My excellent colleague Johan Christensen has a new book out with Stanford on the power of economists in government. I talk to him about it here for the podcast we made for my research methods class with Natascha van der Zwan. New episodes will be up soon.
Gephi has a tool to import tweets and map them as a network through hashtags. This is how it looks with the search key “Trump” just after the news came out in the Washington Post that Trump had allegedly leaked classified information to the Russian Ambassador and Foreign Minister. One can clearly see where the action is (in the orange area), but there seems to be a pro-Trump galaxy around the “MAGA” hashtag.
The graph above shows the level of electoral support for British political parties by ideological self-positioning, based on wave 10 of the British Electoral Study (data collected in November and December 2016). I have left out people who refuse to place themselves. The curve obviously declines as you go right for Labour, and increases for the Conservatives (with a decline on the far right as UKIP goes up). What is the most interesting is the slope of the two main parties: electoral support for Labour declines faster as you go right, and the Conservatives rise earlier, meaning that the Conservative are stronger on the center ground. The Conservatives do much better than Labour among people who consider themselves in the very middle of the political spectrum (that is, choose a 5 out of 10). 27% of voters in this category intend to vote Conservative, against only 18% who plan to vote for Labour. As the histogram below shows, this is the largest category of voters in the electorate, among those who are willing to place themselves (23% don’t). The sample is 30’319.
The network below shows the country participants in yesterday’s Eurovision song contest connected to their top 3 recipients of votes (both jury and call-in votes). The colour of the arrows corresponds to the recipient. We can see the two main contenders Portugal and Bulgaria clearly standing out.
Emmanuel Macron and Marine Le Pen will be competing in the election run-off of the French presidential election on May 7th. It is fair to say that Macron and Le Pen propose completely opposed views of what France should be. Macron is a liberal cosmopolitan, is pro-EU and wants to relax employment protection, and Le Pen is a nationalist who champions a stronger role for the state. In recent work we have done with Line Rennwald, we found that the Front National had been the European Party that had moved the furthest to the left when it came to welfare state issues, at least when it came to its agenda (what they actually do is a different story).
In line with this, the social classes that they appeal to are also completely opposed: Le Pen does the best among blue collar workers, while Macron is the most popular among managers and what sociologist call “socio-cultural professionals”, people in the creative industries, doctors, lawyers, intellectuals, and upper social categories (definitions for French categories are here). In 2002, Jacques Chirac had claimed that he represented “La France d’en bas” (France from the bottom) and not “La France d’en haut” (France from above). If we look at socio-economic categories, Macron is the candidate of the “France d’en haut”, and Le Pen the candidate of the “France d’en bas”.
Using electoral results data from the 1st round combined with socio-economic data on electoral districts (n= 551) gathered by the French National Statistics Office, we can get an idea of how the Le Pen and Macron votes related to the social composition of districts, and notably the share of manual workers and socio-cultural professionals. I have run a few simple regressions to look at this. An obvious problem in this kind of analysis is that one risks making ecological fallacies: with socio-economic data at the district level, we don’t actually know who votes for Le Pen or Macron. But it can still give some useful information.
We know from opinion polls (e.g data above) that Marine Le Pen scores highest among manual workers. This is in line with recent research in other European countries. Indeed, the share of manual workers in electoral districts is clearly positively correlated with the strength of the Le Pen vote. Not controlling for anything else, a 1 prcentage point increase in the share of manual workers in French constituencies is associated with a 1.45 percentage point increase in the Le Pen vote share. The share of manual workers explains 37% of the variation in the Le Pen vote share, and the relationship is statistically significant at the 99.9% level. This relationship is the opposite for the Macron vote. The greater the proportion of manual workers in a district, the lower the vote share of Emmanuel Macron. For a 1 percentage point increase in the share of manual workers, on average the Macron vote share decreases by 0.85 percentage points (R2=0.34). This relationship is not statistically significant for the Macron vote if we control for the share of managers and immigration, however (click graphs to enlarge).
If we look at the social category of managers and socio-cultural professionals (the core base of the Macron electorate), or what we call “les cadres” in French, this relationship is reversed. The greater the proportion of cadres in an electoral district, the greater the Macron vote share, and the lower the Le Pen vote share.
Another way to look at this not to look at vote shares, and look at how the social composition of districts affects the probability of one candidate winning over the other. For this, I have created a dummy variable that captures who between Le Pen and Macron wins in electoral districts. The graphs below show how this probability evolves at different shares of working class voters in an electoral district. Logically, the probability to win increases with the share of working class voters for Le Pen, and decreases for Macron.
Finally, we can also look at the relationship between economic indicators such as unemployment and the vote shares of the two candidates. I couldn’t find data at the district level, so we can use the level of the department (France’s main administrative division). There is a positive relationship between unemployment and the vote share of Marine Le Pen (R= 0.58), and a negative one for Macron (-0.59) (N=102).
In the table below, I have regressed Le Pen and Macron vote shares by the socio-economic variables mentioned above, controlling for the share of foreign citizens per district. Data from INSEE only include people who do not have French citizenship and not French citizens of foreign descent. Not controlling for anything else, the effect of the share of immigrants is pretty small, (r-square 0.05 for Macron and 0.17 for Le Pen; not shown), but its coefficients are negative for both in the overall model, meaning that greater shares of foreign citizens mean lower vote shares for both candidates.
In Panorama, April 2017.
What would vote transfers in the second round of the presidential election look like? In an earlier post, I showed that Jean-Marie Le Pen got very few votes from other parties in the second round of the 2002 election, when he made it to the second round. Voters of all other candidates massively rallied behind Chirac against Le Pen. Will things be different this time around?
The CEVIPOF at Sciences Po Paris and IPSOS have released a poll yesterday (n= 11 601) which asks people who they would vote for in the first and the second round, presenting different scenarios. The report shows the proportion of Fillon voters who would vote for Macron if Fillon doesn’t make it to the second round, the proportion who would vote for Le Pen, and so forth. This could all be presented in a matrix.
I have used this data and polls to project possible voter flows between the first and the second round in the different scenarios. Basically, I have taken the current poll numbers of the candidates and reallocated their share of the vote to the two candidates in the second round based on the report.
Now, this should be considered with a pinch of salt because one has to make a lot of – sometimes quite strong – assumptions. First, I assume that polls are accurate, and the level of support of the different candidates in the first round will be close to the real first round result. Second, I assume that voters of a successful candidate in the first round will stick with them in the second one, and none of them switches or abstains. Third, since people were given the opportunity not to express a preference, I assume that the people without a preference will abstain. This is not realistic and surely overestimates abstention, but the lack of preference surely signifies that there is no likable alternative for these voters. Fourth, and most importantly, these were scenarios presented to voters before they knew who is actually in the final runoff, and preferences may change once the choice before them becomes real. But this gives an overview of how “likable” the different candidates are beyond their own electorate, which is crucial to win a two-round election. Amongst the potential candidates, Macron and Mélenchon seem to have the broadest appeal beyond their own electorate.
I have built a series of Sankey diagrams that present these voter flows in the different scenarios. In brackets, the poll numbers for the different scenarios as given by Ipsos. The numbers in the graph include abstention, so the share of the vote for each candidate is given as a share of the electorate, not of valid votes. One problem is that abstention is not factored in in the first round, but the numbers are only given for people who “are sure they’re going to vote” (people tend to overestimate this). Also, “small” candidates like Philippe Poutou and Nathalie Arthaud are not in the poll, but the vote transfers from these candidates are small anyway.
Scenario 1: Macron (61%) vs. Le Pen (39%)
The biggest addition to the Le Pen vote in a duel Le Pen-Macron would come from Fillon, but an even greater share of the Fillon voters say they would vote for Macron. Macron has the big advantage of drawing from left, right and center.
Scenario 2: Macron (64%) vs. Fillon (36%)
Fillon is a very weak candidate against Macron because he doesn’t draw any votes at all from the left. Mélenchon and Hamon voters loathe him, and the share of Le Pen voters going to him is not big enough.
Scenario 3: Fillon (55%) vs. Le Pen (45%)
Fillon-Le Pen is the tightest race against Le Pen. Fillon would still win, but he is very unpopular among Mélenchon and Hamon voters (a bit less among Macron voters). Because there is no left candidate, this is the race that in theory produces the most abstention (a large share of Hamon and Mélenchon voters don’t express a preference, meaning that for them, Fillon or Le Pen doesn’t make much of a difference). This is quite different from 2002, where Chirac’s more centrist position was less of a repellent for left voters.
Scenario 4: Le Pen (43%) vs. Mélenchon (57%)
Mélenchon-Le Pen is the most polarised race. Mélenchon would win by drawing on his good standing among Macron and Hamon voters, even if he doesn’t get many votes from the right (Fillon). Le Pen, again, could only draw from a transfer of votes from Fillon. It would be too small to beat Mélenchon, however.
Scenario 5: Macron (57%) – Mélenchon (43%)
Macron wins against Mélenchon because he appeals to the right, and could rely on a substantial transfer of votes from Fillon and even Le Pen. There would also be an interesting transfer of votes from Le Pen to Mélenchon, whose radical left position apparently appeals to some of Le Pen’s working class base. Fillon voters wouldn’t transfer much to Mélenchon, preferring to abstain.
Scenario 6: Fillon (42%) – Mélenchon (58%)
Mélenchon is actually quite popular among Macron voters, would draw on almost all Hamon voters, while Fillon only appeals to his own base and some Le Pen voters. As things stand, Mélenchon would beat Fillon even if he is currently polling behind him in the first round.