Beyond prediction: Using big data for policy problems

Athey, Susan

doi:10.1126/science.aal4321

Cited by 444 publications

(278 citation statements)

References 18 publications

Supporting

Mentioning

273

Contrasting

Unclassified

Order By: Relevance

“…If one believes that prediction is a chief objective of modelling, one might look to a different strand of literature on linkprediction, originating in the physical sciences (see Lü & Zhao 2011) and not aiming at explaining processes of change. Today, experts in machine learning can efficiently solve such prediction tasks, but high predictive accuracy comes at the cost of little insight (Athey, 2017;Breiman, 2001). …”

Section: Resultsmentioning

confidence: 99%

Change we can believe in: Comparing longitudinal network models on consistency, interpretability and predictive power

et al. 2018

View full text Add to dashboard Cite

While several models for analysing longitudinal network data have been proposed, their main differences, especially regarding the treatment of time, have not been discussed extensively in the literature. However, differences in treatment of time strongly impact the conclusions that can be drawn from data. In this article we compare auto-regressive network models using the example of TERGMs -an extensions of ERGMs -and process-based models using SAOMs as an example. We conclude that the basic TERGM, in contrast to the ERGM, has no consistent micro-level interpretation, and thus only allows interpretation on the level of the network. Further, parameters in the TERGM are strongly dependent on the interval length between two time-points. Neither limitations is true for processbased network models such as the SAOM. Finally, both compared models perform poorly in out-of-sample prediction compared to trivial predictive models.

show abstract

Section: Resultsmentioning

confidence: 99%

Change we can believe in: Comparing longitudinal network models on consistency, interpretability and predictive power

et al. 2018

View full text Add to dashboard Cite

show abstract

“…This real-time data-abundance revolution has positively impacted public health (Khoury et al, 2014; Murdoch et al, 2013) and public health policy (Athey, 2017; Beam et al, 2018). As a result, the university community has escalated its graduate and undergraduate data-science curriculum to include big-data analysis tools to advance the understanding of the trends in social lifestyle diseases (Elgin et al, 2017; De Veaux et al, 2017).…”

Section: Introductionmentioning

confidence: 99%

Data Talks: Obesity-Related Influences on US Mortality Rates

D’Souza

Bautista

Wentzien

2018

RHS

View full text Add to dashboard Cite

Background: In the US, obesity is an epidemiologic challenge and the population fails to comprehend this complex public health issue. To evaluate underlying obesity-impact patterns on mortality rates, we data-mined the 1999–2016 Center for Disease Control WONDER database’s vital records. Methods: Adopting SAS programming, we scrutinized the mortality and population counts. Using ICD-10 diagnosis codes connected to overweight and obesity, we obtained the obesity-related crude and age-adjusted causes of death. To understand divergent and prevalence trends we compared and contrasted the tabulated obesity-influenced mortality rates with demographic information, gender, and age-related data. Key Results: From 1999 to 2016, the obesity-related age-adjusted mortality rates increased by 142%. The ICD-10 overweight and obesity-related death-certificate coding showed clear evidence that obesity factored in the male age-adjusted mortality rate increment to 173% and the corresponding female rate to 117%. It also disproportionately affected the nation-wide minority population death rates. Furthermore, excess weight distributions are coded as contributing features in the crude death rates for all decennial age-groups. Conclusions: The 1999–2016 data from ICD-10 death certificate coding for obesity-related conditions indicate that it is affecting all segments of the US population.

show abstract

“…I use Kernel Regularized Least Squares (KRLS) estimation, a machine learning approach (Hainmueller & Hazlett, 2013) which constructs algorithms that learn from data, merging these with a randomized experimental set-up. In so doing, the paper answers the call for work showing how statistical prediction approaches can support causal inference (Einav & Levin, 2014;Varian, 2014;Athey, 2017).…”

Section: Introductionmentioning

confidence: 99%

Priming Human-Computer Interactions: Experimental Evidence from Economic Development Mobile Surveys

Opoku-Agyemang

2017

SSRN Journal

View full text Add to dashboard Cite

This paper investigates how citizens from developing countries vocalize controversial topics, combining behavioral economics with human-computer interaction. I examine a priming effort to understand how people decide to discuss controversial local subjects, using the human-computer interaction of people with their mobile phones to quantify how attracted people feel to alternative local political economy topics when randomly asked what they think about international aid. The treatment significantly impacted the choice to discuss sanitation, health, poverty, democracy, individual determination, pro-poor support, and happiness. However, the intervention does not affect subjectively ranked preferences.Human-computer interaction approaches may help policy makers understand experimental outcomes.

show abstract

Beyond prediction: Using big data for policy problems

Cited by 444 publications

References 18 publications

Change we can believe in: Comparing longitudinal network models on consistency, interpretability and predictive power

Change we can believe in: Comparing longitudinal network models on consistency, interpretability and predictive power

Data Talks: Obesity-Related Influences on US Mortality Rates

Priming Human-Computer Interactions: Experimental Evidence from Economic Development Mobile Surveys

Contact Info

Product

Resources

About