Other than the obvious economic repercussions, public finance
also impacts and informs the political environment of a nation. In one of my
first tentative steps into empirical research, I tried to investigate whether
economic factors impact partisanship and gridlock in the United States
Congress.
The idea was straightforward: regress different measures of
polarization and partisanship on different economic measures, controlling for
such factors as time trends, whether elections were held on a given year, whether
the Democratic Party was the majority in Congress for that session, and even
the number of women in Congress, among others. The polarization measures were
the percent of party unity votes in that Congress (a party unity vote defined
as one where more than 50% of one party votes against 50% of the opposite
party), the percent of moderates in that Congress, the difference between the means of the
two parties on a liberal-conservative dimension, the percentage of overlapping
members on a liberal-conservative dimension (where zero overlap means the most
“liberal” Republican is still more conservative than the most “conservative”
Democrat), and both the percentage of Democrat and Republican moderates.
The economic metrics used were the usual suspects: growth,
unemployment, and inflation. I also included the deficit (or surplus), and the
level of debt in the economy. I also included first-differenced variables, such
as the change in the unemployment rate from one year to the next, etc. Data
ranged from the beginning of the Great Depression to the end of 2013.
Of course, the first difficulty was typical of working with
time series: the frequency and span of measurements frequently did not match or
correspond. While annualized measures of unemployment, growth, and debt data
could be easily obtained from existing observations, Congressional data is on a
per-congress basis: meaning measured every two years and spanning those two
years of each given Congress. An obvious—though not particularly ideal
solution—was simply using the Congressional partisanship or polarization
measures for both years of the given Congress. Thus, for the 107th
Congress (January 2001- January 2003), measures of polarization were assumed to
be the same for both 2001 and 2002. This is most appropriate assuming little
turnover or change in the members of Congress for the years that make up a
Congress.
At the time of carrying out this research, my knowledge and
application of time series methods in Economics were both rather limited. As
such, my results were quite inconclusive. In particular, since conducting my
research, I have grown much more informed as to selecting an appropriate ARIMA
model, choosing the model that best fits with Akaike’s or Bayesian Information
Criteria, and most importantly, using a VAR model instead and—even more
critically—potentially a VEC (Vector-Error-Correction) Model when dealing with
multiple, potentially related variables as both independent and dependent
measures.
For my original research, my main results supported a weak
effect: economic measures would have the type of effect on Congress that we
would want to see, in that lower
levels of growth, higher unemployment, and generally extreme levels of inflation
(both low and high) decreased the
level of polarization and partisanship in Congress. Yet, all of these effects
were comparatively weak both in terms of the magnitude of the coefficients and
the degree of significance. Yet, one effect conclusively stood out as
significant across all models and large in magnitude: the level of national
debt. In short, the larger our debt gets, the more polarized and partisan our
Congress, even after controlling for time effects and other potential controls.
However, knowing the other shortcomings of my research, and
now more informed of time series methods, if I were to conduct this research
again I would certainly suspect potential cointegration of some of my
variables, in particular those that showed my strongest results. As such, I
would conduct tests for cointegration and run a VEC model for debt and
polarization methods. Most critically, by doing so I would want to rule out a
spurious regression between these variables. For example, a peculiar result of
my research was that the number of women in Congress seemed to increase
polarization in Congress. It is likely, however, that this result was not
properly capturing underlying time trends and other factors—after all, both the
number of women in Congress and our national debt have risen steadily over
time; a model would have to be very carefully specified to control for all
these effects. Otherwise, a case would have to be made for the polarizing
effect of women in Congress, which seems more a relic of earlier centuries.
Of course, the question is then: were my original results due
to a spurious regression? It’s highly possible. The alternative would be that
further, better-informed tests and VAR models would reveal cointegration of
debt and polarization. In that case, there would be a long-run relationship
between the national debt and the level of polarization where, if a large
increase in the rate of national debt growth occurs, polarization would
decrease to bring back the variables to their long-run equilibrium. This would
imply then, that at moments of rapidly climbing debt, Congress would actually
unite to resolve the issue, which is precisely what we as citizens would expect
and prefer. Alternatively, at times where the debt is increasing less rapidly,
members of Congress may give themselves the luxury of focusing on more
partisan, less immediately pressing issues.
The accuracy of this statement, however, is still left up to
consideration. Are the state of our economy and the state of our Congress
jointly determined? What is the specification of this structural VAR, and its
economic implications? How can it inform possible reforms to the way our
Congress operates under different economic scenarios?
