An interesting book focused on big data and data analysis. The author emphasizes growing importance of big data in social sciences and social science research. The underlying message is that data is everywhere. You can find several sets of data from Google searches, Facebook and other social networks, but also from thousands of pages from published books, pictures, songs etc. All this data can be subject to a scientific research and given their sheer volume, the outcomes should definitely have some statistical importance, in particular compared to a traditional social science research based on a small group of participants (mostly grad students). Stephens-Davidowitz further suggests that such data may additionally be carrying more truth (digital truth serum) as opposed to those submitted by people in traditional questionaires.

The author pointed out some of the interesting findings that arised from his own research. In particular, contrary to what has many tradional public opinion poll makers had drawn from Barrack Obama being elected a president (having interpreted it practically as the end of racism), the Google searches on that topic had strongly indicated otherwise (that it actually raised the level of racism within the US society). Further on, contrary to popular beliefs (as it is, for example, claimed in a recent Netflix documentary Social Dilemma), the author's findings suggests that the proliferation of social media does not significantly participate on polarization of the society by locking the people within their own social groups and preventing them from comming across differing opinions or views. The liberal democrat may enjoy The New York Times as much as the 'average' white supremacist. If true, this would indicate that our world and society is yet again much more complicated than what the popular belief may suggest.

From seeing what could be found in data, it may almost seem that everything is possible with the big data - answer to any question may be found with the help of data analysis. This is why I was very pleased that the author decided to point out also some of the shortcommings of the data-driven approach. To me, one of the biggest issues is what Stephens-Davidowitz describes as 'dimensionality problem'. In a sufficiently large data set, (spurious) correlation could be found between the two observed items without establising the causation between the two. Further on, with the advancement of technological progress, especially in the field of Internet companies, big data play huge role in making the algorithms even more effective at what they do. This may be achieved by the so called 'AB testing' which allows engineers in tech companies to test basically any aspect of their product or service on a large sample of users to find out which version meets its purpose the best. While this may be great for the companies, when mishandled, it may very well end up as a weapon against the customers (tool of surveillance capitalism).

To conclude, data-science and big data certainly will be a big deal in comming years. It already changes several aspects of our lives and it can be expected that this trend will only continue. schovat popis