Schedule - Parallel Session 6 - Big Data
WMG IMC Room 246 - 15:40 - 17:10Does Tv Consumption Affect Health and Well-Being? Evidence from a Natural Experiment
Adrian Chadi; Manuel Hoffmann
Abstract
Standard utility theory fails in the attempt to explain why people voluntarily decide to spend a large amount of time doing something that from multiple perspectives seems to be bad for them: watching television. The explanation given in previous economic discussions is that this is a case of irrational behavior and that people have self-control problems. However, a critical review of the literature reveals that even the most seminal studies on individual well-being do not establish causal evidence. Researchers typically observe all kinds of problems in the lives of those who spent much time on watching TV, but the direction of causality is unclear. The problem here is that it is extremely difficult to manipulate some people’s television consumption in significant ways to measure potential implications in comparison to a random group of unaffected individuals. To establish causality based on quasi-experimental evidence, we argue that regional heterogeneity in the provision of media during their implementation period can trigger differences in TV consumption that are free of selectivity issues. The happenstance that we exploit took place in the mid-1980s when Germany lifted a ban on private television. While TV consumption had already reached high levels in other countries, the average German still watched less than two hours of television per day. In the sequel, the legalization of private television brought up new channels that increased consumption significantly. However, citizens in many areas of the country did not watch any of these new programs due to reception problems, as the responsible public institution failed to establish satellite or cable TV in a timely manner. Hence, the officials of the emerging TV channels looked for other ways to reach the German households, and they found a way that establishes our natural experiment: terrestrial frequencies that by chance were still open. All across Germany, there are dozens of transmitter stations that were built in the 1960s and up to this day provide the country with free terrestrial TV signals. We exploit original data from the official records for all of Germany’s TV transmitters in the late 1980s. We merge our technical calculations on the reach of each station’s signal with data from the German Socio-Economic Panel (SOEP) study, which includes exact hours for someone’s time use on a typical day regarding a broad set of categories (job, housework, television, etc.). Most importantly, the data gives us the great opportunity to analyze the causal effects of watching TV for various standard outcomes in the fields of health and happiness. Our empirical evidence stands in contrast to the common belief and previous studies without (quasi-)experimental evidence. First, we find a significantly positive effect of television consumption on people’s well-being. Second, increased television consumption does not impair people’s health, at least not in the short-run.
Characterization of Privacy Loss
Gail Gilboa Freedman; Rann Smorodinsky; Kobbi Nissim
Abstract
How much privacy is lost by a mechanism? This question is not a new question. However, an un-equivocal answer is still missing in the literature. Our research aims to formalize the conceptual notion of privacy. We suggest a new measure for privacy-loss and rigorously justify this measure. The applicability of our research is for having a standard methodology of prioritizing mechanisms by their level of privacy-loss. We model any privacy-jeopardizing-mechanism abstractedly, as a signaling-matrix. The set of rows represents a set of types (secrets). The set of columns represents a set of signals. In each row there is a distribution, representing the association between the corresponding type and the signals. Publication of such matrices jeopardizes privacy to some level. Consider any two signaling matrices, it is natural to ask: which mechanism is preferable in the context of keeping the type secret? We follow a Decision-Theory-flavored approach and capture the natural properties for such decisions by a list of ordinal axioms. These axioms imply a (subjective) preference relation model. We identify that f-divergence represents this preference relation model, and therefore can serve as a natural measure of privacy-loss. f-divergence is familiar in the realm of information-theory for quantifying the distance between a pair of probability distributions. We also follow a reverse direction of the axiomatization approach in order to characterize differential privacy, which is a standard privacy-loss measure in the computer- science literature. We give a rigorous construction for this measure and examine if it stems from natural behavioral axioms or not.
Behavioural Data Analytics
Ganna Pogrebna
Abstract
Big Data analytics is an established methodology used primarily by statisticians and computer scientists to analyse large masses of data. While the Big Data methodology is helpful for describing data patterns as well as for determining the main trends in the data, it is hard to apply existing Big Data techniques to predict human behaviour observed in the field. In this paper we introduce the concept of behavioural data analytics and show how decision theoretic approach can be combined with Big Data methodology to analyse and predict human behaviour using large datasets. All behavioural data analytics methods used in the paper are supported by the real-world examples.
Expectations and Emotions: A Field Study with Big Data
Sudeep Bhatia; Barbara Mellers
Abstract
Decision affect theory (Mellers et al., 1997) proposes that expectations influence emotional reactions in the presence of uncertainty, with surprising outcomes generating the strongest affective responses (see also Bell, 1985; Loomes & Sugden, 1986). Expectations have also been hypothesized to play a role in determining reference points in economic decisions, in influencing the strength of conditioning in reinforcement learning tasks, and in guiding dopamine-based learning in the brain (Kõszegi & Rabin, 2006; Schultz et al., 1997; Sutton & Barto, 1998). In all of these domains, decision makers’ beliefs about the occurrence of uncertain events affects the ways in which outcomes are processed. Most existing work documenting the above relationship has used laboratory experiments. In this paper, we tested this relationship in the field, with the microblogging platform Twitter.com. Study 1 examined tweets made about National Football League (NFL) teams in the 2014/2015 NFL season. We downloaded all tweets referencing individual teams made within a 24 hour window of games played during this season. We also obtained expectations for these games using point spreads offered by popular betting websites. Our final dataset consisted of 7,515,023 tweets, which we coded for affect using standard sentiment analysis techniques. Our goal was to examine how the affective content of the tweets for a team after each game was influenced by the expected scores (as specified by the point spread) and the final scores. Overall, we found that tweets for teams that strongly beat their expectations greatly increased in affect after the game, and tweets for teams that strongly fell short of their expectations greatly decreased in affect after the game, consistent with decision affect theory. There was little change in tweet affect when teams performed as expected. A formal analysis controlling for the various game outcomes (i.e., winning or losing), as well as other relevant variables, showed a strong effect of expectations on tweet affect. Study 2 tested for the robustness of these results. It obtained 372,981 tweets referencing individual candidates in the 2014 US Senate elections. We examined changes in tweet affect after the election as a function of expectations for the candidates winning or losing. Expectations were formalized using the predictions of political forecasters. As in Study 1, there was a significant effect of expectations, with unexpected winners and losers being associated with the largest changes in tweet affect. Again, this is consistent with decision affect theory. Overall, our results show that expectations strongly influence reward processing in sports games and elections, and demonstrate that existing expectation-based theories of emotion make successful predictions in large-scale real-world settings. In doing so, they show how naturally occurring online data can be used to test psychological decision theories in the field.