Content uploaded by Daniel Nettle
Author content
All content in this area was uploaded by Daniel Nettle on Jul 27, 2018
Content may be subject to copyright.
1
Opportunistic food consumption in relation to childhood and adult
food insecurity: An exploratory correlational study
Daniel Nettle1*, Mona Joly1,2, Eleanor Broadbent3, Chloe Smith3, Ellie Tittle3, Melissa
Bateson1
1. Centre for Behaviour & Evolution and Institute of Neuroscience, Newcastle University,
Newcastle, UK
2. École Normale Supérieure, 45 rue d’Ulm, Paris, France
3. School of Psychology, Newcastle University, Newcastle, UK
* To whom correspondence should be addressed: daniel.nettle@ncl.ac.uk
In press, Appetite, July 2018
Abstract
Food insecurity is associated with high body weight for women but not men in affluent
Western societies. However, it is not currently known what behavioural or psychological
mechanisms drive this association. Moreover, it is also unknown whether only current
experience of food insecurity in adulthood is important, or there are lasting effects of
childhood experience. We carried out a mock ‘taste test’ where 126 adult volunteers had the
opportunity to consume and rate energy-dense snack foods. Current food insecurity was
measured using the standard USDA measure, and in addition, we used a novel measure that
also captures childhood experience of food insecurity. As well as the expected gender-
specific association between current food insecurity and body weight, we found some
evidence for associations between food insecurity and calorie consumption in the taste test,
and liking of one of the foods, chocolate. However, associations between current food
insecurity and the outcomes were moderated by childhood experience of food insecurity, with
greater childhood food insecurity enhancing the positive effect of current food insecurity on
body weight, but attenuating the positive effect of food insecurity on calorie consumption and
liking for chocolate. These findings are exploratory, but they suggest that any effects of food
insecurity in adulthood on eating and the hedonic value of foods may be moderated by
childhood experience.
Keywords: food insecurity, insurance hypothesis, developmental programming, obesity,
liking, food motivation, calorie consumption
2
Introduction
Food insecurity (FI), defined as the limited or uncertain ability to acquire nutritionally
adequate and safe food in socially acceptable ways (Anderson, 1990), is associated with
high body weight for women but not men in affluent Western societies. This pattern has been
found in a number of well-powered epidemiological studies (Adams, Grummer-strawn, &
Chavez, 2003; Hanson, Sobal, & Frongillo, 2007; Martin-Fernandez, Caillavet, Lhuissier, &
Chauvin, 2014; e.g. Townsend, Peerson, Love, Achterberg, & Murphy, 2001), and is
confirmed by a large meta-analysis (Nettle, Andrews, & Bateson, 2017). The association
survives control for socioeconomic status, although socioeconomic status does tend to be
correlated with FI (Gundersen, Kreider, & Pepper, 2011). Although the relationship between
FI and high body weight is often described as ‘paradoxical’ (Crawford & Webb, 2011; Dinour,
Bergen, & Yeh, 2007; Tanumihardjo et al., 2007), Nettle et al. (2017) argue that increased
body weight represents the expected biological response to an insecure food supply. Food-
insecure organisms must insure themselves against periods of shortfall with extra
consumption and energy storage when food is easily available. This principle is supported by
behavioural-ecological models (Higginson, McNamara, & Houston, 2016; Lima, 1986); and in
non-human animals, experimentally imposing food insecurity leads to increased consumption
and/or weight gain (Ekman & Hake, 1990; Li, Cope, Johnson, Smith, & Nagy, 2010; Wilson &
Cantor, 1987; Witter & Swaddle, 1995). Thus, Nettle et al. (2017) argue that the association
between FI and high body weight may be causal: humans—human females at least—may
possess psychological mechanisms that respond to experiences of FI by motivating them to
consume more than they expend when the opportunity is available.
It follows from the energy balance equation for weight gain that there are two (non-mutually-
exclusive) ways an association between FI and high body weight could come about. Either
food-insecure individuals could have higher food intake when eating opportunities present
themselves; and/or food-insecure individuals could have lower energy expenditure. Despite
the large literature on FI and body weight in humans, the direct evidence for higher caloric
intake in food-insecure individuals in sparse. One early study demonstrated poorer diet
quality associated with FI, but without measuring caloric intake (Kendall, Olson, & Frongillo,
1996). A more recent study found that children in food insecure households consumed more
calories overall, and more added sugars in particular, than children in food-secure
households (Sharkey, Nalty, Johnson, & Dean, 2012). However, since the parents rather
than the children will have dictated the available foods, this gives only indirect insight into
how experience of food insecurity alters individual food motivation.
Thus, it remains to be established that food-insecure individuals consume more calories than
non-food-insecure individuals do when opportunities present themselves. If they do, it may
be because FI is associated with higher hedonic value of energy-dense foods, since high
hedonic value of energy-dense foods is related to weight gain (Finlayson, King, & Blundell,
2007).The present study aimed to explore these issues, by presenting adult volunteers with a
standardized opportunity to freely consume energy-dense snack foods, as well as rate how
much they liked them, under the guise of a mock ‘taste test’. Participants drank a preload of
a sugary drink 10 minutes prior to the taste test. This preloading procedure has been used in
previous studies to equalize cues of internal energetic need (Hill, Prokosch, DelPriore,
Griskevicius, & Kramer, 2016; Wang & Dvorak, 2010). Thus, any excess calorie consumption
by food-insecure individuals should indicate that such individuals are prone to consume more
3
even in the absence of cues of energetic need, not just that they were hungrier when they
came to the session.
To measure FI, we administered the standard USDA FI questionnaire (Bickel, Nord, Price,
Hamilton, Cook, et al., 2000), as well as developing novel questionnaire measures of our
own. The motivations for the novel measures were the following. The USDA questionnaire
measures only current FI (i.e. FI in adulthood when used with adults). However, there are
suggestions in the literature that FI or related deprivation experienced during childhood have
a developmental programming effect on adult food-related behaviour (Hill et al., 2016; Olson,
Bove, & Miller, 2007). That is, having experienced childhood FI may be related to adult food
motivation, even after controlling for current FI. Our novel questionnaire separately enquires
about experiences of FI in the past year; and when the respondent was a child. It thus yields
two separate two scores henceforth described as AFI (adult food insecurity) and CFI
(childhood food insecurity). It was not possible to simply reuse the USDA questions but
change their timeframe to childhood in order to capture childhood experience of FI. The
USDA questions specifically probe FI due to lack of financial resources. People may not
know or remember the reasons for their experiences in childhood, though; only that meals
were irregular or that they were sometimes hungry. More generally, if there are evolved
psychological mechanisms that respond to cues or experiences of insecurity, there is no
reason to expect they would only be sensitive to insecurity whose cause is financial. The
psychological response to food unavailability appears to be similar whatever its cause
(Polivy, 1996). Thus, our new measures designed to assess any experience of FI, whatever
its source, in the current time or in childhood respectively.
Our main outcome measures were: participant BMI; consumption of the foods in the taste
test; and rated liking of the foods in the taste test. Our predictor variables were current FI
(either USDA score or AFI, both sets of analyses are presented); childhood FI (CFI), and
gender. Interactions between gender and the FI measures were included in all analyses. Our
predictions, based on the FI-body weight literature, are that FI will be associated with higher
BMI, higher consumption and greater liking of the foods for women, but not for men. Hence
we predict an interaction between gender and current FI, which, taking women as the
reference category, will have a negative sign. We consider the analyses testing these
predictions to be confirmatory: the BMI prediction derives from the known associations of FI
with body weight in women and men, and the consumption and liking predictions follow
directly from the hypothesis that greater consumption and greater hedonic value of food are
mechanisms behind the gender-specific association of FI and body weight.
Based on the arguments for developmental programming effects in the literature, we
expected that childhood FI might explain variation in BMI, consumption, and liking of the
foods, above and beyond that explained by adult FI alone. We considered both main effects
of childhood FI, interactions with gender, and interactions between current FI and childhood
FI. The mode of operation of developmental programming in some cases appears to be
sensitization of individuals to relevant features of their adult situation (see for example
Griskevicius, Tybur, Delton, & Robertson, 2011). This can lead to non-additive interactions
between childhood experience and adult context in predicting behaviour. We consider the
analyses involving childhood FI to be exploratory, since we have no clear basis for predicting
whether there will be additive effects or interactions, or what form interactions might take.
4
Materials & Methods
Ethics
The study was approved by the Faculty of Medical Sciences ethics committee of Newcastle
University, approval no. 1400/15594/2017. All participants provided written informed consent.
Participants
We recruited an opportunity sample of 126 participants (42 male, 84 female, see table 1 for
sample characteristics) through research participation registers held at Newcastle University.
Ninety-four participants were undergraduate students, 8 were post-graduate students, and
the remaining 24 worked at the university or were members of the local community.
Participants were requested not to take part if they had any intolerances or allergies to egg,
gluten, milk or soya.
Procedure
Participants were instructed not to eat anything for 90 minutes before the session.
Participants were tested individually. They were met by a researcher and led into a
laboratory room, where they would be screened from the experimenter's view with curtains.
After instruction and filling in a consent form, the participants were invited to drink the content
of two plastic cups containing 75 ml of Coca-Cola (42 kcal / 100 ml) and 75 ml of Pepsi (44
kcal / 100 ml). All participants did so. Participants then completed a 10-min filler task
(watching video commercials of the two brands and answering questions about the drinks
and the advertisements) that should allow time for changes in blood glucose level to occur
(Wang & Dvorak, 2010). We then measured current subjective hunger (participants bisected
a horizontal line with a mark, position subsequently measured in mm), hours since last meal,
and the content of the last meal, as well as a number of demographic questions. Once these
measures were complete, height and weight were measured using a stadiometer (measured
in inches, precision 1/8 inch, and converted to cm) and digital scales (precision 0.1 kg). BMI
was calculated from these values.
Participants were then presented with pre-weighed standard plates of three food products:
salted popcorn (10 g on plate; 527 kcal / 100 g), ready salted crisps (potato chips; 19 g on
plate; 533 kcal / 100 g), and squares of milk chocolate (68 g on plate; 534 kcal / 100 g). The
three plates were all visible simultaneously. The food amounts were chosen to make the
three plates appear similarly full. Participants were instructed to taste each food, complete a
number of qualitative questions, and rate their overall liking for the food (5-point Likert scale).
On completion of this task, they were notified that they could go on eating as much as they
wished, since the leftovers would go to waste. The researcher left them alone to complete
the three FI questionnaires (see below), whilst eating more of the foods if they wished. On
completion, they summoned the researcher, were thanked and debriefed with the true aim of
the study, and escorted out. The remaining amounts of each food were weighed using digital
scales (precision 1 g), and the amounts consumed converted to calories using nutritional
information provided on the packaging.
5
Materials
Participants completed the individual-related items (questions 2, 3, 4, 4a, 8, 8a, 9, 10, 11, 12
and 12a ) from the standard USDA questionnaire (Bickel, Nord, Price, Hamilton, & Cook,
2000). Cronbach’s α was 0.86 for the USDA score. Sixty-seven participants (53%) had at
least one positive response, and hence some symptom of food insecurity. Thirty participants
(24%) had more than two positive responses, and would thus be considered categorically
food-insecure by USDA guidelines (Bickel, Nord, Price, Hamilton, Cook, et al., 2000).
Our two novel FI measures, the AFI and CFI, each consisted of the same 20 ‘yes/no’
questions (see Appendix). The items were generated through discussion between the
authors of the kinds of experience that would be indicative of FI, regardless of the cause. The
questions balanced those where a ‘yes’ indicated greater FI and those where a ‘no’ indicated
greater FI. The score in each case was the number of ‘high FI’ responses. The difference
between the AFI and CFI was that in the former case, participants were asked to think about
the past year, whereas for the latter, the participant was asked to think about the period of
their lives up until they were 12 years old. Cronbach’s α was 0.83 for AFI and 0.78 for CFI. At
the end of each of the two new scales, the participants were asked on a 5-point Likert scale
how easy it had been for them to recollect the relevant memories.
Data analysis
One participant failed to complete the liking ratings, and hence the sample size is one
smaller for analyses involving liking (n = 125) than other analyses (n = 126). The calorie
consumptions of each of the three foods were highly positively correlated with one another
(all rs > 0.75). Hence, we summed together the three food types and analyse total calories
consumed as the consumption outcome variable. Very similar results are obtained using
calories of chocolate consumed as the outcome variable (chocolate consumption accounted
for more than 95% of the variation in total calorie consumption). The liking ratings for the
three foods were not correlated (all rs < 0.10), and hence are considered separately. Calorie
consumption, CFI, USDA score, and BMI all had skewed distributions and were log
transformed for analysis. AFI score was used untransformed. For plotting, we have retained
variables in their raw state, even where logarithmic transformations are used in the
corresponding statistical models.
In preliminary analyses, we first explored the associations between calories consumed, liking
of the foods, BMI and current hunger; and the associations of AFI and CFI with one another,
with the USDA score, and with current hunger. For the main analyses, we present parallel
analyses using the USDA score as the measure of current FI, and using our AFI score.
For each predictor variable, we first fitted general linear models with current FI (either USDA
or AFI), gender, and the current FI by gender interaction as the predictors. These basic
models, summarised in table 2, are the simplest direct test of our predictions regarding
current FI. To investigate whether CFI had any explanatory value above and beyond current
FI, for outcomes where we had found significant effects involving current FI, we then
compared the basic model to models in which CFI had also been added, in the five possible
ways (namely: only main effect of CFI, main effect plus interaction with gender; main effect
plus interaction with AFI, main effect plus both 2-way interactions; main effect plus both 2-
6
way and 3-way interactions). We retained as our final model that with the lowest AICc value.
These models are summarised in table 3.
For the general linear models, FI predictors were standardized and centred to facilitate
interpretation of interactions. All analyses were run in R 3.3.3. software (R Core Team,
2016). R scripts and raw data from the study are freely available via the Zenodo repository
at: https://doi.org/10.5281/zenodo.1197845.
Results
Preliminary analyses: Consumption and liking
Caloric consumption of the foods during the taste test was highly variable across participants
(see table 1 for descriptive statistics). Calorie consumption was not significantly related to
self-reported current hunger (r124 = 0.11, p = 0.23), and was weakly positively correlated with
BMI (r124 = 0.22, p = 0.01). Calorie consumption was not significantly correlated with liking of
the foods (all rs < 0.15), and liking of the foods was not significantly correlated with current
hunger (all rs < 0.11). Finally, BMI was not significantly correlated with liking of any of the
three foods (all rs < 0.08).
Preliminary analyses: Food insecurity
AFI and CFI were weakly positively correlated with one another (r124 = 0.24, p < 0.01). AFI
scores were significantly higher than CFI scores (paired t-test : t125 = 12.90, p < 0.01; since
the two scales involve exactly the same questions with two different reference periods, this
comparison is meaningful). CFI produced much smaller variability than AFI (table 1).
Unsurprisingly, participants reported finding it more difficult to recollect the relevant
memories for CFI than AFI (CFI: mean 3.81, s.d. 0.71; AFI: mean 4.65, s.d. 0.54; paired t-
test: t124 = -11.78, p < 0.01). The USDA score was significantly positively correlated with both
AFI (r124 = 0.58, p < 0.01) and CFI (r124 = 0.25, p < 0.01). The stronger correlation with AFI is
reassuring, since USDA and AFI are both intended to measure current FI. Neither AFI, CFI,
nor USDA score were significantly correlated with current hunger (AFI: r124 = -0.13, p = 0.16;
CFI: r124 = 0.09, p = 0.32; USDA: r124 = -0.09, p = 0.33).
7
Table 1. Descriptive statistics for the main study variables
Variable
Median
Mean
Standard
deviation
Range
Age
20
24.50
10.98
18 - 72
USDA Score
1
1.68
2.40
0 - 10
AFI
8
8.62
4.29
1 - 19
CFI
2
3.20
3.01
0 – 19
Current hunger
55.10
50.92
22.36
5.21 – 100
BMI
23.13
23.45
3.90
15.43 – 40.37
Calories consumed
93.97
154.25
151.77
5.34 – 562.50
Liking chocolate
4
4.14
0.90
1 - 5
Liking crisps
4
3.71
0.81
1 - 5
Liking popcorn
3
3.10
1.06
1 - 5
Basic models with current FI and gender as predictors
Table 2 summarizes the basic models in which the outcome variables were predicted by
current FI, gender, and their interaction. Using the USDA score as the current FI measure,
there was a significant main effect of current FI on BMI, as well as a significant interaction
between current FI and gender. The FI associations took the predicted form: a positive
association in women, absent in men (figure 1; simple slopes: 0.05 for women, -0.01 for
men). Using the AFI measure instead of USDA, the associations were in the same direction
(simple slopes: 0.03 for women, -0.01 for men), but not significant.
For calorie consumption, using USDA, only the main effect of gender was significant (men
consumed more than women; means (s.d.): 230.16 kcals (164.60) versus 116.29 (130.11)).
Using AFI, there was, as well as the significant main effect of gender, a significant interaction
between current FI and gender. As predicted, this was because of a more positive slope in
women than men. However, the pattern was as much due to a negative relationship between
current FI and consumption in men, as a positive relationship in women (figure 2; simple
slopes: 0.18 for women and -0.21 for men).
For liking for chocolate, results were almost identical using USDA and using AFI: a significant
main effect of current FI, with more food-insecure individuals liking chocolate more (B = 0.02;
figure 3), but the main effect of gender and the gender by current FI interaction non-
significant (though for both measures, the simple slopes were steeper for women than men:
USDA: 0.22 for women, -0.02 for men; AFI: 0.23 for women, -0.03 for men). For liking of the
other two foods, there were no significant predictors in the models.
8
Table 2. Summary of statistical models for each study outcome with current FI, gender, and
their interaction as the predictors. For gender, female is the reference category, and so
parameter estimates for gender represent male deviations from the female outcome.
USDA as the current FI measure
AFI as the current FI measure
Outcome
variable
Predictor
B (s.e.)
t
p-value
B (s.e.)
t
p-value
BMI
Current FI
0.05 (0.02)
3.10
0.002*
0.03 (0.02)
1.89
0.06
Gender
0.04 (0.03)
1.22
0.22
0.04 (0.03)
1.20
0.23
Current FI
*Gender
-0.06 (0.03)
-2.15
0.03*
-0.04
(0.03)
-1.33
0.19
Calorie
consumption
Current FI
0.05 (0.11)
0.43
0.70
0.18 (0.10)
1.72
0.09
Gender
0.80 (0.18)
4.41
<0.001*
0.79 (0.18)
4.36
<0.001*
Current FI
*Gender
-0.07 (0.18)
-0.44
0.66
-0.38
(0.18)
-2.09
0.03*
Liking for
chocolate
Current FI
0.23 (0.10)
2.29
0.02*
0.23 (0.10)
2.32
0.02*
Gender
-0.10 (0.17)
-0.61
0.54
-0.10
(0.17)
-0.56
0.57
Current FI
*Gender
-0.25 (0.17)
-1.47
0.15
-0.26
(0.17)
-1.47
0.14
Liking for
popcorn
Current FI
-0.01 (0.12)
-0.10
0.92
0.00 (0.12)
0.01
0.99
Gender
-0.11 (0.20)
-0.56
0.58
-0.14
(0.20)
-0.67
0.50
Current FI
*Gender
-0.08 (0.20)
-0.40
0.69
-0.19
(0.21)
-0.90
0.37
Liking for
crisps
Current FI
0.03 (0.09)
0.29
0.78
0.00 (0.09)
0.00
1.00
Gender
-0.15 (0.15)
-0.96
0.34
-0.14
(0.16)
-0.90
0.37
Current FI
*Gender
-0.20 (0.15)
-1.29
0.20
-0.00
(0.15)
-0.02
0.99
* p < 0.05
9
Figure 1. Scatterplot of BMI against current food insecurity (USDA score) for female and
male participants. Solid lines represent linear fits through the raw data, with shaded areas
representing the 95% confidence intervals.
Figure 2. Scatterplots of total calories consumed in the taste test against current food
insecurity (AFI measure) for female and male participants. Solid lines represent linear fits
through the raw data, with shaded areas representing the 95% confidence intervals.
10
Figure 3. Scatterplot of rated liking for chocolate by current food insecurity, for female and
male participants. The AFI measure is shown, but very similar results are obtained using the
USDA measure. The AFI by gender interaction is not statistically significant. Points have
been jittered slightly in the vertical dimension to avoid over-plotting given the discrete rating
values. Solid lines represent linear fits through the raw data, with shaded areas representing
the 95% confidence intervals.
11
Adding in childhood FI
For the three outcomes where there were significant effects involving current FI for at least
one of USDA and AFI, we considered all possible models additionally including CFI and its
interactions, retaining the one with the lowest AICc value. Table 3 summarises these models.
For BMI, the best-fitting USDA model (AICc 1.62 units better than the basic model) included
CFI and the interaction between CFI and USDA. The main effect of USDA and its interaction
remained significant in this model. Additionally, the interaction between CFI and USDA was
significant, with a positive sign. This means that exposure to more childhood FI potentiates
the effect of current FI on BMI (simple slopes of BMI on USDA: 0.02 for a woman 1 s.d.
below the mean of CFI; 0.09 for a woman 1 s.d. above the mean of CFI). Using AFI instead
of USDA, no model involving CFI improved model fit compared to the basic model.
For calories consumed, models involving CFI improved model fit both when using USDA and
when using AFI (best model 6.10 AICc units better than basic model using USDA, 1.34 units
using AFI). There were some differences between the best-fitting models in the two cases.
Using USDA, the main effect of CFI was significant, and there was also a significant three-
way interaction between gender, current FI, and CFI. Using AFI, the main effect of CFI was
not significant, and the three-way interaction was not included in the model. For both models,
though, there was a significant two-way interaction between current FI and CFI, with a
negative sign. This means that, opposite to the BMI case, experiencing more childhood FI
attenuates the relationship between current FI and calories consumed (simple slopes using
USDA: 0.23 for a woman 1 s.d. below mean CFI, -0.27 for a woman 1 s.d. above mean CFI;
using AFI: 0.30 for a woman 1 s.d. below mean CFI, -0.05 for a woman 1 s.d. above mean
CFI).
For liking for chocolate, models including CFI improved model fit both for USDA and AFI (by
2.10 AICc units for USDA and 6.06 AICc units using AFI). In the USDA model, only a main
effect of CFI was included. This effect was significantly negative, indicating that greater
childhood FI was associated with less liking for chocolate (the opposite direction to the
significant positive association for current FI). The AFI model concurred in including the
significant negative association between CFI and liking, and the significant positive
association between current FI and liking. It additionally included a significant interaction
between CFI and AFI, with a negative sign (i.e. greater CFI attenuates the positive
association between AFI and liking; simple slopes: 0.46 for a woman 1 s.d. below mean CFI,
0.10 for a woman 1 s.d. above mean CFI).
12
Table 3. Summary of best-fitting statistical models for each study outcome once childhood FI
and all possible interactions are included as predictors in addition to current FI and gender.
For gender, female is the reference category, and so parameter estimates for gender
represent male deviations from the female outcome.
USDA as the current FI measure
AFI as the current FI measure
Outcome
variable
Predictor
B (s.e.)
t
p-value
B (s.e.)
t
p-value
BMI
Current FI
0.06 (0.02)
3.29
0.001*
0.03 (0.02)
1.89
0.06
Gender
0.04 (0.03)
1.29
0.21
0.04 (0.03)
1.20
0.23
Childhood FI
-0.01 (0.01)
-0.66
0.51
-
-
-
Current FI
*Gender
-0.08 (0.03)
-2.83
0.006*
-0.04
(0.03)
-1.33
0.19
Current FI *
Childhood FI
0.03 (0.01)
2.41
0.02*
-
-
-
Calorie
consumption
Current FI
-0.02 (0.11)
-0.16
0.88
0.13 (0.11)
1.18
0.24
Gender
0.61 (0.19)
3.21
0.002*
0.66 (0.19)
3.52
0.001*
Childhood FI
0.22 (0.11)
2.05
0.04*
0.14 (0.09)
1.50
0.14
Current FI
*Gender
-0.23 (0.20)
-1.15
0.25
-0.29
(0.18)
-1.59
0.14
Childhood FI
*Gender
-0.24 (0.19)
-1.26
0.21
-
-
-
Current FI *
Childhood FI
-0.25 (0.11)
-0.24
0.02*
-0.17
(0.09)
-2.01
0.047*
Current FI *
Childhood FI
* Gender
0.57 (0.17)
3.42
0.001*
-
-
-
Liking for
chocolate
Current FI
0.27 (0.10)
2.70
0.01*
0.28 (0.10)
2.83
0.006*
Gender
0.00 (0.18)
0.02
0.99
-0.03
(0.18)
-0.15
0.88
Childhood FI
-0.18 (0.09)
-2.05
0.04*
-0.17
(0.09)
-2.02
0.045*
Current FI *
Childhood FI
-
-
-
-0.18
(0.08)
-2.24
0.03*
* p < 0.05
13
Discussion
It is well documented that FI is associated with high body weight in Western women, but not
men (Nettle et al., 2017; Townsend et al., 2001). However, the mechanisms by which such
an association might come about have been little studied. One possibility is that experience
of FI changes people’s motivation to opportunistically consume energy-dense food when this
is freely available. We investigated this possibility in a mock taste test where volunteers were
given a pretext for eating as much or as little as they liked of three snack foods, and had their
current and childhood FI measured.
We first considered only current FI, in interaction with gender. For BMI, we found the pattern
typical of previous studies in affluent populations: a positive association between current FI
and BMI moderated by gender (a steeper slope for women than men), although the pattern
was only statistically significant when using the USDA measure of current FI, and not our
novel AFI measure. For calorie consumption, we found only weak evidence in support of our
predictions in these first analyses. There was a significant interaction between current FI and
gender in predicting calorie consumption, and the slope was more positive for women than
men. However, this pattern was only statistically significant using our novel AFI measure, not
the USDA score. Moreover, even using the AFI measure, the significant interaction was
driven as much by a negative relationship between FI and consumption in men (where we
predicted a null association), as it was by a positive association between FI and consumption
in women. For hedonic value of food, we found a simple association between current FI (by
either measure) and liking for chocolate: more food-insecure individuals liked chocolate
more, an effect not moderated by gender. There were however no associations between
current FI and liking for either of our other two snack foods.
In a second set of analyses, we additionally considered the possible role of childhood FI.
These were exploratory analyses, since we had no clear predictions regarding whether
childhood FI would have a separate additive effect, or interact with current FI and/or gender.
For all three outcomes (BMI, calorie consumption, liking for chocolate), adding in childhood
FI improved model fit for one or both of the measures of current FI. In four of the five cases
where it did so, there was a significant interaction between current FI and childhood FI. This
suggests that childhood experience of FI might serve quite generally as a moderator of the
impact of current FI on eating and weight. However, the mode of moderation was different for
the different outcomes. For BMI, higher childhood FI made the association between current
FI and BMI stronger, suggesting that childhood FI experiences might sensitize individuals to
similar experiences later in life. For calorie consumption and liking for chocolate, the
moderation was the other way around: higher childhood FI made the association between
current FI and consumption/liking weaker. One way of interpreting this finding is that
individuals who have experienced childhood FI behave as if food insecure in adulthood,
regardless of whether they actually are or not. This interpretation would be compatible with
previous descriptions of the impact of childhood poverty, deprivation or trauma on adult body
weight and eating (Greenfield & Marks, 2009; Hill et al., 2016; Olson et al., 2007). Our study
goes beyond the prior research in specifically measuring and examining childhood FI, as
opposed to more general socioeconomic or psychosocial adversities in childhood. However,
in view of the exploratory nature of the analyses and the contradictory directions of the
findings, we view these result of the current study regarding childhood FI as indications of the
need for future investigation. Why experiencing FI in childhood might enhance the effect of FI
14
experience in adulthood for one outcome, body weight, whilst attenuating it for the linked
outcomes of opportunistic calorie consumption and hedonic value of chocolate, is far from
clear.
Tentative though the conclusions about childhood FI may be, the fact that we observe
moderation by childhood FI does cast a somewhat different light on our main findings
concerning current FI and calorie consumption. The evidence in this study for current FI
predicting calorie consumption is weak when current FI is considered without regard to
childhood FI, being statistically significant only for one of the two current FI measures.
However, in interaction with childhood FI, current FI significantly predicts consumption for
both current FI measures. This suggests that effects of current FI can be masked because of
moderation by childhood experience. This would mean that the moderate associations
between current FI and women’s body weight observed in the literature (Nettle et al., 2017)
may be the amalgam of stronger associations in sub-groups with certain childhood
experiences, and weaker or null associations in others. This is a potentially useful
observation for the literature, and we suggest that measures of childhood experience could
usefully be incorporated into study designs, even where the main study questions concern
current FI.
The conceptual framework of our study is based on the idea that high BMI might be driven by
higher consumption of energy-dense foods, which might in turn be driven by greater liking of
those foods; and indeed, the associations with FI (for women) are all the same direction,
higher FI predicting higher BMI, greater consumption, and greater liking for chocolate.
However, the correlations in the dataset between liking of the foods, calorie consumption and
BMI are very close to zero. Liking for a food is certainly not the only determinant of
consumption (Brunstrom & Rogers, 2009; Drewnowski, 1997; Finlayson et al., 2007), so the
low correlation between liking and consumption is not in itself surprising. The mono-factorial
hypothesis that variation in liking for foods (or energy-dense foods in particular) explains
most of the individual variation in obesity is not well supported by evidence (Drewnowski,
1997). However, there is a case for a more subtle role of individual differences in hedonic
response to particular foods in the pathway to weight gain in at least some individuals
(Finlayson et al., 2007; Salbe, Delparigi, Pratley, Drewnowski, & Tataranni, 2004). For
example, variation in liking may inhibit flavour-specific satiety for particular types of food, or
drive consumption in the absence of homeostatic need (see Nasser, 2001 for review). Thus,
it would be possible for liking of energy-dense foods to be on the causal pathway liking FI to
weight gain, without needing to claim that liking for energy-dense foods is the only or main
determinant of body weight. However, our data to not directly demonstrate such a role: since
liking for chocolate was not significantly related to calories consumed or body weight, it
cannot formally be considered a mediator in this dataset.
An innovation of our study was the development of new questionnaires for adult FI (AFI) and
FI in childhood (CFI). The adult scale broadened the focus of the USDA scale on financial
constraints to obtaining food, and instead encompass the experience of irregularity in the
food supply, regardless of the reasons for it. If there are psychological mechanisms that
respond to irregular food intake, there is no reason to suspect that they are sensitive to the
different reasons such irregularity occurs. Restriction imposed by non-financial constraints
may have very similar consequences to financially-imposed restriction (Nettle et al., 2017;
Polivy, 1996). Our new AFI measure was moderately correlated with the USDA scale.
15
Moreover, it produced a much better distribution than the USDA scale, with a smaller mode
at or near zero. This is probably related to the broader range of experiences that it asks
about. However, we cannot make any strong claim that it had greater predictive utility than
the USDA score. For consumption, it was the AFI measure that produced the significant
association, whereas for BMI it was the USDA score. The parameter estimates for the two
measures (which were directly comparable since variables were standardized for analysis)
were very similar in both cases. The CFI questionnaire produced lower scores and a smaller
range of variation than AFI. Participants reported that recall of the relevant experiences fairly
easy, though less easy than for AFI. This may be because the present sample were mostly
fairly young, and hence were recalling events of only a decade ago.
One limitation of the study is that almost all of the calorie consumption in the taste test was of
the chocolate. This makes it impossible to establish whether any associations with FI are
specific to chocolate: it is variation in consumption of chocolate that drives the statistical
associations between FI and total consumption, but this may be because there is so much
less variation in the consumption of the other foods. Interestingly, the only food for which
liking was associated with FI was also chocolate. Chocolate was the only sweet food, and
the hedonic value of and appetite for sweet foods have been implicated in weight gain
following food restriction (Cabanac, Duclaux, & Spector, 1971; Fantino, Baigts, Cabanac, &
Apfelbaum, 1983; Paradis & Cabanac, 2008). However, since there was no replication of
different kinds of sweet food, and chocolate differs from the other foods in other ways than
just being sweet, our findings do not warrant any strong claims about appetite for sweet
foods being specifically related to FI.
Ours is a correlational study, and from correlational data alone, it is not possible to make
strong inferences about causality. Adult and childhood FI might causally impact eating and
weight, but there are other possible pathways by which the associations we observe could
come about. For example, a liking of or habitual consumption of energy-dense snack foods
may cause differences in regularity of food intake, manifest as a higher FI score.
Alternatively, both FI and liking or consumption may be influenced by some unmeasured
third variable. The only way to advance causal understanding in this area will be to find ways
of experimentally manipulating FI. Recent research has shown that experimental
manipulations of subjective social status can have an immediate causal impact on caloric
consumption (Cardel et al., 2015; Cheon & Hong, 2016). It is possible that these
manipulations are effectively altering implicit or explicit feelings of FI. One of the definitive
characteristics of low social status is more precarious access to resources. Thus, avenues
for future research include determining whether these experimental manipulations alter
perceived FI; whether changes in perceived FI mediate their effects on food consumption;
and whether it is possible to manipulate FI directly by some means other than manipulating
subjective social status. However, the experimental approach is not an available avenue for
investigating the causal impact of FI in childhood. There are animal models where FI
experiences during development can be experimentally manipulated (Andrews et al., 2015;
Bloxham, Bateson, Bedford, Brilot, & Nettle, 2014; Remmers & Delemarre-van de Waal,
2011). For childhood experience of FI in humans, which our findings suggest could be
important, we are effectively restricted to correlational epidemiological methods.
In conclusion, by presenting volunteers who also completed FI questionnaires with a
standardized opportunity to consume energy-dense snack food, our study shed some light
16
on the pathways by which FI may lead to high body weight. Higher-FI women consumed
more calories, and higher-FI participants of both sexes liked chocolate more. We also found
exploratory evidence suggesting that childhood FI, which has not been specifically measured
before, may moderate the individual’s response to experiences of FI in adulthood.
Acknowledgements
This project has received funding from the European Research Council (ERC) under the
European Union’s Horizon 2020 research and innovation programme (grant agreement No
AdG 666669, COMSTAR).
17
Appendix: Food insecurity questionnaires
Instructions:
Childhood food insecurity: Please think back to a period before you were 12 years old when
answering the following questions.
Adult food insecurity: Please think about the past year when answering the following
questions.
All questions have a yes/no response format.
1. Did you have a consistent dinnertime?
2. Did you usually feel full after meals?
3. Did you worry about food supplies running out?
4. Did you ever not eat for a whole day?
5. Were you ever unsure where your next meal would come from?
6. Did you ever find that good quality food was not available?
7. Were your cupboards always full?
8. Were your meals mainly made up from cheap foods?
9. Did you ever miss an evening meal?
10. Did you regularly eat food that you liked?
11. Were you sometimes unsure about the time of your next meal?
12. Did you usually have dessert after your evening meal?
13. Did you overall have a healthy diet?
14. Did you ever miss lunch?
15. Did you ever go to sleep hungry?
16. Could you afford to eat balanced meals?
17. Did you struggle to have three meals on some days?
18. Were your meals a similar size from day to day?
19. Were you ever hungry but didn’t eat because there was not money to buy food?
20. Did you ever miss breakfast?
18
References
Adams, E. J., Grummer-strawn, L., & Chavez, G. (2003). Food Insecurity Is Associated with
Increased Risk of Obesity in California. Journal of Nutrition, 1070–1074.
Anderson, S. A. (1990). Core indicators of nutritional state for difficult-to-sample
populations. The Journal Of Nutrition, 120 Suppl, 1559–1600.
Andrews, C., Viviani, J., Egan, E., Bedford, T., Brilot, B., Nettle, D., & Bateson, M. (2015).
Early-life adversity increases foraging and information gathering in European starlings,
Sturnus vulgaris. Animal Behaviour, 109, 123–32.
http://doi.org/10.1016/j.anbehav.2015.08.009
Bickel, G., Nord, M., Price, C., Hamilton, W., & Cook, J. (2000). Measuring food security in the
United States: Guide to measuring household food security. Washington (DC): USDA.
Bickel, G., Nord, M., Price, C., Hamilton, W., Cook, J., & U.S. Department of Agriculture Food
and Nutrition Service. (2000). h. Washington (DC): USDA.
Bloxham, L., Bateson, M., Bedford, T., Brilot, B., & Nettle, D. (2014). The memory of hunger:
developmental plasticity of dietary selectivity in the European starling, Sturnus vulgaris.
Animal Behaviour, 91, 33–40. http://doi.org/10.1016/j.anbehav.2014.02.025
Brunstrom, J. M., & Rogers, P. J. (2009). How many calories are on our plate expected
fullness, not liking, determines meal-size selection. Obesity, 17(10), 1884–1890.
http://doi.org/10.1038/oby.2009.201
Cabanac, M., Duclaux, R., & Spector, N. H. (1971). Sensory feedback in regulation of body
weight: Is there a ponderostat? Nature, 229(5280), 125–127.
http://doi.org/10.1038/229125a0
Cardel, M. I., Johnson, S. L., Beck, J., Dhurandhar, E., Keita, A. D., Tomczik, A. C., … Allison, D.
B. (2015). The effects of experimentally manipulated social status on acute eating
behavior: A randomized, crossover pilot study. Physiology and Behavior, 162, 93–101.
http://doi.org/10.1016/j.physbeh.2016.04.024
Cheon, B. K., & Hong, Y.-Y. (2016). Mere experience of low subjective socioeconomic status
stimulates appetite and food intake. Proceedings of the National Academy of Sciences,
201607330. http://doi.org/10.1073/pnas.1607330114
Crawford, P. B., & Webb, K. L. (2011). Unraveling the paradox of concurrent food insecurity
and obesity. American Journal of Preventive Medicine, 40(2), 274–275.
http://doi.org/10.1016/j.amepre.2010.11.003
Dinour, L. M., Bergen, D., & Yeh, M. C. (2007). The Food Insecurity-Obesity Paradox: A
Review of the Literature and the Role Food Stamps May Play. Journal of the American
Dietetic Association, 107(11), 1952–1961. http://doi.org/10.1016/j.jada.2007.08.006
Drewnowski, A. (1997). Taste Preferences and Food Intake. Annual Review of Nutrition, 17,
237–253. http://doi.org/10.1146/annurev.nutr.17.1.237
Ekman, J. B., & Hake, M. K. (1990). Monitoring starvation risk: Adjustments of body reserves
in Greenfinches (Carduelis Chloris L.) during periods of unpredictable foraging success.
19
Behavioral Ecology, 1(1), 62–67. http://doi.org/10.1093/beheco/1.1.62
Fantino, M., Baigts, F., Cabanac, M., & Apfelbaum, M. (1983). Effects of an overfeeding
regimen—the affective component of the sweet sensation. Appetite, 4(3), 155–164.
http://doi.org/10.1016/S0195-6663(83)80028-2
Finlayson, G., King, N., & Blundell, J. E. (2007). Liking vs. wanting food: Importance for
human appetite control and weight regulation. Neuroscience and Biobehavioral
Reviews, 31(7), 987–1002. http://doi.org/10.1016/j.neubiorev.2007.03.004
Greenfield, E. A., & Marks, N. F. (2009). Violence from parents in childhood and obesity in
adulthood: using food in response to stress as a mediator of risk. Social Science &
Medicine (1982), 68(5), 791–8. http://doi.org/10.1016/j.socscimed.2008.12.004
Griskevicius, V., Tybur, J. M., Delton, A. W., & Robertson, T. E. (2011). The influence of
mortality and socioeconomic status on risk and delayed rewards: a life history theory
approach. Journal of Personality and Social Psychology, 100(6), 1015–1026.
http://doi.org/10.1037/a0022403
Gundersen, C., Kreider, B., & Pepper, J. (2011). The economics of food insecurity in the
United States. Applied Economic Perspectives and Policy, 33(3), 281–303.
http://doi.org/10.1093/aepp/ppr022
Hanson, K. L., Sobal, J., & Frongillo, E. a. (2007). Gender and marital status clarify
associations between food insecurity and body weight. The Journal of Nutrition, 137(6),
1460–1465.
Higginson, A. D., McNamara, J. M., & Houston, A. I. (2016). Fatness and fitness: Exposing the
logic of evolutionary explanations for obesity. Proceedings of the Royal Society B:
Biological Sciences, 283, 20152443.
Hill, S. E., Prokosch, M. L., DelPriore, D. J., Griskevicius, V., & Kramer, A. (2016). Low
Childhood Socioeconomic Status Promotes Eating in the Absence of Energy Need.
Psychological Science. http://doi.org/10.1017/CBO9781107415324.004
Kendall, A., Olson, C. M., & Frongillo, E. A. (1996). Relationship of hunger and food insecurity
to food availability and consumption. Journal of the American Dietetic Association.
http://doi.org/10.1016/S0002-8223(96)00271-4
Li, X., Cope, M. B., Johnson, M. S., Smith, D. L., & Nagy, T. R. (2010). Mild calorie restriction
induces fat accumulation in female C57BL/6J mice. Obesity, 18, 456–462.
http://doi.org/10.1038/oby.2009.312
Lima, S. L. (1986). Predation risk and unpredictable feeding conditions : determinants of
body mass in birds. Ecology, 67(2), 377–385.
Martin-Fernandez, J., Caillavet, F., Lhuissier, A., & Chauvin, P. (2014). Food insecurity, a
determinant of obesity? - An analysis from a population-based survey in the Paris
metropolitan area, 2010. Obesity Facts, 7(2), 120–129.
http://doi.org/10.1159/000362343
Nasser, J. (2001). Taste, food intake and obesity. Obesity Reviews, 2(4), 213–218.
http://doi.org/10.1046/j.1467-789X.2001.00039.x
Nettle, D., Andrews, C., & Bateson, M. (2017). Food insecurity as a driver of obesity in
20
humans: The insurance hypothesis. Behavioral and Brain Sciences, 40, e105.
http://doi.org/10.1017/S0140525X16000947
Olson, C. M., Bove, C. F., & Miller, E. O. (2007). Growing up poor: Long-term implications for
eating patterns and body weight. Appetite, 49(1), 198–207.
http://doi.org/10.1016/j.appet.2007.01.012
Paradis, S., & Cabanac, M. (2008). Dieting and food choice in grocery shopping. Physiology
and Behavior, 93(4–5), 1030–1032. http://doi.org/10.1016/j.physbeh.2008.01.014
Polivy, J. (1996). Psychological consequences of food restriction. Journal of the American
Dietetic Association. http://doi.org/10.1016/S0002-8223(96)00161-7
R Core Team. (2016). R: A language and environment for statistical computing.
Remmers, F., & Delemarre-van de Waal, H. A. (2011). Developmental programming of
energy balance and its hypothalamic regulation. Endocrine Reviews, 32(2), 272–311.
http://doi.org/10.1210/er.2009-0028
Salbe, A. D., Delparigi, A., Pratley, R. E., Drewnowski, A., & Tataranni, P. A. (2004). Taste
preferences and body weight changes in an obesity-prone, (February), 372–378.
Sharkey, J. R., Nalty, C., Johnson, C. M., & Dean, W. R. (2012). Children’s very low food
security is associated with increased dietary intakes in energy, fat, and added sugar
among Mexican-origin children (6-11 y) in Texas border Colonias. BMC Pediatrics, 12(1),
16. http://doi.org/10.1186/1471-2431-12-16
Tanumihardjo, S. a., Anderson, C., Kaufer-Horwitz, M., Bode, L., Emenaker, N. J., Haqq, A. M.,
… Stadler, D. D. (2007). Poverty, Obesity, and Malnutrition: An International Perspective
Recognizing the Paradox. Journal of the American Dietetic Association, 107(11), 1966–
1972. http://doi.org/10.1016/j.jada.2007.08.007
Townsend, M. S., Peerson, J., Love, B., Achterberg, C., & Murphy, S. P. (2001). Food
insecurity is positively related to overweight in women. The Journal of Nutrition, 131(6),
1738–1745.
Wang, X. T., & Dvorak, R. D. (2010). Sweet future: Fluctuating blood glucose levels affect
future discounting. Psychological Science, 21(2), 183–188.
http://doi.org/10.1177/0956797609358096
Wilson, J. F., & Cantor, M. B. (1987). An animal model of excessive eating: Schedule-induced
hyperphagia in food-satiated rats. Journal of the Experimental Analysis of Behavior, 47,
335–46.
Witter, M., & Swaddle, J. P. (1995). Dominance, competition, and energetic reserves in the
European starling, Sturnus vulgaris. Behavioral Ecology, 6, 343–348.
http://doi.org/10.1093/beheco/6.3.343
