During my stint at graduate school I had a few papers published. Computational social science is a pretty broad field; my work gravitated towards psychology and statistics. However I was fortunate to learn things within other subjects including machine learning, social networks, and complex systems.
Success stories typically focus on just a few examples to try to teach people about success. Unfortunately, this is just a form of cheating with statistics, because a few examples will tell you nothing about a general pattern.
We ran an experiment with over a thousand participants and found that people were strongly influenced by these kinds of anecdotes. By randomly showing participants a few dropouts who became successful startup founders, we caused them to bet on a dropout becoming a successful startup founder (and vice versa for graduates). Participants reported medium to high confidence in their bets and many justified their bets with an explanation for why a dropout (or graduate) might be succeed.
Lifchits, George, Ashton Anderson, Daniel G. Goldstein, Jake M. Hofman, and Duncan J. Watts. 2021. “Success Stories Cause False Beliefs about Success.” Judgment and Decision Making 16 (6): 25. https://sjdm.org/journal/21/210225/jdm210225.pdf.
@article{lifchits_success_2021,
title = {Success Stories Cause False Beliefs about Success},
author = {Lifchits, George and Anderson, Ashton and Goldstein, Daniel G. and Hofman, Jake M. and Watts, Duncan J.},
date = {2021},
journaltitle = {Judgment and Decision Making},
volume = {16},
pages = {25},
langid = {english},
number = {6},
url = {https://sjdm.org/journal/21/210225/jdm210225.pdf}
}
We had access to a unique dataset of thousands of A/B tests on headlines from a popular news website called Upworthy. With this experimental data, we can investigate the linguistic properties that drive the success of text.
This is a registered report in which we declared our hypotheses based on a small exploratory dataset. Since we didn't have access to the full data, we avoid the problems of p-hacking or overfitting. Our hypotheses will be rigorously tested againsted the full data.
Gligorić, Kristina, George Lifchits, Robert West, and Ashton Anderson. 2021. “Linguistic Effects on News Headline Success: Evidence from Thousands of Online Field Experiments (Registered Report Protocol).” PLoS ONE 16, no. 9 (September 15, 2021): e0257091. https://doi.org/10.1371/journal.pone.0257091.
@article{gligoric_linguistic_2021,
title = {Linguistic Effects on News Headline Success: Evidence from Thousands of Online Field Experiments (Registered Report Protocol)},
shorttitle = {Linguistic Effects on News Headline Success},
author = {Gligori\'c, Kristina and Lifchits, George and West, Robert and Anderson, Ashton},
editor = {Lev-Ari, Shiri},
date = {2021-09-15},
journaltitle = {PLoS ONE},
shortjournal = {PLoS ONE},
volume = {16},
pages = {e0257091},
issn = {1932-6203},
doi = {10.1371/journal.pone.0257091},
langid = {english},
number = {9}
}
Every day, humans learn different classes of functional relationships, like linear, periodic, exponential, and so on.
We ran an experiment where we asked participants to learn a function by choosing a limited number of samples. People generally followed a process of sampling where the uncertainty was the highest (similar to a standard Gaussian Process model) and those who followed this strategy more closely were better at learning non-linear functions.
Gelpí, Rebekah A., Nayan Saxena, George Lifchits, Daphna Buchsbaum, and Christopher G. Lucas. 2021. “Sampling Heuristics for Active Function Learning.” Proceedings of the 19th International Conference on Cognitive Modeling. https://mathpsych.org/presentation/568.
@inproceedings{gelpi_sampling_2021,
title = {Sampling Heuristics for Active Function Learning},
author = {Gelpi, Rebekah A. and Saxena, Nayan and Lifchits, George and Buchsbaum, Daphna and Lucas, Christopher G.},
date = {2021},
url = {https://mathpsych.org/presentation/568},
eventtitle = {Proceedings of the 19th International Conference on Cognitive Modeling}
}