A longitudinal study examining the influence of diet-related compensatory behavior on healthy weight management

The aim of the present study was to examine the role of diet-related compensatory behavior in healthy weight management regarding diet quality, physical activity, and body mass index (BMI) over time in a non-clinical general population. Data were based on the first and third waves of the Swiss Food Panel 2.0 survey, which included questions about food consumption frequencies and constructs measuring weight management strategies. Data were examined using principal component analysis and correlation analyses to examine the psychometric properties of the adapted items, and multiple linear regression analyses for longitudinal examination. The adapted items measuring diet-related compensatory behavior were shown to be valid and reliable. On a longitudinal level, the results show that diet-related compensatory behavior was a significant predictor for change in physical activity and diet quality. With a higher tendency for diet-related compensatory behavior, physical activity and diet quality increased after two years. No effect was found for changes in BMI over time. Individuals from a non-clinical population showing diet-related compensatory behavior more frequently seem to have an improved diet quality and an increase in physical activity over time. Therefore, when applied in healthy doses, diet-related compensatory behavior may contribute to the maintenance of a balanced and healthy body weight, but it is not a successful strategy for weight loss over time.


Introduction
Following a healthy diet is known to contribute to the prevention of health risks and diseases such as overweight and obesity (Roberts et al., 2019), cardiovascular illnesses (Lassale, Gunter, Romaguera, & Peelen, 2016;Yu, Malik, & Hu, 2018), and diabetes (Jannasch, Kröger, & Schulze, 2017). The increase in food supply and abundance of choice are linked to the development of obesity on a worldwide level (Vandevijvere, Chow, Hall, Umali, & Swinburn, 2015). Among other high-income countries, Switzerland has had an excess in food energy supply per capita for the last few decades (Vandevijvere et al., 2015). The permanent availability of tempting high calorie food and frequent energy overconsumption potentially hinders the maintenance of a healthy body weight and, thus, further fuels the obesity epidemic (Vandevijvere et al., 2015). Even though these environmental factors play an important role in health behavior (Larson & Story, 2009;Popkin, Duffey, & Gordon-Larsen, 2005), the influence of psychological factors and coping strategies is of high importance as well. On the individual level, food can be consumed to deal with negative emotions and stressful situations (Torres & Nowson, 2007). Caloric overconsumption out of emotional distress is a dysfunctional coping mechanism to deal with these negative emotions and stress, often resulting in weight gain (Torres & Nowson, 2007). To maintain a healthy weight despite caloric overconsumption, the use of strategies to balance out unhealthy dietary behavior might be needed.
One way of coping with these overconsumption triggers is to foster diet-specific compensatory health beliefs (CHBs) (Knäuper, Rabiau, Cohen, & Patriciu, 2004). A CHB refers to a belief one might have about certain healthy behaviors to neutralize or compensate for unhealthy behaviors. For instance, one might hold the belief that skipping breakfast might make up for indulging in sweet dishes (Knäuper et al., 2004). The problem with this kind of coping mechanism is that a belief cannot be directly compared to actual behavior (Kaklamanou, Armitage, & Jones, 2013).
For instance, Kronick, Auerbach, Stich, and Knäuper (2011) looked at the contribution of CHBs to the calorie intake of dieters. They found that stronger CHBs led the participants to be more disinhibited when tempted by indulgence. If CHBs were not followed by the actual compensating behavior, participants were at risk of having a sustained increase in caloric intake (Kronick et al., 2011). The CHB model describes the interplay between emotionally driven states (e.g., craving or desire for unhealthy foods) and goal motivation (e.g., healthy eating, weight loss) (Rabiau, Knäuper, & Miquelon, 2006). If the motivational conflict between a desire and a health goal has activated CHBs, a compensatory behavior intention is formed. If the behavior is not or not fully implemented, the conflict remains and needs to be resolved through a change in risk perception concerning the unhealthy behavior, or the felt discomfort fades with the passage of time (Rabiau et al., 2006). Accordingly, a positive relation between body mass index (BMI) and CHB scores was found (Knäuper et al., 2004), because two problems may arise. First, a lack of commitment in following through with a specific diet-related compensatory behavior would lead to an increase in caloric intake and, in turn, to weight gain (Kronick et al., 2011;Rabiau et al., 2006). Second, weight gain could also occur when the applied compensatory behavior does not accurately counteract the previous excess in caloric intake (Knäuper et al., 2004).
Physical activity is another important factor in maintaining a healthy weight (Goldberg & King, 2007). The associations between CHBs and physical activity have mainly been examined by cross-sectional studies or studies examining changes over short periods of time (Berli, Loretini, Radtke, Hornung, & Scholz, 2014;Radtke & Rackow, 2014). Even though CHBs were positively related to the readiness to change behavior (e.g., increasing physical activity), this relationship was not found between CHBs and actual self-reported stair or elevator use (Radtke & Rackow, 2014). Similarly, in a study examining physical activity in adolescents, CHBs did not have an effect on self-reported physical activity two weeks after the first time point of data collection (Berli et al., 2014). Thus, current studies in physical activity could not find an influence of CHBs on actual behavior.
Thus, compensatory beliefs seem to be applied to license oneself to indulge in an unhealthy behavior (Knäuper et al., 2004;Kronick et al., 2011) rather than having a positive effect on dietary health behavior. For instance, rather than increasing the frequency of physical activity, CHBs might be used as a justification to not change actual eating behavior. The negative effects mostly occur when the compensatory belief is not converted into actual behavior and when the applied behavior does not accurately compensate for the unhealthy behavior (Knäuper et al., 2004). Indeed, the problem with some of the items used in CHB scales is the lack of compensatory accuracy. For instance, even when converted into actual behavior, a healthy diet would not accurately compensate for excessive smoking (Radtke, Scholz, Keller, Knäuper, & Hornung, 2011).
Thus, CHBs do not necessarily reflect actual compensatory behavior (Dohle & Hofmann, 2019;Kaklamanou et al., 2013;Radtke & Scholz, 2012. One might have a certain belief about something without carrying through with the respective behavior (Kaklamanou et al., 2013). But of course, it is also possible to show or execute a specific behavior without having a compensation belief attached to it. Therefore, several studies brought up the necessity of examining behavior-specific compensatory tendencies (Kaklamanou et al., 2013;Radtke & Scholz, 2012;Radtke, Scholz, Keller, Perren, & Hornung, 2013). Compensatory health behavior was first analyzed regarding alcohol consumption (Radtke & Scholz, 2016) where such a behavior was significantly positively predictive of alcohol consumption. Even though participants had the intention to engage in a healthier behavior pattern (e.g., consume less alcohol), compensatory health behaviors might still serve as a justification to engage in an unhealthy behavior (Radtke & Scholz, 2016). As with CHBs, the same accuracy problem occurs with items measuring compensatory behavior. Having healthy eating habits would not accurately compensate for the negative effects of excessive alcohol consumption on health (Radtke & Scholz, 2016).
Due to this inaccuracy and the potential lack of carrying through with a compensatory behavior, the actual effects of CHBs on weight management are inconclusive. Therefore, it is important to examine the frequency of actual compensatory behavior to make statements about the real consequences of such compensation techniques. Specifically, the focus of the present study is on diet-related compensatory behavior presenting realistic compensation techniques (e.g., dietary restriction, food choice change, physical activity) within the domain of weight management and examines its implications in dietary health behavior change and BMI change.
To the best of our knowledge, diet-related compensatory behavior has not yet been examined in detail within a large sample, and longitudinal studies are needed to make observations about changes in health behavior over time (Radtke & Scholz, 2016) within a broader range of ages, including the aging population (Radtke et al., 2013).

Aim of the present study
After the adapted items measuring diet-related compensatory behavior were tested for their psychometric properties, the aim of the present study was to test whether diet-related compensatory behavior was related to actual dietary health behavior change and change in BMI. To this end, it was analyzed whether diet-related compensatory behavior would be a predictor of change in diet quality, physical activity levels, and BMI over time. Thus, the present findings might inform about the link between diet-related compensatory behaviors and healthy weight management.

Participants
The present study used data from the first and third waves of the Swiss Food Panel 2.0, which assesses the eating behaviors and underlying psychological mechanisms of the Swiss population in a longitudinal design. The first data collection wave took place in spring 2017. The participants were randomly selected residents of the Germanspeaking and French-speaking parts of Switzerland. Participants were all 20 years of age or older. Most of the addresses were retrieved from the telephone book. To reach enough participants between the ages of 20 and 39 years, additional addresses were purchased from an address company. For the first wave, 5781 people returned the questionnaire. Thus, the response rate was 25.1%. Participants who did not indicate their sex or age and those who filled out less than 50% of the questionnaire were excluded from the analysis. The baseline sample of the year 2017 finally consisted of 5586 participants. For the second and third waves, the same exclusion criteria applied. Since the analyses included the BMI as a variable, pregnant women were excluded at the baseline wave (T1) and in the third wave (T2) (2017, n = 348; pregnancy in 2017 and/or 2019, n = 175). The final baseline sample in the year 2017 consisted of 5238 participants. The longitudinal sample of 2017-2019, which only included participants who voluntarily remained in the sample over three waves, included 2638 participants. Sociodemographic details of the samples used in this study can be found in Table 1. Due to the drop-out of participants over time, the sample characteristics slightly changed. The proportion of male participants decreased and the proportion of participants with higher education increased, while average BMI did not change. For cross-sectional analyses, the baseline sample from 2017 was used to include the largest number of participants. Medium-term analyses were conducted with the longitudinal sample of 2017-2019.

Measures
Diet-related compensatory behavior. Diet-related compensatory behavior was measured with six items. The items are based on the dietrelated compensatory health beliefs scale by Poelman, Vermeer, Vyth, and Steenhuis (2013). The items were adapted, however, to address compensation behavior instead of beliefs. The focus was on compensation behaviors for having eaten too much in terms of calories or amount. They reflect typical strategies for weight management, such as the monitoring of calories and/or portion sizes, eating frequency adaptation (i.e., breakfast skipping), and physical activity (Keller & Siegrist, 2015c). The items were translated into German and French. Participants had to indicate how often they perform certain behaviors (e.g., "When I eat too much, I will go to the gym/do sports later on") on a response scale from 1 (never) to 7 (always). All items were combined by calculating a mean value. Therefore, a higher score corresponded to a higher frequency of engaging in compensatory behavior following overeating. A list of all the items included in the diet-related compensatory behavior scale can be found in Table 2.
Weight status. BMI was calculated by dividing the self-reported body weight (in kilograms) by the square of the self-reported height (in meters).
Diet quality. For the Diet Quality Index, participants filled out a food frequency questionnaire adapted from the Nurses' Health Study questionnaire (Hu et al., 2016), where they had to indicate their usual consumption frequencies of a number of different types of food and beverages during the past year. Response options for the various foods ranged from more than 4 times per day to 1-3 times per month and less or never. For further details, see previous publications (Hagmann, Siegrist, & Hartmann, 2019). In line with the Swiss dietary recommendations (Swiss Society for Nutrition, 2019), a diet quality index was calculated based on the participants' self-reported answers from the food frequency questionnaire, recoded into weekly consumption frequencies. The following food groups were included: (1) fruit (excluding fruit juice), (2) vegetables and salads (raw and cooked), (3) meat and meat products, (4) wholegrain products, and (5) sweets, savory snacks, sugar-sweetened beverages, and alcohol. Previous studies showed that these food groups either have a positive or negative effect on health (Malik, Willett, & Hu, 2013). One point was assigned for food groups with a positive impact on health (fruit, vegetables and salad, wholegrain products) if the consumption was equal to or higher than the recommended minimum, and 0 points were assigned if the consumption was lower than the recommended minimum. If consumption of all other food groups (e.g., meat) was equal to or below the recommended maximum, one point was assigned. Zero points were assigned if the consumption exceeded the recommended maximum. A summary score was created ranging from 0 (rather unhealthy diet) to 5 (rather healthy diet). Further details can be found in previous publications (Hagmann et al., 2019;Hagmann, Siegrist, & Hartmann, 2020). Physical activity. Physical activity level (PAL) was assessed with a German and French translation of the physical activity questionnaire (Johansson & Westerterp, 2008) measuring physical activity levels at work with a response scale from 1 (very light, e.g., sitting at the computer most of the day or sitting at a desk) to 4 (heavy, e.g., heavy industrial work, construction work, or farming), and during leisure time with a scale from 1 (very light: almost no activity at all) to 4 (very active: strenuous activities several times a week). Based on the indicated activity intensities in these two domains, a PAL value was determined from a scheme that was developed and validated by Johansson and Westerterp (2008). Achievable PAL values reached from 1.4 (very light PAL at work and very light PAL during leisure time) to 2.3 (heavy PAL at work and very active during leisure time).
Restrained eating. Restrained eating was assessed using the diet interest scale of the German translation (Dinkel, Berth, Exner, Rief, & Balck, 2005) of the Restraint Scale (Herman & Polivy, 1980), which was also translated into French. The scale consisted of six items and focused on the participants' diet interest (e.g., "How often do you go on a diet?"). Four items were assessed on a scale ranging from 1 (never) to 5 (all the time). The last two items (e.g., "Would a weight change of 2 kg affect your lifestyle?") were assessed on a scale from 1 (not at all) to 4 (very strongly). A higher mean score indicated a higher tendency for restrained eating. Internal consistency was acceptable (α = 0.65). A more detailed description can be found in previous publications (Keller & Siegrist, 2015a;2015b).
Health consciousness. Participants' diet-related health consciousness was assessed using a German and French version of an adapted scale (Schifferstein & Oude Ophuis, 1998) published in 2014 (Dohle, Hartmann, & Keller, 2014). The scale consisted of four items (e.g., "My health depends on what and how I eat"), with a 7-point response scale ranging from 1 (not at all) to 7 (very much). The mean value of the four items was calculated. A high score indicated high health consciousness.
Sociodemographic data. Sociodemographic data such as age, sex, and educational level were assessed.

Statistical analyses
To test whether the adapted versions of the items measuring dietrelated compensatory behavior would influence the psychometric properties of the construct, a principal component analysis (PCA), Cronbach's alpha, and test-retest reliability using Pearson's r were conducted. Cross-sectional correlation analyses were conducted with diet-related compensatory behavior at baseline and restrained eating and health consciousness to explore the strength of association between diet-related compensatory behavior and the constructs of restrained eating and diet-related health consciousness. Furthermore, to explore long-term relationships between diet-related compensatory behavior and diet quality, physical activity as well as BMI, separate hierarchical multiple linear regression analyses were conducted. This method was chosen to highlight the added value of diet-related compensatory behavior at T1 as a predictor for change in the outcome variables over Note. Educational level was divided into three categories: low-no education, primary and lower secondary school; middle-vocational school; high-higher secondary school, college and university. time. Because the respective outcome variables at T2 are likely to be dependent on their initial values at T1, the respective outcome variables at T1 were introduced as predictors into the regression models (Twisk, 2013). This removes the potential influence of these values at T1, and the effect of the other predictor variables can be interpreted as independent from them (Cohen, Cohen, West, & Aiken, 2003). Age and sex were introduced into the regression models as control variables. Because of the large sample size, missing data were handled within each analysis separately. The statistical analyses were conducted using the statistical software IBM SPSS statistics version 25.

Examination of the diet-related compensatory behavior items
A PCA was performed using the six diet-related compensatory behavior items. The sampling adequacy of the analysis was verified with the Kaiser-Meyer-Olkin (KMO) measure of 0.84, and Bartlett's test of sphericity with (χ 2 (15) = 11267.97, p < .001) indicated that the correlations between the items were large enough to conduct a PCA. One component had an eigenvalue over Kaiser's criterion of 1 and explained 53.1% of the variance. Therefore, the PCA confirmed a one-factor structure. Additionally, all the items showed satisfactory corrected item-total correlations (Pearson's r > 0.2) (Kline, 1986) ranging from 0.26 to 0.73 (Table 2).
Reliability analysis was good with Cronbach's alpha of .81. Additionally, the test-retest reliability of the diet-related compensatory behavior scale from T1 (wave of 2017) to T2 (wave of 2019) was acceptable with Pearson's r = 0.68. Thus, all the items on the scale were found to measure the construct of diet-related compensatory behavior without subscales or items to discard. Details of the analyses, including all the diet-related compensatory behavior items, can be found in Table 2.
Furthermore, a small positive correlation was found between dietrelated compensatory behavior and diet quality (men: r = 0.17, women: r = 0.15, p < .001 respectively). For physical activity, a small significant positive correlation was found for women, but not for men (men: r = < 0.01, n.s.; women: r = 0.07, p < .001). Thus, diet-related compensatory behavior was significantly correlated with variables related to weight management (Table 3 and Table 4).

Changes over time in diet quality, physical activity, and BMI
Separate stepwise multiple linear regression analyses were conducted to analyze the predictive value of diet-related compensatory behavior on diet quality, physical activity, and BMI over the course of two years from T1 to T2 (Table 5). For all three regression models, the respective data from T1, age, and sex were introduced as a first step into the models. In a second step, diet-related compensatory behavior was added. The results show that diet-related compensatory behavior at T1 had a significant predictive effect on diet quality (F change (1,2235) = 8.32, p < .01) and on physical activity levels at T2 (F change (1,2366) = 7.62, p < .01) in the second step of the models. For participants with higher diet-related compensatory behavior scores, diet quality seemed to increase between 2017 and 2019. Even though the cross-sectional analysis between diet-related compensatory behavior and physical activity showed a significant positive correlation for women only (men: r = < 0.01, n.s.; women: r = 0.07, p < .001), an overall increase over time in physical activity predicted by diet-related compensatory behavior was observed (β = 0.04, p < .01), independent of sex. At a crosssectional level, diet-related compensatory behavior and BMI correlated positively for men but not for women (Table 3). Nevertheless, there was no such evidence found for changes in BMI from T1 to T2 predicted by diet-related compensatory behavior introduced as a second step (F change (1,2491) = 1.27, n.s.). Therefore, diet-related compensatory behavior was not a significant predictor for changes in BMI over two years.

Discussion
The present study emphasizes the importance of examining dietrelated compensatory behavior, first by analyzing the psychometric properties of the adapted self-reported diet-related compensatory behavior items within an adult sample, and second by showing the relationship between the scale and health behavior change over time. The diet-related compensatory behavior scale assesses the extent to which compensatory behaviors are executed in response to overeating or the intake of an unbalanced or calorie-rich meal. Internal consistency and test-retest reliability were good. Thus, the scale can be considered to reliably measure the construct of diet-related compensatory behavior and remains rather stable over time.
To tackle the compensation inaccuracy found in other scales (Knäuper et al., 2004;Radtke et al., 2011;Radtke & Scholz, 2016), diet-related compensatory behavior items were specifically reformulated and adapted to address more realistic compensation within the weight management domain (e.g., dietary restriction, food choice change, physical activity). Thus, as an example, overeating one day would be compensated for by exercising the next day. With the focus on weight management, the realistic aspect of compensation of the diet-related compensatory behavior items has been maximized. Therefore, the diet-related compensatory behavior items offer a direct Table 3 Descriptive statistics of the demographic and behavioral variables and dietrelated compensatory behavior at T1.
consequential link between an unhealthy behavior and its planned compensation. Diet-related compensatory behavior appears to be related to diet interest and health consciousness, which seems logical because the items were formulated to investigate weight management and dietary balance strategies.
Furthermore, it was observed that the diet-related compensatory behavior scale positively predicted dietary health behavior change over time. The results showed evidence that diet-related compensatory behavior acted as a predictor for change in diet quality and physical activity levels after two years. Thus, engaging in diet-related compensatory behavior more regularly led to an improvement in diet quality and an increase in physical activity from T1 to T2. An improvement in diet quality could be explained by an increase in fruit and vegetable consumption over time to make up for moments of indulgence or caloric overconsumption. Similarly, one could speculate that physical activity could serve as a compensatory approach to counteract potential weight gain over time. Ideally, through the application of diet-related compensatory behavior, physical activity might become a more inherent and regular part of life. Thus, physical activity can potentially be associated with positive effects on the individual's general health status (Gallaway & Hongu, 2015). Both an improved diet quality and an increase in physical activity through diet-related compensatory behavior might therefore contribute to a healthier lifestyle over time.
In line with previous studies on CHBs and BMI (Knäuper et al., 2004;Kronick et al., 2011), cross-sectional analysis showed a positive relationship between diet-related compensatory behavior scores and BMI. However, there was no evidence for changes in BMI predicted by diet-related compensatory behavior scores after two years. A possible explanation could be that individuals with higher diet-related compensatory behavior scores are inclined to have higher BMIs in general, but their weight does not change significantly when assessed after two years because they might be using diet-related compensatory behaviors to maintain or regulate their weight. Thus, individuals with higher BMIs seem to be more likely to engage in diet-related compensatory behavior, which might help in maintaining weight but does not help with weight loss. Whether the participants had the goal of maintaining their weight or whether they had other dieting intentions was not assessed in the present study. Thus, no statement can be made as to how successful they were in their weight management.
Nevertheless, the findings concerning BMI and diet-related compensatory behavior stress the importance of results from longitudinal in comparison to cross-sectional study designs. An initially positive relationship between two variables on a cross-sectional level is not necessarily maintained over time. Thus, longitudinal data allow for more detailed statements about the development of relationships between variables. Diet-related compensatory behavior seems to have a positive effect on one's diet and possibly health status. This might be due to the promotion of health behavior with improved diet quality and an increase in physical activity levels over time. Diet-related compensatory behavior as a planned oscillation between the selective indulgence in unhealthy eating patterns followed by a form of restraint behavior seems to have better health outcomes without having an influence on BMI changes over the course of two years.
Nevertheless, the findings of the present study need to be interpreted with caution, because there are limitations that need to be addressed. First, diet-related compensatory behavior explained less than 1% of the variances in diet quality and physical activity change over time, which represents a small effect size. Yet, because small effects can accumulate over time and within an individual, these results still have an important stance within the domain of nutrition psychology (Funder;Ozer, 2019;Gignac & Szodorai, 2016). Additionally, in comparison to experimental research with homogenous samples, the effects of population-based survey research can be weaker (Taris, 2000). Furthermore, similar effect sizes were observed in studies about food behavior using comparable longitudinal analysis methods (Hartmann, Dohle, & Siegrist, 2015;van Strien, Herman, & Verheijden, 2014).
Second, participants had to estimate their weekly food intake over one year, which does not account for short-term fluctuations. However, for the purposes of the present study, changes in dietary intake after two years were the most important factors, which makes momentary oscillations irrelevant.
Moreover, self-reported food frequency questions might be distorted by conscious or unconscious over-or underreporting of true consumption (Voss, Kroke, Klipstein-Grobusch, & Boeing, 1997). BMI was calculated from self-reported weight and height. In overweight and obese individuals, underestimation or underreporting of weight is likely to have occurred (Nyholm et al., 2007). Nevertheless, a recent study (Luo et al., 2019) showed that the correlation between self-reported and objectively measured weight is relatively high. Thus, the use of self-reported data, such as BMI, is legitimated, especially because other Table 5 Separate multiple linear regression analyses with the respective outcome variables, age, sex, and diet-related compensatory behavior score at T1 as predictors, and diet quality, BMI, and physical activity at T2 as respective outcome variables. Only the results for the second step (full model) of the analyses are presented. Note. *p ≤ .05, **p ≤ .01, ***p ≤ .001. a Sex: coded 0 = men, 1 = women. b ΔR 2 = .002 for Regression 1 and Regression 2 due to inclusion of diet-related compensatory behavior scores into the models as a second step. assessment methods (e.g., measurements taken in laboratory settings) would barely be feasible and would be very costly within a large sample size. Similar concerns can be addressed regarding self-reported data on diet-related compensatory behavior and physical activity. There is the possibility, as is the case with self-reported data in general, that participants did not report their diet-related compensation or physical activity levels as they really are in their everyday lives. Lastly, even with longitudinal data at hand, it is not possible to confirm a causal relationship between diet-related compensatory behavior, BMI, physical activity, and diet quality with certainty.
In the present study, it was assessed whether people use compensatory behavior following overconsumption and how it is linked to changes in weight-related health behavior over the course of two years. Longitudinal analyses over longer periods of time should be conducted to accumulate long-term data about the associations between dietrelated compensatory behavior, diet quality, physical activity, and BMI.

Conclusion and implications
People from a non-clinical population showing diet-related compensatory behavior more frequently appear to have an improved diet quality and an increase in physical activity after two years. Dietrelated compensatory behaviors might be one way of maximizing dietary pleasure while balancing out the indulgence with more healthoriented behavior in return. With a certain degree of health awareness in mind, a healthy dose of diet-related compensatory behavior could possibly help to contribute to a balanced lifestyle while dealing with the constant exposure to food abundance.

Ethics approval and consent to participate
The study was approved by the Ethics Committee of ETH Zurich (EK 2017-N-19).

Declaration of competing interest
The authors declare that they have no competing interests.