Physiology & Behavior 82 (2004) 317 – 324
How habitual caffeine consumption and dose influence flavour preference
conditioning with caffeine
Elizabeth M. Tinleya, Paula J. Durlachb, Martin R. Yeomansa,*
a
Department of Psychology, School of Life Sciences, University of Sussex, Falmer, Brighton BN1 9QG, UK
b
Army Research Institute, Orlando, FL, USA
Received 14 April 2003; accepted 25 March 2004
Abstract
This study investigated the effects of both habitual caffeine use and dose administered in determining the ability of caffeine to reinforce
conditioned changes in flavour preference. Thirty overnight-withdrawn moderate caffeine consumers and 30 non or low-dose caffeine (non/
low) consumers evaluated five novel-flavoured fruit teas. Subsequently, their median-rated tea was used in four ensuing conditioning
sessions. Either placebo, 1 or 2 mg/kg of caffeine (n=10 consumers, 10 non/low consumers in each condition), was added to the target tea,
and all five teas were reevaluated at a final tasting. Pleasantness ratings over the four conditioning sessions indicated that non/low consumers’
liking increased for the noncaffeinated fruit tea with no change for the tea containing either 1 or 2 mg/kg of caffeine. Among consumers,
pleasantness ratings tended to decrease for the noncaffeinated fruit tea but increased significantly at the 1-mg dose and showed a tendency to
increase at the 2-mg dose. Similar effects were shown in the evaluations made before and after conditioning, with no change in the
nonexposed drinks. These results show that 1.0 mg/kg of caffeine reinforces changes in flavour pleasantness in acutely withdrawn habitual
consumers but not in nonconsumers or nondependent low-caffeine consumers, further endorsing the negative-reinforcement theory of
conditioning with caffeine.
D 2004 Elsevier Inc. All rights reserved.
Keywords: Caffeine; Flavour preference; Conditioning; Reinforcement; Liking
1. Introduction
Recent research has demonstrated that caffeine can
reinforce changes in preference and liking for novel flavours; however, such conditioning with caffeine depends
critically on certain factors. The two most important factors
appear to be the habitual consumption level of the population under test [1] and the caffeine-deprivation state of
consumers when tested (e.g., [2 – 5]). Evidence of the
importance of habitual consumption levels is limited to
one published study to date, yet these data are critical to
our understanding of the nature of caffeine reinforcement. It
is therefore important to establish how robust these findings
are, and the present study was designed to evaluate further
the importance of habitual caffeine consumption levels as a
predictor of changes in liking for novel flavours paired with
caffeine as a function of caffeine dose.
* Corresponding author. Tel.: +44-1273-678617; fax: +44-1273678058.
E-mail address: martin@sussex.ac.uk (M.R. Yeomans).
0031-9384/$ – see front matter D 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.physbeh.2004.03.018
Abstention from caffeine by habitual consumers results
in well-characterised withdrawal symptoms which occur
after a relatively short period of deprivation (approximately
12 h; for review see Ref. [6]). Symptoms include both
physical and affective changes, with headaches being the
most frequently reported symptom (e.g., Ref. [7 –10]) and
tiredness/lethargy (e.g., [11]) also commonly reported.
These withdrawal symptoms are not restricted to high and
moderate caffeine consumers, they have also been reported
by those ingesting as little as 100 mg of caffeine per day
(e.g., Ref. [12]). Recent findings that implicate the importance of habitual caffeine use in the expression of conditioning with caffeine suggest that caffeine acts as a negative
reinforcer by relieving withdrawal symptoms encountered
after acute cessation of caffeine use. In the earliest study [1],
low caffeine consumers (mean caffeine consumption<120
mg/day) expressed an overall increased preference for a
target drink irrespective of caffeine content. However,
moderate consumers (mean caffeine>120 mg/day) only
showed a (nonsignificant) increase in preference for a
caffeinated drink, and a decreased preference if the drink
318
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
was noncaffeinated. Inasmuch as low consumers would not
encounter the dysphoric effects of withdrawal, one explanation for these findings is that caffeine had no reinforcing
effects on drink preference in the absence of withdrawal. In
contrast, the acutely withdrawn moderate consumers would
have experienced relief of withdrawal after consuming the
caffeinated drink, and the association of the beneficial
effects of caffeine could explain the slight increase in
preference for the caffeinated drink. In addition, the noncaffeinated version could have been associated with the
(possibly increasing) withdrawal effects leading to the
subsequent decrease in preference for that flavour. Yeomans
et al. [3] provided further evidence of this and highlighted
the negative reinforcing effect of caffeine. Acutely withdrawn moderate caffeine consumers reported an increased
liking for a caffeinated drink, yet liking for the flavour
decreased if it did not contain caffeine. No such changes in
flavour liking were seen in the absence of caffeine deprivation. Subsequent research further confirmed the sensitivity of such conditioning to deprivation state [5]. Moderate
consumers who had been trained following overnight abstinence and thus in a caffeine-deprived state reported an
increase in liking for a novel-flavoured caffeinated drink
throughout training, but when subsequently tested undeprived (and thus no longer experiencing withdrawal symptoms), liking for that flavour fell significantly.
The importance of the relief of withdrawal symptoms in
self-administration of caffeine was suggested by Hughes et
al. [13] who commented that effects such as headaches
‘prospectively predicted subsequent self-administration of
caffeine’. Further evidence of the importance of withdrawal
relief in reinforcement with caffeine was the finding that
participants were willing to forfeit more money to avoid
placebo than they would pay to receive caffeine [14].
Although the case that caffeine reinforces changes in
flavour preference and liking through reversal of the negative consequences of caffeine abstinence in dependent consumers appears well supported, this conclusion relies
heavily on the findings from one study [1] stating that
nondependent and moderate caffeine consumers responded
differently in a flavour – caffeine conditioning paradigm.
However, whereas subsequent caffeine-conditioning studies
have measured flavour pleasantness, the original study used
a measure of relative preference to assess conditioning.
Although changes in liking and preference covary, a change
in flavour pleasantness need not always reflect a change in
preference and vice versa (see Ref. [15]). A preference
reflects a choice between two or more options, whereas
liking is a pure hedonic evaluation. Thus, for example,
although a person may have a higher hedonic rating for
the flavour of a hot drink than for the flavour of a cold drink,
their preference may be to consume the cold drink depending on ambient temperature. Thus, the finding that the lowand moderate caffeine consumers differed in the way that
they altered their preference for flavours paired with the
absence or presence of caffeine need not necessarily imply
that changes in flavour liking showed the same changes. It
could be, for example, that low consumers increased their
liking for the caffeine-paired drink, but inasmuch as they
prefer not to consume caffeine, this liking was not translated
into a flavour preference. Thus, the present study aimed to
replicate and extend the earlier study by Rogers et al. [1], by
including both a measure of liking (the rated pleasantness of
the conditioned flavour) and preference for that flavour.
Another relevant factor to reinforcement with caffeine is
dose. Among habitual consumers, low doses of caffeine in
the range of 25– 100 mg are liked, whereas higher doses of
caffeine are not. This was demonstrated in studies revealing
increased liking for coffee containing caffeine at doses of 25
and 50 mg expressed by both high [16] and moderate
consumers [17], with both groups having reported a monotonic decrease in liking with doses of 100 – 600 mg of
caffeine. These ‘liked’ doses closely resemble the amount
of caffeine contained in commonly consumed caffeinated
beverages (30 – 125 mg of caffeine).
The present study set out to reexamine the relevance of
habitual caffeine consumption level (non/low consumers,
defined as less than 65 mg of caffeine/day, vs. moderate
consumers, defined as those consuming between 75 and 500
mg/day) on the ability of 0, 1, and 2 mg/kg to reinforce both
changes in liking and preference for the caffeine-paired
flavour. The lower dose (1 mg/kg bodyweight) provided
the equivalent amount of caffeine to that consumed when
drinking tea or instant coffee, and the higher dose (2 mg/kg
bodyweight) approximately equated to that found in filter
coffee.
2. Method
2.1. Design
Using a mixed design, changes in pleasantness across
time (within-subject) were contrasted between six treatment
groups based on consumer status (‘consumer’ or ‘non/low
consumer’) and administered caffeine dose (0 [placebo], 1,
or 2 mg/kg).
2.2. Participants
Sixty participants were recruited from Sussex University,
using e-mail, personal communication, and poster adverts.
All applicants had previously completed a questionnaire
devised at Sussex University, combining the Three Factor
Eating Questionnaire [18] with questions on general food
and drink use. Questions regarding caffeine consumption
were embedded within the section on drinks. Thus, potential
participants were not aware that recruitment was specifically
based on their daily caffeine consumption. Caffeine consumption levels were estimated from self-reported daily
drink consumption using the following approximations of
normal caffeine content: cola=30 mg/can, tea=60 mg of
319
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
caffeine/cup, instant coffee=70 mg/cup and filter coffee=125 mg/cup (adapted from Ref. [19]). Respondents
reporting medical contraindications (diabetes, use of medication [excluding the contraceptive pill], allergies to any of
the breakfast ingredients, or diagnoses of an eating disorder)
were excluded from the study. The first 30 respondents
consuming less than 65 mg of caffeine/day were assigned to
the ‘non/low consumer’ groups (mean=27F5 mg/day), and
30 respondents consuming between 75 – 500 mg of caffeine/
day to the ‘consumer’ groups (mean=221F19 mg/day). The
two groups did not differ in BMI, weight, or age (see Table
1). However, further analysis of habitual caffeine consumption by the consumers, divided into the three treatment
conditions, revealed an unexpected difference between
groups [ F(2,27)=2.99, P=.06], with the group who were
assigned to the 2.0-mg/kg-caffeine condition habitually
consuming significantly less than the group assigned to
placebo ( P<.05), with the 1.0 mg/kg group intermediate.
The protocol was approved by the University of Sussex
Ethics Committee, and the study was conducted according
to the ethical standards laid down in the Declaration of
Helsinki 1964.
2.3. Test drinks
The teas used were Lemon and Ginger (Twinings);
Camomile and Spiced Apple (Twinings); Ginger, Orange,
and Honey (Twinings); Sweet Berry Swirl (London Herb
and Spice); and Camomile and Honey (London Herb and
Spice). These were chosen from a pilot study of 10 fruit
teas, and were those rated most novel (to try to ensure
minimal prior exposure to the flavours) and mid-range for
pleasantness (to allow for both an increase and decrease in
ratings). Drinks were made by steeping a tea bag in 120 mg
of boiling water for 3 min.
2.4. Procedure
Participants attended the laboratory at the Psychopharmacology Unit of the University of Sussex on six nonconsecutive days between 8.30 and 9.30 a.m., having refrained
from eating or drinking anything other than water from
11.00 p.m. on the evening prior to each session. At least one
‘nonexperimental’ day between sessions ensured that participants returned to their normal caffeine consumption
patterns and were not fasting on consecutive nights.
At the first visit (pretest session), participants read and
signed an information sheet and a standard volunteer consent form. The information sheet described the purpose of
the experiment as determining the effect of common drink
ingredients on mood. It included sucrose, glucose, fructose,
aspartame, acesulfame, saccharin, caffeine, natural food
flavourings, and natural colourings as possible additives to
the drinks. The purpose of this deception was to prevent
responses biased by participants’ preconceptions of caffeine
effects. To try to ensure compliance with the overnight food
and drink restrictions, each participant provided a saliva
sample, which they were told would be analysed, but,
however, was not. They were then presented with a sample
of all five test teas, each of which was tasted in an order
chosen by the participant. Each tea was evaluated for
‘pleasant’, ‘novel’, ‘sweet’, ‘bitter’, and ‘sour’ qualities
using 100-mm line scales end-anchored with ‘‘Not at all’’
and ‘‘Extremely’’, and each participant recorded their most
preferred tea. A standard breakfast of 60 g of Crunchy Nut
Cornflakes (Kellogg’s brand) and 160 g of semiskimmed
milk (total 310 kcal) was then given. Once these were eaten,
the participants’ height and weight were measured. Rated
pleasantness for the teas was assessed, and the tea which
received the median score on the pleasantness scale was
designated as the target tea for that participant.
Sessions 2 –5 were designed as conditioning trials (subsequently referred to as Conditioning Days 1– 4) during
which participants consumed their target tea containing
either placebo (2 mg/kg of maltodextrin, Cerestar UK),
which has been previously used in this and other laboratories to maintain the double-blind, or 1 or 2 mg/kg of caffeine
(Courtin and Warner), according to the allocated dosage
group. All drug administration was double-blind. As before,
participants reported to the Psychopharmacology Unit between 8.30 and 9.30 a.m., and saliva samples were taken at
each visit. On each conditioning day, a mood questionnaire
indicating current moods was completed, after which the
target tea was consumed and a tea-rating questionnaire was
completed. The mood questionnaire has been successfully
used to assess subjective moods in relation to caffeine in
previous research [20]. Sixteen adjectives (friendly, angry,
cheerful, dejected, confident, uncertain, clearheaded, muddled, calm, placid, jittery, tense, energetic, lively, drowsy,
and tired) were rated on a 100-mm visual analogue scale
from 0 mm (not at all) to 100 mm (extremely). Also
included were ‘headache’, a recognised symptom of caffeine
Table 1
Mean (FS.E.) habitual caffeine usage, body size and weight, and age of the six groups of participants
Consumer group
Caffeine intake
BMI
Weight
Age
Non/low consumer group
0 mg/kg
1 mg/kg
2 mg/kg
0 mg/kg
1 mg/kg
2 mg/kg
275.5F37.8
21.1F0.5
60.1F2.1
22.6F1.3
223.0F32.5
24.0F1.2
71.0F2.9
21.5F0.8
167.0F21.5
21.1F0.9
60.7F3.4
23.5F2.2
36.0F9.8
21.5F0.5
67.1F2.9
23.6F1.6
21.5F9.3
21.2F0.4
58.7F1.7
20.1F0.7
24.0F8.7
22.12F1.0
65.4F2.5
20.1F1.4
320
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
withdrawal, ‘hunger’, and ‘thirst’. Once the entire fruit tea
was consumed, subjects received breakfast and ate as much
as they wanted. Thirty minutes later, a second mood
questionnaire was completed, again according to current
mood.
On Day 6 (posttest session), subjects again provided a
saliva sample and completed a mood questionnaire. As were
in the pretest, all five tea flavours were rated, this time with
the relevant caffeine or placebo dose added to the target tea.
Again, participants indicated their most preferred tea. Following breakfast, a structured debriefing was conducted with
participants asked what they thought the experiment was
about, whether they thought it involved caffeine, and whether they thought there was caffeine in the fruit tea. Participants were then given a full explanation of the purpose of
the experiment, thanked, and paid for participating.
2.5. Data analysis
The rated pleasantness of the fruit teas was assessed
across the four conditioning days using a mixed model
ANOVA, with group (non/low consumers vs. consumers)
and dose (0, 1, and 2 mg/kg) as between-subject factors, and
conditioning day [1 –4] as a within-subject factor. Inasmuch
as we were specifically predicting an increase in rated
pleasantness for the caffeine-paired flavour with exposure
in the moderate consumer group, we also calculated the
difference in rated pleasantness on Conditioning Day 4
(when change should be maximal) relative to Conditioning
Day 1 and contrasted these change scores between groups
using two-way ANOVA. The same approach was used to
assess change in pleasantness for the target tea from the
pretest and posttest sessions.
Change in ranked preference was assessed from the
pleasantness ratings made on all five teas at the pretest
and posttest sessions. The fruit tea given the lowest pleasantness rating was assigned Rank 1, and the fruit tea with
the highest pleasantness rating assigned as Rank 5. Change
in ranking was assessed by subtracting the rank of the tea at
posttest from the ranking at pretest.
To assess changes in pleasantness for the nontarget teas,
paired t tests were conducted on the pleasantness ratings for
each of the nontarget teas at the Pre- and posttest sessions.
ANOVA was not possible for these data inasmuch as
different participants were exposed to different teas, depending on their initial pleasantness evaluation, and so different
teas were available for analysis for different consumers.
Using separate protected t tests therefore allowed us to
include all data for nonexposed teas before and after
conditioning.
To assess whether caffeine had any effect on mood,
ratings were taken prior to, and 30 min after ingestion of
the tea. Those mood ratings previously shown to be sensitive to caffeine (jittery, energetic, lively, tired, clearheaded,
and headache) as well as hunger and thirst were analysed by
a 2 (group)3 (dose)4 (day) repeated-measure ANOVA
over the four conditioning days. Inasmuch as the number of
possible tests was large, significance was adjusted using the
Bonferoni correction to avoid Type 2 errors. For brevity, we
only discuss significant findings.
3. Results
3.1. Pleasantness and novelty ratings for the target tea
Analysis of the pleasantness ratings over the four conditioning days (Fig. 1) revealed a significant GroupDose
Day interaction, [ F(6,162)=2.47, P<.05]. To simplify analysis of this complex interaction, these changes in pleasantness were examined further by calculating overall changes
in pleasantness across the four conditioning trials, at which
time any changes through association should be maximal
(Fig. 2). Here, a GroupDose interaction [ F(2,54)=5.99,
P<.005] confirmed that non/low consumers rated pleasantness of the noncaffeinated fruit tea increased significantly
[t(9)=2.41, P<.05], whereas consumers rated pleasantness
tended (NS) to decrease for this drink. Conversely, rated
pleasantness of the tea containing 1 mg/kg of caffeine
increased significantly over this time in the consumer group
as predicted [t(9)=2.04, P<.05, one-tailed], but no significant changes in pleasantness were observed in the non/low
consumer group. With 2 mg/kg, there was a tendency for
pleasantness to increase in the consumers and to decrease in
the non/low consumers. As expected, novelty ratings decreased over the 4 days [ F(3,162)=7.512, P<.001].
The changes in pleasantness during the conditioning
period were also reflected in the evaluations made on Days
1 and 6, before and after conditioning. A GroupDose
interaction [ F(2,54)=3.284, P=.045] of the change in
pleasantness for the target tea reflected distinct patterns
in the consumer and non/low consumer groups. In the
latter, pleasantness increased significantly for the noncaffeinated fruit tea [t(9)=3.47, P<.01] and tended to decrease
for the fruit tea containing 2 mg/kg of caffeine, whereas
consumers rated the fruit tea containing 1 mg/kg of
caffeine as significantly more pleasant after than before
conditioning [t(9)=1.84, P<.05, one-tailed], but rated pleasantness for the drinks with 0 or 2 mg/kg did not change
significantly.
The present design relies on all five potential target teas
showing similar changes in liking when paired with the
presence or absence of caffeine. If this was not so and if
choice of target had differed between the six test groups, this
would question the validity of any conclusions we might
make about conditioned changes in flavour evaluation. To
test this assumption, we examined the frequency with which
the five teas were selected as target for the six test groups.
All five teas were used, and the frequency with which the
teas were used did not differ form chance [chi-squared
(4)=8.5, NS]. The frequency with which each drink was
used as target by the consumer and nonconsumer groups
321
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
Fig. 1. Mean (FS.E.) rated pleasantness of the target fruit tea on the four conditioning trials for the non/low consumers (left hand panel) and moderate
consumers (right hand panel). The target drink contained either 0 ( ), 1 ( ), or 2 ( ) mg/kg of caffeine.
was the following: Lemon and Ginger, consumers=6, non/
low consumers=4; Camomille and Spiced Apple, consumers=10, non/low consumers=9; Ginger, Orange, and Honey,
consumers=7, non/low consumers=8; Sweet Berry Swirl,
consumers=2, non/low consumers=4; Camomille and Honey, consumers=5, non/low consumers=5. Thus, there were
no differences between consumers and nonconsumers, and
although numbers in each category were too small to allow
statistical analysis, inspection of target choice across the six
groups did not suggest any systematic differences in the
frequency with which each tea was used as target across
groups.
3.2. Pleasantness of the nontarget teas
Table 2 shows mean pleasantness ratings for each of the
nontarget teas on Day 1 (pretest) and Day 6 (posttest). There
were no significant differences in liking for these teas
between Day 1 and Day 6 [ F(1,59)=3.61, NS]. Although,
there was a nonsignificant trend for an overall increase in
pleasantness (from 40.1F2.7 to 43.8F2.1). Inasmuch as the
only group which increased liking for the target drink when
paired with caffeine was the consumers who received 1 mg/
kg of caffeine, it was possible that liking for the nontarget
teas also increased in that group, but it was also possible that
this change was masked by the lack of change in the
remaining five groups. However, analysis of pleasantness
ratings for the unexposed teas in the consumer/1.0-mg/kg
caffeine group alone did not support this [ F(1,9)=1.26, NS],
with a nonsignificant increase in liking for the unexposed
teas (from 37.4F4.5 to 41.2F2.7) in line with the overall
nonsignificant change in pleasantness.
3.3. Ranked preference for the target tea
The target tea was selected as the median-preferred tea at
the pretest session, and so it was possible to test whether
Table 2
Mean (FS.E.) pleasantness ratings of the unexposed (nontarget) teas on
Days 1 (pretest) and 6 (posttest)
Tea
Fig. 2. Mean (FS.E.) change in rated pleasantness between Conditioning
Days 1 and 4 trials for the non/low consumers ( ) and moderate
consumers ( ) consuming the target drink with either 0, 1, or 2 mg/kg of
added caffeine.
Lemon and ginger
Camomile and spice apple
Ginger, orange, and honey
Sweet berry swirl
Camomile and honey
n
50
41
45
54
50
Pleasantness rating
Day 1
Day 6
39.4F3.8
43.4F4.3
32.8F3.9
47.6F4.3
31.6F4.1
43.7F3.4
50.5F3.4
39.2F4.2
46.2F4.4
35.2F4.2
322
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
exposure to these teas with or without added caffeine during
conditioning resulted in a change in ranked preference at the
posttest session (Day 6). Although the ranked preference
changes were in the same direction as actual changes in
rated flavour pleasantness (Table 3), analysis found no
significant effects of group or caffeine dose on the ranking
measure.
3.4. Mood ratings
Consumers reported higher headache ratings than non/
low consumers [ F(1,54)=10.296, P<.005], indicating acute
caffeine withdrawal effects among the consumer group.
Headaches were relieved by consumption of the fruit tea
[ F(1,54)=12.18, P=.001], however, this was irrespective of
group or caffeine content. Non/low consumers reported
feeling more clearheaded than consumers [ F(1,52)=7.22,
P<.01], and although clearheadedness increased after fruit
tea consumption [ F(1,52)=7.703, P<.01], it was unaffected
by dose [ F(2,52)=0.593, NS]. As expected, there was a
main effect of time on hunger and thirst ratings with
participants less hungry and thirsty after consumption
[ F(1,52)=152.78, P<.001 and F(1,52)=80.29, P<.0001,
respectively], but this was unaffected by group or caffeine
dose.
4. Discussion
This experiment aimed to verify the reinforcing effect of
caffeine between non/low and moderate consumer populations and at different doses. The caffeine doses used in this
study provided ecologically valid amounts, as mean bodyweight of participants was 63.8 kg, thus 1 mg/kg provided
caffeine similar to that contained in tea (60 mg/cup), while 2
mg/kg yielded an amount similar to that found in filter
coffee (125 mg/cup).
A difference in the reinforcing effects of caffeine between non/low consumer and consumer groups was evident.
Increased liking for caffeine-paired flavours only occurred
among regular moderate caffeine consumers, with liking for
the noncaffeinated version reduced among this group.
Conversely, for non/low consumers, pleasantness ratings
tended to decrease for both caffeinated drinks yet increased
significantly for a noncaffeinated drink. This supports and
extends the view that caffeine is primarily a negative
Table 3
Mean (FS.E.) change in rank preference for the target fruit tea between
pretest and posttest sessions as a function of normal caffeine consumption
and level of caffeine during exposure
Group
Non/low
consumers
Consumers
Caffeine dose in drink during conditioning (mg/kg)
0
1
2
0.50F0.55
0.20F0.41
0.00F0.44
0.00F0.38
0.80F0.47
0.15F0.44
reinforcer in this context, as only moderate consumers
would be responsive to the withdrawal alleviation of the
caffeine, while non/low consumer groups may experience
negative effects from ingestion of caffeine resulting in
decreased liking. In the earlier study [1], low consumers
(<120 mg/day) reported an increased preference for the
target drink regardless of caffeine content, whereas the
present study only found an increase in pleasantness for
the drink without caffeine. Inspection of the earlier data [1],
however, suggests that in that study, the change in preferences was also greater for the noncaffeinated than the
caffeinated drink, but this difference was not significant.
Moreover, the different daily caffeine intake levels of the
populations in the two studies may also partly explain the
slight difference between the two studies. Maximal habitual
caffeine use of low consumers in Rogers et al. study [1] was
almost twice that of the non/low consumers in the present
study (<70 mg/day). In the present study, the daily caffeine
intake of the non/low consumer group was below the
minimal level at which withdrawal effects have been
reported (100 mg; [12]), but withdrawal reversal may still
have been an element of preference change in the low
consumer group from the study by Rogers et al. [1].
Inasmuch as it has also been reported that consumption of
two evenly spaced caffeinated drinks per day was enough to
sustain optimal mood and psychomotor performance among
caffeine consumers [21], the low consumer group in the
earlier study [1] may also have incorporated consumers who
regularly experienced beneficial effects from the ingestion
of caffeine. Furthermore, although no detrimental effects of
caffeine consumption were evident from the mood data in
the present study, it is possible that the non/low consumers
experienced some aversive effects from the caffeinated
drink, particularly at the higher caffeine dose, and this
aversive effect was the basis for the tendency for pleasantness to decrease in that group. Thus, the slight differences in
responses between the non/low consumer group in the
present study and that reported by Rogers et al. [1] could
be explained by differences in study power along with
inclusion of some mildly caffeine-dependent consumers in
the low consumer group in the earlier study.
Overall, the present results show a trend towards an
inverse relationship between the reinforcing effect of caffeine and dose. Among the moderate consumers, rated
flavour pleasantness increased, and preference tended to
increase, at the 1-mg/kg dose, but these effects were
marginal at the higher (2-mg/kg) dose, supporting previous
research of similar dose effects. For example, voluntary
intake of coffee was significantly greater at lower (25- and
50-mg) doses than at a higher (100-mg) dose [22]. Higher
caffeine doses have also been reported as less reinforcing
than lower doses (100 and 200 mg; [23]), and coffee with
very high caffeine content is aversive (400 and 600 mg;
[23]). Together, these findings suggest that there is an
optimal dose range at which caffeine reinforces flavour
preferences among moderate caffeine consumers, which
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
falls within levels contained in regular caffeinated drinks. In
the present study, any such negative effects of the 2.0-mg/kg
caffeine dose in the consumer group may have been
exacerbated by them having (by chance) a lower habitual
caffeine consumption than the placebo-treated consumer
group inasmuch as tolerance would be predicted to be
related to habitual caffeine usage.
The current experiment measured change in preference
and change in rated pleasantness for the target drink. The
results are similar to those reported in previous flavour
conditioning studies in that change in rated pleasantness
(e.g., Ref. [2 – 5,24]) provided evidence of conditioned
changes in liking. However, relative preference for the target
drink, the method used in the earliest investigation of
flavour preference conditioning with caffeine [1], resulted
in nonsignificant trends only although these trends were
broadly in line with changes in rated pleasantness. The
difference between changes in pleasantness and changes in
ranked preference could be explained by the sensitivity of
the measurement methods used. Firstly, change in rank
produces ordinal scores ranging from 2 to +2, whereas
pleasantness ratings were measured on a 100-point scale.
Thus, the greater variability in scores permitted by the rating
method allowed for a more sensitive evaluation of change
than with ranked scores. Secondly, in the current experiment, the fruit teas were assigned ranks calculated from
pleasantness ratings on Days 1 and 6, rather than a separate
evaluation of preference. Thus, for a drink to alter in ranked
preference using this method, the increase in rated pleasantness had to exceed the difference in pleasantness between
itself and the next highest ranked tea at pretest. Thus, if the
teas that were ranked one higher than the target tea had
pleasantness ratings that were at least 15 points higher on
the 100-mm rating scale, a change in ranked preference
could only be seen if the rated pleasantness of the tea at
posttest was at least 15 units. In practice, the ranking
procedure used in the present study may not have been
optimal, and an independent ranked preference measure
may have been more appropriate. However, other studies
in this laboratory suggest that where choice is not driven by
factors other than the pleasantness dimension (i.e., where
factors, such as belief, drink temperature, etc, are controlled
for), pleasantness and preference closely match. An interesting question for future research would be to contrast
changes in pleasantness and preference where preference
may be driven by a nonhedonic factor.
Change in pleasantness was specific to the target flavour,
as pleasantness ratings for the four other flavours sampled at
the initial and final sessions did not change significantly
between the pretest and posttest sessions, and similar findings were reported in a related recent study [25]. This was
the case overall and, independently, for the only group who
increased liking for a caffeine-paired flavour (the consumer
group receiving 1.0 mg/kg of caffeine). Together, these
results imply that changes in pleasantness were due to the
postingestive effects of caffeine associated with the specific
323
flavour encountered during conditioning and provides further evidence of Pavlovian conditioning in flavour preference conditioning.
Analysis of the mood data in the present study showed
significant differences between the non/low consumer and
consumer populations for headache (consumers reporting
higher ratings at both time points) and clearheadedness, with
non/low consumers more clearheaded at both time points.
During debriefing, one of the participants from the non/low
consumer group receiving 1 mg/kg commented on feeling
jittery between 1 and 2 h after the conditioning sessions, and
another nonconsumer receiving 2 mg/kg mentioned feeling
jittery half an hour after the sessions finished. Therefore, to
better understand the time-course of these effects and
whether they differ between consumers and non/low consumers, future studies should examine changes in mood
over a more prolonged period postingestion.
Although the data overall clearly demonstrated an increase in liking for a flavour paired with 1.0 mg/kg of
caffeine consumed by overnight-deprived moderate caffeine
consumers, the study suffered form a number of methodological shortcomings. Firstly, the contrast in evaluations of
the drinks pre- and postconditioning was complicated by the
addition of the relevant additives from the training phase
into the drink at posttest. The reasons for this were pragmatic: inasmuch as pretest determined which drink was to
be the target, we could not include the relevant additives at
that stage. However, once participants had become familiar
with the drink with the relevant additive during training,
subsequent omission of that small flavour component could
have affected their response at posttest. In practice, the
effects of having caffeine or placebo present are likely to
have been small, and inasmuch as caffeine is bitter and
therefore mildly aversive in flavour, any effect of the added
caffeine would have been counter to the increase in pleasantness predicted for the caffeine-paired flavour. Indeed, our
past studies have consistently failed to find effects of does of
caffeine up to 100 mg on liking for these types of drinks [3–
5,24]. Where the presence of caffeine may have been
important is in the 2.0 mg/kg group, where the bitter flavour
may have acted to counter any increase in liking through
association and so masked learning in these groups.
In conclusion, the present experiment supported and
extended earlier findings [1,3,5] that flavour preference
conditioning with caffeine is dependent on habitual caffeine
consumption levels and is not seen in participants who
consume very little or no caffeine. This supports the theory
that flavour preference conditioned by caffeine is primarily
the consequence of negative reinforcement in that it negates
the effects of short-term caffeine withdrawal. Interpretation
of the present data as evidence of negative reinforcement is
further supported by the more recent finding that habitual
caffeine consumers who have been fully withdrawn from
caffeine for 2 weeks prior to flavour –caffeine conditioning
developed an aversion to the caffeine-paired flavour, while
consumers who were maintained on caffeine developed a
324
E.M. Tinley et al. / Physiology & Behavior 82 (2004) 317–324
conditioned preference [2]. The present results also suggest
that the reinforcing effects of caffeine on flavour preferences
are stronger at 1 mg/kg (a dose similar to that in most
popular caffeinated beverages) than at 2 mg/kg, where any
tendency to increase preference may be confounded by
mildly aversive effects of higher dose caffeine.
[10]
[11]
[12]
Acknowledgements
[13]
This work was supported by a BBSRC studentship to
Elizabeth Tinley with additional funding from Unilever.
[14]
[15]
References
[16]
[1] Rogers PJ, Richardson NJ, Elliman NA. Overnight caffeine abstinence
and negative reinforcement of preference for caffeine-containing
drinks. Psychopharmacology 1995;120:457 – 62.
[2] Tinley EM, Yeomans MR, Durlach PJ. Caffeine does not reinforce
conditioned flavour liking in fully abstinent caffeine consumers. Psychopharmacology 2003;166:416 – 23.
[3] Yeomans MR, Spetch H, Rogers PJ. Conditioned flavour preference
negatively reinforced by caffeine in human volunteers. Psychopharmacology 1998;137:401 – 9.
[4] Yeomans MR, Jackson A, Lee MD, Steer B, Tinley EM, Durlach PJ,
et al. Acquisition and extinction of flavour preferences conditioned by
caffeine in humans. Appetite 2000;35:131 – 41.
[5] Yeomans MR, Jackson A, Lee MD, Nesic JS, Durlach PJ. Expression of flavour preferences conditioned by caffeine is dependent on
caffeine deprivation state. Psychopharmacology 2000;150:208 – 15.
[6] Fredholm BB, Battig K, Holmen J, Nehlig A, Zvartau EE. Actions of
caffeine in the brain with special reference to factors that contribute to
its widespread use. Pharmacol Rev 1999;51:83 – 133.
[7] Hughes JR, Oliveto AH, Bickel WK, Higgins ST, Badger GJ.
Caffeine self-administration and withdrawal: incidence, individual
differences and interrelationships. Drug Alcohol Depend 1993;32:
239 – 46.
[8] Mitchell SH, De Wit H, Zacny JP. Caffeine withdrawal symptoms and
self administration following caffeine deprivation. Pharmacol Biochem Behav 1995;51:941 – 5.
[9] Lane JD. Effects of brief caffeinated-beverage deprivation on mood,
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
symptoms, and psychomotor performance. Pharmacol Biochem
Behav 1997;58:203 – 8.
Hofer I, Battig K. Cardiovascular, behavioural, and subjective effects
of caffeine under field conditions. Pharmacol Biochem Behav 1994;
48:899 – 908.
Evans SM, Griffiths RR. Caffeine tolerance and choice in humans.
Psychopharmacology 1992;108:51 – 9.
Griffiths RR, Evans SM, Heishman SJ, Preston KL, Sannerud CA,
Wolf B, et al. Low-dose caffeine physical dependence in humans.
J Pharmacol Exp Ther 1990;255:1123 – 32.
Hughes JR, Higgins ST, Bickel WK, Hunt WK, Fenwick JW, Gulliver
SB, et al. Caffeine self-administration, withdrawal, and adverse effects
among coffee drinkers. Arch Gen Psychiatry 1991;48:611 – 7.
Schuh KJ, Griffiths RR. Caffeine reinforcement: the role of withdrawal. Psychopharmacology 1997;130:320 – 6.
Mela DJ, Rogers PJ. Food, eating and obesity: the psychobiological
basis of appetite and weight control. London: Chapman and Hall;
1997. 240 pp.
Griffiths RR, Bigelow GE, Liebson IA. Human coffee drinking: reinforcing and physical dependence producing effects. J Pharmacol Exp
Ther 1986;239:416 – 25.
Griffiths RR, Woodson PP. Caffeine physical dependence: a review of
human and laboratory animal studies. Psychopharmacology 1988;94:
437 – 51.
Stunkard AJ, Messick S. The three-factor eating questionnaire to
measure dietary restraint, disinhibition and hunger. J Psychosom
Res 1985;29:71 – 83.
James JE. Caffeine and health. London: Academic Press; 1991.
Richardson NJ, Rogers PJ, Elliman NA, O’Dell RJ. Mood and performance effects of caffeine in relation to acute and chronic caffeine
deprivation. Pharmacol Biochem Behav 1995;52:313 – 20.
Robelin M, Rogers PJ. Mood and psychomotor performance effects of
the first but not subsequent cup-of-coffee equivalent doses of caffeine
consumed after overnight caffeine abstinence. Behav Pharmacol 1998;
58:611 – 8.
Kozlowski LT. Effects of caffeine on coffee drinking. Nature 1976;
264:354 – 5.
Griffiths RR, Woodson PP. Reinforcing effects of caffeine in humans.
J Pharmacol Exp Ther 1988;246:21 – 9.
Yeomans MR, Ripley T, Lee MD, Durlach PJ. No evidence for latent
learning of liking for flavours conditioned by caffeine. Psychopharmacology 2001;157:172 – 9.
Yeomans MR, Pryke R, Durlach PJ. The effect of caffeine deprivation
on liking for a noncaffeinated drink. Appetite 2002;39:35 – 42.