Response to comment on "Policy impacts of statistical uncertainty and privacy".

We offer our thanks to the authors for their thoughtful comments. Cui, Gong, Hannig, and Hoffman propose a valuable improvement to our method of estimating lost entitlements due to data error. Because we don't have access to the unknown, "true" number of children in poverty, our paper simulates data error by drawing counterfactual estimates from a normal distribution around the official, published poverty estimates, which we use to calculate lost entitlements relative to the official allocation of funds. But, if we make the more realistic assumption that the published estimates are themselves normally distributed around the "true" number of children in poverty, Cui et al.'s proposed framework allows us to reliably estimate lost entitlements relative to the unknown, ideal allocation of funds-what districts would have received if we knew the "true" number of children in poverty.

Policy impacts of statistical uncertainty and privacy.

Funding formula reform may help address unequal impacts of uncertainty from data error and privacy protections.

How privacy's past may shape its future.

An account of privacy's evolutionary roots may hold lessons for policies in the digital age.

Differential discounting and present impact of past information.

How does information about a person's past, accessed now, affect individuals' impressions of that person? In 2 survey experiments and 2 experiments with actual incentives, we compare whether, when evaluating a person, information about that person's past greedy or immoral behaviors is discounted similarly to information about her past generous or moral behaviors. We find that, no matter how far in the past a person behaved greedily or immorally, information about her negative behaviors is hardly discounted at all. In contrast, information about her past positive behaviors is discounted heavily: recent behaviors are much more influential than behaviors that occurred a long time ago. The lesser discounting of information about immoral and greedy behaviors is not caused by these behaviors being more influential, memorable, extreme, or attention-grabbing; rather, they are perceived as more diagnostic of a person's character than past moral or generous behaviors. The phenomenon of differential discounting of past information has particular relevance in the digital age, where information about people's past is easily retrieved. Our findings have significant implications for theories of impression formation and social information processing. (PsycINFO Database Record

Privacy and human behavior in the age of information.

This Review summarizes and draws connections between diverse streams of empirical research on privacy behavior. We use three themes to connect insights from social and behavioral sciences: people's uncertainty about the consequences of privacy-related behaviors and their own preferences over those consequences; the context-dependence of people's concern, or lack thereof, about privacy; and the degree to which privacy concerns are malleable­manipulable by commercial and governmental interests. Organizing our discussion by these themes, we offer observations concerning the role of public policy in the protection of privacy in the information age.

"Heads or tails?"--a reachability bias in binary choice.

When asked to mentally simulate coin tosses, people generate sequences that differ systematically from those generated by fair coins. It has been rarely noted that this divergence is apparent already in the very 1st mental toss. Analysis of several existing data sets reveals that about 80% of respondents start their sequence with Heads. We attributed this to the linguistic convention describing coin toss outcomes as "Heads or Tails," not vice versa. However, our subsequent experiments found the "first-toss" bias reversible under minor changes in the experimental setup, such as mentioning Tails before Heads in the instructions. We offer a comprehensive account in terms of a novel response bias, which we call reachability. It is more general than the 1st-toss bias, and it reflects the relative ease of reaching 1 option compared to its alternative in any binary choice context. When faced with a choice between 2 options (e.g., Heads and Tails, when "tossing" mental coins), whichever of the 2 is presented first by the choice architecture (hence, is more reachable) will be favored. This bias has far-reaching implications extending well beyond the context of randomness cognition; in particular, to binary surveys (e.g., accept vs. reject) and tests (e.g., True-False). In binary choice, there is an advantage to what presents first.

"I cheated, but only a little": partial confessions to unethical behavior.

Confessions are people's way of coming clean, sharing unethical acts with others. Although confessions are traditionally viewed as categorical-one either comes clean or not-people often confess to only part of their transgression. Such partial confessions may seem attractive, because they offer an opportunity to relieve one's guilt without having to own up to the full consequences of the transgression. In this article, we explored the occurrence, antecedents, consequences, and everyday prevalence of partial confessions. Using a novel experimental design, we found a high frequency of partial confessions, especially among people cheating to the full extent possible. People found partial confessions attractive because they (correctly) expected partial confessions to be more believable than not confessing. People failed, however, to anticipate the emotional costs associated with partially confessing. In fact, partial confessions made people feel worse than not confessing or fully confessing, a finding corroborated in a laboratory setting as well as in a study assessing people's everyday confessions. It seems that although partial confessions seem attractive, they come at an emotional cost.

Reputation as a sufficient condition for data quality on Amazon Mechanical Turk.

Data quality is one of the major concerns of using crowdsourcing websites such as Amazon Mechanical Turk (MTurk) to recruit participants for online behavioral studies. We compared two methods for ensuring data quality on MTurk: attention check questions (ACQs) and restricting participation to MTurk workers with high reputation (above 95% approval ratings). In Experiment 1, we found that high-reputation workers rarely failed ACQs and provided higher-quality data than did low-reputation workers; ACQs improved data quality only for low-reputation workers, and only in some cases. Experiment 2 corroborated these findings and also showed that more productive high-reputation workers produce the highest-quality data. We concluded that sampling high-reputation workers can ensure high-quality data without having to resort to using ACQs, which may lead to selection bias if participants who fail ACQs are excluded post-hoc.

Predicting Social Security numbers from public data.

Information about an individual's place and date of birth can be exploited to predict his or her Social Security number (SSN). Using only publicly available information, we observed a correlation between individuals' SSNs and their birth data and found that for younger cohorts the correlation allows statistical inference of private SSNs. The inferences are made possible by the public availability of the Social Security Administration's Death Master File and the widespread accessibility of personal information from multiple sources, such as data brokers or profiles on social networking sites. Our results highlight the unexpected privacy consequences of the complex interactions among multiple data sources in modern information economies and quantify privacy risks associated with information revelation in public forums.