We collected 299 frontal face images of 2017 cabinet ministers from 15 post-Soviet states (Armenia, Azerbaijan, Belarus, Estonia, Georgia, Kazakhstan, Kyrgyzstan, Latvia, Lithuania, Moldova, Russia, Tajikistan, Turkmenistan, Ukraine and Uzbekistan). For each image, the minister’s body-mass index is estimated using a computer vision algorithm.
The median estimated body-mass index of cabinet ministers is highly correlated with conventional measures of corruption (Transparency International Corruption Perceptions Index, World Bank worldwide governance indicator Control of Corruption, Index of Public Integrity).
This result suggests that physical characteristics of politicians such as their body-mass index can be used as proxy variables for political corruption when the latter are not available, for instance at a very local level.
[Keywords: body-mass index, computer vision, corruption, government, post-Soviet states]
…Dataset: We collected 299 frontal face images of cabinet ministers from 15 post-Soviet states who were in office in 2017.2 In case of a cabinet reshuffle, when 2 (sometimes even 3) individuals occupied the same ministerial position in 2017, we collected the image of the individual who occupied this position for the longest period in 2017.3 Country-specific details are presented in the Appendix. For each minister, we conducted a Google image search in the form “Name Surname” + 2017. The minister’s first name and surname were typed in the official language of his or her country (eg. in Cyrillic script for Belarus, Kazakhstan, Kyrgyzstan, Russia, Tajikistan and Ukraine). Whenever possible, we selected a minister’s image that resembled a passport photograph—unobscured frontal face image preferably taken during an event in 2017 (such as an official press conference, an official visit abroad or a meeting with a counterpart minister from another country)
Estimation: For each image in the dataset, the minister’s body-mass index is estimated using the computer vision algorithm recently developed by et al 2017.4 This algorithm is a 2-stage procedure. The first stage is a deep convolutional neural network VGG-Face developed by Parkhi, Vedaldi, and Zisserman (2015). This neural network extracts the features from a deep fully connected neuron layer fc6 for the input image. The second stage is an epsilon support vector regression (1997) of the extracted features to predict body-mass indexes of 3,368 training images (with known body-mass index values) collected by et al 2017.
…Estimated body-mass index for ministers in our dataset is generally quite high. According to the estimated body-mass index, 96⁄299 ministers (32%) are severely obese (estimated body-mass index 35–40). In particular, 13⁄24 Uzbek ministers (54%), 8 out 18 Tajik ministers (44%) and 10⁄24 Ukrainian ministers (42%) are estimated to be severely obese. Another 13⁄299 ministers in our dataset (4%) are very severely obese (estimated body-mass index greater than 40). In particular, 3⁄20 Kazakh ministers (15%) and 2⁄24 Ukrainian ministers (8%) are estimated to be very severely obese. Only 10⁄299 ministers in our dataset (3%) are estimated to have normal weight (body-mass index between 18.5 and 25). In particular, the governments of Azerbaijan, Estonia, Georgia, Kazakhstan, Latvia, Lithuania, Ukraine and Uzbekistan each have one minister with an estimated normal weight. None of the ministers in our dataset is estimated to be underweight (body-mass index below 18.5)
…A visual inspection of Table 1 confirms the intuition presented in §1—as ministers’ images in the third column get progressively more overweight and obese, conventional corruption indicators in the last 5 columns get progressively worse. Our median estimated ministers’ body-mass index is highly correlated with all 5 conventional measures of perceived corruption. The correlation coefficient with Transparency International Corruption Perceptions 2017, World Bank worldwide governance indicator ‘Control of Corruption’ 2017, European Research Centre for Anti-Corruption and State-Building Index of Public 2017, the sub-attribute ‘Absence of Corruption’ of Global State of Democracy 2017 and Basel Anti-Money Laundering is −0.92, −0.91, −0.93, −0.76 and 0.8, respectively.
…Our proposed methodology is widely applicable across countries as photographic data of top public officials are relatively accessible in traditional mass media and social media. This creates the potential of measuring corruption in many regions where administering reliable micro-level surveys is problematic and foreign experts have limited direct access. Our proposed corruption measure can be also applied retrospectively in time. This introduces for the first time, the possibility of measuring corruption from a historical perspective (before the mid-1990s when the first indexes of perceived corruption were constructed).