Living Large or Long: Preference Estimates from Completed-Life Stories

How to cite

Luttmer, Erzo; Chandra, Amitabh; Schwartzstein, Joshua; Jagelka, Tomáš. 2025. "Living Large or Long: Preference Estimates from Completed-Life Stories." AEA RCT Registry. AEARCTR-0015566. https://doi.org/10.1257/rct.15566-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Erzo Luttmer* , Amitabh Chandra , Joshua Schwartzstein , Tomáš Jagelka

* Primary investigator

Registry ID
AEARCTR-0015566
DOI
https://doi.org/10.1257/rct.15566-1.0
First registered
Last updated
Status
In development
Trial dates
2025-03-14 – 2025-12-31
Countries
United States of America ()
JEL codes
I10, H51
Registry record
http://www.socialscienceregistry.org/trials/15566

Abstract

A variety of resource allocation decisions, including in health and environmental policy, depend on how much money people are willing to give up to live an extra year of life. This rate of substitution is typically estimated based on compensating differentials for taking on very small risks of death. We develop an alternative approach to estimate this rate of substitution by presenting respondents on the Prolific platform with pairs of hypothetical life stories, or “vignettes,” that describe complete life courses, and ask them to choose which life they would prefer for themselves. We randomize the lifetime incomes and the ages of death in the vignettes. We present a battery of tests to gauge whether these hypothetical choices measure underlying preferences. The benefit of our method is that it yields estimates of the distribution of preferences for longevity in a representative population, allows for a characterization of heterogeneity in preferences by respondent characteristics, and does not require people to correctly evaluate very small probabilities.

Intervention

The two key randomizations are: the lifetime income and the age of death in the vignettes shown to respondents. The vignettes and survey instrument contain numerous other randomized interventions that allow us to gauge whether choices over vignettes reflect stable preferences for longevity and to rule out possible confounding factors.

Experimental design

The longevities and life-time incomes of the protagonists shown in the vignettes are independently randomized across respondents and vignettes, except that in wave 2 of the survey we present the respondent with one vignette pair that is identical (and has identical randomizations) to one randomly selected vignette pair that the respondent saw in wave 1.

Design details

Randomization method
Randomization done in office by a computer.
Randomization unit
The randomization for longevities and life-time incomes of the protagonists in the vignettes occurs at the (vignette)*(respondent)*(question number) level. Thus, each time a respondent sees a vignette, the longevity and life-time income of the protagonist is randomly drawn for that vignette, that respondent, and that question number. So, if a respondent sees the same vignette more then once (say, in different waves of the survey), the longevity and life-time income of the protagonist will be re-randomized. There is one exception: In wave 2 of the survey, we present the respondent with one vignette pair that is identical (and has identical randomizations) as one randomly selected vignette pair that they saw in wave 1.
Clusters
The planned analysis sample consists of 27,500 choices of a vignette within a vignette pair. Details: We aim for a sample size of 3750 respondents in wave 1. We expect 75% of these respondents, so about 2800 of them, to pass the sample inclusion criteria. The sample inclusion criteria are are pre-specified in the analysis plan. We invite only respondents who passed the inclusion criteria in wave 1 to take wave 2. Based on our estimated response rate, we expect to re-survey about 2400 respondents in wave 2. This brings the sample size in the two waves combined to about 6150 respondent*wave observations. We expect that about 75% of respondent*wave observations pass our inclusion criteria. Thus, the analysis sample will be about 4600 respondent*wave observations. Within a each wave, each respondent is asked six times to choose one vignette out of a pair of vignettes. Hence, we obtain 6 choices per respondent per wave. This brings the expected number of units of observation to about 27,500.
Observations
The planned analysis sample consists of 27,500 choices of a vignette within a vignette pair. Further details: See the explanation given under "planned number of clusters." The choices are made in about 4600 respondent*wave observations by about 2800 unique respondents.
Arms
There are no treatment arms; the longevity for the protagonist in a (vignette)*(respondent)*(question number) observation is randomized between all ages from 60 to 90 and the protagonist's income is drawn randomly from all annual income levels (rounded to the nearest $1000) that fall within 33% of the baseline income in a given vignette (the baseline income level in a vignette is chosen so that it is realistic for the protagonist's occupation).
Primary outcomes
Choice of vignette within a pairs of vignettes (= completed-life stories of about two to three paragraphs in length) where the life-time incomes and longevity of the protagonists in the vignettes is randomized.

Cited by

No citation data available for this registration yet.

Keywords