CoChat
Sign Up
Cited by 185Impact of Technology on AdolescentsDecision-Making and Behavioral EconomicsDigital Media and Philosophy

The DOI matches its Crossref record; 60 of the 61 references it cites were found in a registry or an index.

A computational reward learning account of social media engagement

Open access
Lindström, Björn, Bellander, Martin, Schultner, David T., Chang, Allen, Tobler, Philippe N., Amodio, David M.·2021·Nature Communications
DOI 10.1038/s41467-020-19607-xPMC PMC7910435
Open this paper in CoChat — free accountOpen PDF

Abstract

Social media has become a modern arena for human life, with billions of daily users worldwide. The intense popularity of social media is often attributed to a psychological need for social rewards (likes), portraying the online world as a Skinner Box for the modern human. Yet despite such portrayals, empirical evidence for social media engagement as reward-based behavior remains scant. Here, we apply a computational approach to directly test whether reward learning mechanisms contribute to social media behavior. We analyze over one million posts from over 4000 individuals on multiple social media platforms, using computational models based on reinforcement learning theory. Our results consistently show that human behavior on social media conforms qualitatively and quantitatively to the principles of reward learning. Specifically, social media users spaced their posts to maximize the average rate of accrued social rewards, in a manner subject to both the effort cost of posting and the opportunity cost of inaction. Results further reveal meaningful individual difference profiles in social reward learning on social media. Finally, an online experiment (n = 176), mimicking key aspects of social media, verifies that social rewards causally influence behavior as posited by our computational account. Together, these findings support a reward learning account of social media engagement and offer new insights into this emergent mode of modern human behavior.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-020-19607-x

Cites

60 of 60 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Schematic illustration of the computational hypothesis. a The R¯L model describes how τPost, the latency to the next social media post (denoted by the “camera” icon), is shaped by social rewards. Each post is followed by social reward (denoted by the “heart” symbol), which varies in number. The model adjusts the response policy, or threshold, which determines τPost, to maximize the average net rate of reward. b The R¯L model posits an effort cost to responding (e.g., taking pictures, uploading), which decreases as a function of τPost. The effort cost term penalizes posting in quick succession, because high effort reduces the average reward rate. c The opportunity cost of time increases as a function of the average reward rate R¯. The gradient of red lines indicates increasing values of R¯ (darker colors represent higher values), and thereby higher opportunity cost. d The optimal value of τPost, which maximizes the net reward δ, varies as function of R¯ (darker colors represent higher values). The δ is used to update average reward rate R¯. Note that the optimum, indicated by the peak of the function, moves to shorter response latencies when R¯ is higher because the opportunity cost of time increases with R¯. The horizontal line denotes 0. The figure assumes a constant effort cost C. e Simulated model predictions. The R¯L model predicts that τPost, the latency between successive social media posts, will be shorter with high compared to low average reward rate, R¯. The simulation involved N = 1000 independent synthetic individuals. The prediction is presented as estimated mean ± 99% CI from mixed-effects regression.
Fig. 1 Schematic illustration of the computational hypothesis. a The R¯L model describes how τPost, the latency to the next social media post (denoted by the “camera” icon), is shaped by social rewards. Each post is followed by social reward (denoted by the “heart” symbol), which varies in number. The model adjusts the response policy, or threshold, which determines τPost, to maximize the average net rate of reward. b The R¯L model posits an effort cost to responding (e.g., taking pictures, uploading), which decreases as a function of τPost. The effort cost term penalizes posting in quick succession, because high effort reduces the average reward rate. c The opportunity cost of time increases as a function of the average reward rate R¯. The gradient of red lines indicates increasing values of R¯ (darker colors represent higher values), and thereby higher opportunity cost. d The optimal value of τPost, which maximizes the net reward δ, varies as function of R¯ (darker colors represent higher values). The δ is used to update average reward rate R¯. Note that the optimum, indicated by the peak of the function, moves to shorter response latencies when R¯ is higher because the opportunity cost of time increases with R¯. The horizontal line denotes 0. The figure assumes a constant effort cost C. e Simulated model predictions. The R¯L model predicts that τPost, the latency between successive social media posts, will be shorter with high compared to low average reward rate, R¯. The simulation involved N = 1000 independent synthetic individuals. The prediction is presented as estimated mean ± 99% CI from mixed-effects regression.CC BY · PubMed Central
Fig. 2 Behavior on Instagram is explained by reward learning (Study 1). a Model comparison shows that the R¯L model explained behavior on Instagram (N = 2,039 independent individuals) better than a model without learning. The AICW expresses the relative likelihood for each model, and are presented as means ± 99% CI. The horizontal line at 0.5 represents the chance level of no difference between models. The distribution of AICW is displayed in Supplementary Fig. 3. Source data are provided as a Source Data file. b The model-derived estimate of R¯, the average reward rate, predicted the latency between posts (N = 2,039 independent individuals). As implicated by reward learning theory, the latency between posts was shorter with high compared to low R¯. Points indicate the corresponding estimates from synthetic data, based on ten generative simulation runs of the R¯L model (see text for details). The colored line denotes the average effect in the simulated data. Results are presented as means (fixed effects regression estimates) +/− 99% CI from mixed-effects regression. c, d Model fit for an example individual. c The posting history of an individual user over 673 days was well approximated by the R¯L model. The model policy (or posting threshold) denotes the average response latency predicted by the model at a given time point. The faded purple lines show 100 simulations of τPost from the estimated model policy, which illustrate the expected degree of variability given that policy, and how the empirical τPost falls within this range. The yellow line indicates the model estimate of the net reward rate, R¯. Note that a higher estimated R¯ is associated with shorter response latencies (τPost). See Supplementary Fig. 4 for additional example individuals. Source data are provided as a Source Data file. d The distribution of τPost for the same individual. The faded purple line shows 100 simulations of τPost from the estimated model policy. Source data are provided as a Source Data file.
Fig. 2 Behavior on Instagram is explained by reward learning (Study 1). a Model comparison shows that the R¯L model explained behavior on Instagram (N = 2,039 independent individuals) better than a model without learning. The AICW expresses the relative likelihood for each model, and are presented as means ± 99% CI. The horizontal line at 0.5 represents the chance level of no difference between models. The distribution of AICW is displayed in Supplementary Fig. 3. Source data are provided as a Source Data file. b The model-derived estimate of R¯, the average reward rate, predicted the latency between posts (N = 2,039 independent individuals). As implicated by reward learning theory, the latency between posts was shorter with high compared to low R¯. Points indicate the corresponding estimates from synthetic data, based on ten generative simulation runs of the R¯L model (see text for details). The colored line denotes the average effect in the simulated data. Results are presented as means (fixed effects regression estimates) +/− 99% CI from mixed-effects regression. c, d Model fit for an example individual. c The posting history of an individual user over 673 days was well approximated by the R¯L model. The model policy (or posting threshold) denotes the average response latency predicted by the model at a given time point. The faded purple lines show 100 simulations of τPost from the estimated model policy, which illustrate the expected degree of variability given that policy, and how the empirical τPost falls within this range. The yellow line indicates the model estimate of the net reward rate, R¯. Note that a higher estimated R¯ is associated with shorter response latencies (τPost). See Supplementary Fig. 4 for additional example individuals. Source data are provided as a Source Data file. d The distribution of τPost for the same individual. The faded purple line shows 100 simulations of τPost from the estimated model policy. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 3 Signatures of reward learning on three social media sites (Study 2). a–c Model comparison shows that the R¯L model explained behavior on the three social media sites (total N = 2,127 independent individuals. a N = 543, b N = 773, c N = 813) better than a model without learning. The AICW expresses the relative likelihood for each model, and are presented as means +/− 99% CI. The horizontal line at 0.5 represents the chance level of no difference between models. The exceedance probability for the R¯L model was 1 in all three datasets. The distribution of AICW is displayed in Supplementary Fig. 3. Source data are provided as a Source Data file. d–f The model derived estimate of R¯, the average reward rate, predicted the latency between posts on each social media platform (d N = 543, e N = 773, f N = 813 independent individuals). In line with reward learning theory, the latency between posts was shorter with high compared to low R¯. The colored points indicate the corresponding estimates from simulated data, based on ten generative simulation runs of the R¯L model (see text for details). The colored lines show the average effect in the simulated data. Results are presented as means (fixed effects regression estimates) ± 99% CI from mixed-effects regressions.
Fig. 3 Signatures of reward learning on three social media sites (Study 2). a–c Model comparison shows that the R¯L model explained behavior on the three social media sites (total N = 2,127 independent individuals. a N = 543, b N = 773, c N = 813) better than a model without learning. The AICW expresses the relative likelihood for each model, and are presented as means +/− 99% CI. The horizontal line at 0.5 represents the chance level of no difference between models. The exceedance probability for the R¯L model was 1 in all three datasets. The distribution of AICW is displayed in Supplementary Fig. 3. Source data are provided as a Source Data file. d–f The model derived estimate of R¯, the average reward rate, predicted the latency between posts on each social media platform (d N = 543, e N = 773, f N = 813 independent individuals). In line with reward learning theory, the latency between posts was shorter with high compared to low R¯. The colored points indicate the corresponding estimates from simulated data, based on ten generative simulation runs of the R¯L model (see text for details). The colored lines show the average effect in the simulated data. Results are presented as means (fixed effects regression estimates) ± 99% CI from mixed-effects regressions.CC BY · PubMed Central
Fig. 4 Computational phenotypes in reward learning on social media. a Cluster analysis (n = 4168 independent individuals from Study 1 & 2) of the estimated R¯L model parameters indicated that there were four distinct individual difference profiles in reward learning on social media. For illustration, the cluster assignments are plotted on the two first principal components (PC). Source data are provided as a Source Data file. b Profile of median R¯L model parameter values (standardized) for each cluster; ɑ = learning rate, P = initial response policy, C = effort cost sensitivity. Source data are provided as a Source Data file.
Fig. 4 Computational phenotypes in reward learning on social media. a Cluster analysis (n = 4168 independent individuals from Study 1 & 2) of the estimated R¯L model parameters indicated that there were four distinct individual difference profiles in reward learning on social media. For illustration, the cluster assignments are plotted on the two first principal components (PC). Source data are provided as a Source Data file. b Profile of median R¯L model parameter values (standardized) for each cluster; ɑ = learning rate, P = initial response policy, C = effort cost sensitivity. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 5 Experimental manipulation of social reward rates (Study 3). The estimated effect of social reward rate condition on posting latencies in the online experiment (n = 176 independent individuals, see Supplementary Fig. 5 for a design overview). Results are presented as means (fixed effects regression estimates) ±1 SE of the regression estimates from from mixed-effects regression.
Fig. 5 Experimental manipulation of social reward rates (Study 3). The estimated effect of social reward rate condition on posting latencies in the online experiment (n = 176 independent individuals, see Supplementary Fig. 5 for a design overview). Results are presented as means (fixed effects regression estimates) ±1 SE of the regression estimates from from mixed-effects regression.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 61 references

Cited by

Sorted by citations

Related

See all

Metadata

Authors
Lindström, B., Bellander, M., Schultner, D. T., Chang, A., Tobler, P. N., & Amodio, D. M.
Affiliations
  • University of Amsterdam (NL)
  • Karolinska Institutet (SE)
  • Boston University (US)
  • University of Zurich (CH)
  • New York University (US)
Corresponding author
Lindström, Björn
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2021-02-26
Volume
12
Issue
1
Pages
1311-1311
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-020-19607-x
PMC
PMC7910435
OpenAlex
W3130507978
Cited by
185
References
61
Citations by year
2026: 49 · 2025: 43 · 2024: 33 · 2023: 33 · 2022: 18 · 2021: 8
FWCI
31.88
Type
journal article
Language
en
Update notices
correction (2021-03-16)
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • VU Research Portal (publishedVersion, other-oa)
  • UvA-DARE (University of Amsterdam) (acceptedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Zurich Open Repository and Archive (University of Zurich) (submittedVersion, cc-by)
  • Data Archiving and Networked Services (DANS) (submittedVersion, other-oa)
Funders
  • National Science Foundation
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (100019, 176016)
  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek (016.Vici.185.058)
Topics
  • Social Sciences › Sociology and Political Science › Impact of Technology on Adolescents
  • Social Sciences › Decision Sciences › General Decision Sciences › Decision-Making and Behavioral Economics
  • Social Sciences › Communication › Digital Media and Philosophy
Keywords
social media engagement, reward learning, reinforcement learning, social reward, likes, posting behavior, effort cost, opportunity cost, individual differences, computational modeling, online experiments
MeSH terms
Female, Humans, Learning, Male, Reinforcement, Psychology, Reward, Social Behavior, Social Media

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review