 It started really by accident. I just read in the news that in Ukraine they elected a president who apparently had very little political experience, but he was playing a role on a TV show of a president who fights corruption. So I got intrigued and I thought, wow, this is quite a gamble. And I started digging into this literature on corruption and I realized that the problem is not so much corruption itself, but the lack of objective measures of corruption. That became already an academic research question. So I thought, what could I do? How can we develop instrumental variables that could be used to measure corruption when there are no other better measurements? One of the things that I read about corruption is that one form of corruption is to invite public officials to lavish banquets, let's say with excessive consumption of food and beverages. So I thought, OK, if somebody is going too often to those things, they might risk gaining weight. And once you gain extra weight, it might be difficult to get rid of those extra pounds. And so I hypothesized that maybe there is a positive correlation between obesity of politicians and the level of corruption in their country. And that's how the project was born. I collected an obscured frontal images. So this is like a passport photograph from the Internet of ministers of 15 post-Soviet republics. It's really simple. I just did a Google search, name, surname plus 2017, that's when the study was done. And why I focused on 15 post-Soviet republics? Initially I was much more ambitious and I wanted to do it for all countries. But very soon I realized that there is a practical limitation. It's easy to find pictures of a prime minister or minister of foreign affairs in international media. Once you look for pictures of ordinary politicians, like minister of agriculture, you've got to go to local media that are often not in English but in local language. So I picked up post-Soviet republics because I can read surnames in Cyrillic and that just made the job much easier. I wouldn't be able to do it, say, in Southeast Asia. So yeah, I collected those images and then what we did, I ran them through a computer vision algorithm. So this is essentially artificial neural network that is trained to recognize human faces on scanned photos. And then once it recognizes the faces, it is trained to learn how to associate a face with BMI index. So I had a database of many, many images with known BMI index. So the network was trained on those images. Once the training was over, there were a couple of millions of parameters to estimate as usual in these programs. Then I showed to the program the pictures of the ministers for whom I didn't know the BMI index and it estimated obesity of these cabinet ministers. So what did I find? The first surprise in finding was that estimated body mass index of cabinet ministers is generally quite high. About one third of them are severely obese with BMI index between 35 and 40. So it's really already almost a threatening medical condition. But the second more important finding was that estimated obesity of cabinet ministers was positively correlated with the level of perceived corruption in their country, according to conventional measures of corruption. So there is indeed a positive correlation between the two variables. Now, I only found correlation. I cannot really say anything on causation. Is it because they are ministers, so they may be more corrupt when they're elected to the office? And this is what's causing this severe obesity? Or it could be that they are elected already obese and then maybe they think they will never be elected again. So they try to profit from a short period when they are in power and take bribes. So causality can go either way. I say really nothing about it. I just found strong correlation. And funny enough, there was a slight negative correlation between obesity of politician and the obesity of general population. Because one argument that I often heard was, well, maybe in some countries they have these obese politicians because the general population is very obese. So they just elect people who more or less represent how the population is. So this graph, guys, really is a snapshot of main results. You can see here the scatterplot of corruption perception index, which is probably the most famous conventional measure of corruption published each year by Transparency International and the estimated median body mass index of cabinet ministers that I did with this computer vision software. And you see that there is really strong correlation between the two. So it's puzzling. But we could use this. For example, when conventional measures of corruption are not available, one example or one possible application might be local governments when we want to measure corruption of mares in little towns or villages. And on that local level, we normally don't have any measures of corruption. So we can try to measure the extra weight of local politicians. And that might give us some idea of what is the level of corruption there.