Available via license: CC BY 4.0
Content may be subject to copyright.
RESEARCH ARTICLE
Carbon innumeracy
Amir Grinstein
1,2
*, Evan Kodra
3
, Stone Chen
3
, Seth Sheldon
4
, Ory Zik
5
*
1Associate Professor of Marketing, D’Amore-McKim School of Business, Northeastern University,Boston,
MA, United States of America, 2Associate Professor of Marketing, School of Economics and Business
Administration, VU Amsterdam, The Netherlands, 3risQ Inc., Cambridge, MA, United States of America,
4Sheldon Data, Athens, OH, United States of America, 5Oryzik.com, Brookline, MA, United States of
America
*a.grinstein@neu.edu (AG); zikory@gmail.com (OZ)
Abstract
Individuals must have a quantitative understanding of the carbon footprint tied to their every-
day decisions to make efficient sustainable decisions. We report research of the innumeracy
of individuals as it relates to their carbon footprint. In three studies that varied in terms of
scale and sample, respondents estimate the quantity of CO
2
released when combusting a
gallon of gasoline in comparison to several well-known metrics including food calories and
travel distance. Consistently, respondents estimated the quantity of CO
2
from gasoline com-
pared to other metrics with significantly less accuracy while exhibiting a tendency to under-
estimate CO
2
. Such relative absence of carbon numeracy of even a basic consumption
habit may limit the effectiveness of environmental policies and campaigns aimed at chang-
ing individual behavior. We discuss several caveats as well as opportunities for policy
design that could aid the improvement of people’s quantitative understanding of their carbon
footprint.
Introduction
While political action is already underway in multiple countries in response to urgent calls for
reductions in greenhouse gas emissions such as the widely discussed carbon footprint [1,2,3]
recent political developments in the U.S.[4] create significant uncertainty regarding policies to
combat climate change. Thus, as 58% of Americans worry about climate change [5] and since
consumers and households significantly contribute to greenhouse gas emissions, it is becom-
ing even more important that part of the burden be carried at the individual level [6,7]. How-
ever, individuals are not clear about the relative contribution to greenhouse gas emissions of
their behavior [8]. Metrics such as carbon footprint remain unclear, intangible byproducts of
day-to-day consumer activities, as well as activities that connect to the consumer through com-
plex material and energy supply chains. Such metrics are largely irrelevant for personal deci-
sions by many individuals.
In this research, we measure the degree to which people’s carbon numeracy is tied to their
consumption behaviors. We define carbon numeracy as one’s ability to approximate a correct
value of one’s carbon footprint without resorting to an explicit calculation. Familiar examples
are food calories and ambient temperature, where people can interpret a quantitative signal
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 1 / 14
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Grinstein A, Kodra E, Chen S, Sheldon S,
Zik O (2018) Carbon innumeracy. PLoS ONE 13(5):
e0196282. https://doi.org/10.1371/journal.
pone.0196282
Editor: Juan J. Loor, University of Illinois, UNITED
STATES
Received: June 21, 2017
Accepted: April 10, 2018
Published: May 3, 2018
Copyright: ©2018 Grinstein et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All necessary data
are within the manuscript and Supporting
Information files.
Funding: This study was supported in part by a
research budget provided by Northeastern
University to AG. Additionally, risQ Inc. and
Sheldon Data provided support in the form of
salaries for authors EK, SC, and SS but did not
have any additional role in the study design, data
collection and analysis, decision to publish, or
preparation of the manuscript. The specific roles of
these authors are articulated in the ‘author
contributions’ section.
(e.g., the temperature) to make decisions (e.g., clothing), without performing a calculation [9].
The motivation for our work is the notion that when the belief system and assumptions that
guide climate-related decisions are grounded in physical reality (e.g., in the form of labeling),
decisions are likely to be more effective [10,11]. This requires numeracy. As far as we are
aware, research thus far has not attempted to measure the degree of carbon numeracy of peo-
ple’s consumption behavior. This paper’s main goal is therefore to be the first step toward
addressing this knowledge gap. Specifically, through three studies, this research measures and
analyzes respondents’ ability to estimate the direct carbon emissions of driving a car due to
gasoline consumption. We present a simple test that involves comparing the ability to estimate
the carbon impact of combusting a gallon of gasoline to other daily metrics, such as food calo-
ries in a gallon of whole milk and travel distance. We select gasoline consumption and a gallon
of gas as they reflect one of the simplest and most consistent activities and measures in the
daily life of many people. We believe that the analogy to food calories is especially useful since
calories have been extensively used as a nutrition metric for battling obesity, a role analogous
to carbon footprint in mitigating climate change [12]. First, we report two small-scale surveys
of the general population and university students (studies 1a and 1b, respectively). Next, we
present a large-scale survey that corroborates the initial results among the general population
and addresses multiple constraints of the small-scale studies (study 2).
The rest of this paper is structured as follows. First, we review relevant literature on carbon
numeracy and human decision making. Then we report findings and their implications and
caveats. Finally, we discuss theoretical and policy implications of this work.
Better understanding of the nature of people’s estimation of the carbon footprints of their
behaviors will help policymakers to strengthen greenhouse gas reduction policies and create
more effective communication that is aimed at changing consumer behavior [9,13,14].
Numeracy, estimation error and bias: Theoretical background and
predictions
People’s numerical understanding of the environmental consequences of their consumption
behavior is often viewed as critical for more responsible consumption behavior [15]. Essen-
tially, numeracy creates a context for behavior (i.e., “is my environmental impact low or
high?”). Such numerical understanding when grounded in physical reality (e.g., in the form of
labeling) can transform environmental values into actions [10,11]. For example, it can lead to
an effective behavioral change that can promote more energy savings that offer both environ-
mental and economic benefits [16].
However, the numeracy of individuals as it pertains to decision-making domains is prone
to error [17]. In the specific case of understanding the carbon footprint consequences of one’s
consumption behavior, a recent study [9] suggests that people’s ability to quantify and capture
the impact of their actions on the environment is very limited. The study argues that when
making decisions, people typically rely on common systems and quantitative signals (e.g.,
price, temperature, food calories) that are consistent (i.e., accurate) and that are often prac-
ticed. Often, people’s familiarity and suitable approximations of these metrics–their consis-
tency and the fact that they are often practiced–are used to support decisions [17,18]. Such
numeracy about sustainability is, for the most part, apparently absent. In a recent survey of
UK consumers, only 14% stated that they knew their own personal carbon footprint and 89%
thought carbon labeling is confusing [19].
Lack of numeracy about carbon footprint is tied to consumption decisions and has two
consequences. First, lack of intuition at the individual level is likely to be translated to a highly
variable distribution of perceptions and attitudes among people at the aggregate level [20,21].
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 2 / 14
Competing interests: The authors have read the
journal’s policy and have the following conflicts: EK
and SC received support in the form of a salary
from risQ Inc. and SS received support in the form
of a salary from Sheldon Data. This does not alter
our adherence to all the PLOS ONE policies on
sharing data and materials.
This would require policymakers to design multiple interventions that are aimed at different
consumer groups and invest more in educating individuals. Second, many people will tend to
underestimate the carbon implications of their consumption choices. In a survey of American
households, Attari et al. [22] found that the average respondent underestimated potential
usage of energy and water by a factor of nearly three. Similarly, Gardner and Stern [6] reviewed
multiple situations where consumers underestimate energy usage and its consequences. A pos-
sible explanation is that people tend to view themselves and their behavior in a positive light
[23], underestimating the potential negative consequences of their undesirable behavior
[24,25]. This may be especially pronounced in domains related to environmental issues where
moral norms play a role [26].
Within the context of this research, we predict that innumeracy would be evidenced by car-
bon estimations with very large distribution (i.e., error). We further predict that the distribu-
tion will be biased downward, demonstrating clear underestimation of carbon footprint.
These errors and biases in carbon estimation are expected to be much higher than estimations
that suffer less from innumeracy such as food calories or travel distance.
Materials and methods
Ben-Gurion University’s Guilford Glazer School of Business’ Human Subjects Research Com-
mittee has approved the project “Carbon Illiteracy: Demonstrating a Prevalent Lack of Quanti-
tative Environmental Intuition” (AG_05022015). Written informed consent was obtained
from participants in all our studies.
Study plan
We designed and performed three survey-based studies to analyze the current state of people’s
innumeracy about the carbon footprint associated with using one gallon of gasoline, as com-
pared to other metrics that are ubiquitous in daily decisions and consumption (i.e., food calo-
ries, distances) or that are less ubiquitous (i.e., car weight). Further, the choice of one gallon is
motivated by the desire to compare responses for equivalent volumes using a unit that is most
familiar to consumers in the United States, where the studies were conducted. The studies,
complement one another in several ways (a summary of the key characteristics of the studies
appear in Table 1). First, they involve different participants–from the general population as
well as students. A key reason to conduct the study among students (i.e., not only among the
general population) is the fact that the former are considered more pro-environmental [27],
which make their study a conservative test of our predictions. Second, the studies involve
online and in-person data collection efforts to control for the possible influence of external
information on respondents’ estimations. Third, the studies involve different comparisons to
carbon footprint estimation (e.g., comparison with food calories). Fourth, while the two initial
Table 1. The key characteristics of the three studies.
Study Sample size Population Data collection approach Key measures/comparisons
1a 175 General population Online -CO
2
of 1 gallon of gasoline
-Calories in 1 gallon of whole milk
1b 100 Students Face-to-face -CO
2
of 1 gallon of gasoline
-Calories in 1 gallon of whole milk
-Travel distance from NYC to LA
2 961 General population Online -CO
2
of 1 gallon of gasoline
-Calories in 1 gallon of whole milk
-Travel distance from NYC to LA
-Weight of an average family car
https://doi.org/10.1371/journal.pone.0196282.t001
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 3 / 14
studies involve relatively limited sample sizes (N<200), the third survey targeted 961 partici-
pants. Finally, the large-scale study addressed various potential methodological concerns of
the initial studies, as discussed below.
S1 Text (Methodological Appendix A) provides the detailed surveys for each of the three
studies.
Analytical approach
In each study, every respondents’ estimates were divided by the actual value to obtain the fol-
lowing estimation ratios (study 1a: CO
2
and milk, study 1b: CO
2
, milk and distance, study 2:
CO
2
, milk, distance and car weight):
Rc¼est:CO2
true CO2
;Rm¼est:calories
true calories ;Rw¼est:weight
true weight ;and Rd¼est distance
true distance :
From these ratios, two metrics are defined and calculated: estimation error and estimation
bias.
We define estimation error as the absolute value of the log
10
transformed estimation ratios:
jlog10ðRcÞj;jlog10ðRmÞj;jlog10ðRwÞj;and jlog10ðRdÞj:
This definition is justified as follows: a perfect estimate will result in an estimation error of
zero, |log
10
(1)| = 0, and any underestimate or overestimate will result in a positive number.
The log
10
transformation aims to help reduce the influence of large outliers in subsequent anal-
ysis. Estimation error defined here provides a measure of the innumeracy of a given respon-
dent regarding any of the studied metrics (CO
2
, milk calories, distance or car weight).
Estimation bias is defined and calculated by taking the log
10
of the ratios: log
10
(R
c
),
log
10
(R
m
),log
10
(R
w
),log
10
(R
d
), but without taking their absolute value. These log
10-
transformed
ratios allow us to treat underestimates and overestimates symmetrically. For example, an esti-
mation ratio of 10 (ten times higher than actual) becomes log
10
(10) = 1 and a ratio of 0.1 (one-
tenth of actual) becomes log
10
(0.1) = -1. These values are reported in S1 Text (Methodological
Appendix B). The mean biases and associated 95% confidence intervals are reported as a per-
centage of the actual value. Mean biases and their associated confidence intervals are com-
puted by converting back to original units. For example, a mean bias of log
10
(0.1) = -1 is
reported 10
−1
100% = 10% of the actual value.
In all studies, estimation error and bias are both separately analyzed using a mixed effects
ANOVA model. To determine if there is a difference in the mean estimation error, a mixed-
effects ANOVA model was used. The estimation error was fitted with the factor (CO
2
, milk
calories, distance or car weight) as a fixed effect and the subject as a random effect. See S1 Text
(Methodological Appendix B) for details.
Study 1a: Online study of the general population
Respondents. The first study involved a survey of N = 175 participants (73 women, mean
age = 35) from the U.S. They were recruited online using Amazon Mechanical Turk–an online
tool used widely to crowd source data from the general population. Specifically, Amazon
Mechanical Turk is a web service that coordinates the supply and demand of different tasks
(including academic studies) that require human intelligence to complete. It is an online labor
market where employees (called workers) are recruited by employers (called requesters) for
the execution of the tasks in exchange for a wage (called a reward) while both workers and
requesters remain anonymous [28].
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 4 / 14
The initial sample for the study involved 208 participants. However, we excluded from the
analysis 33 participants who did not follow the instructions: 22 assessed CO
2
estimation across
all measurement units when asked to estimate using only a single unit and an additional 11 did
not estimate both CO
2
and food calories. We note that removing the 33 respondents does not
qualitatively change the nature of the core findings. Data set can be found in S2 Table.
Procedure. We asked respondents to use only their intuition and estimate 1) the amount
of CO
2
emitted by burning one gallon of gasoline in their car, and 2) the food calories con-
tained in 1 gallon of standard dairy milk. In the case of CO
2
emissions, we allowed for answers
in U.S. customary units (e.g., pounds) and metric units (e.g., kilograms) to make estimation
easy across American and non-American participants. The actual values are approximately 9
kg of CO
2
[29], and 2,400 Calories, respectively. Specifically, we asked: “How much CO2 do
you think is emitted in the production/consumption of 1 gallon of standard gasoline in your
car?” and, “How much calories do you think are contained in 1 gallon of standard dairy milk?”
For each respondent, the sequence of these questions was randomized to minimize any
order effects. We also asked about general environmental involvement, based on the Con-
sumer Involvement Scale [30]. The scale is a five-point semantic differential scale that consists
of five adjectives relating to the statement–“I find environmental and ecological issues. . .”:
“not important/important,” “not essential/essential,” “not valuable/valuable,” “not interesting/
interesting,” and “not significant/significant.” Finally, we asked respondents about potential
educational background that might influence their results (e.g., background in sustainability,
chemistry or physics).
Analysis and results. The main finding is that mean estimation error in the case of CO
2
is
significantly higher than for calories as determined by mixed-effects ANOVA (see Tables 2
and 3). Specifically, holding all else constant, the mean estimation error for calories in milk is
0.450 (95% confidence interval (CI) of 0.38–0.517) and 1.145 for CO
2
from gasoline (95% CI
of 0.994–1.295). Secondly, holding all else constant, there is a tendency to underestimate the
amount of CO
2
from gasoline with a mean bias of -0.750, which translates to 17.7% of the
actual value (95% CI of 11%-28%), similar to that for calories in milk, which holding all else
constant has a mean 44% of the actual value (95% CI of 37–53%).
Importance of environmental issues and potentially relevant educational background were
not factors that had a significant effect on either estimation error or bias. We also report in S1
Table the results in percent bias of the real value for all studies.
Study 1b: Face-to-face study of the student population
In study 1b, we sought to further explore the above observations but also to extend our analysis
in three ways. First, a concern with study 1a is that participants could have consulted external
information during estimation (e.g., online carbon footprint calculators), although they were
explicitly instructed not to. Thus, study 1b adopts an in-person moderated procedure with no
Internet access that eliminates this concern. Second, study 1a involved a general population
sample. In study 1b, we examine whether the findings hold with a sample of university students,
potentially characterized by higher levels of environmental awareness [31]. Third, in study 1a
we compared estimates of CO
2
in standard gasoline to those of calories in whole milk. One may
argue that respondents could have especially high interest in developing a strong numeracy
about calories because they have a direct personal impact on one’s health. Therefore, in study
1b, apart from a comparison to calories in a gallon of milk, we also prompt participants to esti-
mate a less personal, but nevertheless practiced, metric of travel distance. Travel distances, espe-
cially long ones, are not likely to be familiar to most people, but we might expect respondents to
have relatively higher numeracy about distance in general compared to carbon footprint.
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 5 / 14
Respondents. One hundred student participants were recruited by a research assistant on
the campus of a large American university. Our sample is comprised of students from 22 dif-
ferent majors, attesting to the academic diversity of the students participating in this study.
Data set can be found in S3 Table.
Procedure. As in study 1a, we first asked the respondents to use only their intuition and
estimate both the amount of CO
2
emitted by burning one gallon of gasoline in their car in one
of the following measurement units–pounds, kilograms, grams or metric tons as well as the
number of calories in one gallon of whole milk. We used the same questions as in study 1a. The
students were then asked to estimate the travel distance from Los Angeles to New York City in
miles or kilometers (approximately 3900 km). We used the same questionnaires as in study 1a
(however, gender and age information were not collected) including one item from Mittal’s
[30] importance of environmental issues scale (“I find environmental and ecological issues. . .
not important/important”) and a question about the student’s major at the university.
Analysis and results. As a starting point, we tested the assumption that the student popu-
lation from study 1b would view environmental topics as more important than the general
population studied in study 1a. A two-sided t-test suggests that the student population rates
environmental importance (EI) significantly higher on average. The mean estimate of EI for
study 1b’s student respondents was 4.46, while the EI for the respondents in study 1a was 4.15
(t = 3.05, p = 0.003).
Similar to the first study, holding all else constant, we find significantly larger estimation
error for CO
2
than calories or estimates of distance from Los Angeles to New York as deter-
mined by the mixed-effects ANOVA model (p-value <0.001; findings are reported in Tables 1
and 2, as well as Fig 1).
The mean estimation error of CO
2
emission from burning gasoline is 2.364 (95% CI of
2.062–2.665 and for calories in whole milk is 1.179 (95% CI of 1.095–1.264), corroborating
results from Study 1a. Further, the mean estimation error in kilometers to travel from Los
Angeles to New York City is 0.742 (95% CI of 0.614–0.869).
As with study 1a, there is a significant underestimation tendency for the amount of CO
2
released from combusted gasoline, with a mean estimation bias of 0.44% of the actual value
(95% CI of 0.22%-0.89%) as well as calories in milk with a mean estimation bias of 6.7% of the
actual value (95% CI of 5.5%-8.1%).
Estimates of carbon from gasoline vary significantly more than estimates of travel distance.
Interestingly though, in contrast to CO
2
and calories, there is a tendency to overestimate dis-
tance with a mean estimation bias of 250% of the actual value (95% CI of 166%-377%). This
finding is in line with the notion that people overestimate objects or realities that lack visibility
like travel distances [21].
Table 2. Mixed-effects ANOVA outputs for estimation error.
Factor Mean Std. Error DF t-value p-value
Study 1a CO
2
in gasoline 1.145 0.059 174 19.269 <0.000
Calories -0.695 0.082 174 -8.494 <0.000
Study 1b CO
2
in gasoline 2.364 0.100 198 23.743 <0.000
Distance -1.622 0.141 198 -11.522 <0.000
Calories -1.184 0.141 198 -8.412 <0.000
Study 2 CO
2
in gasoline 0.867 0.019 2880 45.933 <0.000
Distance -0.525 0.026 2880 -20.338 <0.000
Calories -0.169 0.026 2880 -6.564 <0.000
Car Weight -0.460 0.026 2880 -17.808 <0.000
https://doi.org/10.1371/journal.pone.0196282.t002
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 6 / 14
Study 2: Large-scale online study of the general population
Respondents. Study 2 involved a survey of N = 961 participants (496 women, mean
age~ = 42) from the U.S. As with study 1a, they were recruited online using Amazon Mechani-
cal Turk. The initial sample for the study involved 1,046 participants. However, we excluded
from the analysis 70 participants that did not put an appropriate estimate (e.g., “I don’t know,”
“little,” “5 percent”). An additional 15 were removed for estimating 0 for either pounds of
CO2, calories of milk, distance between NY and LA, and weight of a car. Data set can be found
in S4 Table.
Procedure. We first asked respondents to use only their intuition and estimate both the
amount of CO
2
emitted by burning one gallon of gasoline in a motor vehicle (to avoid confu-
sion we only asked participants to use pounds; “What quantity of CO2 do you think is emitted
by consuming 1 gallon of standard gasoline when driving a motor vehicle?”), as well as the
number of calories in one gallon of whole milk (in calories; “How many calories do you think
are contained in 1 gallon of whole milk?”), the distance between Los Angeles and New York
City (in miles; “What do you think is the distance between Los Angeles and New York (coast-
to-coast)?”) and another estimation that unlike the previous ones has no direct personal rele-
vance to day-to-day lives of consumers and is less practiced: the weight of an average family
car (in pounds; 1,850 kg [32]; “What do you think is the weight of an average family car in the
U.S.?”). The estimation requests were presented in a random order.
We then included a single item that checked participants’ attention (“You need to check the
third answer in the following question”). All participants successfully passed the attention
check. To examine if results depend on pro-environmental attitudes, we included a single item
about environmental importance (based on Mittal [30]). Finally, to examine if results depend
on political orientation (conservative vs. liberal) we included an additional set of items [33,34].
Demographic questions concluded the survey; these can be found in detail in S1 Text (Meth-
odological Appendix A).
Table 3. Summaries of mean estimation error and bias with 95% CIs (all experiments).
Factor Mean Std. Error Lower Bound Upper Bound
Study 1a CO
2
in gasoline (bias) -0.75 0.1 -0.95 -0.55
Calories -0.35 0.04 -0.43 -0.27
CO
2
in gasoline (error) 1.15 0.08 0.99 1.3
Calories 0.45 0.03 0.38 0.52
Study 1b CO
2
in gasoline (bias) -2.36 0.16 -2.66 -2.05
Distance 0.4 0.09 0.22 0.58
Calories -1.18 0.04 -1.26 -1.09
CO
2
in gasoline (error) 2.36 0.15 2.06 2.67
Distance 0.74 0.07 0.61 0.87
Calories 1.18 0.04 1.1 1.26
Study 2 CO
2
in gasoline (bias) -0.58 0.03 -0.64 -0.51
Distance 0.12 0.02 0.09 0.16
Calories -0.62 0.02 -0.66 -0.58
Car weight -0.27 0.02 -0.3 -0.23
CO
2
in gasoline (error) 0.87 0.02 0.82 0.91
Distance 0.34 0.02 0.31 0.37
Calories 0.7 0.02 0.66 0.74
Car weight 0.41 0.02 0.38 0.44
https://doi.org/10.1371/journal.pone.0196282.t003
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 7 / 14
Analysis and results. As with the first two studies, the difference in mean estimation
error was determined with a mixed-effects ANOVA model. The model determined that mean
CO
2
estimation error was significantly larger than mean estimation error of milk, distance and
car weight (p-values all <0.001; findings are reported in Tables 1and 2, as well as Fig 1).
From largest to smallest, the mean estimation error for each was: 0.867 for CO
2
from gaso-
line, 0.698 for calories in milk, 0.408 for car weight, and 0.342 for distance from Los Angeles to
New York City.
As with studies 1a and 1b, there is a tendency to underestimate the amount of CO
2
from a
gallon of gasoline with an estimate of 27% of the actual value (95% CI of 23%-31%) and calo-
ries in a gallon of milk with a mean estimate of 24% of the actual value (95% CI of 22%-26%).
The weight of a car also tends to be underestimated with a mean estimate of 55% of the actual
Fig 1. Results of studies 1a, 1b, and 2 –carbon versus calorie estimations.
https://doi.org/10.1371/journal.pone.0196282.g001
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 8 / 14
value (95% CI of 50%-59%). Again, distance in miles from Los Angeles to New York City
tends to be overestimated with a mean estimate of 134% of the actual value (95% CI of 122%-
145%).
Fig 2 describes the findings of study 2 in a visual, more simplified manner.
Discussion and conclusions
Results from three studies provide evidence supporting the hypothesis that people have a
higher degree of carbon innumeracy tied to consumption than they have for other more com-
monly practiced and understood metrics like calories and distances (and even more than less
practiced metrics like the weight of an average family car). We define estimation error and
bias measures and analyze those through mixed effect ANOVA models. Results show that on
Fig 2. A simplified visualization of study 2’s results.
https://doi.org/10.1371/journal.pone.0196282.g002
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 9 / 14
average respondents consistently estimate the amount of CO
2
from one gallon of standard gas-
oline significantly less accurately than they are able to estimate the number of calories in a gal-
lon of whole milk, the travel distance from Los Angeles to New York City, or the weight of an
average family car. The higher estimation error may not come as a surprise since, unlike food
calories or distance that are practiced by individuals in everyday life [17], measurements of
CO
2
are not part of everyday decisions. This innumeracy is also evident in our final survey
where CO
2
estimates fall short even when compared to respondents’ ability to estimate the
weight of an average family car, even though car weight is not a measure that is practiced
commonly.
Overall, the results provide evidence that, despite increasing public awareness and discourse
about climate change, environmental impact, and carbon footprinting [5,6] people cannot esti-
mate their carbon footprint associated with the common practice of driving a motor vehicle.
Further, even respondent samples that on average are more pro-environmental (students in in
study 1b) tend to exhibit lower carbon numeracy compared to calories and distance.
We also consistently find a statistically significant tendency to underestimate the amount of
CO
2
from combusting gasoline. This bias is directionally consistent with prior research of
American households, in which respondents underestimated potential usage of energy by a
factor of nearly three [22] or four [6]. This underestimation bias could be explained by people’s
tendency to underestimate the potential negative consequences of their undesirable behavior
[24,25].
Remaining questions include whether general carbon innumeracy also characterizes
experts and policymakers, not only laymen, and to what extent their carbon numeracy is corre-
lated with the climate efficacy of their decisions. This may be especially valuable in the context
of battling climate change as policymakers and sustainability experts are the ones designing
policies and campaigns that aim to make a change [6,16]. Future research should address this
question rigorously, for example, surveying environmentally savvy participants such as stu-
dents enrolled in a sustainability programs or environmental consultants.
Conclusions and policy implications
We have provided multiple cases of heuristic empirical evidence for carbon innumeracy,
exemplified using a gallon of gasoline as the simplest way of capturing the consumption habit
of using a car.
Our findings point to the need by policymakers, responsible marketers, NGOs, and aca-
demic scholars to remedy the ways in which individual carbon and other environmental foot-
prints are quantified and communicated. One way to interpret Kahneman’s work [17] is that
numeracy (and broadly quantitative intuition) will require both consistency and practice. En-
hancing carbon numeracy may be achieved by multiple, non-mutually exclusive approaches,
which we discuss next.
As a starting point, it is critical that the available carbon information is consistent. For
instance, given the complex sources of energy in the U.S. (e.g., coal, solar, hydro, wind) it is
not often clear how efficient would be an electric car [35]. Practicing an inconsistent signal will
be less efficient in reducing the estimation error and will not contribute to carbon numeracy.
Further efforts by scientists and policy makers should focus on the consistency of carbon foot-
print calculations in various systems, especially those that involve complex, location dependent
systems such as the emissions tied to electricity consumption [36].
As far as practicing carbon footprint, the traditional approach is to create more knowledge
about, and awareness for, carbon footprint and its meaning. For example, creating an educa-
tional campaign that establishes carbon as a metric for environmental impact. Environmental
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 10 / 14
campaigns however are costly and there is limited evidence as to their effectiveness [37]. A dif-
ferent way to implement this approach is to integrate carbon footprint information in consum-
ers’ physical space. For example, providing carbon footprint information on consumer goods
packaging. Prior research suggests this can influence consumers’ decision making significantly
[10,11]. A related example is to implement at various instant feedback mechanisms (e.g.,
energy use indoor digital meters, mobile apps). A prerequisite for an effective implementation
of any policy involving product labeling or other awareness building mechanisms is carbon
numeracy. Our work exposes the gap that we should bridge.
A second approach may be to establish a more intuitive alternative metric than carbon foot-
print, based on preexisting numeracy that rigorously represents carbon emissions as a proxy.
For example, one can conceive of augmenting carbon footprint with “climate points,” “energy
points,” or equivalencies that reflect negative monetary externalities that are equal to or associ-
ated with a kilogram of CO
2
or the embodied energy of a gallon of gasoline, respectively. This
concept is similar to the way Weight Watchers induced dietary behavior using a points system
to augment calories [38]. While the use of calories leads to various biases [18,39] and it may be
more effective for some populations [40], overall there is a positive, albeit modest, positive
effect of calorie food labeling on food consumption [41]. Essentially, food calories and labeling
can more broadly work by filling an information gap that may have an effect on subsequent
decision making [40].
The advantage of the latter approach is having a metric that accounts for carbon but is less
prone to estimation errors and not as dependent on expensive educational campaigns that are
limited in effectiveness. This approach might be able to establish an environmental parallel of
“calories” to quantify the impact of resource consumption of a given product or service. Such a
metric will be especially valuable as it will enable people to make better environmental deci-
sions when comparing alternative behaviors with environmental impact that carries different
metrics (e.g., installing solar panels vs. eating local food). One promotional approach expected
to be relatively low cost and effective would be to display this type of metric or environmental
informatics in the general marketplace (e.g., near gasoline prices at gas stations; [42]).
Limitations and future research
We highlight several caveats that can be addressed by future research. First, gasoline and milk
were selected because of their simplicity, ubiquity, consistency and similarity in price per vol-
ume, but we did not independently determine whether gasoline and milk are representative of
decisions that involve carbon and calories, respectively.
In addition, considering the issue of consistency, we chose a gallon of gas as it is perhaps
the simplest and most consistent example. Thus, across our studies respondents did not have
to estimate the carbon footprints tied to other products or services that vary in composition.
Other consumption staples such as food, water, and electricity indirectly produce greenhouse
gas consequences that vary by space and time, and are thus less suggestive to our study but
should be included in future studies.
Second, in the current study we examined respondents’ absolute intuitive knowledge about
carbon footprint strictly as it relates to standard gasoline. Future work may study relative
knowledge on carbon footprints, for instance, by asking people to rank by impacts of various
consumption behaviors from low to high on their estimated carbon footprint, as this may be
highly relevant to people’s decision making.
Third, we should point out that our measurement approach was deliberately U.S.-centric
(using gallons for example), but can be expanded to other countries. We nevertheless make
sure to report our findings in metric units to enable generalizability.
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 11 / 14
Fourth, the question about CO
2
in surveys 1a and 1b asks about the “production/consump-
tion” of 1 gallon of standard gasoline in one’s car. Initially, we considered including the indi-
rect carbon emissions associated with the gallon of gasoline itself (i.e., crude oil extraction, oil
refining, prior to combustion by the driver), but ultimately decided to focus on direct emis-
sions for simplicity. Unfortunately, the artifact was not caught before the surveys were admin-
istered. Given the context and low probability that the respondents would understand the
phrase to be asking about emissions outside of the boundary of a car engine, and the consistent
in results with survey 2, in which the error was addressed, we think that the impact on the
results is minimal.
Finally, future work may find it useful to study more rigorously whether estimation error
and bias also characterize sustainability experts, not only laymen.
Supporting information
S1 Table. The results in percent bias of the real value for all studies.
(XLSX)
S2 Table. Data_study1a.
(XLSX)
S3 Table. Data_study1b.
(XLSX)
S4 Table. Data_study2.
(XLSX)
S1 Text. Methodological Appendix: A, B.
(DOCX)
Acknowledgments
The authors are grateful for the research assistance of Shruti Naseri and for helpful suggestions
from Dan Ariely, Michael Toffel, Ron Milo, Ann Kronrod, Auroop Ganguly, Matt Eckelman,
and Steven Scyphers.
Author Contributions
Conceptualization: Ory Zik.
Formal analysis: Evan Kodra, Stone Chen, Seth Sheldon.
Methodology: Stone Chen, Seth Sheldon.
Project administration: Amir Grinstein.
Validation: Evan Kodra.
Visualization: Evan Kodra, Stone Chen.
Writing – original draft: Amir Grinstein, Evan Kodra, Seth Sheldon, Ory Zik.
Writing – review & editing: Stone Chen, Seth Sheldon.
References
1. Allen MR, Vicente RB, Broome J, Cramer W, Christ R, Churchet JA, et al. IPCC Fifth Assessment Syn-
thesis Report-Climate Change 2014 Synthesis Report. 2014. (available at http://www.citeulike.org/
group/15400/article/13416115).
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 12 / 14
2. Grant D, Bergstrand K. Running effectiveness of U.S. state policies in reducing CO2 emissions from
power plants. Nature Climate Change. 2014; 4(11): 977–982.
3. Williams JH, DeBenedictis A, Ghanadan R, Mahone A, Moore J, Morrow III WR, et al. The technology
path to deep greenhouse gas emissions cuts by 2050: The pivotal role of electricity. Science. 2012; 335
(6064): 53–59. https://doi.org/10.1126/science.1208365 PMID: 22116030
4. https://www.nytimes.com/2017/06/01/climate/trump-paris-climate-agreement.html?_r=0
5. http://climatecommunication.yale.edu/visualizations-data/ycom-us-2016/
6. Gardner GT, Stern PC. The short list: The most effective actions US households can take to curb cli-
mate change. Environment:Science and Policy for Sustainable Development. 2008; 50(5): 12–25.
7. Tukker A, Cohen MJ, Hubacek K, Mont O. The impacts of household consumption and options for
change. Journal of Industrial Ecology. 2010; 14(1): 13–30.
8. Whitmarsh L, Seyfang G, O’Neill S. Public engagement with carbon and climate change: To what extent
is the public “carbon capable”? Global Environmental Change. 2011; 21(1): 56–65.
9. Zik O, Kulatilaka N. The sustainability babel fish. Sustainability Science. 2013; 8(2): 295–300.
10. Cohen MA, Vandenbergh MP. The potential role of carbon labeling in a green economy. Energy Eco-
nomics. 2012; 34:S53–S63.
11. Guenther M, Saunders CM, Tait PR. Carbon labeling and consumer attitudes. Carbon Management.
2012; 3(5):445–455.
12. Carels RA, Konrad K, Harper J. Individual differences in food perceptions and calorie estimation: An
examination of dieting status, weight, and gender. Appetite. 2007; 49(2):450–458. https://doi.org/10.
1016/j.appet.2007.02.009 PMID: 17428574
13. Sterman JD, Sweeney LB. Understanding public complacency about climate change: Adults’ mental
models of climate change violate conservation of matter. Climatic Change. 2007; 80:213–238.
14. Pandey D, Agrawal M, Pandey JS. Carbon footprint: Current methods of estimation. Environmental
Monitoring and Assessment. 2011; 178(1):135–160.
15. Kahan DM, Kahan DM, Peters E, Wittlin M, Slovic P. The polarizing impact of science literacy and
numeracy on perceived climate change risks. Nature Climate Change. 2012; 2(10): 732–735.
16. Dietz T. Narrowing the U.S. energy efficiency gap. Proceedings of the National Academy of Sciences.
2010; 107(37): 16007–16008.
17. Kahneman D. Thinking,Fast and Slow. 2011. Macmillan.
18. Brown RE, Canning KL, Fung M, Jiandani D, Riddell MC, Macpherson AK, et al. Calorie estimation in
adults differing in body weight class and weight loss status. Medicine and Science in Sports and Exer-
cise. 2016; 48(3): 521–526. https://doi.org/10.1249/MSS.0000000000000796 PMID: 26469988
19. Gadema Z, Oglethorpe D. The use and usefulness of carbon labelling food: A policy perspective from a
survey of UK supermarket shoppers. Food Policy. 2011; 36(6): 815–822.
20. Caverni JP, Fabre JM, Gonzalez M (Eds.). Cognitive Biases. 1990; Vol. 68. Elsevier.
21. Tversky A, Kahneman D. Judgment under uncertainty: Heuristics and biases. In Utility,Probability,and
Human Decision Making. 1975; 141–162. Springer Netherlands.
22. Attari SZ, DeKay ML, Davidson CI, De Bruin WB. Public perceptions of energy consumption and sav-
ings. Proceedings of the National Academy of Sciences. 2010; 107(37): 16054–16059.
23. Taylor SE, Brow JD. Illusion and well-being: A social psychological perspective on mental health. Psy-
chological Bulletin. 1988; 103: 193–210. PMID: 3283814
24. Dickerson CA, Thibodeau R, Aronson E, Miller D. Using cognitive dissonance to encourage water con-
servation. Journal of Applied Social Psychology. 2003; 22(11): 841–854.
25. Krizan Z, Windschitl PD. The influence of outcome desirability on optimism. Psychology Bulleting.
2007; 133(1): 95–121.
26. Stern PC. Towards a coherent theory of environmentally significant behavior. Journal of Social Issues.
2000; 56(3): 407–424.
27. Shen J, Tatsuyoshi S. Reexamining the relations between socio-demographic characteristics and indi-
vidual environmental concern: Evidence from Shanghai data. Journal of Environmental Psychology.
2008; 28(1):42–50.
28. Paolacci G, Chandler J, Ipeirotis P. Running experiments on Amazon Mechanical Turk. Judgment and
Decision Making. 2010; 5(5): 411–419.
29. Khachatryan H, Joireman J, Casavant K. Relating values and consideration of future and immediate
consequences to consumer preference for biofuels: A three-dimensional social dilemma analysis. Jour-
nal of Environmental Psychology. 2013; 34: 97–108.
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 13 / 14
30. Mittal B. A comparative analysis of four scales of consumer involvement. Psychology and Marketing.
1995; 12(7): 663–682.
31. Diamantopoulos A, Schlegelmilch BB, Sinkovics RR, Bohlen GM. Can socio-demographics still play a
role in profiling green consumers? A review of the evidence and an empirical investigation. Journal of
Business Research. 2003; 56(6): 465–480.
32. https://www.epa.gov/fueleconomy/trends-report
33. Nail PR, McGregor I, Drinkwater AE, Steele GM, Thompson AW. Threat causes liberals to think like
conservatives. Journal of Experimental Social Psychology. 2009; 45(4): 901–907.
34. Kidwell B, Farmer A, Hardesty DM. Getting liberals and conservatives to go green: Political ideology
and congruent appeals. Journal of Consumer Research. 2013; 40(2): 350–367.
35. How Green is Your Electric Car? 2014. The Wall Street Journal, Aug. 29.
36. Kodra E, Sheldon S, Dolen R, Zik O. The North American electric grid as an exchange network: An
approach for evaluating energy resource composition and greenhouse gas mitigation. Environmental
Science and Technology. 2015; 49(22):13692–13698 https://doi.org/10.1021/acs.est.5b03015 PMID:
26473284
37. Abrahamse W, Steg L, Vlek C, Rothengatter T. A review of intervention studies aimed at household
energy conservation. Journal of Environmental Psychology. 2005; 25(3): 273–291.
38. https://www.google.com/patents/US20060165756
39. Chandon P, Wansink B. The biasing health halos of fast-food restaurant health claims: Lower calorie
estimates and higher side-dish consumption intentions. Journal of Consumer Research. 2007; 34
(3):301–314.
40. Hawkes C, Smith TG, Jewell J, Wardle J, Hammond RA, Friel S, et al. Smart food policies for obesity
prevention. Lancet; 385:2410–2421. https://doi.org/10.1016/S0140-6736(14)61745-1 PMID:
25703109
41. Kiszko KM, Martinez OD, Abrams C, Elbel B. The influence of calorie labeling on food orders and con-
sumption: A review of the literature. Journal of Community Health. 2014; 39(6):1248–1269. https://doi.
org/10.1007/s10900-014-9876-0 PMID: 24760208
42. Ariely D. Eyes off the price. 2008; N.Y.Times.
Carbon innumeracy
PLOS ONE | https://doi.org/10.1371/journal.pone.0196282 May 3, 2018 14 / 14
