Design Tidbits

back Tidbits Overview

Tools for Design and Visualization

Simplification

Interaction Design

Screen Design

Hierarchies

Visual Design

Design Process

Web, Web Applications, Web Design

Performance

Miscellany

 

 

Eurovision Song Contest 2012 and NodeXL: A Second Brief Encounter

By Gerd Waloszek, SAP AG, SAP User Experience – July 3, 2012, updated July 13, 2012

NodeXL is "a free, open-source template for Microsoft® Excel® 2007 and 2010 that makes it easy to explore network graphs" (from the NodeXL Website). Last year, I wrote a two-part article about this visualization tool and applied it to the European Song Contest (ESC) 2011 voting data. This year, I would like to add one more article on this topic to my collection, in which I apply the tool to the ESC 2012 voting data. The main goal of this article is not to explain the tool itself (which I did in the previous two articles), but to look for some differences and similarities between both years' voting data – and to inspire readers to conduct further explorations and comparisons of their own. Nevertheless, I will also present variations of the graphs, particularly of the 2012 graph, to illustrate how you can play around with the tool to highlight certain aspects of the data.

 

The 2011 Voting Data

First and as a reference, I will present a few graph variations of the 2011 voting data. For details, see my articles Eurovision Song Contest 2011 and NodeXL: A Perfect Match – Part I and Part II.

Automatic Layout

The following two figures show graphs created with the Fruchterman-Rheingold algorithm, which arranges nodes (called vertices in NodeXL) automatically (the algorithm creates a different layout with each refresh of the graph). In Figure 2, dynamic filters are additionally applied to the edges so that only the top votes (12 points) are shown prominently. The other votes are not completely suppressed, which is the default, but shown with a filter opacity of 15%. This manipulation highlights the distribution of the top votes and still provides an impression of the overall voting behavior.

Figure 1: Graph after calculating graph metrics and using them for autofill: Vertex shape set to label, vertex fill color determined by degree (connectedness); line color, width, and opacity determined by edge weight (voting points)

Figure 2: Vertex coloring based on a cluster algorithm and degree; dynamic filters used for edge weight (12-point votes only); filter opacity = 15%

The countries at the periphery are mostly the ones that did not participate in the final. Switzerland [CH] is an exception – it received only a very small number of votes and was therefore placed at the periphery.

Using a Map as Background

In the following two figures, the country nodes (= vertices) were placed manually on a background map (from Wikipedia). This arrangement was expected to give a better impression of whether geographical relationships show up in the voting. Figure 3 shows all votes, whereas Figure 4 shows only the top votes (12 points).

Figure 3: Map used as a background; vertices placed manually; Sweden selected

Figure 4: Map used as a background; vertices placed manually; dynamic filter for edge weight (12-point votes); Sweden selected


In Figure 4, we can see, for example, that there are some connections in the following geographical regions:

  • Scandinavian and also Baltic countries (not very strong)
  • Southern countries, including France (common Romance languages?)
  • Balkan countries (ex-Jugoslavia)
  • Russia, ex-Russian, and Caucasian countries

Of course, these are not all-or-nothing relationships. There are numerous deviations, and the relationships are not symmetrical either.

In the small version of Figure 4 it is hard to see, but the larger image shows more clearly that the winner Azerbaijan (AZ) received only three top votes, which came from two neighboring countries and from Malta. Actually, Bosnia Herzegovina (BA) got the most top votes (5), which is even harder to see in the map version (it is easier to verify in Figure 2).

 

The 2012 Voting Data

I will now turn to the ESC 2012 voting data and present several variations of the resulting network graph, first with country nodes arranged automatically by the Fruchterman-Rheingold algorithm (as I already mentioned, these graphs change with each refresh), and then with the country nodes placed manually on a map* background.

*) Note: For reasons of simplicity , I reused the map from 2011, although there are a few differences with respect to the participating countries (the important aspect for me was the location of the countries).

Automatic Layout

In the following automatically laid out graphs, I vary the number of displayed votes: all votes, the three top votes (8, 10, and 12 points), and the top votes only (12 points). I also contrast graphs where there is no node selected with graphs where the winner Sweden (SE) is selected. The latter highlights the votes for the winner.

Figure 5: Graph after calculating graph metrics and using them for autofill: Vertex shape set to label, vertex fill color determined by betweenness centrality; line color, width, and opacity are determined by edge weight (voting points)

Figure 6: Ditto, winner Sweden selected

Figure 7: Graph showing only the three top votes (8, 10, and 12 points)

Figure 8: Ditto, winner Sweden selected

Figure 9: Graph showing only the top votes (12 points)

Figure 10: Ditto, winner Sweden selected

Figure 11: Graph showing only the top votes (12 points); filter opacity = 10%

Figure 12: Ditto, winner Sweden selected

Figures 9 to 12, in particular, demonstrate that the winner, Sweden (SE), received a large number of top votes. Again, the countries at the periphery are mostly the ones that did not participate in the final. Denmark [DK] and Norway [NO] are exceptions – they lie at the periphery because they received only a small number of votes.

Using a Map as Background

In the following four figures, the country nodes were placed manually on a map background (from Wikipedia; 2011 map) to give a better indication of geographical relationships in the votings. Figures 13 and 14 show all votes, whereas Figures 15 and 16 show only the top votes (12 points). Like Figure 2, Figure 16 also shows the other votes in a subdued style (opacity = 20%)

Figure 13: Map used as background; vertices placed manually, all votes, Sweden selected

Figure 14: Ditto, but only the three top votes shown (8, 10, and 12 points)

Figure 15: Ditto, but only the top votes shown (12 points)

Figure 16: Ditto, filter opacity = 20%

In Figure 15, we can see, for example, that there are some connections in the following geographic regions:

  • Balkan countries
  • Russia, ex-Russian states, Caucasian countries

Potential preferences within Scandinavian and Baltic countries are this time hidden by the fact that Sweden (SE) received many top votes from all over Europe.

 

Comparison of 2011 and 2012 Results

Finally, I will briefly compare the ESC voting results from the years 2011 and 2012.

Automatic Layout

Both years' graphs show countries that neither participated in the final nor received many votes at the periphery, whereas countries that received many votes (and from many different countries) are located at the center. This is the basic similarity between the two graphs.

However, there is a big difference between both years' voting results: Last year's winner, Azerbaijan (AZ), did not receive many top votes (3) and was also not the most connected country (Sweden was the most connected country in 2011). In 2012, however, the winner, Sweden (SE), received a large number of top votes (18) and was also the most connected country. The Sweden (SE) node really looks like a focus point in the 2012 graph.

Figure 17: Vertex coloring based on cluster algorithm and degree; dynamic filters used for edge weight (12-point votes only); filter opacity = 15%

Figure 18: Graph showing only the top votes (12 points); filter opacity = 10%

Using a Map as Background

The map-based 2012 graph also shows Sweden (SE) as a focus point. The 2011 graph, on the other hand, indicates more geographical relationships than the one from 2012: Scandinavia/Baltic, Romance-language countries (Portugal, Spain, Italy, France, San Marino), Balkan, Caucasus/Russia/Ex-Russia. This year, these effects, as already mentioned, are probably hidden, because the winner, Sweden, received a large number of top votes (18) from all over Europe. Only in the Balkan and Caucasus/Russia/Ex-Russia regions do – some connections show up in the 2012 graph.

Figure 19: Map used as a background; vertices placed manually; dynamic filter for edge weight (12-point votes); Sweden selected

Figure 20: Ditto, but only the top votes shown (12 points)

All in all, it is interesting to see that the map-based graphs from the two consecutive years 2011 and 2012 look so different. On closer inspection, however, you can see that some of the top votes are the same for both years: Portugal > Spain, Malta > Azerbaijan, Turkey > Azerbaijan, Cyprus > Greece, and so on). There also seem to be a few "thank you" relationships, that is, "thank you" votes in 2012 for a 2011 top voting: Greece > Cyprus, Azerbaijan > Turkey, Italy > Albania, and so on. Of course, it is easy to interpret too much into these graphs...

 

Final Remarks

This explorative article is not an attempt to to engage in deeper analyses of the ESC voting data. It is just meant to inspire readers to start their own explorations with NodeXL and with whatever data they are interested in.

 

References

Aftermath

Ben Shneiderman encouraged me to post samples of my map graphs in the NodeXL Graph Gallery. For this reason, I had to download a new version of NodeXL that easily allows to export graphs to the gallery. However, when starting the new version with my data, I had a few issues with color values that needed to be recreated and also with the graph's scaling, which did no longer fit the map. I also decided to update the map and use the 2012 map for the new graphs. Here is a recap of my final comparison based on top votes alone:

Figure 19a: Map used as a background; vertices placed manually; dynamic filter for edge weight (12-point votes); Sweden selected

Figure 20a: Ditto, but only the top votes shown (12 points) (new map)

 

By the way, when I exported my graphs to the NodeXL Graph Gallery, I encountered another scaling issue:

Gallery graph

I contacted Ben Shneiderman who forwarded my e-mail to the NodeXL team (see below for a reply). My first investigations indicated that the map is exported in its original size, meaning that the graph does not seem to scale properly.

Addendum July 13, 2012

I received the following reply from Tony Capone (July 12, 2012):

NodeXL's support for background images is rudimentary. It can't scale a background image, which is why the larger image on the Graph Gallery didn't display the map the way you wanted. We've discussed making background images more flexible, but we haven't gotten to that yet.

I've gone ahead and deleted your two exported graphs. Sorry for the trouble.

Thus, at the moment, the NodeXL Graph Gallery has to do without my samples. The only way to export graphs with nodes laid out according to a background is by taking screenshots (as I did for the article).

 

top top