second
election and, to render the parameters we must estimate identifiable,
let us assume that any increased turnout comes exclusively from the
ranks of previous nonvoters. What we want to learn, now, is the
percentage of voters who supported $A$ in both elections, the
percentage if any who voted for party $B$ in the first election but
switched to $A$ in the second, and, assuming that turnout increased
from one election to the next, the percentage who abstained from
voting in the first election but voted for $A$ in the second. Formally,
then, we are interested in estimating the coefficients $a_{A}^{A}$,
$a_{B}^{A}$, and $a_{0}^{A}$ of the following equation:
\[
x^{A}_{2}=a_{A}^{A} x^{A}_{1}+a_{B}^{A}
x^{B}_{1}+a_{0}^{A}(x^{0}_{1}-x^{0}_{2}).
\]
Notice that this expression subtracts the share of the electorate that
abstains in both elections. Since our example assumes that turnout
increased, and since to render our model identifiable we must assume
that anyone who voted in the first election voted as well in the second,
we treat new voters as a separate party -- as a part of the electorate
that `voted' not to vote in the first election.

Unfortunately, no econometric method guarantees unbiased estimates
of the variables in this model (King 1996). The primary difficulty is
aggregation error, which arises whenever people in one observation act
differently than people in some other. Nevertheless, we can estimate
the elasticities of the coefficients if we accept the same assumptions
that are implicit in any attempt, journalistic or otherwise, to infer
substantive meaning from aggregate election returns. Specifically, we
must assume that the value of coefficients do not vary across
observations (or equivalently, across any subset of observations) and 
that our independent variables do not correlate significantly.

\noindent
Now notice that since 
\[x^{A}_{1}+x^{B}_{1}+x^{0}_{1} = 100,\]
or, equivalently, since
\[x^{0}_{1} = 100 - x^{A}_{1}-x^{B}_{1},\]
we can rewrite that expression as,
\[ x^{A}_{2}=K + a_{A}^{A} x^{A}_{1}+a_{B}^{A} x^{B}_{1}.\]
Thus, the elasticities of $x^{A}_{1}, x^{B}_{1}$,
and $x^{0}_{1}$, respectively, are, $ a_{A}^{A} + K, a_{B}^{A} +
K$, and $K$.\footnote{Notice that the usual form of a model intended to
be estimated by standard econometric techniques would include an error
term, usually additive and assumed to be independently and normally
distributed. Here, however, since our variables must sum to a constant, the
only source of stochastic error, contrary to our assumption, is variation in
the values of our coefficients across the data. A priori, then, we know that
any error term will not satisfy the usual assumptions employed to ensure
unbiased estimates of coefficients. Unfortunately, there is no way to know
the direction of this bias, nor do we know of any procedures for
eliminating it. However, dividing our sample into various subpopulations
(see this essay's last section) and rerunning our regressions will provide a
partial check on the severity of the problem as well as a check for
aggregation error.}

Suppose now that we have estimates of our coefficients and we ask:
what inferences can we make about individual decisions. Naturally, if
the model's assumptions are not satisfied -- if aggregation error biases
or otherwise renders our estimates meaningless -- the answer is
`nothing'. But notice that there are several checks on the consistency
of those estimates. First, the estimated coefficients should not sum to a
number significantly different from 1.0, since presumably supporters of
$A$, $B$, and nonvoters exhaust the potential electorate (excluding `hard
core' nonvoters). Second, no individual coefficient should be significantly
less than zero or greater than 1. Finally, we need to check the assumption
that coefficients do not vary from one region to another. The particular
danger here is aggregation error -- estimates, for instance, that are, say,
significant within various subpopulations but insignificant when the entire
data set is considered. The usual method here is to divide our sample into
various subpopulations -- subpopulations that are most likely to occasion
different values for the coefficients. In the case of Russia, the most likely
candidate is the divide between urban and rural populations, since there
is considerable evidence that urban voters, owing to differences in
information, education, and economic opportunities, respond differently
to the parties than do rural voters (see Hough et al 1996, Clem and
Craumer 1993, 1995, and Slider et al 1994).\footnote{This is not to say
that we assume that urban voters have the same preferences as rural ones.
That assumption is patently false. Rather, aggregation error can pose a
problem if, for instance, urban voters, when then grow disenchanted with
reform and Yeltsin, defect to different parties than do rural voters.}
Hence, in the last section of this essay we divide our data into `urban' and
`non-urban' subpopulations and rerun several of our regressions to see if
at least our qualitative conclusions are affected.

To see now how we can interpret our estimates, suppose for purpose of an
example that we estimate $a_{A}^{A} + K$ at 0.7. In this case we would
infer that if party $A$'s support increases in the first election by one
percent, then {A}'s support in the second election should increase by .7
percent. An equivalent interpretation is that 70% of those who voted for
$A$ in the first election voted for $A$ in the second. The caution that
needs to be applied here about either interpretation, though, is that if
$A$'s vote in the first election averaged, say, 40 percent, it is not
necessarily the case that supporters in a district that gave him a mere 10
percent acted the same as supporters in one that gave him precisely 40
percent or the same as voters in a district that gave him 90 percent. Our
model is linear and, especially when treating variables constrained to
values between 0 and 100 percent, it may not be true that extreme cases
are like average ones.

Notice now that our regressions delete one party, candidate, or
referendum response and allow its coefficient to be estimated by the
constant term. We do this since, if all possible choices (including not
voting and casting an invalid ballot) are included, our variables sum
to 100 percent, our independent variables are linearly dependent and
our estimated coefficients meaningless. But in deleting a variable, we
must delete a `significant' one whose support does in fact vary across
observations, since otherwise, our remaining variables will be
approximately linearly dependent, and our estimates correspondingly
unreliable. Some care, though, must be taken when choosing which
variable to delete. If we delete one that bears a strong negative
correlation to one that remains in the sample, then the mathematics of
regression analysis tell us that some of our estimates will be strongly
biased downwards. We should note here, however, that these problems
do not generally arise here unless there are a `great many' independent
variables, which is to say that they tend to arise only when we
examine the relationship between party list voting in the December
1993 and 1995 parliamentary elections.


\section{Results}

Since we want to learn how voters who supported, say, reform, act
subsequently, our dependent variables pertains to choices made in the
second (most recent) election in the pair, and independent variables
concern choices made in the first election. However, since the
methodology is limited by the level of aggregation of our data and
allows only estimates of the flow of votes, we must focus on the
"major" issues, such as support and opposition to reform, and primary
candidates or parties. Thus, we do not attempt to estimate the sources
of support of a party that gathers a mere one or two percent of the
vote; in these instances we group minor parties together according to
a priori judgements about their ideological persuasions.

Turning first to the analysis of the flow of votes between June 1991
and April 1993, we first divide the set of 1991 presidential candidates
into four groups: Yeltsin, Ryzhkov, Zhirinovsky, and `others'. The
first thing we want to measure is the extent to which Yeltsin's
supporters in 1991 supported him in April 1993 by answering Yes to
the referendum's first question. In addition, we also want to see who
was hurt most be the decline in turnout between these two elections.
Thus, an additional dependent variable is the percentage of the
electorate choosing not to vote in each rayon in April 1993 minus the
percentage who did not vote in 1991.


The data in Table 2 show the results of our regressions, and presents
our results in two forms.\footnote{Although the issue does not arise
here, to simplify subsequent tables, coefficients that are not
significantly different from zero are reported as 0. Appendix A offers
all estimates and their associated statistics} Looking first at the column
under `Yeltsin', the first number tells us that of those who voted for
Yeltsin in 1991, 73 percent voted Yes on the first referendum
question, 8 percent voted No, and 18 percent failed to vote. The
second number in parentheses tells us that of those who voted Yes in
our data, 42.29 percent -- or fully four fifths -- came from people
who voted for Yeltsin in 1991.In contrast, 87 percent of Ryzhkov's
supporters in 1991 voted No and accounted for nearly one half the No
vote in April. Zhirinovsky's voters and those who voted for other
candidates, on the other hand, tended to split nearly evenly between
voting Yes and No, with an edge given to an anti-Yeltsin vote of No.
This is our first indication (but not our last) that we cannot easily
characterize Zhirinovsky's base of support vis-a-vis their attitudes
toward reform.\footnote{Since turnout is greater in 1991 than in April
1993 and since we assume, in effect, that all those who voted in 1993
voted in 1991, there is no `nonvoter' column in Table 2 -- those who
failed to vote in 1991 are assumed to have not voted in 1993. Also,
turnout in a few rayons (less than 50 in any pair of elections) move in
a direction opposite that of the national average and, thus, in a
direction inconsistent with our assumption. These rayons are deleted
from the analysis.}


Of the things Table 2 reveals, two warrant emphasis. First, note the
stability of that part of the electorate polarized between pro and anti-
Yeltsin positions. Our estimates suggest that few members of the
electorate -- less than six percent of those who voted in 1991 --
switched from a pro-Yeltsin vote in 1991 to a No vote in 1993 or from
a pro-regime (Ryzhkov) vote in 1991 to a Yes vote in 1993. In this
respect at least, the bipolarity McFaul infers with respect to the
electorate between December 1993 and 1995 (see hypothesis H1) is not
peculiar to those parliamentary contests; a core of polarized voters
appear to exist as early as 1991. If any fundamental shift occurred, it
occurred within the pro-Yeltsin camp as nearly one sixth of Yeltsin's
earlier support disappeared into the ranks of nonvoters. If people were
less enthusiastic about reform in April 1993, their moderated
enthusiasm did not lead them to switch sides but led instead to
growing indifference between both polar positions. Conversely, few
of Ryzhkov or Zhirinovsky's voters in 1991 stayed home in April
1993. As a consequence, Yeltsin's support in 1991 of nearly 58 percent
fades in April 1993 to a bare majority.

Turning next to the Russian electorate between April and December
1993 -- to the elections that bracket Yeltsin's coup against the old
Congress -- we now let responses to the April referendum's first
question correspond to our independent variables and for dependent
variables we consider both the constitutional referendum and voting
on the party list elections to the State Duma.\footnote{Since invalid
ballots in April 1993 are such a small share of the electorate, we delete
this column from our presentation.} Because turnout continued its
decline we identify ``nonvoters in December who voted in April" as a
separate dependent variable, and since December also witnessed a
sharp increase in the numbers of invalid ballots, we also create the
category ``voters who cast valid ballots in April but invalid ones in
December'' as an additional dependent variable to be explored.


Table 3 summarizes the results of our regressions and reveals several
patterns that correspond to what we already know about the December
election. Perhaps most importantly, though, although few (4 percent)
of Yeltsin's April supporters appear to have opposed his constitution,
the share of those who voted in April and supporting reform in
December by voting for the constitution or for parties we might
reasonably classify as reformist or centrist continues to decline as
additional Yes voters from April enter the ranks of nonvoters. By our
estimate, fully one quarter of those who voted Yes in April failed to
appear at the polls in December. In contrast, although nearly one fifth
of those voting No in April voted for the constitution, little (2
percent) of that opposition to Yeltsin's chose not to vote in December.
Thus, the progressive `bleeding' of reformers into the ranks of
abstainers, which began in April, continues and perhaps even
accelerates in December. Aside from this erosion and despite the
events of September and October, the bipolar stability that we observe
in Table 3 and that corresponds to McFaul's first hypothesis is
maintained: Of those supporting the constitution, 85 percent supported
Yeltsin in April whereas among those voting against the constitution,
fully 92 percent opposed Yeltsin in April. Overall, then, although it
appears that few voters crossed over in their preferences, Yeltsin's
support declines from a bare majority in April to something just above
41 percent in December.


Looking now at voting for the party lists, the division of April Yes
and No voters offers few surprises: votes for Russia's Choice and
Yabloko come exclusively from those who voted Yes in April whereas
votes for Communist and Agrarian parties come exclusively from
those who voted No. Women of Russia and Travkin's DPR, true
perhaps to the `centrist' label often given them, derive their support
from both Yes and No voters, although since there are more Yes than
No voters, their support comes predominantly from Yes voters.
Shakrai's list and those of the parties that failed to surpass the 5
percent threshold secure votes primarily from Yes voters. Finally, just
as Zhirinovsky's 1991 voters divided approximately 1:2 against Yeltsin
in April, support for the LDPR in December comes from both pro and
anti-Yeltsin camps in approximately the same ratio. Once again, then,
we cannot put the LDPR into the same opposition and anti-reform
category as we might place Communists and Agrarians.

Perhaps the most interesting question about December concerns
Zhirinovsky's vote. The numbers in Table 3 suggest the following: if
we assume that Zhirinovsky succeeded in keeping his 1991 voters,
then approximately one quarter (3.12 percent of those who voted in
April) of that additional support came from people who supported
Yeltsin in April and three quarters (8.98 percent) came from No
voters. This estimate, though, leaves unanswered the subsidiary
question as to why Yes voters who defected from Yeltsin defected
only to Zhirinovsky (as opposed to Communists or Agrarians).
Myagkov and Sobyanin's (1995) answer is that this apparent defection
(as well as an equal number from the No category) represents election
fraud -- ballots added to the total in order to increase turnout and
render the constitutional referendum legitimate. Indeed, if we assume
that 3.12 percent of those who voted Yes in April but for Zhirinovsky
in December as well as 3.12 percent who voted No in April but for
Zhirinovsky as well are falsified ballots, we come to an estimate of
fraud's magnitude (6.24 percent times an April turnout of 64.2 million
= 4 million ballots falsified for Zhirinovsky) that is in the ball park of
Sobyanin's (1995) initial estimate of 6.3 million.
Admittedly, however, such speculations take us beyond the confines
of our analysis, and the overall picture painted by Table 3 is of an
electorate that continues to melt away from pro-reform positions into
the ranks of nonvoters and the absence of crossovers between reform
and anti-reform (communist) camps. Although McFaul formulates his
seventh hypothesis about the impact of party proliferation for the
December 1995 election, it applies here as well if we classify the five
small parties that failed to qualify for seats as pro-reform, since they
secured virtually all of their votes from those who voted Yes in April.
Moreover, although Travkin's DPR and Women of Russia won votes
from both Yes and No April voters, the plurality of Yes over No
means that both parties secured a majority of their vote from Yes
voters. The implication, then, is that if their Yes voters are like the
rest, the presence of Women of Russia and DPR on the ballot also
reduced the relative support recorded by Russia's Choice and Yabloko.
On the other hand, had Zhirinovsky not been in the race, and if his
Yes and No voters were polarized between pro and anti-reform
positions like all the rest, elimination of the LDPR from the contest
would have contributed more to Communist and Agrarian party totals
than to any other.


When we turn to the December 1995 parliamentary election we
confront some new methodological problems that derive not only from
the great number of parties that competed then (43), but also from the
large number (13) that competed in December 1993. The specific
problem is that our estimates become increasing unreliable to the
extent that our independent variables (the support given to each party)
are linearly dependent. In our case the set of all such variables, taken
together, are dependent in that they must sum to 100 percent.
However, recall that we delete one variable from each regression and
allow its coefficient to be determined by the constant term. If this
variable accounts for a `significant' part of the eligible electorate and
if it varies `sufficiently' in the data, then multicolinearity is
manageable. But if there are a many independent variables and if no
one of them accounts for a significant share of the vote or has support
that fails to vary significantly across observations, multicolinearity
re-emerges. Our approach here, then, is to first group the parties that
competed in 1993 into four categories: the Communist Party plus
Agrarians, Zhirinovsky's LDPR, centrist and reform (Travkin's DPR,
Women of Russia, Russia's Choice, Yabloko, Shakrai's Unity and
Accord, etc.), invalid ballots, and new voters. After ascertaining which
groups contributed support to each significant party in 1995, we
subdivide those groups and estimate each party's contribution within
the group, but only after we re-cluster the groups that are not
indicated as providing significant support by the first regression.

Table 4, then, gives the results of our first pass at the data (since these
estimates are preliminary, we do not offer any estimate in parentheses
of the division of the full electorate across party lists). Notice the
considerable increase in the number of coefficients estimated to be
negative and significant -- a sure sign that our positive estimates are
biased upwards. Nevertheless, we do see in this table approximately
the same pattern we saw earlier with respect to the bipolarity of the
electorate. For example, of those that voted for `Democrats, center
parties, and against all' in 1993, none are indicated as voting for the
LDPR, Communist or Agrarian parties, Anpilov or Rutskoi. Instead,
their votes are spread out nearly uniformly across the remaining
parties, including various clusters of parties that individually received
only a small fraction of the vote. Conversely, those who voted for the
Communist or Agrarian parties in 1993 voted much the same way in
1995. But notice here that this communist vote, although dissipated
somewhat across five parties, concentrates itself in the Communist
Party. Thus, although the communist vote is dissipated somewhat by
the presence of Anpilov and Rutskoi on the ballot, it is far less
uniform than the `democratic and centrist' vote. McFaul's seventh
hypothesis about the damage done to democrats by the proliferation
of parties, then, is largely borne out by this preliminary analysis.
Insofar as Zhirinovsky's lost support is concerned, Table 4 suggests
that much of this support went to the Communist Party, with smaller
shares going to Rutskoi, Anpilov, Lebed and Govorukin. The primary
beneficiaries of the increased turnout are Yabloko, Derzhava, and the
Communists. Thus, Table 4 at least, fails to support for McFaul's fifth
hypothesis about the basis of support for Anpilov's Working Russia.


Finally, consider the column `invalid ballots', which attempts to
estimate the beneficiaries of the relative decline in the number of
invalid ballots. Owing to the considerable number of statistically
significant negative estimates, our positive estimates cannot be
regarded as reliable; nevertheless, the suggestion here is that the
primary beneficiaries of this decline were `other democrats', Women
of Russia, the LDPR, the Communist Party, Agrarians, and Anpilov.
This is indeed a curious mix, but before we attempt any interpretation
of things, let us turn to the refinement of our analysis in which we try
to dissect further the basis of each party's support. Table 5 presents
our results, except that here, rather than present a multitude of
columns with statistically insignificant estimates, the last column
reports only those estimates that are statistically significant for the
parties we label `Democrats and center' in Table 4 (as before, the
numbers in parenthesis denote the overall percentage of the vote given
to parties from the 1993 electorate).


Looking first at the first nine rows of this table, notice that Russia's
Choice receives only about one quarter of its 1993 vote; in conformity
with McFaul's sixth and seventh hypotheses, the remainder is spread
across a broad mix of parties, including not only Our Home Is Russia,
but also the party lists of S. Federov, Popov, Lebed, and Govorukin,
and our aggregation of fifteen insignificant parties (row 9). Yavlinski's
support, in contrast, is remarkably stable: Yabloko retains all of its
1993 vote, and picks up additional small shares from Gaidar (.72
percent), Shakrai's Unity and Accord (.86 percent), and from new
voters (1.68 percent). The fifteen small parties in row 9 get only .56
of the vote from Yabloko, whereas the party lists associated with
Lebed, Govorukin, S. Federov, and Popov get none. The picture here,
then, is of one party, Russia's Choice, that hemorrhaged in nearly
every direction, and of another, Yabloko, that maintained its base of
support but failed to attract significant votes from anywhere else.

Turning now to the opposite end of the ideological spectrum, to rows
10 through 14 of Table 5, notice first that no coefficient in the last
column under `democratic + centrist' is significant or positive; the
bipolarity of the electorate is sustained. But if there is a parallel to
Russia's Choice in the `opposition' camp, it is the LDPR, whose 1993
vote dissipates across the LDPR, Rutskoi and company, the
Communist Party, and Anpilov. Indeed, contrary to hypotheses H3 and
H5, nearly half of Anpilov's vote comes from the LDPR, as well as
nearly half of Rutskoi's and the Communist Party's. In contrast, the
only significant loss experienced by the Communist Party is to
Rutskoi's list.

No consider the columns denoted `invalid ballots' and `new voters'.
The primary beneficiary of increased turnout is the Communist Party,
which secures nearly one third of all new voters, and overall,
opposition parties secure nearly 60 percent of all such votes. However,
although twelve percent of this increase is `wasted' on Anpilov and
Rutskoi et al's lists owing to their failure to qualify for seats in the
Duma, democratic and centrist parties waste even more -- all but the
12 percent of increased turnout that goes to Yabloko and the 5 percent
that goes to Our Home Is Russia. Thus, the combination of
fractionalization among democrats and the general edge given to
opposition parties among new voters gives the Communists a 2-to-1
edge over democrats among new voters whose votes counted. Perhaps
more importantly, though, recall that as turnout declined between
1991 and December 1993, the lost votes came almost exclusively from
the reformist camp -- from voters who voted for Yeltsin in 1991 and
who voted Yes on the April referendum's first question. But those
voters who `reappear' in 1995 do not return by voting for reform;
instead, a clear majority vote for opposition parties. The evidence,
then, is consistent with the hypothesis that a significant part of the
electorate -- over 8 percent of those voting in 1995 -- moved from
supporting reform in 1991 to oppose it in 1995, but only after they
passed through a period of indifference to all parties and positions.

The analysis of invalid ballots, the third column in Table 5, is
interesting in and of itself. The difficulty here is that we are dealing
with a relatively small fraction of the electorate, less than 5 percent,
which helps to make our estimates less reliable than usual. This fact is
borne out by the proliferation of negative estimates in the
corresponding column of Table 5. Nevertheless, accepting that our
positive estimates are in all likelihood biased upwards, Notice that the
primary beneficiary of the decline in invalid ballots is the communist
party, with the LDPR and Our Home Is Russia a close second and
third, respectively. This is indeed curious, since normally we would
expect invalid ballots to be spread uniformly across the parties. It is as
if only the `party of power' and the primary opposition parties can
attract previously `stupid' voters. We might, of course, attempt to
fashion a variety of stories to explain these estimates, but here we note
simply that if the sharp rise in invalid ballots in 1993 is somehow a
byproduct of election fraud (Sobyanin 1995), our estimates here
suggest that that fraud was not a uniform phenomena but concentrated
somehow in rayons that gave the LDPR, the Communist Party, and
any party associated with the Kremlin special support.


Finally, Table 5 allows us to reconsider McFaul's fourth hypothesis
about the vanishing center in Russian politics. If we take the center in
1993 to be Shakrai's Unity and Accord, Travkin's DPR, Women of
Russia, and those parties that failed to qualify for seats, we cannot
confirm H4. Our estimates at least suggest that the DPR's vote, for
instance, dissipates across Our Home Is Russia, the lists offered by S.
Federov, Popov, Lebed and Govorukin, plus other democratic parties
such as those headed by Pamfivoliva and Boris Federov. Women of
Russia appear to have given their lost votes primarily to Pamfivoliva,
B. Federov and our collection of 15 small parties, whereas Shakrai lost
votes primarily to Rybkin, with smaller shares given to Our Home Is
Russia, Yabloko, and a variety of insignificant parties. Indeed,
regardless of how we re-constitute our categories of parties, we find
no evidence that the votes lost by any of these parties went to any
opposition party.


In summary, then, of the seven hypotheses that summarize McFaul's
analysis, four are confirmed here but three are not. Hypotheses H1,
H2, H6, and H7 are strongly supported; on the other hand we find
little evidence consistent with H3, H4, and H5. It is not so much that
we see a migration of centrist voters into the opposition camp or a
bevy of new voters supporting extremist lists such as Anpilov's.
Instead, if we see change it is primarily in the form of a segment of
the electorate -- perhaps as much as 10 percent -- making the
transition from supporting reform to indifference to voting with the
opposition.

\section{Long-term Trends and Conclusions}
Thus far, we have examined pairs of adjacent elections, but, as a
check on our conclusions, two other comparisons should be considered
-- comparisons that allow us to trace the flow of votes directly from
the `heyday' of reform, 1991, to December 1993 and 1995. Looking
first, then, at the comparison of 1991 to 1995, Table 6 gives our results
in the same form as before.\footnote{However, to simplify
presentation, we delete the column corresponding to invalid ballots or
ballots cast against all candidates since they constitute an insignificant
part of the data, less than 4 percent.} Three things in particular are
apparent from this table:

First, those who voted in 1991 for the communist candidate Ryzhkov
continue to vote against reform by supporting opposition parties in
1995. Indeed, those who voted for any candidate other than Yeltsin in
1991 vote, with but some small exceptions, for opposition lists in 1995.

Second, Yeltsin's original support in 1991, although continuing to
favor parties that are not explicitly opposed to reform (the parties
corresponding to rows 1 through 9 in Table 6) 2 to 1 over parties
opposed to reform, is nearly uniformly spread across party lists.
Moreover, although Zhirinovsky's LDPR gains 7 percent of this vote,
the Communist Party gains even more, 10 percent. In a word, Yeltsin's
original support has dissipated to nearly all points.

Finally, notice that although some of Ryzhkov's voters (approximately
15 percent of his original vote) have entered the ranks of nonvoters,
some 20 percent of Yeltsin's far larger share are nonvoters in 1995.
Hence, even though turnout increased in 1995 over the levels it
achieved in 1993, a significant part of the electorate that favored
Yeltsin in 1991 continued to stay away from the polls four years later.

Turning now to Table 7, which compares 1991 with December 1993,
we see a somewhat clearer picture of the flow of Yeltsin's original
support. In particular, notice that although the LDPR gets 10 percent
of Yeltsin's original vote, the Communist and Agrarian parties, in
contrast to 1995, together get none of it! Moreover, fully 37 percent
of Yeltsin's vote enters the nonvoting ranks whereas only 30 percent
vote for Russia's Choice, Yabloko, Future of Russia or RDDR and
another 16 percent vote for the remaining (centrist?) parties on the
ballot. Insofar as the constitutional referendum is concerned, 47
percent of Yeltsin's original supporters vote for his constitution --
presumably those who voted for reform and centrist parties --
whereas 77 percent of those supporting Ryzhkov voted against the
constitution. Interestingly, notice that, in keeping with his campaign
advice, Zhirinovsky's supports also give majority approval to the
constitution.

Tables 6 and 7 together, then, are consistent with the view offered
earlier of a `reform' electorate that erodes into ranks of nonvoters, and
reappears in 1995 as part of the support garnered by the opposition to
Yeltsin's policies and reform. This finding about the ebb and flow of
nonvoters, along with the finding that the original support received by
Communist and opposition candidates has largely held firm, is perhaps
this essay's core conclusion. However, before we can give full confidence
to such conclusions we must check the potential for aggregation error --
the possibility that, contrary to our assumption that the same parameters
characterize all rayons, different subpopulations act differently.

As a partial check against such error, the data at our disposal allows us to
check for the possibility that urban and rural voters act differently. We
know, of course, that such voters not only confront different economic
and social circumstances and that they are impacted by Yeltsin's policies
differently, but that they also vote differently, with urban voters being
much more supportive of Yeltsin and reformist parties than rural voters
(see, for example, Hough et al 1996, Clem and Craumer 1993, 1995, and
Slider et al 1994) and with rural rayons reporting much higher rates of
turnout than urban ones. The question, however, is whether, for instance,
rural voters who supported Yeltsin in 1991 defected in different ways
subsequently than did voters who came from cities. 

Table 8, then, reproduces the coefficients reported in Table 6 except that
each cell reports two coefficients -- the first for rayons with 70% or more
urban residents and the second for their `non-urban' counterparts. Perhaps
the two most important things revealed by this table, at least from the
perspective of the conclusions we offer earlier, is, first, that regardless of
whether we look at urban or non-urban subpopulations, Yeltsin's 1991
vote disperses itself across the ideological spectrum whereas Ryzhkov's
vote holds firm against reform in 1995. Second, twenty percent or more
of Yeltsin's original vote, both urban and rural, became nonvoters in 1995
whereas none of Ryzhkov's urban support and only 13 percent of his rural
support failed to vote in Russia's second parliamentary election. 

This is not to say that there are no differences between urban and rural
voters. Notice, for example, that if we look down Yeltsin's column, the
first coefficient is greater than the second in the first nine rows, but
thereafter (excepting the last row, which corresponds to the difference in
turnout) the second number is greater. The implication, then, is that in
urban constituencies, the majority (but not all) of Yeltsin's vote moved to
reformist or centrist parties, whereas in rural rayons that vote moved
predominantly to anti-reform parties. Here, however, we can only infer
general tendencies since more often than not the coefficient for urban and
nonurban subpopulations are not statistically significantly different -- a
fact that gives us some confidence in believing that aggregation error does
not significantly impair our previous conclusions. We should, though,
offer one note of caution with respect to Zhirinovsky's 1991 supporters.
Although we are concerned here with less than 8 percent of the electorate,
the coefficients for those voters seem especially unstable, with significant
negative values appearing in three of our regressions (the regressions
corresponding to Russia's Choice, the Communists, and nonvoters). A
coefficient of 60 for urban rayons and -20 for rural ones with respect to
nonvoters suggests the possibility of significant aggregation error in our
overall estimate of the coefficient for Zhirinovsky with respect to this
regression in Table 6. Notice that a similar problem exists with respect to
Ryzhkov's supporters. Although the difference there, 0 versus 13, is less
pronounced than for Zhirinovsky, neither of these numbers is as great as
the estimate Table 6 reports for the share of Ryzhkov voters who became
nonvoters in 1995. This disparity suggests that the percentage given in
Table 6, 15 percent, overestimates the share of the communist vote that
failed to appear at the polls. Of course, this possibility merely strengthens
our conclusion that new nonvoters in 1993 and 1995 came predominantly
from Yeltsin's original base of support. 


Overall, then, the analysis offered here paints a remarkably consistent
picture of hard-and-fast opposition, a part of the electorate that moves
from supporting reform to opposing it after first becoming indifferent to
all candidates, and a reform core that succeeded in both parliamentary
elections in dividing itself up among a plethora of candidates and parties.
It remains to be seen, of course, whether reform candidates in the future
can recapture some of their lost support or whether the erosion witnessed
over the last four years will continue. Our analysis does not sanction such
predictions. But it does permit us to reject the idea that Russian voters are
unlike voters elsewhere or that they are an unstable lot that move easily
between reform and anti-reform positions.

\section{References}
\parindent=-1.0em

Clem, Ralph S and Peter R. Craumer. 1993. ``The Geography of the April
25, 1993, Russian Referendum,'' {\it Post-Soviet Geography} 34, 8:481-96.

Clem, Ralph S. and Peter R. Craumer. 1995. ``A Rayon-Level Analysis of
the Russian Election and Constitutional Plebiscite of December 1993.'' {\it
Post-Soviet Geography} 36, 8:459-75.

Filippov, Mikhail, and Peter Ordeshook. 1996. ``Fraud or Fiction: Who
Stole What in Russia's December 1993 Elections.'' Social Science
Working Paper No. 963, California Institute of Technology.

Hough, Jerry F., Evelyn David Heiser and Susan Goodrich Lehmann. 1996.
{\it The 1996 Russian Presidential Election}, Washington, D.C.: Brookings.

King, Gary. 1996. ``A Solution to the Ecological Inference Problem.''
 Manuscript, Harvard University.

McFaul, Michael, and M. Steven Fish. 1996. ``Russia Between Elections.''
{\it Journal of Democracy} 19:90--118.

Myagkov, Mikhail, and Alexander, Sobyanin. 1995. ``Irregularities in
the 1993 Russian Election.''  Social Science Working Paper, California
Institute of Technology. 

Slider, Darrell, Vladimir Gimpleson and Sergi Chugrov. 1994. ``Political
Tendencies in Russia's Regions: Evidence from the 1993 Parliamentary
Elections.'' {\it Slavic Review}, pp. 711-32.

Sobyanin, Alexander, and Vladimir Sukhovolsky. 1995.
``Democracy Bounded by Falsifications.'' Manuscript, Moscow.

\end{document}