Ron Berman

The Role of Search Engine Optimization in Search Marketing

This paper examines the impact of search engine optimization SEO on the competition between advertisers for organic and sponsored search results. The results show that a positive level of search engine optimization may improve the search engines ranking quality and thus the satisfaction of its visitors. In the absence of sponsored links, the organic ranking is improved by SEO if and only if the quality provided by a website is sufficiently positively correlated with its valuation for consumers. In the presence of sponsored links, the results are accentuated and hold regardless of the correlation. When sponsored links serve as a second chance to acquire clicks from the search engine, low-quality websites have a reduced incentive to invest in SEO, giving an advantage to their high-quality counterparts. As a result of the high expected quality on the organic side, consumers begin their search with an organic click. Although SEO can improve consumer welfare and the payoff of high-quality sites, we find that the search engines revenues are typically lower when advertisers spend more on SEO and thus less on sponsored links. Modeling the impact of the minimum bid set by the search engine reveals an inverse U-shaped relationship between the minimum bid and search engine profits, suggesting an optimal minimum bid that is decreasing in the level of SEO activity.

Provable Unlinkability against Traffic Analysis

Chaum [1, 2] suggested a simple and efficient protocol aimed at providing anonymity in the presence of an adversary watching all communication links. Chaum’s protocol is known to be insecure. We show that Chaum’s protocol becomes secure when the attack model is relaxed and the adversary can control at most 99% of communication links.

Provable Unlinkability Against Traffic Analysis with Low Message Overhead

Rackoff and Simon proved that a variant of Chaum’s protocol for anonymous communication, later developed as the Onion Routing Protocol, is unlinkable against a passive adversary that controls all communication links and most of the nodes in a communication system. A major drawback of their analysis is that the protocol is secure only if (almost) all nodes participate at all times. That is, even if only n≪N nodes wish to send messages, allN nodes have to participate in the protocol at all times. This suggests necessity of sending dummy messages and a high message overhead.Our first contribution is showing that this is unnecessary. We relax the adversary model and assume that the adversary only controls a certain fraction of the communication links in the communication network. We think this is a realistic adversary model. For this adversary model we show that a low message overhead variant of Chaum’s protocol is provably secure.Furthermore, all previous security proofs assumed the a priori distribution on the messages is uniform. We feel this assumption is unrealistic. The analysis we give holds for any a priori information on the communication distribution. We achieve that by combining Markov chain techniques together with information theory tools in a simple and elegant way.

p-Hacking and False Discovery in A/B Testing

We investigate to what extent online A/B experimenters “p-hack” by stopping their experiments based on the p-value of the treatment effect, and how such behavior impacts the value of the experimental results. Our data contains 2,101 commercial experiments in which experimenters can track the magnitude and significance level of the effect every day of the experiment. We use a regression discontinuity design to detect the causal effect of reaching a particular p-value on stopping behavior. Experimenters indeed p-hack, at times. Specifically, about 73% of experimenters stop the experiment just when a positive effect reaches 90% confidence. Also, approximately 75% of the effects are truly null. Improper optional stopping increases the false discovery rate (FDR) from 33% to 40% among experiments p-hacked at 90% confidence. Assuming that false discoveries cause experimenters to stop exploring for more effective treatments, we estimate the expected cost of a false discovery to be a loss of 1.95% in lift, which corresponds to the 76th percentile of observed lifts.

The Impact of Time Horizon on the Effect of Diversification

Although investors are often advised to diversify their investment portfolios as well as to consider rebalancing them periodically, research has shown that they often ignore this advice. We try to determine if this behavior is rational by analyzing a risk-averse investor who chooses between buy-and-hold portfolios comprised of assets with dynamic uncertain returns. The assets in the portfolio evolve according to multiplicative random walks, distinguishing them from the traditional one-shot or additive models. Solving for the optimal choice, we find an interaction between diversification and the time horizon an investor is facing. This interaction results in conditions for which an optimal portfolio in one time horizon becomes suboptimal in a longer (or shorter) horizon. Moreover, we find that rebalancing may be suboptimal if the portfolio is diversified enough. Such effects are a consequence of the non-ergodicity of the value of assets that follow multiplicative dynamics. Thus we are able to provide a rational explanation for observed behavior of investors and subjects in lab experiments who choose to not diversify their portfolios or do not rebalance as often as the standard theory would prescribe.

The Impact of Curation Algorithms on Social Network Content Quality and Structure

Social platforms often use curation algorithms to match content to user tastes. Although designed to improve user experience, these algorithms have been blamed for increased polarization of consumed content. We analyze how curation algorithms impact the number of friends users follow and the quality of content generated on the network, taking into account horizontal and vertical differentiation. Although algorithms increase polarization for fixed networks, when they indirectly influence network connectivity and content quality their impact on polarization and segregation is less clear. We find that network connectivity and content quality are strategic complements, and that introducing curation makes consumers less selective and increases connectivity. In equilibrium, content creators receive lower payoffs because they enter into a contest leading to a prisoner’s dilemma. Filter bubbles are not always a consequence of curation algorithms. A perfect filtering algorithm increases content polarization and creates a filter bubble when the marginal cost of quality is low, while an algorithm focused on vertical content quality increases connectivity as well as lowers polarization and does not create a filter bubble. Consequently, although user surplus can increase through curating and encouraging high quality content, the type of algorithm used matters for the unintended consequence of creating a filter bubble.