Conventional wisdom suggests that primary elections attract a loyal base of partisans, and consequently, political candidates must take more extreme positions to secure the nomination before moderating for the general electorate. Yet, the academic literature contains little empirical evidence on candidate positioning over the electoral process. I address this gap by using congressional candidates' tweets to quantify ideological rhetoric during the 2020 election using three different approaches. First, I adopt a data-driven approach to select the most partisan bigrams and specify a multinomial model of speech; second, I use a theoretically-derived dictionary to measure the frequency of moral values associated with political convictions; finally, I specify a natural language model using a deep learning architecture. I provide one of the first empirical analyses on the evolution of candidates' ideological messaging over the entire election cycle. I find statistically significant evidence of moderation among Republican but not Democratic candidates, with mixed support for greater movement in competitive general elections. I conclude that Republicans likely face stronger incentives than Democrats to employ extreme rhetoric in primaries and thus to moderate in general elections.

We study the effect of Facebook and Instagram access on political beliefs, attitudes, and behavior by randomizing a subset of 19,857 Facebook users and 15,585 Instagram users to deactivate their accounts for 6 weeks before the 2020 U.S. election. We report four key findings. First, both Facebook and Instagram deactivation reduced an index of political participation (driven mainly by reduced participation online). Second, Facebook deactivation had no significant effect on an index of knowledge, but secondary analyses suggest that it reduced knowledge of general news while possibly also decreasing belief in misinformation circulating online. Third, Facebook deactivation may have reduced self-reported net votes for Trump, though this effect does not meet our preregistered significance threshold. Finally, the effects of both Facebook and Instagram deactivation on affective and issue polarization, perceived legitimacy of the election, candidate favorability, and voter turnout were all precisely estimated and close to zero.

Several factors influence the decision to apply for the Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI) programs, which provided disability benefits to 13 million non-elderly individuals in the U.S. in April 2021. In this study, we examine how changes in the minimum wage affect the number of applications to both programs. Theoretically, a higher minimum wage could either increase or reduce the number applying for disability benefits. We leverage changes in the effective minimum wage across states and over time during the 2000-2015 period to estimate this causal effect. We find that the effect of the minimum wage is positive and statistically significant in our main specification: a one dollar increase in the minimum wage increases the total application rate for SSI and SSDI combined by 0.04 percentage points, an increase of 2 percent. Our estimates suggest that application rates to both programs would increase by approximately 10 percent if the federal minimum wage were increased to $15, with the largest increases expected in states that currently have the lowest minimum wages.

A central assumption in the behavioral sciences is that choice behavior generalizes enough across individuals that measurements from a sampled group can predict the behavior of the population. Following from this assumption, the unit of behavioral sampling or measurement for most neuroimaging studies is the individual; however, cognitive neuroscience is increasingly acknowledging a dissociation between neural activity that predicts individual behavior and that which predicts the average or aggregate behavior of the population suggesting a greater importance of individual differences than is typically acknowledged. For instance, past work has demonstrated that some, but not all, of the neural activity observed during value-based decision-making is able to predict not just individual subjects’ choices but also the success of products on large, online marketplaces – even when those two behavioral outcomes deviate from one another – suggesting that some neural component processes of decision-making generalize to aggregate market responses more readily across individuals than others do. While the bulk of such research has highlighted affect-related neural responses (i.e. in the nucleus accumbens) as a better predictor of group-level behavior than frontal cortical activity associated with the integration of more idiosyncratic choice components, more recent evidence has implicated responses in visual cortical regions as strong predictors of group preference. Taken together, these findings suggest a role of neural responses during early perception in reinforcing choice consistency across individuals and raise fundamental scientific questions about the role sensory systems in value-based decision-making processes. We use a multivariate pattern analysis approach to show that single-trial visually evoked electroencephalographic (EEG) activity can predict individual choice throughout the post-stimulus epoch; however, a nominally sparser set of activity predicts the aggregate behavior of the population. These findings support an account in which a subset of the neural activity underlying individual choice processes can scale to predict behavioral consistency across people, even when the choice behavior of the sample does not match the aggregate behavior of the population.

Patent applications and acceptances are a useful domain for assessing the state of innovation across various fields, including biomedical sciences, artificial intelligence, software services, and more. The number of patents filed per year has nearly doubled since 2000 with over 650,000 patent filed in the 2020 fiscal year. Until now, no large-scale corpus of patent filings exists for ML and NLP practitioners to leverage. The Harvard USPTO Patent Dataset (HUPD), consisting of over 4.5 million English-language utility patent applications filed between 2004 and 2018 is the first example of such a corpus. Unlike other, smaller but similar corpora, this dataset contains the inventor-submitted versions of patent applications as opposed to the final versions of granted patents, allowing for the usage of NLP techniques at the time of filing. Taking advantage of this rich data, we vary the metadata inputs to a number of NLP models to conduct an ablation study on the binary classification of filed patents (i.e. acceptance or rejection). Our best metadata-augmented model achieves 63.32% binary classification accuracy, outperforming the best language models from the HUPD paper as well as our baseline models. Yet, for some text fields our best model still cannot outperform bag-of-words models, likely due to specific qualitative linguistic features of these fields.

COVID-19 reshaped American housing markets by causing workers with the ability to work-from-home to leave city centers for the suburbs. Economists typically study the drivers of such changes using tabular economic data, like population density. But this misses out on much of urban life that has not been tabulated, such as the presence of skyscrapers or upscale amenities. We propose this context can be visually learned. Using a dataset of over 60,000 images of American cities from Google Street View, we predict post-Covid housing market performance using a variety of neural methods and show three main results. First, a pre-trained vision transformer fine-tuned on our data, is able to predict an ordinal post-Covid housing performance with 32% accuracy, which is substantially better than other vision models (for context, randomly picking labels achieves 5% accuracy). Second, saliency maps suggest that our models captures key urban features like skylines, the distance to large buildings, and sidewalk corners. Third, an ensemble model that combines the scores from our vision transformer with tabular economic data outperforms either approach individually, achieving an accuracy of 51%. Our results suggest urban imagery contain unique information relevant to how the pandemic affected housing markets.

Accurately predicting which areas require additional attention and assistance in the medical world is a critical task for augmenting global health outcomes. Doing so with limited data is of particular interest, as the least supported areas are least likely to have infrastructure in place to facilitate the acquisition of additional healthcare resources. In this paper, we use satellite images from NASA's Landsat database to predict malnourishment (operationalized by mean BMI) and child mortality rates. We assess the performance of a variety of computer vision models, task-agnostic MOSAIKS feature vectors, and metadata-enhanced fusion models on ordinal discretizations of our outcome variables. Our best model, a fine-tuned Vision Transformer pre-trained on ImageNet, achieves a balanced accuracy of 66.7%, doubling random chance, on a CDC-derived ordinal encoding of Body Mass Index (BMI), and an accuracy of 50.3%, a 51% improvement over the random baseline, on an ordinal encoding of child mortality rates. We provide suggestive evidence that the model attends to sociologically relevant aspects of the images, such as road networks and city/village layouts. Future work should entail further enhancements of the most promising Vision Transformer model, further qualitative analysis and visualization of the model's predictions, and extensions to other measures capturing a region's health outcomes.

Using over 220,000 project proposals scraped from Kickstarter, we predict whether a Kickstarter project proposal will succeed or fail in achieving its fundraising goal using only information from project launch. We evaluate the performance of various machine learning models for these predictions based on the project category, fundraising goal, and short descriptions of the proposed product. We achieve an accuracy rate of 83% with our best model.

California’s recidivism rate is among the highest in the nation. Over 65% of individuals released from prison return within three years; 73% of recidivists commit a new crime within a year of release. Steady employment is one of the strongest predictors that a formerly incarcerated individual will maintain a distance from crime. Employment provides both a reliable source of income and a predictable daily routine. Unfortunately, the stigma associated with having a criminal record remains a significant barrier to employment across industries. The unemployment rate among the formerly incarcerated is 27% nationally, meaning over one in four of those once in prison are unable to find employment. Improving employment prospects for those with criminal records can be done through two mechanisms — increasing in-prison job training or incentivizing employers to hire those formerly incarcerated. We see numerous drawbacks to vocational job training programs. Instead, we propose a California-based version of the federal Work Opportunity Tax Credit (WOTC) targeted to employers hiring the formerly incarcerated. The Formerly Incarcerated Persons Opportunity Tax Credit (FIP-OTC) would scale by county factors, targeting tax credits to areas where economic opportunity for the formerly incarcerated is most dire.

We propose Raising Incentives for Student Education (RISE). This program, administered by the Department of Education (DoE), will provide all federally unemployed individuals, aged 18-24, with a voucher to attend participating community colleges in their state for free. To participate, community colleges will apply to receive a grant to cover the expense of enrolling these students and designing COVID-19 flexible learning infrastructure. RISE serves multiple goals: it will provide unemployed youth with increased education, decrease the labor force supply glut, train individuals for industries in which they would create the greatest value post-COVID-19, and provide an injection of funding into the community college system. The current economic crisis is the result of systemic issues in our nation. That is why RISE will attempt to address the root causes of the current instability rather than simply create a stop-gap measure to stabilize the country until the end of the pandemic.

The United States is once again facing the “Curse of Bigness." American industry has become increasingly consolidated, allowing private power to grow unfettered. With the acquisitions of Instagram and WhatsApp and the cloning of Snapchat’s distinctive features, Facebook has positioned itself as the monopolist of the social networking and digital advertising industries. This monopoly status grants Facebook tremendous power over the American people with no mechanism for accountability. Although Facebook does not set exploitative prices as the traditional monopolist does, its tepid commitment to the protection of user data privacy and its role in endangering the sanctity of American democracy has induced significant non-pecuniary negative externalities on the American people. Furthermore, its market dominance has suffocated innovation, leading to the establishments of kill zones for aspiring startups. Legislation seems ill-suited to redress the repercussions of Facebook’s monopoly status. Instead, the United States government must return to its great tradition of decentralized power and embrace robust antitrust enforcement.

What is the role of the U.S. government in labor strikes? Should the federal government protect workers’ rights and allow —and perhaps even promote—unionization, as was advocated for by labor activists and progressives? Or, as industrialists and the American elite of the 19th century had hoped, should the government support corporate interests in quashing labor strife? Or, alternatively as free-market conservatives and capitalists would wish, should the government remain wholly neutral in disputes, intervening only when absolutely necessary? Is the latter even possible? American public policy prior to 1894 and George M. Pullman would certainly dictate that the government support corporate and commercial interests. Yet, the American Railway Union (ARU), formed in 1893 and championed by Eugene V. Debs vowed to push for labor reform. One of the first paternalistic company towns in the United States—Pullman, Illinois—was founded in 1881 and constructed by George Pullman to house the employees of his railroad car company, the Pullman Palace Car Company (PPCC). The Panic of 1893 resulted in a decreased demand for railroad cars and continental travel, and so Pullman cut worker wages, choosing to pay his shareholders instead. In response, the ARU initiated the Pullman Strike of 1894. Originally limited to the city of Pull- man, the strike capitalized on the widespread dissatisfaction that laborers felt toward their employers, spreading throughout the country. Although an immediate failure for the workers residing in Pullman, the strike ultimately advanced the position of labor in the United States. The strike led to the demise of “government by injunction,” thereby restricting federal intervention in dismantling labor strikes and disputes, and brought forth widespread labor reform, including the establishment and the strengthening of labor unions.