Gabriel J. Michael / gmichael at gwu dot edu
Update 12/12: This post is now available in a significantly less technical form at The Washington Post’s Monkey Cage Blog. If you notice differences between the graphs, you’re paying attention. They exist for two reasons. First, in order to keep things understandable, I used simpler axis scales in the other piece. Second, MDS uses a simulation process to generate the positions of points on the graph. When combined with jitter, points shift slightly each time I reproduce the graphs, although not enough to affect substantive interpretation.
I could have normalized distances across chapters by dividing by the number of proposals per chapter. This would permit all graphs to use the same scale. However, it might also understate/overstate distances.
Please share widely and repost with attribution. This content is licensed CC BY-SA 3.0. Comments, suggestions, and criticism are welcome.
For those wondering about units or labels for the axes, if you don’t want to read the details, just note that absolute distances are arbitrary. Only relative distances matter. The axes are simply X and Y axes on the coordinate plane.
Early Monday morning, Wikileaks released a second set of documents pertaining to the ongoing Trans-Pacific Partnership negotiations. Unlike the previous leak, this one does not contain a draft text, and instead consists of a series of comments and analysis by an unidentified negotiating party (clearly not the U.S.), as well as a table reporting each negotiating party’s position on specific issues in each chapter. The documents come from the Salt Lake City round of negotiations, which took place from November 19 to November 24, 2013.
The comments are brief, but well worth reading, as they indicate a degree of frustration with the lack of progress during the Salt Lake City round, and explicitly charge the United States with intransigence (e.g., on the topic of agricultural export subsidies, the author writes “All TPP countries except the U.S. commit to eliminate them”; likewise, with respect to financial services, he/she writes “United States shows zero flexibility.”)
An additional round of negotiations was held from December 7 to December 10 in Singapore. We learned yesterday that contrary to previous expectations, the TPP will not be concluded by year’s end, and instead at least one additional meeting will be held in January 2014.
Following up on my previous post, which analyzed the leaked intellectual property chapter by using network graphs, in this post I analyze overall negotiating positions across the entire agreement, as well as in each individual chapter. This analysis is based on the leaked table reporting negotiating positions.
Visualizing Negotiating Positions
The following graphs take the approach of plotting negotiating “distances” between countries. That is, the more dissimilarity there is between two countries’ negotiating positions, the further apart the countries will appear on a graph. Distances are derived directly from the leaked position table. The centroid is given by the origin point (0,0). Absolute distances are not particularly meaningful, as they depend on the coding scheme used; what’s important is the relative distances between countries. In most cases, both axes use the same scale, but in a few cases (which I identify) they are different. The axes don’t have any labels because they are simply X and Y axes – we’re plotting countries on a coordinate plane. Scales vary between graphs. All the graphs use “jitter,” which helps avoid overlapping labels, but makes distances slightly inexact. For technical details, see the “Nuts and Bolts” section towards the end.
This first graph represents the overall negotiating position distances between countries across all available chapters. There is a striking separation between the United States and all other TPP negotiating parties. Australia comes closest to the U.S. position, although it is not any further from the centroid than Peru. Overall, this graph serves to provide a simple visual confirmation of the leaked comments and other news reporting on TPP negotiations: the U.S. position is quite distant from the consensus point of all other countries.
Since my area of interest is intellectual property, I look at the IP chapter next. The results suggest my previous analysis was relatively accurate. The U.S. position in the intellectual property chapter is again farthest from the centroid. Australia is also quite far from the centroid, but also far from the U.S. All the other parties are clustered relatively close together, although we can still see differences within this cluster: e.g., Vietnam and Brunei are quite near one another, as are Canada and Malaysia (I had noted both these connections in the previous analysis).
In the remainder of this section, I consider each leaked chapter in alphabetical order.
The competition/state-owned enterprises graph tells a very different story than the previous two graphs. Here we immediately see two clusters on either side of the centroid: a United States/Canada/Australia/Mexico cluster, and a Japan/New Zealand/Singapore/Brunei/Chile cluster. Peru, Vietnam, and Malaysia fall in the middle. Note that the relative level of disagreement when compared to other chapters is low, since the data source reports positions only on a single proposal (sub-national coverage).
In the customs chapter, we see agreement amongst all parties except Japan (who has a “reserved position”) and the United States (who is the only party accepting the proposal). Again, the relative level of disagreement when compared to other chapters is low, since the data source reports positions only on a single proposal (a de minimis exception of $200).
The e-commerce chapter also provides a very different picture. There is a clear United States/Japan/Mexico cluster, which Peru also joins. Canada is nearby, but Australia is not. Brunei and Vietnam, normally close pairs, are quite far apart in this chapter.
The environment chapter also shows a distant United States position. Canada, Australia, Japan, Mexico, and Brunei cluster together quite close to the centroid, while Peru, Chile, Vietnam and Malaysia (all lower-income and middle-income countries) appear on the other periphery.
In government procurement, Mexico and Malaysia are the outliers. Both have rejected the proposal for sub-national coverage of this chapter. Relative distances are still small, given that there are only two proposals shown in the source table.
In the investment chapter, the United States appears again as the country furthest from the centroid. Japan is also relatively distant. Australia/Canada/Mexico cluster together, as do Brunei/New Zealand/Peru/Malaysia/Chile. One of the biggest debates in this chapter surrounds so-called investor-state dispute settlement, which would permit foreign firms to sue governments over alleged trade agreement violations. This type of provision is what has permitted tobacco company Philip Morris to sue the Australian government over its plain packaging legislation.
On labor issues, Australia and the United States cluster together far from the centroid. Mexico is also quite distant. Malaysia and Canada form a pair, but closer to the centroid. Mexico’s peripheral position is due to its rejection of the proposal on forced labor. The other two proposals concern sub-national coverage and dispute settlement.
The legal chapter includes a variety of topics such as the medicines “transparency” annex (i.e., the U.S. taking aim at foreign drug price controls), provisions exempting tobacco regulations from challenge, the “cultural exception” (bonjour Québec!), and issues concerning when the agreement will enter into force, inter alia. Interestingly, we have almost a circle of positions around the centroid, with the exception of the United States, which is again farthest from the centroid.
The market access chapter is what most people typically think of when discussing trade agreements. Here, we see a wide variety of positions, representing significant disagreement. However, yet again, the United States appears farthest from the centroid. As noted above, the U.S. is the lone holdout on the proposal to eliminate agricultural subsidies. A recent Washington Post article on the U.S. sugar industry suggests that such subsidies aren’t going anywhere. I suppose that’s why the TPP is called a “partnership agreement”, rather than a “free trade agreement.”
Rules of origin are rules defining where a product is deemed to come from. Given that anything other than the simplest products will incorporate materials or parts from multiple countries, such rules are critically important in determining what products benefit from tariff reductions. We can see significant disagreement between most parties, with the United States position yet again farthest from the centroid.
With respect to trade in services, the United States and Canadian positions are identical, but farthest from the centroid. Peru and Chile are also peripheral, while all other parties cluster together near the centroid. The U.S. and Canadian positions are defined by their rejection of the “necessity test” proposal. Necessity tests require domestic regulation of services to be limited to only what is “necessary” to achieve a party’s policy objectives. Peru’s position is defined by its rejection of the open skies (air travel regulation) proposal.
SPS refers to sanitary and phytosanitary measures, which deal with food safety and plant and animal health regulations. If you’re wondering what this has to do with free trade, think of Japan or the European Union banning imports of U.S. beef: such bans may be motivated by legitimate health and safety concerns, but they may also simply provide a convenient cover for protectionism. The graph shows a significant amount of disagreement amongst most parties. In this chapter, the U.S. remains on the periphery, as does Japan, although there is no major consensus cluster amongst the other parties.
TBT stands for technical barriers to trade. Such provisions in trade agreements are designed to ensure that regulations, standards, and testing or certification requirements do not unduly burden free trade. Australia and the United States are clustered near one another, but this time it is a Peru/Chile cluster that is farthest from the centroid. All the Asian members plus New Zealand also form a cluster.
Without access to the text, it’s sometimes difficult to know exactly what the various proposals mean. However, what we do know is that in the majority of chapters for which we have data, the United States appears quite far from the centroid position, and often by itself. This isn’t really news, but I think it’s interesting to systematically consider the distances between negotiating positions, and to note that the distances and clusters vary significantly by chapter and issue.
As for what it means, I think it’s fair to conclude that the TPP – unlike something like ACTA – is by no means an agreement amongst “like-minded countries.” Now, contention and disagreement isn’t always bad; on the contrary, it’s part and parcel of any negotiation. But one has to wonder about which countries are going to end up shifting positions, and in what direction, in order to make consensus and a final text possible. I suspect it won’t be the United States offering the compromises.
There are important issues at stake in the TPP negotiations, affecting access to medicines, national health policies, and national sovereignty inter alia. Yet the negotiating parties are being pushed to come to an agreement within the next two months. I won’t delve into any more detail here about the substantive issues, other than to recommend this Guardian guide to the most contentious issues in the TPP negotiations.
Nuts and Bolts
The approach I use here is called multidimensional scaling (MDS) which visualizes similarity and differences between cases as distances in N-dimensional space. I have to give a hat tip to Zhou Fang, who suggested this approach to me.
The leaked table of negotiating positions lends itself perfectly to MDS. In fact, the hardest part was probably retyping the scanned Wikileaks document. I coded an “accepted” position as 1, a “rejected” position as a 0, and a “reserved position” as 0.5. Note that coding is somewhat arbitrary; e.g., I could have chosen 1, -1, and 0 instead. Different coding will affect the absolute distances, but not the relative distances (unless the coding employed unequal intervals).
R’s dist function creates a distance matrix. The R MASS library provides the isoMDS function. isoMDS chokes when cases are identical (zero distance), so I had to add a nominal amount of distance between otherwise identical negotiating positions. Plotting is done with ggplot2, which makes adding jitter very easy. I adjust jitter on a case-by-case basis in order to avoid misleading distortions in the graphs. Most of the graphs use identical X and Y axes, but in a few cases the Y axis would have been highly compressed, so I expand it to make visual differentiation possible. Note carefully the scale changes between graphs.
Data and code available upon request. I’m currently in South Africa, so between the time difference and sporadic access to the Internet, I might not get back to you quickly, but I will do so eventually.