Background Gun violence research is characterized by a dearth of data available for measuring key... more Background Gun violence research is characterized by a dearth of data available for measuring key constructs. Social media data may offer a potential opportunity to significantly reduce that gap, but developing methods for deriving firearms-related constructs from social media data and understanding the measurement properties of such constructs are critical precursors to their broader use. Objective This study aimed to develop a machine learning model of individual-level firearm ownership from social media data and assess the criterion validity of a state-level construct of ownership. Methods We used survey responses to questions on firearm ownership linked with Twitter data to construct different machine learning models of firearm ownership. We externally validated these models using a set of firearm-related tweets hand-curated from the Twitter Streaming application programming interface and created state-level ownership estimates using a sample of users collected from the Twitter ...
Harvard Kennedy School Misinformation Review, 2021
When U.S. presidential candidates misrepresent the facts, their claims get discussed across media... more When U.S. presidential candidates misrepresent the facts, their claims get discussed across media streams, creating a lasting public impression. We show this through a public performance: the 2020 presidential debates. For every five newspaper articles related to the presidential candidates, President Donald J. Trump and Joseph R. Biden Jr., there was one mention of a misinformation-related topic advanced during the debates. Personal attacks on Biden and election integrity were the most prevalent topics across social media, newspapers, and TV. These two topics also surfaced regularly in voters’ recollections of the candidates, suggesting their impression lasted through the presidential election.
Abrupt changes in mortality rates and life expectancy allow us to understand how shocks like COVI... more Abrupt changes in mortality rates and life expectancy allow us to understand how shocks like COVID-19 can exacerbate health inequalities across groups. We look at Washington, D.C., a major city with a diverse population and long-standing socio-economic divisions, to describe the all-cause mortality trends from 2015 to 2021 by age, sex, race, and ward of residence. We report differences in cause-specific mortality pre- and post-COVID-19 outbreak and estimate the Years of Life Lost (YLL) attributable to COVID-19. We compute death rates using information from death certificates and the Census, and we calculate YLL using the life table approach, comparing the life expectancy of people with and without COVID-19. We find that in 2020 and 2021, there were respectively 1,128 and 629 excess deaths (158 per 100K and 94 per 100K) compared to the annual average over the previous five years, and 689 and 363 deaths in 2020 and 2021, respectively (97 per 100K and 54 per 100K) listing COVID-19 as a...
The social sciences face growing demand for reproducible tools for processing massive troves of o... more The social sciences face growing demand for reproducible tools for processing massive troves of often-complex text data (political speeches, medical notes, etc.). In response, we aim toward computational literature review by developing an inductive method of applying expert-built dictionaries for automated analysis of complex texts. Our workflow begins with developing dictionaries from foundational texts and domain expertise. Next, we apply text-analytic methods of differential domain-specificity and complexity to create vector-space representations of texts. Finally, we compare the validity of these methods by using regression models to evaluate relationships between their representations and ground truth. Taking as our use case a large corpus of academic articles in organizational science, we find that domain-specific, relatively simple embeddings were most valid--while the more sophisticated models were very weak. Thus, we suggest that social science workflows for learning from c...
Research shows that charter schools are more segregated by race and class than traditional public... more Research shows that charter schools are more segregated by race and class than traditional public schools. I investigate an under-examined mechanism for this segregation: Charter schools project identities corresponding to parents’ race- and class-specific parenting styles and educational values. I use computational text analysis to detect the emphasis on inquiry-based learning in the websites of all charter schools operating in the 2015-16 school year. I then estimate mixed linear regression models to test the relationships between ideological emphasis and school- and district-level poverty and ethnicity. I thereby transcend methodological problems in scholarship on charter school identities by collecting contemporary, valid, population-wide data, as well as by blending text analysis with hypothesis testing. Findings suggest that charter school identities are both race- and class-specific, lending weight to arguments for further regulating charter school enrollments. This project c...
Industrial Relations: A Journal of Economy and Society
Is employee involvement universally either good or bad, a “best practice” or an exploitative tool... more Is employee involvement universally either good or bad, a “best practice” or an exploitative tool—or do its effects depend on context? To shed light on this issue, I ask the following question: Do organizational–cultural factors determine whether employees are stressed by membership in teams? By constructing mixed-effects models from a large mid-1990s survey of U.S. steel employees, I find that team membership is linked to increased stress only when implemented in cultural contexts of conflict and distrust. I conclude that the unintended consequences of institutionalized formal practices depend on organizationally specific cultural conditions.
Industrial Relations: A Journal of Economy and Society
Is employee involvement universally either good or bad, a “best practice” or an exploitative tool... more Is employee involvement universally either good or bad, a “best practice” or an exploitative tool—or do its effects depend on context? To shed light on this issue, I ask the following question: Do organizational–cultural factors determine whether employees are stressed by membership in teams? By constructing mixed-effects models from a large mid-1990s survey of U.S. steel employees, I find that team membership is linked to increased stress only when implemented in cultural contexts of conflict and distrust. I conclude that the unintended consequences of institutionalized formal practices depend on organizationally specific cultural conditions.
Background Gun violence research is characterized by a dearth of data available for measuring key... more Background Gun violence research is characterized by a dearth of data available for measuring key constructs. Social media data may offer a potential opportunity to significantly reduce that gap, but developing methods for deriving firearms-related constructs from social media data and understanding the measurement properties of such constructs are critical precursors to their broader use. Objective This study aimed to develop a machine learning model of individual-level firearm ownership from social media data and assess the criterion validity of a state-level construct of ownership. Methods We used survey responses to questions on firearm ownership linked with Twitter data to construct different machine learning models of firearm ownership. We externally validated these models using a set of firearm-related tweets hand-curated from the Twitter Streaming application programming interface and created state-level ownership estimates using a sample of users collected from the Twitter ...
Harvard Kennedy School Misinformation Review, 2021
When U.S. presidential candidates misrepresent the facts, their claims get discussed across media... more When U.S. presidential candidates misrepresent the facts, their claims get discussed across media streams, creating a lasting public impression. We show this through a public performance: the 2020 presidential debates. For every five newspaper articles related to the presidential candidates, President Donald J. Trump and Joseph R. Biden Jr., there was one mention of a misinformation-related topic advanced during the debates. Personal attacks on Biden and election integrity were the most prevalent topics across social media, newspapers, and TV. These two topics also surfaced regularly in voters’ recollections of the candidates, suggesting their impression lasted through the presidential election.
Abrupt changes in mortality rates and life expectancy allow us to understand how shocks like COVI... more Abrupt changes in mortality rates and life expectancy allow us to understand how shocks like COVID-19 can exacerbate health inequalities across groups. We look at Washington, D.C., a major city with a diverse population and long-standing socio-economic divisions, to describe the all-cause mortality trends from 2015 to 2021 by age, sex, race, and ward of residence. We report differences in cause-specific mortality pre- and post-COVID-19 outbreak and estimate the Years of Life Lost (YLL) attributable to COVID-19. We compute death rates using information from death certificates and the Census, and we calculate YLL using the life table approach, comparing the life expectancy of people with and without COVID-19. We find that in 2020 and 2021, there were respectively 1,128 and 629 excess deaths (158 per 100K and 94 per 100K) compared to the annual average over the previous five years, and 689 and 363 deaths in 2020 and 2021, respectively (97 per 100K and 54 per 100K) listing COVID-19 as a...
The social sciences face growing demand for reproducible tools for processing massive troves of o... more The social sciences face growing demand for reproducible tools for processing massive troves of often-complex text data (political speeches, medical notes, etc.). In response, we aim toward computational literature review by developing an inductive method of applying expert-built dictionaries for automated analysis of complex texts. Our workflow begins with developing dictionaries from foundational texts and domain expertise. Next, we apply text-analytic methods of differential domain-specificity and complexity to create vector-space representations of texts. Finally, we compare the validity of these methods by using regression models to evaluate relationships between their representations and ground truth. Taking as our use case a large corpus of academic articles in organizational science, we find that domain-specific, relatively simple embeddings were most valid--while the more sophisticated models were very weak. Thus, we suggest that social science workflows for learning from c...
Research shows that charter schools are more segregated by race and class than traditional public... more Research shows that charter schools are more segregated by race and class than traditional public schools. I investigate an under-examined mechanism for this segregation: Charter schools project identities corresponding to parents’ race- and class-specific parenting styles and educational values. I use computational text analysis to detect the emphasis on inquiry-based learning in the websites of all charter schools operating in the 2015-16 school year. I then estimate mixed linear regression models to test the relationships between ideological emphasis and school- and district-level poverty and ethnicity. I thereby transcend methodological problems in scholarship on charter school identities by collecting contemporary, valid, population-wide data, as well as by blending text analysis with hypothesis testing. Findings suggest that charter school identities are both race- and class-specific, lending weight to arguments for further regulating charter school enrollments. This project c...
Industrial Relations: A Journal of Economy and Society
Is employee involvement universally either good or bad, a “best practice” or an exploitative tool... more Is employee involvement universally either good or bad, a “best practice” or an exploitative tool—or do its effects depend on context? To shed light on this issue, I ask the following question: Do organizational–cultural factors determine whether employees are stressed by membership in teams? By constructing mixed-effects models from a large mid-1990s survey of U.S. steel employees, I find that team membership is linked to increased stress only when implemented in cultural contexts of conflict and distrust. I conclude that the unintended consequences of institutionalized formal practices depend on organizationally specific cultural conditions.
Industrial Relations: A Journal of Economy and Society
Is employee involvement universally either good or bad, a “best practice” or an exploitative tool... more Is employee involvement universally either good or bad, a “best practice” or an exploitative tool—or do its effects depend on context? To shed light on this issue, I ask the following question: Do organizational–cultural factors determine whether employees are stressed by membership in teams? By constructing mixed-effects models from a large mid-1990s survey of U.S. steel employees, I find that team membership is linked to increased stress only when implemented in cultural contexts of conflict and distrust. I conclude that the unintended consequences of institutionalized formal practices depend on organizationally specific cultural conditions.
Uploads