Overview

This paper studies the causal impact of social media exposure during college on later earnings, using the staggered introduction of Facebook across U.S. universities as a natural experiment. The main finding is that greater exposure to Facebook during undergraduate studies is associated with lower wages after graduation, with effects that are strongest early on and fade over time.


Introduction

Social media changed how students communicate and spend time, but its effect on long-run labor market outcomes is still not fully understood. This paper asks a simple question:

How did access to Facebook during college affect graduates’ earnings one, five, and ten years later?

The paper exploits a key institutional fact: Facebook was rolled out across colleges at different times between 2004 and 2006, before becoming publicly available in 2006. This staggered expansion creates quasi-experimental variation in social media exposure during university years.


Data

The analysis combines two data sources:

  1. Facebook college rollout dates (from replication files used in prior work, based on archived data)
  2. PSEO (Post-Secondary Employment Outcomes) Earnings, which reports aggregate earnings by institution, major, cohort, and percentile (25th, 50th, 75th), measured one, five, and ten years after graduation.

Matched sample

After merging sources, the paper matches 111 universities that received Facebook before it became public and that can be linked to PSEO data. The final analysis focuses on undergraduates, motivated by high Facebook take-up rates among that group in the early rollout period.

Exposure measure

Because individual Facebook usage is not observed, the paper constructs an institution-cohort exposure variable:

  • normalized from 0 (no exposure during undergraduate years)
  • to 1 (full exposure during undergraduate years)

This variable measures the share of undergraduate years in which a cohort at a given institution had access to Facebook.


Empirical Strategy

The core empirical design is a difference-in-differences style specification using quasi-experimental variation from staggered Facebook rollout.

The outcome is:

  • log(earnings) at the 25th, 50th, and 75th percentiles,
  • measured 1, 5, and 10 years after graduation.

The main regressor is the cohort-level Facebook exposure measure, with controls for:

  • graduation cohort fixed effects,
  • institution and major effects,
  • and institution–major structure (as in the paper’s fixed-effects setup).

Identification assumptions

The interpretation relies on:

  1. Stable selection into institutions and majors across cohorts
  2. Parallel trends in wages across treated and untreated institutions (within major), absent Facebook exposure

The baseline specification compares pre-treatment cohorts (2001–2003) to post-treatment cohorts (2007–2009), excluding the intermediate 2004–2006 cohorts because exposure is harder to assign cleanly in those grouped data.


Main Results

The paper finds a negative and persistent effect of Facebook exposure on wages.

1 year after graduation

Full undergraduate exposure to Facebook is associated with wage declines of roughly:

  • -3.7% at the 25th percentile
  • -3.9% at the median
  • -3.6% at the 75th percentile

5 years after graduation

The negative effect remains but is smaller (around -1.1% at the median).

10 years after graduation

The effect is close to zero and no longer statistically meaningful in the baseline estimates.

Nonlinear exposure effect

The paper also estimates specifications with a quadratic exposure term. The positive coefficient on exposure-squared suggests that the negative wage effect grows more slowly at higher exposure levels (i.e., diminishing marginal harm). A possible interpretation is that the initial arrival of social media is more disruptive, while students later adapt their routines.


Interpretation

The paper interprets the results through a simple work–leisure framework:

  • Facebook increases time spent on social activity / distraction
  • Students study less
  • Academic performance (e.g., GPA) falls
  • Lower academic performance weakens labor market bargaining power and reduces wages

This mechanism is not directly observed in the data, but it provides an intuitive explanation for the short- to medium-run earnings effects.


Heterogeneity Analysis

The paper groups majors into:

  • STEM
  • Humanities
  • Social Sciences
  • Others

and estimates a triple-differences specification to test whether effects vary by field.

Main takeaway

There is no single subgroup that cleanly drives the overall result, but the point estimates suggest:

  • Humanities students may be the most affected
  • Social Sciences appear less affected (in point-estimate terms)

However, standard errors overlap substantially, so subgroup differences should be interpreted cautiously.


Placebo Tests and Sensitivity Analysis

The paper performs two robustness exercises.

1) Placebo cohort tests

It assigns “synthetic” Facebook exposure to cohorts that should not exhibit treatment variation (or where treatment timing should not matter) and re-runs the main specification.

Result: placebo estimates are generally null / insignificant, which supports the identification strategy and parallel-trends assumption.

2) Randomization inference

For the 1-year outcome, exposure is randomly reassigned across universities (10,000 times), and the regression is re-estimated.

Result: the observed estimates are far from the randomization distribution, suggesting the main effect is very unlikely to be due to chance.


Limitations

The paper highlights several limitations:

  1. Three-cohort aggregation in PSEO

    • This blurs treatment timing (especially for cohorts around 2009)
    • It weakens the precision of exposure assignment
  2. Limited control for the 2008 financial crisis

    • Cohort fixed effects absorb only an average cohort shock
    • Residual differential effects may still bias estimates

These limitations suggest the estimates are informative but should be interpreted with appropriate caution.


Conclusion

Using the staggered rollout of Facebook across U.S. colleges, this paper provides quasi-experimental evidence that social media exposure during college reduced post-graduation wages in the short run, with effects that persist for several years and largely fade by ten years after graduation.

The results are consistent with a distraction-based mechanism and suggest that digital technologies can have economically meaningful effects on human capital accumulation and early-career labor market outcomes.