Prediction and explanation in social systems

From AcaWiki
Jump to: navigation, search

Citation: Jake M. Hofman, Amit Sharma, Duncan J. Watts (2017/02/03) Prediction and explanation in social systems. Science (RSS)
DOI (original publisher): 10.1126/science.aal3856
Semantic Scholar (metadata): 10.1126/science.aal3856
Sci-Hub (fulltext): 10.1126/science.aal3856
Internet Archive Scholar (search for fulltext): Prediction and explanation in social systems
Tagged:

Summary

In social science there is an emphasis on explanation ("identification of interpretable causal mechanisms" and "un-biased estimation of model parameters") rather than prediction. Two trends are changing this:

  • concern that unthinking "search for statistical significance" has contributed to replication crisis
  • interest of computational scientists in social science topics

Distinguishing between exploratory and confirmatory studies, with pre-registration of the latter, can mitigate well known pitfalls.

Open problem of theoretical limit to predictive accuracy for complex social systems should be interesting for both evaluating causual explanations and guiding whether further work on a predictive model or collection of more data would be valuable.

Overly emphasizing prediction could lead to spurious predictions. Authors suggest explanation and predictive models are not necessarily in conflict and suggest a hybrid approach, including minimizing generalization error through entire modeling process, and then searching for simpler and more interpretable versions of the model.