1
Sunk-cost judgements across the child to adult lifespan: Supplementary Material

Table of Contents

1. Deviations from preregistration..................................................................................................................................................................... 2

2. Experiment 1 Supplementary Results ........................................................................................................................................................... 6
a. Demographics............................................................................................................................................................................................. 6
b. Sensitivity Analysis .................................................................................................................................................................................. 10
c. Sunk-cost decision type analysis .............................................................................................................................................................. 13
d. Exploratory analysis excluding participants age <5 ................................................................................................................................. 15
3. Experiment 2 Supplementary Results ......................................................................................................................................................... 16
a. Demographics........................................................................................................................................................................................... 16
b. Post-hoc power simulations...................................................................................................................................................................... 20
c. Preregistered categorical age analysis ...................................................................................................................................................... 21
d. Preregistered decision type analyses ........................................................................................................................................................ 26
e. Exploratory memory error analyses ......................................................................................................................................................... 36
f.

Exploratory analysis excluding participants age <5 ................................................................................................................................. 41

2
1. Deviations from preregistration

Summary of Preregistration Deviations
Deviations
No.
1
Type
Reason
Timing

2

Type
Reason

Details
Typo/Oversight
Timing of
preregistration
After results known

Original Wording
“Thus far we
conducted one study
in 610 individuals
ages 3-97 using a
single vignette to
measure the SCE.”

Sample Size
Preregistered plan
not possible or

“We plan to collect a
total sample of N =
213.”

Deviation Description
In our preregistration for Experiment 2,
we report the Experiment 1 sample size
as 610, whereas the final Experiment 1
sample size reported in the manuscript is
682. This discrepancy is due to the fact
that Experiment 1 is part of a larger
ongoing longitudinal lifespan cognition
study, for which data collection
continued after the registration of
Experiment 2 in 2019. Over the course of
conducting Experiment 2 (and multiple
manuscript revisions), we obtained a
larger Experiment 1 sample than we
initially had. Because we had not
preregistered Experiment 1, and because
our lifespan analyses required as large a
sample size as possible (i.e., enough data
points across the full age range for a
continuous age analysis), we opted to use
the current full Experiment 1 sample size
available to us at the time of analysis.
In our preregistration, our target sample
size was N = 213 across our three age
groups. Our final sample size was

Reader Impact
The impact of this deviation
is minimal—the overall
pattern of results at N = 610
and N = 682 do not
qualitatively differ (see
figure below table for an
analogue of our primary
analysis conducted at N =
610 for a conference
presentation nearer to the
time of our initial
preregistration). In both
analyses, we observed no
SCE in young children, an
SCE in adolescents and
adults, and some evidence
for a lower SCE in older
adults.

Our decision to collect a
larger sample than initially
planned was not motivated

3

Timing

inappropriate
because of data
After data access

substantially larger—N = 378
participants. There were two reasons for
this substantial deviation. First, when we
examined the obtained age ranges prior
to and at our target sample size, although
we had obtained roughly equal ns in our
three age groups, there were substantial
gaps in the continuous age range (i.e., we
had many children age 3-12, but not
many adolescents 13-17). We collected
additional data to ensure that the age
distribution for Experiment 2 was at least
close to the one obtained for Experiment
1. Second, data collection was
“messy”—we recruited participants inperson at community centres and a
science centre where we could not
anticipate the exact number of
participants for a given booked day or
data collection session (and it was not
feasible to do ‘live’ sample size analyses
on-site). Because of this practical
constraint, we opted to collect as much
data as possible in the in-person sessions.
The decision to terminate data collection
coincided with the end of our data
collection sessions in these facilities.
Finally, our pre-registered sample size
planning was primarily based on
categorical age group analyses. We opted
to report only our pre-registered
continuous age analyses in-text (partly
based on cautions against dichotomizing

by results for our primary
hypothesis tests (indeed, we
did not perform the analyses
until data collection for the
full sample had been
completed). As such, the
impact of this deviation is
minimal. It is certainly
possible that the results may
have been different in a
smaller sample, but such a
sample would have arguably
been underpowered to
detect effects in a moreappropriate continuous age
analysis mirroring the one
conducted in Experiment 1.
And, given the gaps in our
age range (particularly
around the seeminglycritical period of
adolescence), it is likely that
such results would have
been misleading.

4
continuous variables like age, e.g.,
McCallum et al., 2002, and to better
match the reported non-preregistered
analyses for Experiment 1). For this
continuous analysis, we conducted posthoc, non-preregistered power simulations
similar to those conducted for
Experiment 1 using our exact sample
size, age range, and estimated regression
residuals. From these simulations, we
determined that we had sufficient power
(≥ .80) in the 1st-person perspective
condition (n = 209) to detect a linear age
effect of |β| ≥ .008, and assuming that
minimum linear age effect, a quadratic
age effect of |β| ≥ .0005. There were
slightly fewer participants in the 3rdperson perspective condition (n = 169),
resulting in sufficient power (≥ .80) in
this condition to detect a linear age effect
of |β| ≥ .01, and assuming that minimum
linear age effect, a quadratic age effect of
|β| ≥ .0015.
Note. Table created using template proposed by Willroth & Atherton (2024)

5
Analogue of Experiment 1 primary analysis conducted at N = 610

Relevant references:
MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of dichotomization of quantitative variables.
Psychological methods, 7(1), 19.
Willroth, E. C., & Atherton, O. E. (2024). Best laid plans: A guide to reporting preregistration deviations. Advances in Methods and
Practices in Psychological Science, 7(1), 25152459231213802.

6
2. Experiment 1 Supplementary Results
a. Demographics
Participants in each age group
Age group
n
3-17
348
18-59
204
60+
130

Sample age breakdown

7
Participant sex (self- or parent-reported)
Sex
n
Female
414
Male
257
Unreported 11
Participant ethnicity (self- or parent-reported)
Ethnicity
White
Asian
Unreported
South Asian
Caucasian
Black
Indian
Hispanic
East Indian
Other
Southeast Asian
Arab
Asian/white
Chinese
Filipino
Persian
1/2 white 1/2 south asian
Afghan
Asian & White
Caucasian-Asian
East Asian
Indo-Canadian

n
243
129
138
29
14
12
11
7
6
5
5
3
3
3
3
3
2
2
2
2
2
2

8
Iranian
Melanesian (Pacific Islander)
Mixed
Multi-Ethnic
Punjabi
South Asian + White
Whtie
1/2 Black 1/2 White
1/2 white + 1/2 hispanic
3/4 Black 1/4 Irish
50% white, 50% hispanic
Acadian/Mohowk
Afghan-Iranian
African
African (Egyptian)
Arabic
Asian Panjab (India)
Asian/ Black, Indian, White
Asian/Indian, Black, White
Asian/White
Asian/white mix
Biracial
British/Vietnamese
Caucasian (English)
Egyptian
English & Chilean
First Nations, Asian English
First Nations, Irish
Half Asian / White
half Romanian, quarter japanese, quarter
english
Hispanic, White

2
2
2
2
2
2
2
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

9
Hispanic, White,French, Aboriginal
Indian, Pakistani, African
Indigenous
Indigenous (Cree)
Metis
Middle East
mixed (white/asian)
Originally from Egypt but he Born in
Canada
Pacific Islander
Pacific Islander/Melanesian
Sikh
White-caucasian
White & Jewish
White&Jewish
White(Caucasian
White/Aboriginal
White/First nations
White/Indian

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

10
b. Sensitivity Analysis
Power curve for linear age effect

Simulations using Experiment 1 data and obtained regression coefficients. Observed linear β = .01

11

Power curve for quadratic age effect

Simulations using Experiment 1 data and obtained regression coefficients (i.e., the observed linear β). Observed quadratic β = -.0007.

12

Power curve for cubic age effect

Simulations using Experiment 1 data and obtained regression coefficients (i.e., the observed linear and quadratic βs). Observed cubic β
= .000007.

13
c. Sunk-cost decision type analyses
We were also interested in the relationship between age and specific response types in sunk-cost vignettes (i.e., Heuristic,
Analytic, Other). The table below shows descriptive statistics of decision types for children/adolescents, adults and older
adults.
Experiment 1 Decision Types by Age-Group
Age Group
Heuristic

2-17 yrs.
33%

18-59 yrs.
47%

60+ yrs.
58%

Analytic

36%

44%

31%

Other

31%

9%

11%

Note. Table presents the percentage of participants in three age-groups that made Heuristic (eat more costly than free pizza), Analytic
(eat the same amount of costly and free pizza) and Other decisions (eat more free than costly pizza). Two- to seventeen-year-olds are
the only age group that makes all three decisions with equal frequency
To formally test age-related differences in these response types (i.e., whether the probability of making each response
type increases and/or decreases with age), we conducted binomial logistic regressions—regressing the response types on
polynomial transformation of age. Because we were primarily interested in the probability of making a Heuristic versus
Analytic response, and how this probability changed with age, we conducted two separate logistic regressions. The first
predicted the probability of a Heuristic versus Analytic response (excluding Other responses), and the second predicted the
probability of making an Other response versus any other response types 1.
As with continuous age, we predicted relative response probabilities from linear, quadratic, and cubic age. The
probability of making a Heuristic response relative to an Analytic response (excluding participants who made an Other
response) increased linearly with age, z(538) = 3.11, p = .002 (neither quadratic nor cubic terms were significant, zs < .76, ps >
.45). The probability of making an Other response (relative to any other response type) was highest in early childhood and
We conducted these two analyses instead of three separate regressions (i.e., predicting each decision type as a “Yes”/”No” dichotomous variable from age)
because the three decision types were not fully independent (i.e., if a participant made a Heuristic decision, they could not have made an Other decision).

1

14
decreased with age, with significant linear (z(678) = 5.71, p < .001) and quadratic (z(678) = 2.68, p = .007), but not cubic
(z(678) = .41, p = .68) terms. For further interpretation of these analyses, see the figure below for a plot of predicted logistic
regression lines for the final chosen response models.

Note. Regression band = 95% CIs on the fit lines. For the left panel, higher values on the vertical axis indicate a higher
probability of making a Heuristic response versus an Analytic one. For the right panel, higher values on the vertical axis
indicate a higher probability of making an Other response (versus a Heuristic or Analytic one). The probability of making a

15
Heuristic decision (vs. an Analytic decision) increased with age, and the probability of making an Other decision decreased
with age.
d. Exploratory analysis excluding participants age <5

In Experiment 1, under-5s accounted for roughly 5% of the sample. In this exploratory analysis, we excluded these participants and
examined the continuous age trajectory again, specifically noting whether there were differences in when the SCE “emerged” (i.e.,
regression band lower-bound first excluded 0/no SCE) and if at all, “disappeared” (i.e., when the regression band lower-bound
includes 0/no SCE after being previously positive). When excluding these participants, the SCE ‘emerged’ at 12 and disappeared at 92,
similar to what we observed in the analysis which included these participants.

16
3. Experiment 2 Supplementary Results
a. Demographics
Age group
n
3-17
180
18-59
105
60+
93

Sample age breakdown
Participant gender (self- or parent-reported)
Gender
Female
Male

n
218
156

17
Non-binary/other
Unreported

2
2

Participant ethnicity (self- or parent-reported)
Ethnicity
white
White
asian
Asian
W
white
White
other
a
asian
w
black
caucasian
Caucasian
caucasian
Indian
mixed (asian and white)
Canadian
east indian
hispanic
Hispanic
O
south asian
A

0

n
200
156
56
38
24
20
20
12
8
8
8
6
6
6
6
6
6
6
4
4
4
4
4
4
2
2

18
a6
Aboriginal
aboriginal, viking,
Anglo saxon
AsiamNN
Asian
Asian and middle eastern and
white
asian and white
Asian and white
asian caucaisian
asian causaisian
asian hispanic
Asian white
asian, middle eastern and white
Asian/white
Asiian
Australia
Black
Canada
Canadian
Caucasian
Caucasion
caucasuian
Causasian
Chinese
chinese and russian
East Indian
east indian
east indian cauacasian
Eurasian
european

2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2

19
farsi
Female
Fidst nations and white
fijian
first nation
First nation and white
First nations
half asian half swedish
half asian half white
Half asian half white
Hispanic
indian
indian
indo canadian
Indo Canadian
Jewish
M
Middle eastern
Mixed
Mixed (European/Asian)
mixed (south asian and persian)
mixed (white and asian)
Mixed asian white
Mixed filipino hungarian
nepalise, indian, scotish
New Zealand European
Other
Persian
Punjabi
South Asian
South Asian (india) but european
citizen

2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2

20
South Indian Fijian/White
Very mixed.
whiTE
White and Hispanic
white, asian, hispanic
Whte

2
2
2
2
2
2

b. Post-hoc power simulations
Because our sample size for Experiment 2 was smaller than that of Experiment 1, our power analyses were based on our
primary pragmatic objective of detecting the presence versus absence of age effects in each of our Vignette × Perspective conditions
(e.g., whether there was no age-SCE relation in any of our four Vignette × Perspective conditions).Via post-hoc power simulations
similar to those conducted for Experiment 1 2 using our exact sample size, age range, and estimated regression residuals, we
determined that we had sufficient power (≥ .80) in the 1st-person perspective condition (n = 209) to detect a linear age effect of |β| ≥
.008, and assuming that minimum linear age effect, a quadratic age effect of |β| ≥ .0005. There were slightly fewer participants in the
3rd-person perspective condition (n = 169), providing sufficient power (≥ .80) in this condition to detect a linear age effect of |β| ≥ .01,
and assuming that minimum linear age effect, a quadratic age effect of |β| ≥ .0015. Our pre-registered sample size planning was
primarily based on categorical age group analyses. Because we opted to report only our pre-registered analyses in-text, we conducted
these post-hoc, non-pre-registered power simulations 3. These power analyses suggested that our Experiment 2 sample size was
adequately powered to detect age effects slightly smaller than the effects observed in Experiment 1. Thus, any null age effects in
Experiment 2 can be interpreted as a failure to replicate the patterns we observed in Experiment 1.

Our pre-registered sample size planning was primarily based on categorical age group analyses. Because we opted to report only our continuous age analyses
in-text, we conducted these post-hoc, non-pre-registered power simulations to provide context for the results.
2

3

Code for these simulations can be found in our Experiment 2 analysis script file.

21

c. Preregistered categorical age analysis
i. NHST linear models
Descriptive statistics by Age Group, Vignette, & Perspective
Age Group
3-17
3-17
3-17
3-17
18-59
18-59
18-59
18-59
60+
60+
60+
60+

Vignette
Pizza
Pizza
Puzzle
Puzzle
Pizza
Pizza
Puzzle
Puzzle
Pizza
Pizza
Puzzle
Puzzle

Perspective
N
SCE m
SD
SE
CI
rd
3 -person
80 0.4875000 1.4582263 0.1630347 0.3245124
st
1 -person
100 0.1900000 1.4750107 0.1475011 0.2926741
rd
3 -person
80 -0.0250000 1.5342091 0.1715298 0.3414215
st
1 -person
100 0.2500000 1.2339884 0.1233988 0.2448501
3rd-person
47 1.2553191 1.4813050 0.2160705 0.4349274
st
1 -person
58 0.7758621 1.4270950 0.1873868 0.3752355
rd
3 -person
47 0.4042553 1.5415911 0.2248642 0.4526281
1st-person
58 0.3793103 1.3089112 0.1718685 0.3441607
rd
3 -person
42 1.4523810 1.3828634 0.2133805 0.4309306
st
1 -person
51 0.9019608 1.0247907 0.1434994 0.2882270
3rd-person
42 0.5000000 1.4356098 0.2215194 0.4473675
st
1 -person
51 0.3921569 0.9813956 0.1374229 0.2760220

Categorical age final model: Pairwise comparisons
Comparison
3-17 vs. 18-59
3-17 vs. 60+
18-59 vs. 60+
Pizza vs. Puzzle (1stperson)

t-value
-3.6935796
-4.7883046
-0.7678587
1.6849545

p-value
0.0002498
0.0000023
0.4430314
0.0934971

Wilcox pvalue
0.0001622
0.0000025
0.4214404
0.0545671

d
-0.3229936
-0.4157594
-0.0766177
0.1581147

d 95% CI lower
-0.4945753
-0.5948526
-0.2746447
-0.0274926

d 95% CI upper
-0.1514120
-0.2366663
0.1214093
0.3437220

22
Pizza vs. Puzzle (3rdperson)

4.8992424

0.0000022

0.0000039

0.4733023

0.2729110

0.6736936

Vignette * Perspective interaction results
Comparison
Pizza vs. Puzzle (1stperson)

t-value
1.684954

p-value
0.0934971

Wilcox p-value
0.0545671

d
0.1581147

d 95% CI lower
-0.0274926

d 95% CI upper
0.3437220

Pizza vs. Puzzle (3rdperson)

4.899242

0.0000022

0.0000039

0.4733023

0.2729110

0.6736936

23

SCE by age group and vignette for 3rd-person perspective vignettes. Error bars = 95% CIs (Between-subjects)

24

SCE by age group and vignette for 1st-person perspective vignettes. Error bars = 95% CIs (Between-subjects)

25
ii. Bayesian linear models

Simulated distributions of SCE by Age Group, Vignette, and Perspective drawn from the prior distribution. Priors are broad, but age
group priors are drawn from the results of the previous experiment. We also allow more variability in the effects of the Puzzle
Vignette and the Self/1st-person perspective.
Model comparison results
Step
Add vignette main effect
Add perspective main effect
Add age group main effect

p-value
0.0000061
0.1456250
0.0000079

Significant
BF Evidence for effect
Yes
3.376413e+03 Yes
No
9.130000e-05 No
Yes
1.571802e+07 Yes

26
Add vignette * perspective interaction
Add vignette * age group interaction
Add perspective * age group interaction
Add vignette * perspective * age group interaction

0.0070519
0.0476084
0.4114347
0.9466934

Yes
No
No
No

6.304673e+00
2.710917e+00
3.878093e-01
7.025207e-01

Yes
No
No
No

Bayes Factors for effects (Savage-Dickey density ratio method)
Difference
3-17 vs. 18-59
3-17 vs. 60+
18-59 vs. 60+
Pizza vs. Puzzle (3rd-person)
Pizza vs. Puzzle (1st-person)

BF01
BF10
0.0073807 1.354893e+02
0.0008779 1.139026e+03
3.8979056 2.565480e-01
0.0000000 7.280780e+10
6.0907417 1.641836e-01

Summary. Analyses with categorical age provided evidence for main effects of age group and vignette, and an interaction between
vignette and perspective. For the age group main effect, children showed lower SCE than adults and older adults, and the latter two
groups showed similar SCE. For the vignette main effect, the Puzzle vignette resulted in lower overall SCE. For the vignette by
perspective interaction, the 1st-person vignette attenuated differences between the Pizza and Puzzle vignettes.
d. Preregistered decision type analyses
We investigated the effects of Age, Vignette, and Perspective on the relative probability of making a Heuristic response (versus an
Analytic response), and the relative probability of making an Other response (relative to any other response type). See the table below
for descriptive statistics.

27
Experiment 1 Decision Types by Age-Group
Age Group
Heuristic Pizza
Heuristic Puzzle

2-17 yrs.
39%
32%

18-59 yrs.
57%
36%

60+ yrs.
61%
38%

Analytic Pizza
Analytic Puzzle

39%
44%

34%
51%

36%
51%

Other Pizza
Other Puzzle

22%
24%

9%
13%

3%
11%

Note. Table presents the percentage of trials in which participants in three age-groups made Heuristic, Analytic and Other
decisions for pizza and puzzle vignettes.
i. NHST linear models
Our preregistered analytic strategy was the same as that used for the SCE magnitude analysis reported in the previous
section with the following exceptions: We used mixed-effects binomial logistic regressions to account for the dichotomous
nature of our response variables and the multiple measurements for each participant.
Heuristic Versus Analytic Responses
First, we tested the effects of Age (orthogonal polynomial linear and quadratic terms), Vignette, and Perspective on the
probability of making a Heuristic response over an Analytic one. We found significant main effects of Vignette (χ2(1) = 13.68,
p < .001) and Perspective (χ2(1) = 17.42, p < .001), and a significant Vignette × Age interaction (χ2(2) = 10.45, p = .005) 4. All
other terms were non-significant (χ2 < 4.43, ps > .10). The figure below depicts the predictions of the final model.

4

The quadratic age term was significant: z(621) = 3, p = .003, while the linear age term was not: z(621) = 1.92, p = .06.

28
Relative Probability of Making a Heuristic vs. Analytic Response by Age, Vignette, and Perspective

Note: Regression bands = 95% CI on the fit lines. Higher values on the vertical axis indicate a higher probability of making a
Heuristic response (versus an Analytic one).
The probability of making a Heuristic compared to an Analytic response increased from childhood to adulthood and decreased
into older adulthood for the Pizza vignette. The opposite pattern was found for the Puzzle vignette. These analyses show a
more striking Age × Vignette interaction than that observed with SCE magnitude, and also show that participants were less
likely to make a Heuristic response relative to an Analytic response if the vignette was in 1st-person rather than 3rd-person. We

29
also found that the probability of making a Heuristic response decreased after around age 50 (but we again exercise caution
about the results in our “old-old” age group (i.e., 80+) due to the lower n in this age range 5).
Other Responses
We performed the same analysis for the probability of making an Other response (relative to any other response type). We
found significant main effects of Age (χ2(2) = 26.39, p < .001) 6 and Perspective (χ2(1) = 4.20, p = .04), and a significant
Vignette × Perspective interaction (χ2(1) = 9.61, p = .002). The figure below depicts the predictions of the final model.

An exploratory analysis excluding participants age 80+ revealed a more gradual increase from childhood to adulthood and stability in adulthood for the Pizza
vignette, and a gradual linear decrease from childhood to adulthood for the Puzzle vignette. For these decision-type results, it is possible that the smaller sample
in Experiment 2 resulted in the imprecise estimates of “old-old” decisions exerting increased influence on the regressions.

5

6

Both polynomial Age terms were significant, for linear: z(743) = 2.54, p = .01, for quadratic, z(743) = 2.12, p = .03.

30
Relative Probability of Making an Other Response by Age, Vignette, and Perspective

Note: Regression bands = 95% CI on the fit lines. Higher values on the vertical axis indicate a higher probability of making an
Other response (versus a Heuristic or Analytic one). The probability of making an Other response decreased with age.
These analyses support the data pattern we observed in Experiment 1: a higher probability of Other responding in children and
a sharp decrease with age 7. Again, we observed reduced vignette differences and a lower probability that participants will
make Other responses with 1st-person vignettes relative to 3rd-person vignettes.

7

We observed nearly identical results when excluding participants age 80+, likely due to the larger and more consistent effect of age on Other responding.

31
ii. Bayesian linear models
Priors

32

Simulated draws from the prior distributions for p(Heuristic vs. Analytic decision) as a function of Age effects, Vignette effects, and
Perspective effects. Age priors are based on the results of Experiment 1, all other priors are broad/uninformative.
Bayes Factors for effects (Savage-Dickey density ratio method) in full model
Effect
Linear age main effect
Quadratic age main effect
Vignette main effect
Perspective main effect
Perspective X linear Age interaction
Perspective X quadratic Age interaction

BF01
2.8935440
1.8370854
0.0191440
0.1795317
1.0560956
1.3713790

BF10
0.3455970
0.5443405
52.2357768
5.5700461
0.9468840
0.7291930

33
Vignette X linear Age interaction
Vignette X quadratic Age interaction
Vignette X Perspective X linear Age interaction
Vignette X Perspective X quadratic Age interaction
Vignette X Perspective interaction

0.7607867
0.8634161
0.9648918
0.9546586
0.2670308

Bayes Factors for effects (Savage-Dickey density ratio method) in final model
Effect
Linear age main effect
Quadratic age main effect
Vignette main effect
Perspective main effect
Vignette X linear Age interaction
Vignette X quadratic Age interaction
Vignette X Perspective interaction

BF01
2.9615515
1.5232002
0.0204028
0.0146163
0.4930452
1.0984552
0.1829440

BF10
0.3376608
0.6565125
49.0129349
68.4168521
2.0282115
0.9103694
5.4661523

1.3144289
1.1581901
1.0363856
1.0474949
3.7448867

34
Summary. NHST analyses suggested main effects of vignette (such that the Puzzle vignette resulted in a lower probability of making a
Heuristic decision relative to an Analytic one) and perspective (such that the 1st-person perspective resulted in a lower probability of
making a Heuristic decision relative to an Analytic one), an interaction between age and vignette (such that we observed a reverse-Ushaped pattern for the Pizza vignette and a U-shaped pattern for the Puzzle vignette), and an interaction between vignette and
perspective (such that vignette differences were attenuated for 3rd-person perspective vignettes). However, we only observed
compelling Bayesian evidence for the vignette and perspective main effects. Also, given that we had a smaller sample than Experiment
1 and the estimates of the fit line were quite variable, we advise caution in interpreting these results.
Other decisions

Simulated draws from the prior distributions for p(Other decision) as a function of Age effects, Vignette effects, and Perspective
effects. Age priors are based on the results of Experiment 1, all other priors are broad/uninformative.

35
Bayes Factors for effects (Savage-Dickey density ratio method) in full model
Effect
Linear age main effect
Quadratic age main effect
Vignette main effect
Perspective main effect
Perspective X linear Age interaction
Perspective X quadratic Age interaction
Vignette X linear Age interaction
Vignette X quadratic Age interaction
Vignette X Perspective X linear Age interaction
Vignette X Perspective X quadratic Age interaction
Vignette X Perspective interaction

BF01
0.0000000
1.4024904
2.0909893
0.0923181
0.7316204
1.2200428
1.2860774
1.1763555
0.9490999
1.0256644
0.5337656

Bayes Factors for effects (Savage-Dickey density ratio method) in final model
Effect
Linear age main effect
Quadratic age main effect
Vignette main effect
Perspective main effect
Vignette X Perspective interaction

BF01
0.0000000
1.0565215
2.2973948
0.0150357
0.4592188

BF10
2.866845e+12
9.465023e-01
4.352757e-01
6.650833e+01
2.177611e+00

BF10
3.864815e+14
7.130174e-01
4.782425e-01
1.083211e+01
1.366829e+00
8.196434e-01
7.775582e-01
8.500832e-01
1.053630e+00
9.749778e-01
1.873482e+00

36
Summary. Like in Experiment 1, analyses suggest a sharp decrease in Other responding from childhood to adulthood. We did observe
evidence for a perspective main effect (such that the 3rd-person perspective resulted in higher rates of Other responding) and a vignette
by perspective interaction (such that vignette differences were attenuated in the 1st-person perspective relative to the 3rd-person
perspective). However, Bayesian evidence for/against effects was generally weak, with the exception of extreme evidence for the age
effect.
e. Exploratory memory error analyses
i. Memory errors by Age group, Vignette, and Perspective
We also measured memory errors for each vignette (i.e., whether a participant correctly remembered whether a given vignette
included sunk-costs (i.e., baked pizza, bought puzzle) or not. Preliminary analyses revealed a high rate of memory errors in our
youngest age group (~40% made an error on at least on vignette). As such, we were interested in 1) the potential relationship between
Age (and our other variables) on memory errors, and 2) whether the amount of memory errors made affects the SCE.
We first assessed the relationship between categorical age and memory errors. To do so, we conducted simplified versions of
our main analysis: single NHST and Bayesian models including all terms. NHST analyses revealed that the only significant effect to
survive Bonferroni correction was an Age group main effect (p < .001). This was corroborated with a simple Bayesian ANOVA using
default prior specifications (Rouder, Morey, Verhage, Swagman, & Wagenmakers, 2016), BF10 = 1.91 * 107. Follow-up pairwise
NHST and Bayesian t-tests using default prior specifications (Rouder, Speckman, Sun, Morey, & Iverson, 2009) revealed that children
had higher memory errors than adults (p < .001, BF10 = 63.66) and older adults (p < .001, BF10 = 2.01 * 103), but that adults and older
adults did not differ (p = .44, BF01 = 6.79). For continuous age, NHST analyses did not reveal any significant effects (ps > .02), but a
simple Bayesian regression revealed evidence for linear and quadratic age effects (BF10s > 1.22 * 109). An examination of the
regression pattern supported the categorical analyses, but was also suggestive of a potential increase in memory errors in older
adulthood. Despite this, memory errors were low on average (0 to .25 out of 2 possible errors per vignette).
1. Categorical age
Pairwise comparisons for Age group main effect
Comparison
3-17 vs. 18-59

t-value
-3.6935796

p-value
0.0002498

Wilcox pvalue
0.0001622

BF10
63.6663895

d
-0.3229936

d 95% CI lower
-0.4945753

d 95% CI upper
-0.1514120

3-17 vs. 60+

-4.7883046

0.0000023

0.0000025

2009.7830884

-0.4157594

-0.5948526

-0.2366663

18-59 vs. 60+

-0.7678587

0.4430314

0.4214404

0.1473534

-0.0766177

-0.2746447

0.1214093

37

Memory errors by Age group and Vignette for 3rd-person perspective. Error bars = 95% CIs.

38

Memory errors by Age group and Vignette for 1st-person perspective. Error bars = 95% CIs.

39
2. Continuous age

Memory errors by continuous Age and Vignette for 3rd-person perspective. Ribbon = 95% CI on the fit line.

40

Memory errors by Age and Vignette for 1st-person perspective. Ribbon = 95% CI on the fit line.
Summary. Memory errors (i.e., forgetting the sunk-cost nature of a particular trial) were higher in children than in adults, and potentially higher in
older adults than adults, but low overall.

ii. Memory errors as a predictor of the SCE
We assessed the relationships between Memory errors and the SCE by adding Memory errors as a predictor (main effects and
interactions) to our final categorical and continuous age models to see if its inclusion improved model fit. The NHST model
comparison was not significant (χ2 p = .48), and the Bayesian model comparison provided very strong evidence against the inclusion
of Memory error effects (BF01 = 50). For continuous age, the NHST model comparison was not significant (χ2 p = .16), and the
Bayesian model comparison provided extreme evidence against the inclusion of Memory error effects (BF01 = 406.50). Thus, although
memory errors seem to change with age, it is unlikely that they affect the SCE.

41
f. Exploratory analysis excluding participants age <5

For Experiment 2, for the 1st-person Pizza vignette, the SCE ‘emerged’ at age 9 (vs. previously age 11), and ‘disappeared’ at age 88
(vs. previously age 86). For the 3rd-person Pizza vignette, the SCE ‘emerged’ at age 5 (vs. previously age 4), but now even the oldestold showed some SCE on average (vs. previously ‘disappeared’ at age 92). For the 1st-person Puzzle vignette, the SCE ‘emerged’ at
age 20 (vs. previously age 13, the only big change but still in line with our conclusions), and for the 3rd-person Puzzle vignette, the
SCE ‘emerged’ at age 45 (vs. previously age 49), and now ‘disappeared’ at age 87 (previously did not ‘disappear’).