NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
Evaluating the Sunk Cost Effect
Initial registration date
July 04, 2019
July 15, 2019 10:54 AM EDT
This section is unavailable to the public. Use the button below
to request access to this information.
University of Warwick
Other Primary Investigator(s)
University of Warwick
University of Oxford
Additional Trial Information
sunk cost effect
sunk cost fallacy
Raven's progressive matrices
international cognitive ability resource
cognitive reflection test
Big Five Personality Inventory
Experimental evidence is used to measure the sunk cost effect, and to evaluate the relationship between it and different measures of intelligence and reflective thinking, including fluid intelligence (Raven's progressive matrices), crystallised intelligence (International Cognitive Ability Resource: verbal reasoning item), the cognitive reflection test and openness from the Big Five Personality Inventory. In addition, the introduced experimental measure is used to validate an accessible scale-based (eighteen-item) measure for the sunk-cost effect based on self-reported choices in hypothetical scenarios.
Subjects either complete the core "treatment" condition or one of two control conditions where there is no real-effort task (designed to control for the endowment effect and a general failure of individuals to maximize pecuniary outcomes, respectively).
Intervention Start Date
Intervention End Date
Primary Outcomes (end points)
A binary variable equal to 1 if subjects stick with a lottery they earn via the real-effort task, or 0 if they switch to a dominant lottery.
We also collect information on different measures of intelligence (fluid and crystallised), reflective thinking, and openness, to correlate with the sunk-cost effect. In addition, we aim to validate a scale-based measure for the sunk-cost effect based on self-reported choices in hypothetical scenarios.
Primary Outcomes (explanation)
To measure the sunk-cost effect, subjects are first asked to complete a task involving real effort (letter-counting) where their performance in the task determines whether or not they earn a lottery (which offers a positive amount of money in expectation). They are then presented with a choice to either keep the lottery they earned or to switch to a different lottery that (stochastically) dominates the one they earned. A maximizer of expected pecuniary outcomes would switch lotteries, but because of the effort sunk into the acquirement, some subjects may choose to keep the inferior (but earned) lottery.
Fluid intelligence (cognitive ability) is measured by subjects' (incentivized) answers to Raven's progressive matrices. Crystallised intelligence is measured by (incentivized) answers to the verbal reasoning item of the International Cognitive Ability Resource. The propensity for reflective thinking is captured by two measures: the Cognitive Reflection Test (incentivized); and the Openness scale from the Big Five Personality Inventory.
The scale-based measure of the sunk-cost effect is given by subjects' choices in a series of 18 hypothetical scenarios where they indicate on a 6-point Likert scale their preference between two hypothetical alternatives (where one corresponds to the sunk cost effect).
Secondary Outcomes (end points)
Age, gender, race, income, education, and political affiliation.
Additionally, two separate conditions account for the endowment effect and concerns that individuals may not maximize expected pecuniary payoffs. This allows us to see how much of any sunk-cost effect is due to these factors.
Secondary Outcomes (explanation)
To measure the role of the endowment effect in the lottery-choice, subjects are first told they have been endowed with the dominated lottery before being asked to make the choice between the dominant and dominated lottery. Concerns regarding whether subjects maximize expected pecuniary outcomes are assessed by giving subjects a straight choice between the two lotteries.
US subjects will be recruited online through Amazon's Mechanical Turk. The primary part of the design requires subjects to carry out a real-effort task. If they perform well enough, they earn a lottery. They are then presented with the choice between the lottery they earned and a dominant lottery. They exhibit the sunk-cost effect if they choose to stick with the earned (but dominated) lottery. Subjects then complete fluid and crystallised intelligence tests, a reflective thinking test, an 18-item scale that measures the sunk-cost effect via stated choices in hypothetical scenarios, and some intelligence, personality and demographic questions.
Experimental Design Details
Randomization carried out by the software Qualtrics.
Randomization is at the individual level
Was the treatment clustered?
Sample size: planned number of clusters
Sample size: planned number of observations
Sample size (or number of clusters) by treatment arms
numbers are approximate
320: Main sunk-cost condition;
180: endowment effect control; 30: basic expected-pecuniary-payoff-maximizer check.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Given a 60/35/5% split between the conditions (as ordered above), and hypothesized effect sizes (the proportion who choose to keep the dominated lottery) of 0.2/0.1/0, n=506 subjects are required in total (two-sided tests for differences in proportions; 80% power; 5% significance) - 304/177/25 in each condition. Rounding up, we will gather 320/180/30, a total of n=530.
INSTITUTIONAL REVIEW BOARDS (IRBs)
University of Oxford Economics Departmental Research Ethics Committee
IRB Approval Date
IRB Approval Number
University of Warwick Humanities and Social Sciences Research Ethics Committee (HSSREC)
IRB Approval Date
IRB Approval Number