The Origins and Breakdown of the Washington Consensus
A Machine Learning Approach to 1984 Congressional Voting Data
History doesn’t repeat itself, but it does rhyme. — Mark Twain
For many, watching the election results coming in on November 8th, 2016 left an indelible impression. The victory of Donald Drumpf still puzzles many people. Pundits, and major news network described Donald Drumpf’s win as one of the biggest political upsets in US history. They described this singular event as surprising, or stunning right after the election. They pointed to states like Wisconsin, Michigan, the so called Rust Belt, where many traditional unionists, and working class voted for Barack Obama for the past two elections, defected to Drumpf. I’ve been following the ins and outs of American politics for a while, and I believe that Drumpf’s victory is not stemmed from a swift political turnaround or even a realignment, but rather the result of a simmering process developed over the years. I was careful not let my biases stand in my way, but rather I would using a voting record from the past and see whether I could find any insights that could explain this shift of the political landscape.
Several years ago, I came across an article that describes Republican and Democratic parties as coalitions of various political forces, and diverse elements of society. In other words, they are not monolithic entities that always vote together on certain issues, but rather groups made of fractious elements whose ultimate message is a result of many backroom compromises, and consensus-buildings. Under this premise, I thought a dataset of congressional voting records from the past could help me to discern any hidden pattern. Are there any faction within a party who didn’t necessarily vote together with the establishment. In other words, I want to know whether there are contrasting voices within in a political party that belied the façade of unity.
Thanks to the riches of data from the Internet, I found a dataset from UCI Machine Learning Repository that is befitting to my goal. It’s a voting records on 16 key issues from members of the House of Representatives. Moreover, the data also contains 435-member’s party affiliation labels. The binary “yea and “nay” on 16 issues makes it easy to do a clustering exercises using unsupervised learning algorithms like k-means. I wanted cluster these Congress members by their votes on issues and compare the label produced by k-means algorithm with their original party labels of the dataset to discern interesting patterns. Moreover, I thought the politics of 80s were not as polarized as it is today, and the congresspeople then would more likely act as delegates to the voters. Their stances on issues would be a barometer of the public sentiment at large.
Data cleanup
I read in the dataset as a data frame and provided it with headers. It’s in wide table format with 17 columns for party label and issues. To better inspect to dataset, I unstacked the dataset, and inspected the top issues supported by Republicans and Democrats based on rough counts of “yeas.” The results are interesting. Their voting pattern largely met people’s expectations of the typical image of two parties. For example, Republicans supported bill that was tough on crimes and endorsed American military interventions in South America while Democrats demurred. Moreover, it was apparent that partisanship centered around the adoption of democrats-penned budget resolution which few Republicans voted for. However, I found many “?” as placeholders for missing values in the dataset. I thought theses missing values could best be treated as 0s, but doing so would distort certain bills as to make them seem less favorable. To compensate for that, I decided to create a function that normalize the votes for each issues.
Clustering
After doing preprocessing, I stripped off the party labels and used k-means to cluster the Congress members based on their votes on issues into 2 and 3 clusters. These plots looked promising to me. For the cluster of 2, k-means successfully clustered most Democrats into one group. 218 Democrats and and 6 Republicans were put into a cluster, which was not bad at all.
For the cluster of 3, k-means has a cluster that captures all Democrats. I was also happy with the graphs my program generated. For the cluster of 3, when I unselected the middle group, and the political chasm between the color red, and blue dots (representing Republican and Democratic Congress members) were held in high relief. It was a great way to visualize the political divide at the time.
Analysis
However, simply inspecting the graphs wouldn’t tell the full story. I then used the original party affiliation label to check whether machine-learning techniques characterize Congress people differently from their original labels. To my surprise, I got some starling results that goaded me thinking more deeply on the political development of past thirty years. The analysis of my discovery consists of three part:
- the emergence of a “new” group of Democrats,
- the rise of the Third Way,
- the ascendency and breakdown of the Washington Consensus.
The Emergence of “new” Group of Democrats
One thing that stroke me as startling was that the 2 cluster-instance labeled 49 Democrats as Republicans. The result suggested that this group of Democrats which made up about a quarter of Democratic Congress members had very similar and consistent voting patterns as Republicans. I didn’t make haste to attribute their voting patterns to Reagan’s sway on House democrats. I then zoomed in to the issues they supported. I first singled out their k-means clustering labels and used them to select the the top issues they supported from my transformed dataset. The result is attached below.
There are three interesting observations from the votes of this group of Democrats. Domestically, they sided with Republicans and enthusiastically supported Religious groups in schools, a bill that was lobbied for Christian groups who wanted to ensure students the right to conduct Bible study programs. Moreover, at odds with the majority of Democrats, they were as harsh on crime and law enforcement as Republicans. ( Coefficient of 0.116 v.s. 0.04 of traditional Democrats). I suspected their supports of the Crime Bill could be the determining factor that tricked the machine into clustering them as Republicans. Internationally, this group of Democrats sided with Republicans in supporting economic and military interventions in Latin America.
As for me, this emergent group of Democrats, albeit only made up a quarter of Democrats, voted distinctly different from the mainstream Democrats at the time. Although their presence in the early 80s were still marginal, and they signaled a profound reconfiguration of the Democratic party. The dominant group of Democrats in the 80s, the so-called Social Democrats, regarded Franklin D. Roosevelt as their model. To them, FDR was the yardstick of their liberal ambitions. They continued to pass on the mantra of FDR, that supports social welfare expansion, and even some forms of economic interventions to promote social justice. The voting pattern of this emerging group of centrist Democrats suggested some dissension with the former. Although they still maintained the labels of Democrats, they didn’t often go along with their social democratic peers. To understand their different voting patterns, I would like to obtain more detailed information of these Congressmen, but I was only able to find a geographic distributions of House members based on political affiliations in 1984.
Given the presence of many Democratic House seats in Mid-West and South, I thought this group of Democrats definitely were pressured by voters into favoring Reagan’s more socially conservative policies. Also Internationally, they were more willing to use and and fund the military to chime in with Reagan’s optimistic message to Americans. Unlike their more dovish Democratic peers, they supported US’s more robust presence in the Third World countries as form of trade liberalization and military intervention to further U.S. Interests.
The Rise of the Third Way
The discovery of this latent splinter 49-member-group of Democrats was the most interesting points this dataset revealed. In 20/20 hindsight, it’s stunning to imagine that the ideas of this small group of centrist Democrats which tricked computer into labeling them as Republicans could prevail. The rise of Bill Clinton in the 90s cemented this group of Centrist Democrats as main voice of the Democratic Party, replacing their older, more left-leaning peers. Bill Clinton was the only Democrat who won two consecutive terms to his day, and many of his policies aligned with the top agendas of the Centrist Democrats of ’84. Clinton’s promise of welfare reform in the 1992 presidential campaign and its subsequent enactment, epitomized a rightward shift of Democratic positions. It’s interesting to see that many of Clinton’s policies seemed a natural continuation of stances of emergent Democrats in 1984. His Defense of Marriage Act, and Religion Freedom Restoration Act were aimed at luring over socially conservative working class voters. And his 1994 Omnibus Crime Bill, which vastly enhanced law enforcement apparatus was more potent than the 1984 Crime Bill introduced by Reagan.
Drumpf and the Breakdown of Washington Consensus
What’s the big deal then? In what ways the developments of Democratic Party related to the politics today? You may ask. I think the Rise of the Centrist Democrats changed the political landscape in three important ways. One is the convergence of Republican and Democratic positions on trade and fiscal polices as if they are two brands of conservatism. For one thing, they share similar attitudes on trade liberalization and potential military intervention in developing countries. These ideas, and bipartisan consensus were best exemplified by the Washington Consensus, a group of 10 policies promoted by Washington-based economic institutions such as IMF, and the World Bank as their standard prescription to developing countries seeking their aid. The 10-point Washington Consensus facilitates the outsourcing and promotion non-American manufacturing jobs while in the name of trade liberalization. The convergences of bipartisan support with regard to trade, and deregulation certainly created a sense of resentment from the working class who saw their income falling in real terms despite the unprecedented global economic growth.
Second, as a reaction to the shift to the center of the Democrats, the Republican Party also shifts further to the right of the spectrum that undermines its ability to govern. A telling example is the recent healthcare fiascos. Many Republicans from the Freedom Caucus thought the health care bill was not conservative enough, and opposed their own party’s own legislation. The fringe yet decisive groups like Freedom Caucus doesn’t have as much the mindset of a governing political party as a radical insurgency that only cares what it can get. Thus, the latest health care drama also underscores one of the implications of the rightward shift of the Democratic Party.
The third is that the convergence of the two parties also contributes a perceived sense of political gridlock and decay especially among rural whites in the sense that something can’t be done in Washington. Moreover, with the decimation of social democrats which used to be make up the bulk of the Democratic Party, many rural people and union members find the channels to express their interests largely cut off. It gives to a sense that both parties don’t care them anymore. Therefore this segment of American population becomes very susceptible to populist agitations and protectionist rhetorics. These people find hope in people like Drumpf whose rhetoric against both “draining the swamp”, and “Job-stealing China” speaks to their hearts. The white working class supported Drumpf for they see Drumpf a solution to their downtrodden economic conditions and much of the political illness of Washington. Washington Consensus which has promulgated by Washington elites over the years as the guiding economic ideal from the West for more than two decades, was dealt a severe blow by Drumpf.
Conclusion
From the machine learning result I got from 1984 voting data, I was be able to tie the growth and ascendency of the Centrist Democrats to the current political developments. Just as Mark Twain once quipped that “history doesn’t repeat itself, but it does rhyme,” history does move in cycles. With Drumpf’s victory, many millennials are becoming more politically conscious. They begins to paying attention to politics and to the part of America who is not usually visible or well-represented. With this growing political consciousness, It would interesting to know where we are going from here.
