Experimental research summary

“Willing but Not Clicking”

Donor Choice and the Intention-Action Gap in Charitable Giving

Group 3 Oskar Herrera, Ella O'Connell, Leo Wu, Justin Gill, Donghyeok Lee
The gap

Choice raised the willingness to give, but click-through to the donation page did not move. The gap lives between saying and doing.

1 · Background

Background & Research Question

Background

Charities reach many donors, but interest rarely turns into action. Identifiable stories and a sense of control are two layers thought to drive giving.

What We Did

Showed each participant one of three appeals: 1. Control (plain ask), 2. Statistics + Story, 3. Statistics + Story + Choice, then measured real clicks on a live UNICEF button.

Hypothesis

Letting donors choose where their $10 goes increases giving - people act more on a gift they control.

Interest
↓
Action?

2 · Treatment arms

Three Appeals, Randomly Assigned

T1 - Control
Control appeal: a basic ten-dollar UNICEF donation request.

Basic donation request only.
No statistics, no story.

T3 - Stats + Story
Statistics and identifiable story appeal featuring Miracle in Sierra Leone.

Donation statistics + an identifiable story (UNICEF: “Miracle,” Sierra Leone).

T4 - Donor Choice
Donor choice question with five possible program areas.

Same as T3, plus letting the donor choose where their $10 goes.

Held constant (all arms): $10 amount · UNICEF · identical “Go to Donate / No thanks” decision & outcome questions - only the appeal layer differs.

3 · Method

Design

Randomize176 participants,
assigned individually
Show AppealT1 / T3 / T4
(59 / 58 / 59)
MeasureClick-through
(primary outcome)
  • Primary outcome: click on the UNICEF donation button (link_clicked).
  • Secondary: stated intention, emotional impact, credibility, hypothetical amount.
  • Random assignment at the individual level (Qualtrics randomizer).
  • Hypothesis (pre-registered): H1 T3 > T1 · H2 T4 > T3 - T1 vs T4 exploratory.
  • Balance (pre-randomization): prior donors 32% / 34% / 36% - χ² p = 0.93 (balanced).
  • Sample: N = 176 (59 / 58 / 59) via one anonymous link - lab vs. social-network sources not separable.

4 · Primary outcome

Primary Outcome: Click-Through Rate

Primary Outcome: Donation Click-Through
Observed click-through rate by appeal arm · error bars show ±1 SE
Donation click-through rate by appeal arm T1 control 11.9 percent, T3 statistics and story 3.4 percent, T4 donor choice 13.6 percent. No comparison is statistically significant. 0% 5% 10% 15% 20% Click-through rate 11.9% 3.4% 13.6% T1Control T3Stats + Story T4Donor Choice Appeal arm
All proportion comparisons: Fisher's exact because of small cell counts. Error bars = ±1 SE.

T1 11.9% · T3 3.4% · T4 13.6%

  • No significant difference between arms.
  • T4 vs T1: +1.7pp, p = 1.00 (Fisher, n.s.).
  • T3 vs T1: -8.4pp, p = 0.16 (Fisher, n.s.).
  • T3 below T1 is counterintuitive but n.s. - only 2 clicks (small cell, noise).
  • Logit (click ~ arm + prior, n = 174): no arm effect - T4 OR 1.17, T3 OR 0.28 (both n.s.).

Click-through to the donation page didn't move.

5 · Intention-action gap

But Intention Tells a Different Story

Attention-check fail rates differed by arm (T1 7% / T3 25% / T4 19%); excluding fails, T4 intention still > T1 (47.9% vs 25.5%, p = .024).

Intention vs. Action - The Gap
Stated intention compared with observed click-through
Stated intention and click-through action by appeal arm Intention rates are 25, 28, and 46 percent for T1, T3, and T4. Action rates are 11.9, 3.4, and 13.6 percent. Intention Action 0% 15% 30% 45% 60% Rate 25% 12% 28% 3% 46% 14% T1Control T3Stats + Story T4Donor Choice Appeal arm
Blue bars show stated intention; gold bars show observed action.

Stated intention to donate
T1 25% · T3 28% · T4 46%

  • T4 vs T1: +20.3pp, p = 0.034 (Fisher, significant).
  • Choice raised willingness - but the red bars stay low.
  • Pre-registered: H1 (T3 vs T1) n.s. (+2.2pp); H2 (T4 vs T3) +18.2pp, p = 0.055.
  • T3 ≈ T1 - the intention lift is attributable to choice, not the story.
  • Logit (n = 174): T4 OR 2.49, p = .024 - holds controlling for prior (OR 2.30).

Intention rose - action did not follow.

6 · Follow-through

Conversion: Intention Doesn't Carry Over

Conversion: Of Those Willing, Who Clicked?
Share of “yes” respondents who followed through
Conversion among respondents who said yes Among willing donors, 47 percent in T1 clicked, 12.5 percent in T3 clicked, and 30 percent in T4 clicked. 0% 15% 30% 45% 60% Willing donors who clicked 47% 12.5% 30% T1Control T3Stats + Story T4Donor Choice Appeal arm
Descriptive conversion among respondents who stated an intention to donate.

Of those who said “yes,” who actually clicked:
T1 47% · T3 12.5% · T4 30%

  • T4 attracted the most willing donors (27), but most of them still didn't follow through (only 30% converted).
  • The extra intent choice created did not convert into action.
  • Descriptive only - “willing” is shaped by treatment, not a causal arm comparison.

The gap lives between saying and doing.

7 · Secondary outcomes

Attitudes Rose Sharply, Behavior Didn't

Attitudes Rose Across Arms (T1 < T3 < T4)
Mean ratings on a 1-7 scale · error bars show ±1 SE
Emotional impact and credibility ratings by appeal arm Emotional impact rises from 2.51 to 3.40 to 3.76. Credibility rises from 2.82 to 3.93 to 3.95. Emotional impact Credibility 0 1 2 3 4 5 6 7 Mean rating (1-7) 2.51 2.82 3.40 3.93 3.76 3.95 T1Control T3Stats + Story T4Donor Choice Appeal arm
Mean ratings on a 1-7 scale. Error bars represent ±1 SE and are visually reconstructed from the source figure because the underlying numeric SE values were not reported.

Every attitude measure rose T1 < T3 < T4:

  • Emotional impact: 2.51 → 3.76 (T1 → T4), p < 0.001.
  • Credibility: 2.82 → 3.95 (T1 → T4), p < 0.001.
  • Hypothetical donation: $11.8 → $14.3 (n.s., p = 0.31).
  • Emotional & credibility rose significantly; hypothetical rose but n.s.

Only behavior stayed flat.

8 · Exploratory heterogeneity

Story Alone Backfires for Non-Donors

Click-Through by Prior Donation Experience
Exploratory split by past donation experience
Click-through by prior donation experience Prior donor click rates are 5.3, 10.5, and 9.5 percent. Non-donor click rates are 15.0, 0.0, and 15.8 percent for T1, T3, and T4. Prior donors Non-donors 0% 5% 10% 15% 20% Click-through rate 5.3% 15.0% 10.5% 0.0% 9.5% 15.8% T1Control T3Stats + Story T4Donor Choice Appeal arm
The 0 / 37 non-donor cell in T3 is exploratory and likely amplified by small n.

Split by past donation experience:

  • Prior donors: 5% → 11% → 10% (both treatments above control).
  • Non-donors: T3 collapsed to 0% (0 / 37).
  • Adding choice (T4) restored non-donor engagement to control level (~16%).
  • 0 / 37 is a null cell, likely amplified by small n; treat as exploratory.
  • Choice → intention effect did not differ by gender (arm x gender interaction p = .16); click-level & year subgroups too sparse to test.
  • T4 allocation choice (n = 59): top 2 categories drew 76% - choice lever actively engaged.

9 · Precision

Statistical Power

Power Calculation - T4 vs T1
Observed effects, Cohen's h, achieved power, and required sample
ComparisonIntentionClick
Observed (T4 vs T1)+20 pp+2 pp
Cohen's h0.43 (medium)0.05 (negligible)
Power at n ≈ 5965%6%
n / arm for 80% power85~6,000
Power calculation - T4 vs T1.

Power Calculation - T4 vs T1

  1. Effect size (Cohen's h)
h = 2(arcsin√p₂ - arcsin√p₁)
  • Intention → 0.43 (medium).
  • Click → 0.05 (negligible).
  • Power at n ≈ 59: Intention 65% · Click 6%.
  • Required n / arm for 80% power: 85 vs ~6,000.

Bottom line

  • Intention effect is real and medium-sized: ~85 / arm would confirm it (we already hit p = 0.034 at n ≈ 59).
  • Click effect is negligible: detecting it needs an impractical sample, so behavior stays inconclusive.

10 · Interpretation

Limitations

  • Small sample (59 / 58 / 59) → underpowered for the behavioral (click) outcome.
  • Low-stakes survey context, participants knew this was a class survey, not a personal donation ask, so may have withheld the click even when willing (depresses behavior, not intention).
  • Cannot fully separate a true gap from a survey artifact.
  • Spillover possible (classmates) - mitigated by online + individual randomization + short window; not fully ruled out.
  • Differential attrition: completion T1 81% / T3 95% / T4 100% (χ² p < 0.001) - Click & intention measured at the decision point (pre-dropout) → unaffected - Only secondary outcomes affected; survivor bias is conservative (shrinks gaps).

Internal validity

Randomization holds - balanced arms, valid comparisons.

External validity

Anonymous-link respondents - not a defined population; planned lab-only primary analysis not feasible (single link).

11 · Application

Implications for Practice

  • If the gap is real (not a survey artifact): converting intent → action needs friction removal, immediacy, a direct path.
  • Choice architecture is a low-cost way to raise donation intent.
  • Campaigns should pair intent-building appeals with a frictionless donation path.
  • Future work: test in a real-stakes campaign with a larger sample.
  • Scale: a 5-option selector is near-zero cost - no change to suggested amount, tone, or design.
Choice raises the willingness to give - turning that into action means removing what stands between intent and the click.