Article Summary
Introduction
As political polarization in U.S. politics continues to rise, a common recipient of blame is the internet and social media. As these modern technologies become more ubiquitous in our daily lives, so too may their alleged ills; the creation of echo chambers, prioritization of sensationalized news, and dissemination of misinformation. Does the propensity to use the internet, then, potentially increase political polarization?
Boxell, Gentzkow, and Shapiro attempt to answer this question in their 2017 paper. Empirical literature suggests the mechanism behind the potential link between internet usage and polarization is online user segregation. Access to the internet appears correlated with consumption of partisan news and subsequent polarization. There is mixed evidence as to the effectiveness of cross-cutting exposure moderating such polarization. By examining the link between internet usage propensity and polarization over time, the authors contribute to existing literature by considering how high vs. low internet exposure moderates polarization longitudinally.
Analytical Approach
The authors utilize survey data from the ANES and Pew Research Center, pulling across multiple studies and focusing on the time period between 1972-2016 for trends in polarization and 1996-2016 for trends in internet usage. The main indicator for internet use is self-reported internet usage from the ANES.
The authors are agnostic on the theoretical meaning of polarization, and therefore create an index of 8 existing measures (spanning affective, ideological, partisan-ideology, perceived partisan-ideology, and other forms of polarization) as their main dependent variable. The index is simply the average of these three measures normalized by the value in 1996.
The main set of analysis involves the measurement of trends in polarization, internet usage by age, and polarization by age and predicted internet usage, with the main purpose to show whether changes in polarization are greatest in groups with higher rates of internet usage. The full DAG of the described relationships are shown in Figure 1.
Finally, the authors measure the proportion of the rise in polarization attributable to internet usage by age groups by fitting a linear model of predicted polarization by group as a function of a linear time trend and group internet usage. The model is fit with and without the constraint that the effect of group internet usage is 0, allowing for the difference in the linear effect to be compared. Standard errors are calculated with a non-parametric bootstrap at the respondent level with 100 replicates.
Main Findings
Polarization and internet usage are on the rise, but polarization has grown most significantly in groups with lower rates of internet usage, and a negligible share of over-time polarization is attributable to internet usage.
All 8 indicators of polarization (including the summary measure) have increased steadily since 1972. This trend appears consistent pre- and post-1996, when most measures of internet usage began. Similarly, the proportion of respondents reporting internet usage has also risen steadily from 1996 onward, although unequally between demographic groups; younger respondents report much higher internet usage than older respondents.
Analyzing trends in polarization by age, respondents age 65+ showed a significantly greater increase in polarization between 1996 and 2016 than respondents age 18-39. Here, age is supposed to act as a proxy for internet usage, showing the group least likely to use the internet appears to have the greatest increase in polarization.
The authors then use a measure of predicted internet use (based on a linear specification using demographic predictors) and measure the trend in polarization among respondents with higher and lower predicted internet usage. This analysis again shows the group with lower predicted internet usage showed the greatest increase in polarization.
Turning to their analysis of the attributability of growing polarization to internet usage, the authors find no significant differences in the constrained versus unconstrained models. This suggests very little (at maximum, 17%) of the over-time increase in polarization is associated with rising internet usage.
Implications
This paper contributes to a growing literature on the connection between internet usage and political polarization. Namely, gives an over-time understanding of the (lack of) connection by showing how increases in polarization do not appear aligned with baseline levels of internet usage, and how increases in polarization do not appear attributable to increases in internet usage within an age group. The paper expands upon prior studies by considering multiple manifestations of polarization and by leveraging a massive set of survey data.
In summary, the findings suggest the internet plays a limited role in the rise of polarization. The authors do not give any indication as to where they believe such drivers may exist, but their analyses suggest a more positive outlook on the relationship between being online and political polarization.
Questions left unanswered
The authors are careful to frame their question as “how trends in political polarization relate to respondents’ propensities to obtain information online or from social media” (emphasis added). But in terms of the broader, causal question of whether internet usage increases political polarization, the authors do not provide an answer.
Furthermore, the hypothesized mechanism of online segregation leading to polarization is not tested in this paper. Only internet/social media usage as a binary measure is analyzed, meaning we cannot distinguish casual internet users from those embedded in potential echo chambers, as have been previously hypothesized
Methods and Analysis
Was the study and its analyses pre-registered?: No
Did the study rely on proxy variables to measure polarization?: N/A
Were standard p-value thresholds used (p<.05 or 95% Confidence Intervals that don’t overlap zero)?: Yes
- Largest p-value presented as significant: 0.05
Were correlational results interpreted with causal language?: No
Limitations / Weaknesses
The authors utilize a large set of data over a long time period with many measures of political attitudes, and this breadth should be commended. However, it does mean the implications of the work are somewhat ambiguous. While many of these measures of polarization are highly correlated, they are theoretically, attitudinally, and behaviorally distinct, meaning the average over all such attitudes likely biases against a finding. For example, the findings in the supplementary material show that respondents with high predicted internet usage show greater increases in straight-ticket voting and issue consistency than low-internet usage respondents, an inversion of the averaged finding in the main manuscript. These have distinct consequences for polarization and democracy writ large.
Additionally, the analysis that might be expected from the title of the paper (polarization ~ internet usage) is never actually estimated. The authors instead opt for predicted internet usage using a very limited set of covariates and a simple linear specification. Beyond it being unclear why the authors opted for predicted versus reported behavior (perhaps to account for differences between surveys), the actual prediction model could likely be improved dramatically using modern statistical learning techniques. It is also unclear whether the uncertainty implicit within these predictions was propagated through the rest of the model.
Open Data & Analyses
Does the article make the replication data publicly available?: Yes
Does the article make the replication analysis scripts publicly available?: Yes
Article Citation
Boxell, L., Gentzkow, M., & Shapiro, J. M. (2017). Greater internet use is not associated with faster growth in political polarization among us demographic groups. Proceedings of the National Academy of Sciences, 114(40), 10612–10617. https://doi.org/10.1073/pnas.1706588114
Bibtex
1
2
3
4
5
6
7
8
9
10
11
12
13
@article{
doi:10.1073/pnas.1706588114,
author = {Levi Boxell and Matthew Gentzkow and Jesse M. Shapiro },
title = {Greater Internet use is not associated with faster growth in political polarization among US demographic groups},
journal = {Proceedings of the National Academy of Sciences},
volume = {114},
number = {40},
pages = {10612-10617},
year = {2017},
doi = {10.1073/pnas.1706588114},
URL = {https://www.pnas.org/doi/abs/10.1073/pnas.1706588114},
eprint = {https://www.pnas.org/doi/pdf/10.1073/pnas.1706588114}
}