Uncontrolled eating is associated with higher impulsiveness, risk taking and novelty-seeking

Previous studies suggest that trait disinhibition as measured by the Three Factor eating Questionnaire (TFEQ) is related to selected measures of impulsivity and risk taking. However, the factor validity of the original trait disinhibition measure has been questioned, and a revised scale of uncontrolled eating consequently developed. To date few studies have revisited the relationship between impulsivity and the uncontrolled and emotional eating scales of the revised TFEQ. In the present study, 283 participants (208 women) completed the revised TFEQ alongside a battery of measures of impulsivity and risk taking in an online study. The total and all subscale scores on the Barratt Impulsiveness Scale were significantly associated with scores on the uncontrolled, but not emotional or restrained, TFEQ scale. Likewise, risky behaviour indexed by the average number of pumps per trial on the Balloon Analogue Risk Task were also associated with uncontrolled, but not emotional or restrained, eating, and the same pattern of associations were also found for the novelty, but not intensity, subscales of the Arnetts Inventory of Sensation Seeking. Overall these data suggest that uncontrolled eating is related to wider personality traits of risky decision making, novelty seeking and wider cognitive impulsivity, which may in turn increase of future weight gain.


Introduction
Although the incidence of obesity has increased worldwide (Blüher, 2019), there is clear evidence that people vary considerably in their propensity to gain weight (Goossens, 2017).Consequently, studies have explored the ideas that individual differences in sensitivity to reward may increase the risk of overeating (see Stice & Burger, 2019).This is supported by studies showing heightened reactions to food cues in overweight and obese individuals (Hendrikse, et al., 2015).Since reward sensitivity is a key component of general impulse control (e.g.Franken & Muris, 2006;Smillie & Jackson, 2006), individual differences in general impulsivity may therefore be a key behavioural trait underlying susceptibility to obesity (Giesbrecht, Rycroft, Rowson, & De Bruin, 2010;Martin & Potts, 2004).Moreover, if sensitivity to reward is a risk factor for subsequent obesity, then evidence for a relationship between reward sensitivity and measures of poor control of appetite should be evident in the wider population, especially a younger population who may not have developed obesity.
To date, many studies have examined how eating and different measures of impulsivity are related in non-obese populations, with varied outcomes (reviewed by Loxton, 2018;Maxwell, Gardiner, & Loxton, 2020).Studies have also examined a wide range of aspects of impulsivity, including both self-report and objective measures, and encompassing impulsive decision making, risk-taking and inhibitory control.Notably one aspect of impulsivity that has received very little attention is sensation-seeking is, and the current paper aims to rectify that shortcoming.
The primary aim of the present study was therefore to investigate further whether the inconsistencies in the literature relating uncontrolled eating and impulsivity might have been in part due to weaknesses in the construct of the TFEQD scale.The original TFEQD factor was formulated based on analyses of the factor weightings of individual items in the original paper which developed the TFEQ (Stunkard & Messick, 1985).Subsequent studies however questioned that factor structure, noting that both the TFEQD and the hunger factor (TFEQH) were unstable (Hyland, Irvine, Thacker, Dann, & Dennis, 1989), that three items had low communalities, and analysis of the remaining items suggested that TFEQD was itself a combination of three components (Bond, McDowell, & Wilkinson, 2001).Moreover, many of the items on the original TFEQD required simple dichotic yes/no responses on items where actual behaviour was unlikely to be bivariate.That led to the development of a modified 18-item version of the TFEQ with a validated revised factor structure (TFEQ18: Karlsson, Persson, Sjostrom, & Sullivan, 2000).That version collapsed the original disinhibition and hunger scales into a single scale defined as uncontrolled eating (TFEQ18uc) and introduced a new third factor of emotional eating (TFEQ18e), since emotional eating was not specifically defined in the original TFEQ.There has long been evidence that emotional eating is a risk factor for obesity (Faith, Allison, & Geliebter, 1997;Ganley, 1989) and related disordered eating, especially binge eating (Ricca, et al., 2009;Zeeck, Stelzer, Linster, Joos, & Hartmann, 2011).Previous studies have reported that emotional eating (measured using TFEQ18e or the Dutch Eating Behaviour Emotional Eating scale) and impulsiveness (BIS-11) were positively associated (Ebneter, Latner, Rosewall, & Chisholm, 2012;Jasinska et al., 2012;Lattimore, Fisher, & Malinowski, 2011).
Given the inconsistencies in the literature exploring how TFEQD was related to impulsivity, the primary aim of the present study was to reexamine these relationships using the validated TFEQ18.To date we are aware of just two studies with similar designs.In the first (Lattimore, et al., 2011), TFEQuc was positively correlated with the total and all three BIS subscales, and that was replicated in a more recent study (Goodwin, Butler, & Nikčević, 2023).The correlations in both those studies were all moderate in size (0.31-0.38 in , and numerically greater than those seen in earlier studies using the older TFEQD measure.This suggested that the revised TFEQuc measure might better capture impulsive influences on eating.However, the more recent study (Goodwin, et al., 2023) only tested the TFEQuc subscale of the TFEQ18.The present study therefore aimed to re-evaluate associations between all three sub-scales of the TFEQ18 (uncontrolled eating, emotional eating and restrained eating, TFEQ18R), and measures of impulsiveness (BIS-11 and risk-taking (BART).
The secondary aim of the present study was to extend the exploration of which facets of impulsivity relate to uncontrolled and emotional eating to include sensation seeking.Sensation-seeking refers to the tendency to seek out varied, complex, novel, and intense experiences (Zuckerman, 1971).In relation to eating behaviour, sensation seeking would be predicted to promote greater variety in the diet since sensation-seekers would look for greater novelty and intensity in the foods they choose.Food variety has long been shown to be associated with greater food intake (Johnson & Wardle, 2014;Raynor & Epstein, 2001).We therefore explored how the three TFEQ18 factors were related to sensation seeking, predicting that uncontrolled eating in particular would be positively associated with sensation seeking measured using the Arnett Inventory of Sensation-seeking (AISS: Arnett, 1994).

Design
The study used a regression model to test the associations between each of the three sub-scales of the TFEQ18 and measures of impulsiveness (BIS-11), risk-taking (BART), and sensation-seeking (AISS), while controlling for effects of age, sex and body-size.

Participants
The final sample were 283 volunteers (206 women) who responded to adverts for the study posted on social media (Facebook and Instagram), and on student participation sites at the University of Sussex.The study was advertised as an investigation of the relationship between personality and eating behaviour and was open to anyone aged 18 or over who had not been diagnosed with an eating disorder.Potential participants accessed the study using an online link which opened the study information sheet and consent form.The final sample were those who completed all three sections of the study (consent, BART task and questionnaires).Participation was encouraged through prize draws with a £25 prize: those who completed the online data collection had an opportunity to enter that draw by leaving their email through a second linked survey which ensured that no personal details were linked with the recorded data.The study protocol was approved by the University of Sussex School of Psychology Ethics committee, with separate approvals for data collection in 2020-2021 (ER/IEB21/1) and 2021/2022 (ER/ NK/430/1).

Eating behaviour
Eating behaviour was assessed using the eighteen item TFEQ-R18 orginally developed by Karlsson et al. (2000), a shortened version of the original 51-item TFEQ (Stunkard & Messick, 1985) designed to overcome concerns with the factor structure and validity of the original TFEQ (Bond, et al., 2001;Karlsson et al., 2000).In the TFEQ18, some of the items originally used in the original TFEQ disinhibition and hunger factors were combined into a single factor labelled uncontrolled eating (TFEQuc).Karlsson et al. (2000) also introduced an emotional eating subscale (TFEQe) to measure eating in response to negative affect, while they retained cognitive restraint (TFEQr) as the third factor based closely on the original TFEQ questions.Another key difference between the TFEQ18 and the original TFEQ was the response structure: whereas the majority of items in the original TFEQ were scored on a dichotomous True/False basis, most items on the TFEQR18 use four-point Likert Scales, so increasing variability in overall scores despite reducing the number of items.The TFEQ18 scale has been shown to have high internal consistency (Martins, da Silva, Maroco, & Campos, 2021;Mostafavi et al., 2017), and that was confirmed for the three subscales in this study: Cronbach alpha overall was 0.81, and for the three subscales was 0.79 (TFEQ18uc), 0.86 (TFEQ18e), and 0.71 (TFEQ18r).

Measures of impulsivity and risk-taking
Impulsivity and risk-taking were assessed by three widely used measures.(BIS-11: Patton et al., 1995) General behavioural impulsiveness was measured using the BIS-11, which consists of 30 items each scored on a 4-point Likert scale ranging from "Rarely/never" to "Almost/always".Items describe a wide M.R. Yeomans et al. range of impulsive responses.Factor analysis consistently identifies three subscales (Lejuez, et al., 2002) in the responses, which have been defined as attentional, motor and non-planning impulsiveness.The BIS11 has been used in studies of normal and abnormal eating behaviour (e.g.Bénard et al., 2019;Lundahl, Wahlstrom, Christ, & Stoltenberg, 2015), and more widely in relation to general risk taking.Cronbach alpha overall was 0.86 for the BIS-11 total score in the present sample.

Barratt Impulsiveness Scale
2.4.2.The Balloon Analogue Risk Task (BART: Lejuez et al., 2002) The BART task was developed as a novel behavioural model of risk taking (Lejuez, et al., 2002).In the BART task, participants are presented with an on-screen representation of a balloon, alongside a box indicating the value of that balloon.Initially the value is 0, but in a series of discrete trials the participant then can inflate the balloon by completing an action (here pressing the keyboard spacebar).Every action ("pump") increases the value of that balloon (in our version by £0.05).The participant then has the choice of either collecting the money received in the current trial into their bank, or to keep pumping and increasing their potential gains.If the participant decides to collect the money, that trial ends.However, there is a risk that any pump may cause the balloon to explode, and all the money gained on that trial is then lost.Participants were made aware that the balloon might burst at any time, and they completed 30 trials in total.The version of the task used here was presented using Inquisit software.Risk taking on the BART can be indexed by a number of measures, of which the most commonly used are the average number of pumps made on each trial, and the total number of balloon explosions.These measures were calculated automatically by the Inquisit version.The BART has been widely used in studies of risk taking, especially in relation to addiction and drug-use (e.g.Lejuez, Aklin, Jones, et al., 2003;Lejuez, Aklin, Zvolensky, & Pedulla, 2003), and has also been used to assess individual differences in eating behaviour (Goodwin, et al., 2023;Leitch et al., 2013;Leslie, Leppanen, Paloyelis, Nazar, & Treasure, 2019;Yeomans & Brace, 2015).Arnett, 1994) The final measure used was the AISS, included as a standardised measure of sensation seeking.The AISS is a self report measure with 20 items all scored on a 4-point Likert scale, and with two subscales (10 questions for each) of novelty and intensity.The AISS was developed as a more reliable measure than the original Zuckerman sensation seeking scale (Zuckerman & Aluja, 2015), and has good internal consistency (Arnett, 1994).It has been widely used in research into personality, but has been less widely used in relation to eating behaviour (Alley & Potter, 2011;Byrnes & Hayes, 2016).In our sample, while overall consistency was adequate (Cronbach alpha = 0.76), the two sub-scales were less consistent (Cronbach alpha's: novelty = 0.66, intensity = 0.62).

Procedure
Data were collected between November 2020 and July 2022.Participants accessed their survey using a link in the online survey advertisement, which automatically opened the study information sheet and consent form (presented using Qualtrics (Qualtrics, Provo, UT) for those who completed the study in 2020/2021 and Inquisit (Inquisit 6, Millis econd.com),for those completing the survey in 2022).Once potential participants had selected on-screen options confirming they were aged 18, that they not been diagnosed with an eating disorder, that they formally consented to participate, and they were accessing the study on a computer (since the BART task is not designed for mobile devices), they were automatically presented with the BART task.On completion, the participant was automatically redirected to a survey presented using Qualtrics, where they completed the BIS-11, TFEQ18 and AISS in that order.They then provided basic demographic information: their age (in years), biological sex (options female, male or decline to answer), their weight and height (in their preferred units).They were then debriefed about the aims of the study and given a final chance to withdraw their data before the opportunity to enter their email in a separate prize draw, with a £25 prize (one draw for each of the two surveys).

Data analysis
The planned data analysis was preregistered through the Open Science framework (DOI 10.17605/OSF.IO/9MSHU) prior to the two survey datasets being merged into the final dataset.
BIS-11, TFEQ18 and AISS were all scored following the standard methodology for each measure.Height and weight data were converted to kilograms and metres respectively, and these measures used to generate estimated body-mass index (BMI) as weight/(height) 2 .For the BART, the adjusted average pumps per balloon and total explosions were extracted from the summary data automatically provided by the Inquisit version of the task.
Prior to analysis, all variables were checked, missing data noted and variables examined for outliers and violations of normality.Six participants declined to provide weight data, and three declined to provide their age or sex, and these were coded as missing data.Data were missing for three participants on the BART task and the TFEQr score was missing for one participant.In analyses, we included all available data in order to maximise overall study power.
To assess the inter-relationships between the three TFEQ factors and the BIS-11 (total and sub-scale scores), BART (pumps and explosions) and AISS measures (sub-scale scores), separate linear models were constructed with the relevant TFEQ factor as dependent measure.At step one we ran the simple regression (TFEQ ~ <impulsivity measure>), and then to check for possible confounding effects age, BMI and sex were added at step 2. Q-Q plots found no evidence of violation of normality of residuals, and/or evidence of violation of homoscedasticity at step 1.Because of outliers in BMI and age, step 2 analyses were repeated using robust models (using the robust package in R): those analyses did not improve model fits, or affect the overall analysis outcome and so are not reported (for brevity), but can be found in the published dataset at 10.25377/sussex.23642778.For clarity, we report the standardised Beta for all factors.All analyses were conducted using R (version 4.3.0)run through the RStudio framework (version 2023.06.0 + 421) on Macintosh computers.

Summary data
The dataset is summarised in Table 1.Both BMI and age were skewed: for BMI, 14 participants were significant outliers, and all of these had BMI in the obese range (32.98-43.27),while for age 22 participants were outliers with ages of 47 or over.

Emotional eating (TFEQe)
There were no significant associations between emotional eating and any of the measures of impulsivity or risk taking in this study: for brevity we summarise these regression models in Table 2.

Restrained eating (TFEQr)
There were also no significant associations between restrained eating and any of the measures of impulsivity or risk taking: again, for brevity we summarise these regression models in Table 2.

Discussion
The present study found clear evidence that uncontrolled eating, as indexed by the TFEQuc factor of the TFEQ18 questionnaire, was associated with higher scores in impulsiveness (BIS-11 scale overall), including all three BIS-11 subscales (attention, motor and nonplanning), with higher risk taking on the BART task, and with higher novelty seeking, but not intensity seeking, on the AISS.In contrast, neither the emotional eating (TFEQe) or restraint (TFEQr) factors were significantly related to any of the impulsiveness, risk-taking or sensation-seeking measures in this study.
The key findings that uncontrolled eating was significantly related to impulsiveness as measured by all three subscales of the BIS-11 and greater risk taking on the BART extend earlier work using the original 52-item TFEQD factor.In particular, the present finding that all four measures of the BIS-11 (total and all three subscales) were related to uncontrolled eating replicated other studies using the TFEQuc measure (Goodwin, et al., 2023;Lattimore et al., 2011), and contrasted with the earlier findings using the older TFEQD measure that never found any relationship with the non-planning BIS-11 factor (Leitch, et al., 2013;Lyke & Spinella, 2004;Yeomans et al., 2008).Notably the correlations with overall impulsiveness using the TFEQuc measure were consistently stronger than were seen with the older TFEQD measure, suggesting that the refinement of the TFEQ score better captures the impulsive nature of uncontrolled eating.
A key novel finding in the present study was the association between uncontrolled eating and the AISS novelty seeking scale, the first study we are aware of to show sensation seeking as part of the uncontrolled eating phenotype.One earlier study did assess how restraint (measured by the original restraint scale (Herman & Mack, 1975),) was related to sensation seeking (Jansen, Klaver, Merckelbach, & van den Hout, 1989).Their results suggested that high restraint scores were related to sensation seeking.In contrast in the present study, using a larger sample and a more focussed restraint measure, there was no evidence that restrained eating was associated with either measure of AISS sensation seeking.A further study explored sensation seeking in relation to binge-eating in adolescents, and found some evidence that sensation seeking was Fig. 1.The relationships between (A) the total score on the Barratt Impulsiveness Scale, (B) average number of pumps per trial on the Balloon Analogue Risk Task and (C) the novelty subscale of Arnett Inventory of Sensation-seeking, and scores on the uncontrolled eating scale from the 18-item Three Factor Eating Questionnaire.
M.R. Yeomans et al. associated with binge-eating in males, but not females (Laghi, Pompili, Baumgartner, & Baiocco, 2015).Here, sex had no influence on the association between TFEQuc and novelty seeking and given that our study population was disproportionately female that suggests that uncontrolled eating, which is a much broader concept than binge-eating, may include some component of being attracted to novel foods.We therefore hypothesise that sensation seeking might operate here by increasing dietary variety, with that variety then driving intake, and that idea now needs to be tested.
The present study was also the first to show that risk-taking indexed by the BART task was associated with higher uncontrolled eating, replicating an earlier finding using TFEQD in our laboratory (Yeomans & Brace, 2015: but see Leitch et al., 2013).This suggests that risky decision making is another aspect of the trait that underlies impulsive food choice, and so could be a novel target for intervention.The finding here found uncontrolled eating related to the average number of pumps on each BART trial, but not to the number of consequent balloon explosions may seem inconsistent: more pumps should, on average, lead to more explosions.The explanation looks to be that the explosions measure is based on a single value, whereas the average pumps is a consolidated measure across trials.Thus the explosions measure has less accuracy and variance, making weaker relationships harder to detect.We included both measures as they have been used widely in research on risk-taking, although the average pumps measure is now the more frequently used one.
A more surprising finding in the present study was the absence of any significant relationships between emotional eating and the impulsiveness, risk-taking or sensation-seeking measures we tested.The only previous study that examined how TFEQe and BIS-11 impulsivity were related however did find associations with all factors other than the BIS non-planning scale (Lattimore, et al., 2011), and the population tested (mainly students at a UK university) was similar to ours.Why we did not replicate that earlier finding is not clear, although we note that the TFEQe scale only has three items, and that is likely to restrict variance and that could impact study outcomes.But based on the current data we conclude that there was no evidence that emotional eating related to wider general measures of impulsivity, risk taking or sensation-seeking.
Finally, although one previous study suggested restrained eating may be related to impulsivity, (e.g.Nederkoorn, Van Eijs, & Jansen, 2004), the majority of previous work has found no association (Leitch, et al., 2013;Yeomans & Brace, 2015;Yeomans et al., 2008), and here we also found no evidence here restraint was related to any of the measures of impulsiveness, risk taking or sensation-seeking.Thus unlike uncontrolled eating, restrained eating (which is likely to arise as a response to a history of overconsumption) does not seem to be reliably associated with impulsivity.This is in some ways surprising: if uncontrolled eating causes overconsumption and consequent weight gain, it might be predicted that uncontrolled eating should also lead to higher restraint (and thereby restraint to indirectly related to impulsivity).Although this was not a primary aim of the current study, we therefore did some exploratory analysis of how TFEQr and TFEQuc were inter-related, and how these related to BMI in our sample.In our sample, TFEQr was unrelated to TFEQuc (r(281) = − 0.04), and neither measure was significantly correlated with BMI.Thus in our sample, uncontrolled eating was associated with greater risk taking, novelty seeking and higher impulsiveness, but there was no evidence that those behaviours had resulted in higher body-size or greater attempts to restrict intake in this younger healthy population.
In the wider literature, it has also been suggested that impulsivity relates to body-size, mainly based on studies reporting higher impulsivity in people with obesity relative to those whose BMI falls in the normal range (e.g.Braet, Claus, Verbeken, & Van Vlierberghe, 2007), but also with studies suggesting impulsivity is related to BMI in cross-sectional data (Bénard, et al., 2017).In the present study, none of the measures of impulsiveness or risk taking were related with BMI.Our population had an average BMI in the normal range and were mainly young adults, with less than 5% were in the obese range.It may be that our population therefore underestimated any wider association with BMI at population level.However, given that there were significant associations between impulsivity measures and uncontrolled eating, our data perhaps also suggest that the associations with BMI in other studies may reflect the outcome of impulsive behavioural choices leading to weight gain.We should however add a note of caution in interpretation of BMI, as the present studies relied on self-report of height and weight, although the wider literature does suggest self-report data are reasonably accurate (Davies, Wellard-Cole, Rangan, & Allman-Farinelli, 2020;Lin, DeRoo, Jacobs, & Sandler, 2012).Future studies should look to re-explore these findings with an older population with higher incidence of obesity and higher restraint to test whether impulsive overeating then emerges as a risk for weight gain that was not evident in our younger, healthier sample, and using actual measures of height and weight to ensure greater accuracy.
One limitation in the current study was that it was conducted entirely online, and so that meant we could not control for participants behaviour prior to completion of the study.This may be important since the relationship between eating and impulsivity is acutely sensitive to food restriction prior to the impulsivity task (Leitch, et al., 2013) or acute exposure to food cues prior to the impulsivity test (Yeomans & Brace, 2015).It would therefore now be useful to reinvestigate the relationships we tested here while manipulating food deprivation and acute food cue exposure.As we were focussed on what aspects of impulsivity were related to aspects of eating that are seen in disordered eating, we excluded potential participants with an existing eating disorder.That decision in turn will likely have limited the severity of the eating scores in our sample, although we did see examples of scores across the whole range of all three TFEQ subscales.As the key factor associated with impulsivity was uncontrolled eating, which is a key component of binge-eating, a follow-up study including a sample of participants diagnosed with binge-eating disorder would be valuable.
The implication of the current findings is that higher impulsiveness, risk-taking and novelty-seeking may all be key traits that are all associated with uncontrolled eating.This suggests that interventions that can improve impulse control may help mitigate the risk either of weight gain through over-eating or development of eating disorders involving binge-eating in those scoring high in uncontrolled eating, and suggests in particular that the focus should be on uncontrolled eating, rather than restrained or emotional eating.
In summary, the present study further investigated individual differences in eating behaviour and their association with impulsivity and risk taking.The key findings were that uncontrolled eating was related to higher impulsiveness, greater risk taking and greater novelty seeking.

Table 1
Summary data for all variables measured in this study.

Table 2
Summaries of all regression models relating emotional and restrained eating and the different measures of impulsivity and risk taking.All data are standardised β with confidence intervals from the Step 1 regressions.*p < 0.05, **p < 0.01, ***p < 0.001.