Bookwise Social Activation Study

Bookwise Social Activation

Page 5 of 7 — Does social engagement predict reading above and beyond what past reading alone predicts?

The Confound

Reading momentum (past sessions predicting future sessions) has the strongest effect size in our analysis (0.805). This raises a critical question: when we see that users with more social connections read more, is that because social engagement CAUSES more reading, or because people who were already going to read a lot happen to accumulate more social connections along the way? Two statistical approaches help disentangle this.

Stratified Analysis — Apples to Apples

Following ≥3 stratified by first-week reading

Among users who read the SAME amount in their first week, those who followed ≥3 people read 2.3× more in the following month — but only in the 1–15 sessions range. Users who never read (0 sessions) don't benefit from social connections. Power readers (16+) are going to read regardless. The sweet spot for social activation is the mid-range reader who has shown intent but hasn't formed a habit yet.

Other Social Metrics (Stratified)

Kudos Received ≥1

Kudos Given ≥1

Following ≥1

Across every social metric, the pattern holds: social engagement matters most for the 1–15 sessions cohort. Kudos received shows the strongest stratified lift (6.7× in the 1–5 group), while following ≥1 shows an 11.7× lift driven by the fact that nearly all non-followers in this cohort have minimal future reading.

Partial Correlations

Raw correlation vs. partial correlation after controlling for past reading.

After statistically removing the effect of past reading, kudos received retains the strongest independent signal (r=0.391). This means kudos are NOT just a proxy for reading volume — receiving engagement from others has genuine, independent predictive power for future reading. Following (r=0.182) and followers (r=0.122) have smaller but real independent effects.

The Target Segment

275
Mid-range readers
(1–15 sessions in week 1)
125
Have <3 social connections
(the target)
150
Already socially active
(following ≥3)

125 mid-range readers — users who showed reading intent but haven't built social connections — are where experiments will have the highest marginal impact. These are the users who respond to social activation: they've demonstrated they want to read, they just haven't found their community yet.

Implications

  1. Social engagement has real independent value. Partial correlations confirm it predicts future reading even after controlling for past reading.
  2. Target the middle, not the extremes. Users with 0 sessions can't be socially activated (nothing to engage with), and power readers don't need it. The 1–15 session cohort is where interventions matter.
  3. Kudos have the strongest independent signal (r=0.391). The existing finish-post → notification → kudos loop is genuinely working. Amplifying it is high-leverage.
  4. Following is the most actionable lever. While kudos have a stronger correlation, following is something we can directly create through experiments (A2–A5). Getting someone to follow 3+ people in their first week predicts a 2.3× reading lift among mid-range readers.