Abstract
At the end of a diet, even a successful one, people often return to overeating. One potential reason is that the behavioral inhibition that people learn while dieting might not readily transfer outside the context in which it is learned: Basic research indicates that after a behavior is inhibited, a return to the conditioning context or simple removal from the treatment context can cause the behavior to return (i.e., to renew). Can states of hunger and satiety play the role of context? In two experiments, rats learned a food-seeking response that earned sucrose or sweet, fatty food pellets while they were satiated. Responding was then inhibited (i.e., extinguished) while the rats were hungry. On the rats’ return to the satiated state, their food seeking was renewed. Additional results suggest that associations with hunger or satiety stimuli were learned more readily than associations with other potentially useful exteroceptive stimuli. The findings have implications for understanding the role of interoceptive contexts in controlling the inhibition of motivated behavior.
Obesity-related death and disease have become a worldwide public-health concern. The prevalence of obesity has doubled since 1980 and now accounts for more preventable deaths than malnutrition (World Health Organization, 2016). Unfortunately, maintaining the weight loss achieved through dieting is often difficult (Elfhag & Rössner, 2005). One possible contributor to the problem may be that the behavioral inhibition that people learn while dieting may be expressed primarily in the context in which it is learned (e.g., Bouton, 2002, 2014). Thus, an individual on a diet may learn to inhibit overeating in the context of hunger. If the inhibition learned is context-specific, then it would be lost to some extent if the individual encountered fullness cues. However, when overeating is prevalent, fullness cues themselves may also become a context for eating. Thus, dieting overeaters may learn, paradoxically, to eat when full and inhibit eating when hungry, leading to an unproductive cycle of inhibition and overeating.
Research on basic behavioral processes suggests that behavioral inhibition in the form of extinction is highly specific to the context in which it is learned (e.g., Bouton, 2002, 2004; Bouton & Todd, 2014). For example, renewal is a type of behavioral relapse that occurs when the context is changed after a behavior has been suppressed or inhibited (e.g., Bouton, Todd, Vurbic, & Winterbauer, 2011; Nakajima, Tanaka, Urushihara, & Imada, 2000). In the instrumental-learning laboratory, renewal experiments often involve reinforcing an instrumental behavior in one context (Context A) and then extinguishing it in a second one (Context B). Behavior that is inhibited through extinction typically returns to performance (i.e., renews) when the response is then tested again in the original context (ABA renewal) or in a new context (ABC renewal). Moreover, testing in a new context also promotes renewal when behavior was both acquired and extinguished in the same context (AAB renewal). Overall, the findings suggest that behavioral inhibition is relatively specific to the context in which it is learned (Bouton & Todd, 2014).
Renewal experiments typically use contexts that differ in terms of exteroceptive stimuli (e.g., visual, tactile, and olfactory cues). However, many different types of stimuli, including, but not limited to, drug states, mood states, and time, may play the role of context (e.g., Bouton, 2002, 2010). There is also evidence that the interoceptive cues provided by hunger and satiety may play the role of context under some conditions (e.g., Davidson, 1993). In one study, Davidson (1987) found that rats could learn to use daily alternations in deprivation stimuli (i.e., no deprivation vs. 24-hr food deprivation) to anticipate whether a foot shock would occur at the end of an experimental session. More recently, researchers have found that animals can also learn to associate alternating deprivation conditions with the delivery of free food reinforcers (Sample, Jones, Hargrave, Jarrard, & Davidson, 2016). However, it is not yet clear that deprivation stimuli can play the role of context in renewal designs that involve relatively few shifts in the deprivation conditions and relatively extensive extinction treatments.
Experiment 1
Our first experiment was therefore designed to determine whether interoceptive deprivation states can function as contexts in an ABA renewal design applied to instrumental food seeking. While satiated, rats learned to press a lever to obtain highly palatable sweet or sweet, fatty pellets over a series of daily sessions. Then, after being made hungry by 23 hr of food deprivation on each of several days, the rats had the opportunity to learn that pressing the lever no longer yielded food pellets. In final tests, the rats’ lever pressing was tested when they were in the hungry and satiated states (in a counterbalanced order). Our hypothesis was that behavior would be renewed when responding was tested in the satiety state after extinction in the hungry state. Although such a recovery of food seeking during satiation (as opposed to hunger) would violate intuition as well as traditional ideas about how hunger motivates instrumental behavior (e.g., Hull, 1943), it would be consistent with the idea that the inhibition of food seeking is specific to the deprivation context in which it is learned.
Method
Subjects
The subjects were 32 naive female Wistar rats, purchased from Charles River Laboratories (St. Constance, Quebec, Canada), that were between 75 and 90 days old at the start of the experiment. The rats were individually housed in a room maintained on a cycle of 16 hr of light and 8 hr of dark. The experiment was run each day during the light period of the cycle. A power analysis informed by data obtained from Bouton et al. (2011) determined that our sample size would be sufficient to provide power of .80 to detect a small to medium-sized renewal effect.
Apparatus
Two sets of four conditioning chambers, or boxes, housed in separate rooms of the laboratory were used. Each box was in its own sound-attenuation chamber. All boxes were of the same design (Model ENV-008-VP; Med Associates, St. Albans, VT). The side walls and ceilings were made of clear acrylic plastic, whereas the front and rear walls were made of brushed aluminum. A recessed 5.1- × 5.1-cm food cup was centered in the front wall approximately 2.5 cm above the level of the grid floor. Retractable levers (Model ENV-112CM; Med Associates) were positioned to the left and right of the food cup; in the present experiments, the left lever was used. Each box was illuminated by one 7.5-W incandescent bulb that was mounted to the ceiling of the sound-attenuation chamber (approximately 35 cm above the grid floor) at the front wall of the box. Ventilation fans provided background noise of 65 dB (A-weighted). Food rewards consisted of 45-mg sweet, high-fat pellets (38% kcal from fat; 45-mg LabTreat OmniTreat enrichment tablet 5TCY; TestDiet, Richmond, IN) or 45-mg sucrose pellets (sucrose reward tablet 5TUT; TestDiet). The apparatus was controlled by computer equipment located in an adjacent room.
Procedure
Feeding schedules
Before each experimental session, the rats received one of two feeding schedules in the home cage. In the deprived condition, the rats received 1 hr of access to chow that ended 23 hr before the beginning of the session. In the sated condition, rats received 23 hr of continuous ad libitum access to chow before the session.
Magazine training
During each of the first 2 days, each rat received two 30-min sessions of training with the food magazine, one session in the sated condition and the other in the deprived condition. The rats were placed in an operant chamber, and after a 2-min delay, received free pellets on a 30-s random time schedule (i.e., a reinforcer was delivered every 30 s on average). This resulted in the delivery of approximately 60 pellets. Half the rats received sucrose pellets throughout the experiment, and the other half received sweet, fatty pellets throughout the experiment.
Acquisition
Every day for the next 12 days, the rats were put through one 30-min session in which they were reinforced for pressing the lever while they were satiated. Each session began when the lever was inserted, which occurred 2 min after the rat was placed into the chamber. (The lever then remained in the chamber for the remainder of the session.) Reinforcers for lever presses were available on a variable-interval (VI) schedule. On Days 1 and 2, we used a 10-s VI schedule (i.e., a pellet became available every 10 s, on average, for the next response). On Days 3 and 4, we used a 20-s VI schedule. Finally, on Days 5 through 12, we used a 30-s VI schedule. The rats learned to press the lever without any further shaping. However, given the inclusion of a satiation condition (i.e., the absence of hunger motivation during training or, indeed, very little experience with hunger), not all the rats learned to press the lever reliably. Ten rats were excluded after the fourth training session (3 that had been receiving the sweet, fatty pellets and 7 that had been receiving sucrose pellets) because they had failed to average a single response per minute (M = 0.22). The remaining 22 rats averaged 5.45 responses per minute in that session.
Extinction
Extinction began when rats were in the deprived condition. Extinction sessions were the same as acquisition sessions except that food pellets were no longer available after the lever was pressed (i.e., responses had no programmed consequences). There were four extinction sessions, one per day, and each session lasted 30 min.
Renewal test
Over the 2 days that followed the last session of extinction, each rat received two extinction tests of lever pressing, one while it was sated and another while it was deprived (counterbalanced order). As usual, the lever was inserted after a 2-min delay; lever presses were then recorded for the next 10 min. Pellets were not delivered.
Results
The results are summarized in Figure 1. The rats learned to press the lever at similar rates whether the reinforcer was sucrose pellets or sweet, fatty pellets (Fig. 1a). A Pellet Type × Session analysis of variance (ANOVA) found a significant main effect of session on the number of lever presses, F(11, 220) = 29.72, MSE = 3.37, p < .001, η p 2 = .60, but no effect of pellet type or interaction, Fs ≤ 1.09. Responding also declined at similar rates for the two pellet-type groups during extinction (Fig. 1b). A Pellet Type × Session ANOVA found a significant main effect of session on the number of lever presses, F(3, 60) = 73.13, MSE = 0.77, p < .001, η p 2 = .79, but no effect of pellet type or interaction, Fs < 1.
Results from Experiment 1: mean number of lever responses per minute (a) during each 30-min acquisition session, when the rats were sated; (b) during each extinction session, when the rats were food deprived; and (c) during each 10-min renewal test session (deprived and sated conditions). Results are shown separately for rats reinforced with sweet, fatty pellets and those reinforced with sucrose pellets. Error bars represent ±1 SEM.
In the test sessions, the clear finding was that the rats pressed the lever more when they were sated than when they were deprived (Fig. 1c). Further, the increase in responding when sated did not depend on the type of reinforcer that had been used during training. A Pellet Type × Session ANOVA found a main effect of session on the number of lever presses, F(1, 20) = 8.89, MSE = 0.26, p = .007, η p 2 = .31, confirming that rats made more responses when they were sated, mean difference = 1.20, 95% confidence interval (CI) = [0.36, 2.03]. However, there was neither a significant main effect of pellet type nor a significant interaction between session and pellet type, Fs < 1.
Discussion
Instrumental food seeking that was acquired while the rats were sated (i.e., in Context A) and then extinguished while the rats were deprived (i.e., in Context B) renewed when the rats were sated again. This result suggests that satiety and hunger can play the role of Contexts A and B in a simple ABA renewal design. Furthermore, the type of reinforcer had little effect on the rate of acquisition, extinction, or renewal of lever responding. For simplicity, the remaining experiments therefore used only sweet, fatty pellets.
Experiment 2
The goal of Experiment 2 was to replicate the ABA renewal effect and to determine whether a simple shift in deprivation state after extinction could produce another form of renewal that has been observed in standard exteroceptive contexts: AAB renewal. In this form of renewal, when behavior is learned in one context (Context A) and then extinguished in that context (Context A), it can renew when the response is tested in a new context (Context B). In Experiment 2, we thus explored whether renewal also occurred when food seeking was trained and extinguished in the context of deprivation (Context A) and then tested in the context of satiation (Context B). Such a counterintuitive increase in responding after a shift from hunger to satiation would be worth knowing about.
Method
Subjects and apparatus
The subjects were 32 naive female Wistar rats of the same age and from the same stock as in Experiment 1. The apparatus was also the same.
Procedure
All the rats received the sweet, fatty pellets as reinforcers. In addition, all the rats received only 1 hr of daily access to chow for the 7 days before the beginning of the experiment. Magazine training then occurred in each deprivation state according to the procedure used in Experiment 1. The rats were then randomly assigned to one of two groups that differed according to their deprivation states during 12 daily acquisition sessions and 4 daily extinction sessions. Like the groups in Experiment 1, the rats in the first group were sated (S) before the acquisition sessions and deprived (D) before extinction sessions (the SDS group). In contrast, the rats in the second group were deprived before each acquisition session and before each extinction session (the DDS group). As in Experiment 1, all the rats then received one session in which lever pressing was tested while they were sated and another session in which lever pressing was tested while they were deprived (counterbalanced order). The acquisition, extinction, and testing sessions otherwise followed the procedures described for Experiment 1. Perhaps because the rats received extended experience with food restriction before the beginning of training, all of them learned to press the lever while satiated in this experiment.
Results
During acquisition (Fig. 2a), the deprived rats (DDS group) made more responses than the sated rats (SDS group). This was confirmed by a Group × Session ANOVA that found significant main effects of session, F(11, 330) = 33.46, MSE = 14.95, p < .001, η p 2 = .53, and group, F(1, 30) = 14.06, MSE = 460.15, p = .001, η p 2 = .32, on the number of lever presses, along with a significant Group × Session interaction, F(11, 330) = 77.78, MSE = 14.95, p < .001, η p 2 = .72.
Results from Experiment 2: mean number of lever responses per minute (a) during each 30-min acquisition session, when rats in the sated-deprived-sated (SDS) group were sated and rats in the deprived-deprived-sated (DDS) group were deprived; (b) during extinction sessions, when both groups were food deprived; and (c) during each 10-min renewal test session (deprived and sated conditions). Error bars represent ±1 SEM.
In the four extinction sessions (Fig. 2b), the group difference in response rate established during acquisition persisted, at least initially. A Group × Session ANOVA revealed a significant effect of session on the number of lever presses, F(3, 90) = 81.79, MSE = 6.30, p < .001, η p 2 = .73, as well as a significant interaction of group and session, F(3, 90) = 62.64, MSE = 6.30, p < .001, η p 2 = .68. The main effect of group on the number of lever presses was not significant, F(1, 30) = 2.00, MSE = 33.09, p = .168. Follow-up comparisons indicated that during the first extinction session, the DDS group made more responses than the SDS group, F(1, 30) = 9.60, MSE = 26.31, p = .004, η p 2 = .24. In the renewal test sessions, there was a clear renewal effect in the SDS group, as in Experiment 1, but not in the DDS group (Fig. 2c). A Group × Session ANOVA found a significant interaction, F(1, 30) = 7.40, MSE = 1.32, p = .011, η p 2 = .20. The main effect of group on the number of lever presses approached significance, F(1, 30) = 3.82, MSE = 8.74, p = .060, η p 2 = .11, whereas the effect of session did not, F(1, 30) = 1.76, MSE = 1.32, p = .194. Follow-up comparisons confirmed that the SDS group responded more when tested in the sated condition than when tested in the deprived condition, t(30) = 2.86, MSE = .406, p = .008, mean difference = 1.16, 95% CI = [0.33, 1.99], η2 = .22. There was no such effect in the DDS group (p = .333).
Discussion
The SDS group responded more in a test session in which they were sated than in an identical session in which they were deprived, successfully replicating the ABA renewal effect of Experiment 1. An AAB renewal effect was not observed in the DDS group. It is worth noting that although AAB renewal can be a robust effect when using exteroceptive contexts, it is often smaller in magnitude and may therefore be more difficult to detect than ABA renewal (e.g., Bouton et al., 2011). Moreover, in the present design, in which food deprivation played the role of Context A, deprivation demonstrably increased the amount of responding that occurred during extinction in the AAB condition (i.e., in the DDS group); an increased amount of responding during extinction may produce a more durable extinction effect (e.g., Rescorla, 1997; see also Bouton, Trask, & Carranza-Jasso, 2016). We further discuss the lack of DDS renewal in the General Discussion.
The ABA renewal effects in Experiments 1 and 2 provide strong evidence consistent with the idea that inhibitory extinction learning is especially dependent on the context in which it is learned. Moreover, the fact that satiation produced an increase rather than a decrease in food-seeking behavior continues to suggest that this is an especially interesting and robust effect.
Experiment 3
Experiment 3 was designed to test whether the deprivation conditions that produced ABA renewal in the first two experiments worked as truly interoceptive contexts. The feeding schedules used in the sated and deprived conditions of those experiments confounded the interoceptive states of satiety and hunger with whether food was present or absent in the home cage before the experimental session. That is, in the ABA condition (i.e., in the SDS group), chow was available in the home cage before sessions in which lever pressing was reinforced, but not when lever pressing was extinguished. Therefore, the rats could have used the presence or absence of food cues in the home cage to predict whether pressing the lever would be reinforced or not. Experiment 3 was designed to differentiate contextual control of instrumental food seeking by true interoceptive deprivation cues from control by exteroceptive food cues.
Method
Subjects and apparatus
The subjects were 24 naive female Wistar rats of the same age and from the same vendor as before. The apparatus was also the same. A power analysis using the renewal data from Experiments 1 and 2, which was the best approximation of the potential discrimination that would be observed in each group, indicated that this sample size would provide 0.8 power to detect a small to medium-sized effect.
Procedure
As in Experiment 2, the rats received daily 1-hr access to chow for the 7 days before the beginning of the experiment. All the rats then received 40 daily sessions in the experimental chambers; the sessions alternated between ones in which lever pressing was reinforced (R+ sessions) and those in which it was not reinforced (R− sessions). The rats were randomly assigned to one of three groups (n = 8 per group). The groups were fed on different schedules that were designed to provide potential deprivation cues or home-cage food cues (or both) that could signal the reinforcement conditions (Fig. 3).
Design of Experiment 3: illustration of food availability in the home cage before R+ (reinforced) and R− (nonreinforced) training sessions.
The food-cue group always received ad libitum access to chow in the home cage, except during the 2 hr before R− sessions. In this way, the presence of chow immediately before a session signaled an R+ session, and the absence of chow signaled an R− session. For the deprivation-cue group, chow was removed from the home cage 23 hr before each R− session and 2 hrs before each R+ session. For this group, food was never present in the home cage immediately before an R+ or R− session, but differential interoceptive cues of satiety or hunger could still signal an R+ or R− session, respectively. Finally, the rats in the both-cues group were fed on a schedule similar to that of the deprivation-cue group, except that their chow was not removed before R+ sessions. In this way, rats in the both-cues group could potentially treat interoceptive deprivation cues or the presence or absence of food in the home cage (or both) as a signal of the upcoming R+ or R− session.
As in the prior experiments, on each of the first 2 days of training, the rats received 30-min sessions of magazine training. Each group received one magazine training session after each of their previously described feeding schedules. Over the next 40 days, there was one daily 30-min session in the operant chamber. As noted earlier, these alternated between R+ and R− sessions. The schedule of reinforcement gradually increased over the first several R+ sessions; in the first two sessions, lever pressing was reinforced on a 10-s VI schedule; in the second two sessions, lever pressing was reinforced on a 20-s VI schedule; and for the remainder of the experiment, lever pressing was reinforced on a 30-s VI schedule. The reinforcer was the sweet, fatty pellet.
Results
The rats in the deprivation-cue and both-cues groups gradually learned to anticipate whether lever pressing would be reinforced or not. This was confirmed by analyzing the latency between insertion of the lever into the chamber and the first response on it in each session. Note that the first response in each session occurred before the delivery (or nondelivery) of the reinforcer, and thus the latency of this response could differ only if the rat had learned to anticipate R+ and R−. (Analysis of response rates throughout the sessions would have been confounded by that factor.) Figure 4a summarizes the latencies, collapsed over the final 24 sessions of training. The data suggest that the deprivation-cue and both-cues groups responded with a shorter latency in the R+ sessions than in the R− sessions. A Group × Session Type ANOVA confirmed a significant main effect of session type, F(1, 21) = 8.73, MSE = 0.064, p = .008, η p 2 = .29, and, crucially, a significant interaction of group and session type, F(2, 21) = 3.91, MSE = 0.064, p = .036, η p 2 = .27. Follow-up comparisons confirmed that log latencies were shorter in the R+ sessions than in the R– sessions for both the deprivation-cue group, t(21) = 2.35, p < .029, mean difference = 0.30, 95% CI = [0.034, 0.559], η2 = .31, and the both-cues group, t(21) = 3.29, p < .004, mean difference = 0.41, 95% CI = [0.152, 0.667], η2 = .51. In contrast, the rats in the food-cue group responded with a similar latency in the R+ and R− sessions, p = .61. Figure 4b provides a summary of how the differences in latency between the R+ and R− sessions developed over training.
Results from Experiment 3: latency of the first response. The bar graph (a) shows mean latency of the first response (collapsed over the final 24 sessions of training) for each group, separately for sessions in which lever pressing was reinforced (R+; 12 sessions) and those in which it was not reinforced (R−; 12 sessions). Error bars represent +1 SEM. The line graph (b) shows average mean difference in latency of the first response (R+ minus R− sessions) for each block of eight sessions (four R+ and four R−). Error bars represent ±1 SEM.
Discussion
Rats that had only the presence and absence of food in the home cage to signal reinforcement and extinction sessions (food-cue group) never learned to anticipate the reinforcement contingency. In contrast, rats that were sated before R+ sessions and hungry before R− sessions did learn (deprivation-cue group and both-cues group). It is noteworthy that the addition of home-cage food cues to the deprivation cues did not make the discrimination more rapid in the both-cues group than in the other two groups. Apparently, the animals learned to discriminate between satiation and deprivation with very little input from the availability of food in the home cage. The results strongly suggest that interoceptive states of hunger and satiety can control instrumental food seeking. Note further that satiety once again served, counterintuitively, as a cue that increased the motivation to lever press for food, as in Experiments 1 and 2.
General Discussion
The results suggest that interoceptive food-deprivation stimuli can play the role of context in controlling food seeking. In Experiments 1 and 2, regardless of whether the reinforcer was sucrose pellets or sweet, fatty pellets, food seeking that was learned while the animal was sated, and then inhibited (extinguished) while it was hungry, was renewed when the rat was sated again. The results of Experiment 3 further confirmed that interoceptive cues connected with deprivation and satiation provide important discriminative cues. The presence or absence of food in the home cage was not effective at signaling the reinforcement contingency.
As mentioned previously, the lack of renewal in Experiment 2 for rats that had been deprived for acquisition and extinction training and sated for the first time at test could be attributed to the smaller magnitude of the AAB renewal effect. It is also possible that the invigorated responding during extinction allowed the development of stronger response inhibition (Bouton et al., 2016; Rescorla, 1997). The failure to observe AAB renewal might also point to features of interoceptive satiety and deprivation cues that make them different from the exteroceptive contexts usually used in the renewal literature. The evidence that interoceptive hunger and thirst states can serve as discriminative cues was once controversial (e.g., Bolles, 1975); one possible complication, among others, was that the discriminative effects of states of deprivation and satiety interact with their motivating effects (e.g., Capaldi, Viveiros, & Davidson, 1981). In our experiments, renewal always took the form of satiety increasing the level of food-motivated responding. In contrast, in the rat’s prior experience, interoceptive satiety cues had presumably been associated with cessation (rather than initiation) of feeding (e.g., Davidson, 1993), perhaps making renewal in the presence of satiety cues difficult without the explicit discrimination between satiety and deprivation that is provided by the ABA procedure.
The results of Experiment 3 indicated that interoceptive deprivation cues rather than exteroceptive food cues were most likely responsible for the discriminations observed. First, the rats’ performance showed a shorter latency of the first response while sated (during R+ sessions), which indicates that interoceptive satiety cues provided a stimulus that signaled reinforcement. In contrast, the presence of home-cage food alone did not enable similar anticipatory responding. And when both food cues and deprivation cues were available, the food cues provided little additional support in helping the rats learn the discrimination. It is worth noting that pharmacological treatments purported to induce satiety or hunger can also produce effects that generalize to satiety and deprivation states created through feeding manipulations. In such methods, after training with a discrimination of hunger and satiety, animals receive tests under identical food-deprivation conditions with and without the pharmacological treatment. The fact that responding after a treatment generalizes to that shown during discrimination training suggests that generalization occurs along the dimension of interoceptive state (Davidson et al., 2005; Kanoski, Walls, & Davidson, 2007).
The present findings are consistent with a research literature pointing to the role of conditioning and learning processes in eating, appetite, their disorders, and treatment of those disorders (e.g., Boutelle & Bouton, 2015; Bouton, 2011; Jansen, 2016). To our knowledge, however, the present results provide the first evidence that satiety can play the role of context and produce renewal in a typical ABA design that provides relatively few shifts between deprivation states. Some authors have proposed that food restriction itself may be a causal factor in the development of maladaptive overeating (e.g., Polivy & Herman, 1985, 2002). However, the present data and other recent analyses (Jansen, 2016) suggest that restrained or inhibited eating alone is not sufficient to cause excessive eating later: Rats that never learned to seek food while sated (the DDS group in Experiment 2) did not exhibit increased food seeking when they were switched from the extinction context of hunger to the context of satiety. Only behaviors that had been reinforced while the rats were satiated were renewed when hunger was interrupted and satiety was resumed (the SDS group). Satiety and hunger states can come to control the excitation and inhibition of food seeking in much the same way that other kinds of contexts do (e.g., Bouton, 2010).
Footnotes
Action Editor
Steven W. Gangestad served as action editor for this article.
Declaration of Conflicting Interests
The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article.
Funding
This research was supported by National Institute on Drug Abuse Grant R01-DA033123 (to M. E. Bouton).
Open Practices
All data have been made publicly available via the Open Science Framework and can be accessed at https://osf.io/enuv3/. The complete Open Practices Disclosure for this article can be found at https://journals-sagepub-com.web.bisu.edu.cn/doi/suppl/10.1177/0956797617719084. This article has received the badge for Open Data. More information about the Open Practices badges can be found at 
.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
