Demand Characteristics in Episodic Future Thinking II: The Role of Cues and Cue Content in Changing Delay Discounting

Abstract

Episodic future thinking (EFT), or imagining experiencing one’s future, reduces discounting of future rewards, but the mechanisms of this effect are unclear. We examined the role of cues to engage in EFT, possible demand characteristics inherent to those cues (prompting awareness of the hypothesized effects of EFT), and changes in temporal horizon (how far one thinks into the future) in these reductions in delay discounting. In Experiment 1, cues prompting participants to engage in EFT during the discounting task were presented or withheld: EFT decreased discounting only when cues were present. In a time-perspective task in which cues were not presented, EFT did not broaden temporal horizon; however, episodic recent thinking (the putative control activity) shortened temporal horizons. In Experiment 2, cues were manipulated in a way designed to make them either more theoretically relevant (prompting episodic thought) or more prone to demand characteristics (prompting attention to the correspondence in delays between EFT and discounting tasks). Only theoretically relevant cues reduced delay discounting. These results do not support the hypothesis that EFT reduces delay discounting because of demand characteristics. Instead, they suggest limitations on the generalizability of EFT effects in uncued contexts, and suggest that mechanisms other than temporal horizon may be responsible for its effects.

Extreme devaluation of delayed outcomes is one mechanism underlying chronic preference for smaller-sooner over larger-later rewards (Madden & Johnson, 2010; Odum, 2011). Because such steep delay discounting is correlated with detrimental health-impacting behaviors, such as cigarette smoking (Baker, Johnson, & Bickel, 2003; Bickel, Odum, & Madden, 1999; Friedel et al., 2014), licit and illicit drug abuse and dependence (Kirby & Petry, 2004; Heil et al., 2006; Petry, 2001; Vuchinich & Simpson, 1998) and non-drug, addiction-related behaviors such as pathological gambling (Alessi & Petry, 2003) and obesity (Amlung et al., 2016), discounting is increasingly discussed as a target for therapeutic intervention (Bickel et al., 2012; Koffarnus et al., 2013). Such interventions may help prevent the development of, or improve the efficacy of interventions designed to combat these harmful behaviors (Gray & MacKillop, 2015; Volkow & Baler, 2015).

One method for reducing delay discounting in humans is Episodic Future Thinking (EFT), or the pre-experiencing of one’s future (Atance & O’Neill, 2001). Many studies indicate that EFT produces consistent reductions in discounting (e.g., Benoit, Gilbert, & Burgess, 2011; Daniel, Stanton, & Epstein, 2013a; Dassen et al., 2016; Peters & Büchel, 2010; Stein et al., 2018; see Rung & Madden, 2018b for review and meta-analysis), as well as problematic behaviors related to steep discounting. For instance, EFT reduces calorie and/or junk food consumption both in (Daniel, Said, Stanton, & Epstein, 2015; Daniel, Stanton, & Epstein, 2013b; Dassen et al., 2016) and outside of the lab (O’Neill, Daniel, & Epstein, 2016; Sze et al., 2015). It also reduces real and hypothetical cigarette smoking in laboratory tasks (Stein et al., 2018, 2016), as well as hypothetical alcohol consumption (Bulley & Gullo, 2017; Snider, LaConte, & Bickel, 2016). The success of EFT interventions is often attributed to the expansion of the participant’s temporal horizon, or the time frame over which individuals tend to consider the consequences of their behavior (Lin & Epstein, 2014; Snider et al., 2016).

An alternative account of these effects of EFT is that they are due to demand characteristics (Orne, 1962); i.e., procedural features that inadvertently reveal the experimental hypothesis to the participant. In a typical EFT experiment, participants are first encouraged to vividly imagine events occurring at specific future times. For each of these imagined events, a cue is created for the purpose of later prompting the participant to engage in EFT while they complete the delay discounting task. For example, if the participant episodically thought about going on vacation in 6 months, their cue might read, “In 6 months I will be on the beach, enjoying vacation”. Then, when completing the discounting task and choosing between $50 now and $100 in 6 months, the “In 6 months I will be on the beach, enjoying vacation” cue is presented along with the reward alternatives. In most EFT experiments, the cues are, in this way, delay-matched to the larger-later (or “self-controlled”) outcome, which, in this example, is also delayed by 6 months. These cues may reveal the experimenter’s hypothesis that EFT will increase preference for the larger-later alternative. Awareness of the hypothesis is problematic because participants may exhibit a good-subject effect by which their behavior becomes biased so as to confirm the hypothesis (e.g., Nichols & Maner, 2008), with some individuals making a conscious effort to do so (e.g., Goldstein et al., 1972).

Rung and Madden (2018a) reported that individuals who read a description of typical EFT procedures (but did not complete EFT themselves) deduced the experimenter’s hypothesis that EFT would reduce impulsive choice (Experiment 1) and consumption of unhealthy food (Experiment 2). Participants who read about typical EFT-control procedures deduced no systematic hypothesis. Similarly, Stein et al. (2018) found that participants actually engaging in EFT—not just reading about the procedures—were more likely than controls to deduce its hypothesized effects. However, the effect of EFT on delay discounting remained significant when participants’ awareness of the hypothesis was controlled for in statistical analyses. This null finding may have been influenced by how hypothesis-awareness was scored. High awareness scores were given for statements merely indicating that EFT and discounting were related, without specifying how they might be related (i.e., in what direction). A good-subject effect, or other expectancy-related bias, should influence discounting rates only if participants were aware of the hypothesized direction in which EFT should affect decision-making. Thus, the scoring technique used by Stein et al. may have obscured the mediating effects of demand characteristics on the effect of EFT on delay discounting.

The goal of the present research was to further evaluate the hypothesis that demand characteristics play a role in EFT effects on delay discounting. In Experiment 1, EFT cue availability during the delay-discounting task was manipulated. If an EFT effect was absent when cues were omitted, this effect could be explained by demand characteristics or by a failure to engage in EFT if not prompted to do so. Experiment 1 also evaluated extension of temporal horizon as a mediator of EFT-produced reductions in discounting. In Experiment 2, components of the EFT cues were manipulated under the hypothesis that some components are primarily responsible for producing engagement in EFT, and others a source of demand characteristics.

Experiment 1

A candidate source of demand characteristics in EFT studies are the delay-matched cues displayed during the discounting task. In Experiment 1, participants completed EFT activities and then immediately completed the delay-discounting task either with or without these cues. To date, no known research has evaluated the necessity of cueing to obtain reductions in discounting following EFT. Experiment 1 also evaluated the mediating role of temporal horizon in EFT effects on discounting. To this end, participants completed the Future Time Perspective task, which is designed to measure the breadth of a participant’s temporal horizon (Petry, Bickel, & Arnett, 1998).

Method

Participants

Participants were between the ages of 19 and 60 years old and were recruited via Amazon Mechanical Turk ® (MTurk), an online crowdsourcing platform in which Requesters (researchers, organizations, etc.) post web-based tasks (called human intelligence tasks, HITs) that registered users can complete for monetary compensation. To be eligible, Amazon MTurk users were required to reside in the United States and have previously completed ≥ 100 HITs, with ≥ 95% of these previously completed HITs approved by their Requesters (i.e., paid for adequate completion).

Participants learned of the study through a brief description on MTurk, which only eligible participants were able to view. The description indicated the task duration (up to approximately 50 minutes) and activities (complete psychological assessments for the purpose of determining population norms). The advertised study duration was based on the maximum time anticipated to complete all study procedures; median times were much lower (see Results for details). All study procedures were approved by the Institutional Review Board at Utah State University (Experiment 2 in protocol #8170, titled “Demand Characteristics in Episodic Future Thinking”); and informed consent was obtained from all individual participants included in the study.

Participants were compensated $3.30 for their participation (approximately $6.20/hour based on median survey durations). At the end of the survey, participants were debriefed and given the option to have their data discarded, which was done upon request. Participants were provided a survey completion code after this debriefing.

Design

This experiment used a 2 × 2 between-subjects design, with independent variables of Temporal Perspective (future [EFT] and recent past [episodic recent thinking; ERT]), and Cue Presence (present and absent); and dependent variables of degree of delay discounting and temporal horizon. Participants were randomly assigned to one of four groups: EFT-Cued, EFT-Uncued, ERT-Cued, or ERT-Uncued. All participants, regardless of group assignment, completed the tasks in the order they appear below.

Measures

At the start of the survey, all participants were prompted to vividly describe three events. The type of event (future or recent [control]) was dependent on group assignment. Participants in the EFT groups were asked to identify positive, realistic events that could occur at delays of 6 months, 1 year, and 5 years; the latter time (which was not used in the discounting task) was included to increase the probability that EFT would expand temporal horizons. The ERT group was asked to identify positive events occurring yesterday between 4 PM - 8 PM, 12 PM - 4 PM, and 8 AM - 12 PM. EFT participants indicated their identified events by completing the sentence starting with, “In [delay], I will…”, and ERT participants by completing the sentence starting with, “Between [time frame], I was…”. The order of presentation of these time-frames was randomly determined.

Participants were asked to describe each event in detail using four prompts based on those in Snider et al. (2016). Specifically, participants were asked: (1) “For [event], what (will you be/were you) doing?” (2) “For [event], who (will you be/were you) with?”, (3) “For [event], where (will you be/were you)?”, and (4) “For [event], how (will you be/were you) feeling?”. For each event, participants were prompted to write at least 2–3 sentences, and were provided examples of “good” (highly specific) and “bad” sentences (vague descriptions of the event). Participants typed their responses to each question on separate screens. On a fifth screen, participants rated the event for its vividness, details, importance, positivity, and negativity using a 5-point Likert scale. These five screens were completed in this order at each of the three delays (EFT) or recent times (ERT).

Responses made in the EFT task were considered valid if they (1) answered the question (i.e., answers to …what will you be [were you] doing?” had to specify an activity), (2) described a coherent episodic event, and (3) were not copied and pasted from the “good” and “bad” sentences. For example, in response to, “…how will you be feeling?” one response classified as invalid was, “Make breakfast a priority on your weight-loss plan…” and, in response to “…where were you?” another invalid response was, “floor was the very tasty and behaviour of that working into this survey below i was also in the kitchen briefly.”

Delay discounting

An adjusting amount task in which participants chose between pairs of hypothetical monetary alternatives was used (Du et al., 2002). One alternative was always an immediate reward and the other a larger reward delayed by either 6 months or 1 year (delay varied across blocks of trials). Participants indicated their preference across 6 pairs of alternatives for a single block/delay, with the amount of the immediate reward titrated based on the participant’s preference. Within a block, participants first chose between $50 now and $100 following the delay; for subsequent questions, the immediate amount was increased (decreased) after selection of the delayed (immediate) alternative. The first adjustment was $25, and subsequent adjustments decreased by half across trials. Degree of delay discounting was quantified as the adjusted immediate amount that would appear on the seventh trial if one was programmed to occur (i.e., the indifference point). The order in which the blocks were completed (i.e., 6-month or 1-year delay completed first) was counterbalanced across participants.

Presentation of an attention check question was dependent on participant’s responses within a block of trials. If participants chose the immediate (delayed) reward exclusively within a block, a seventh trial occurred in which the participant chose between $0 now and $100 now ($500 now and $100 in 1 year). If participants failed to pass one or more attention checks (by choosing the smaller reward amount), their data were excluded from analyses.

Temporal horizon

To assess changes in temporal horizon following EFT/ERT, all participants completed the first portion of the Future Time Perspective task (the FTP; Wallace, 1956). Participants were instructed to list five events likely to happen during the rest of their lives and their age at each of these events. Two dependent measures were calculated: maximum extension (the longest projection to an event; future-event age minus current age) and median extension (the median extension across all five events). Responses made in the FTP task were considered valid if (1) the participant’s age at each of the five events was greater than or equal to their current age, (2) the maximum anticipated age at an event did not surpass 110 years¹, (3) entries could reasonably be interpreted as events (e.g., the single-word responses “athletic” and “tomorrow” were considered invalid), and (4) none of the events were repeats. Participants who provided invalid data on the FTP task were omitted from analyses pertaining to measures from the FTP, but retained in analyses of discounting data if they passed other inclusion criteria specific to those tasks.

Social desirability scale

The Marlowe-Crowne Social Desirability Scale (SDS; Crowne & Marlowe, 1960) was used to assess participants’ tendency to present themselves in a socially desirable way. We hypothesized that those presenting themselves as socially desirable would be more likely to behave in accord with demand characteristics (e.g., Nichols & Maner, 2008); that is, social desirability would moderate the effects of EFT on delay discounting. The SDS questionnaire consists of 33 true/false statements pertaining to idealistic behaviors (e.g., “No matter who I’m talking to, I’m always a good listener”). The dependent measure was the number of socially desirable behaviors endorsed as true.

Experimental history

At the end of the survey, participants were asked, via a series of yes/no questions, whether they had previously completed an online delay discounting task or EFT/ERT. Individuals reporting prior history with EFT/ERT were compensated but excluded from analyses. There were two reasons for this exclusion: (1) all (known) published EFT studies recruiting MTurk workers have used cues (Stein et al., 2017, 2018; Sze et al., 2017) and this experience could undermine the integrity of the manipulation of cue presence/absence, and (2) participants with prior experience may have been debriefed on the hypothesized effects of EFT.

Demographics

Participants provided their age, sex, race and ethnicity, in addition to income and highest level of education achieved.

Procedures

After accepting the HIT and providing informed consent, participants completed the tasks in the order shown above. For those in the EFT-Cued and ERT-Cued groups, at the top of each screen of the discounting task, a cue prompted participants to think about an event they had written about during the EFT/ERT task; e.g., “Now, imagine in 6 months, I will be on the beach, enjoying vacation while answering…/Now, imagine yesterday between 4 PM and 8 PM, I was…”. The delay to the cued event always matched the delay to the larger-later reward. Those in the ERT groups were cued to imagine the event they reported occurring between 4 PM - 8 PM yesterday when the large reward was delayed by 6 months, and the event occurring between 12 PM and 4 PM when the delay was 1 year. Participants in the EFT-Uncued and ERT-Uncued groups completed the discounting tasks without cues. Cues were not provided during any subsequent component of the experiment.

Data analysis and Sample Size Determination

The targeted sample size for Experiment 1 (N = 350) was the necessary N to detect a significant small-to-medium interaction (f = 0.15) in a two-way ANOVA with .80 power and an alpha of .05. A small-to-medium effect size was chosen since it was unknown if cueing would interact with EFT/ERT; this magnitude of effect approximates that which may be due to placebo effects in psychotherapies (see Wampold, Minami, Tierney, Baskin, & Bhati, 2005 for discussion).

Prior to analyses, demographic differences across groups were examined using chi-squared tests, ANOVAs, etc., as appropriate; and potentially relevant covariates were examined within-groups using correlations (e.g., correlations between indifference points and income, age, vividness of episodic prospection, etc.). Many variables in correlation analyses showed a high degree of skewness and/or lacked bivariate normality; thus, for the sake of consistency, Spearman correlations were used. All of these analyses (and below) were conducted using R (R Development Core Team, 2013). Unless otherwise noted, the functions used for conducting statistical tests were those included in the basic R software.

For all statistical tests, the normality of distributions was assessed using Shapiro-Wilks tests. For all regression analyses, typical diagnostics were performed (e.g., assessing normality of residuals, screening for overly influential observations, outliers, etc.). When overly influential points were found (based on Cook’s D values), they were excluded from analysis and the model was re-run. All model results reported are those without overly influential values; instances in which their inclusion produced discrepant results are noted.

Analyses of Cue Presence

Beta regression was used to assess group differences in discounting and the mediating and moderating roles of temporal horizon and socially desirable proclivities, respectively. Indifference points for the 6-month and 1-year delays were transformed into proportions (i.e., divided by 100) and used as the dependent variable for these analyses. Beta regression is a type of generalized linear model that assumes a beta distributed dependent variable, with values bounded between 0 and 1 (Ferrari & Cribari-Neto, 2004). The bounded nature of the indifference points prior to transformation, in addition to the fact that beta distributions can take on a variety of different shapes, made this analysis the most suitable and powerful choice (e.g., contrasted with nonparametric techniques which cannot accommodate moderators, mediators, or covariates more generally). A unique aspect of beta regression is that it includes parameterization of the variance of the outcome (i.e., the variance can be modelled separately with its own regression equation; Cribari-Neto & Zeileis, 2010). In the absence of explicit parametrization of the variance (specification of predictors), the variance component contains a single precision coefficient (ϕ). Greater values of ϕ indicate greater precision of model estimates, which can be used as an additional tool for assessing model fit. Beta regression models were conducted using the betareg package (Cribari-Neto & Zeileis, 2010).

First, effects of the independent variables (Temporal Perspective, Cue Presence) and their interactive effects on delay discounting were examined using beta regressions. The beta regression was conducted once with each indifference point (at the 6-month and 1-year delays) as the dependent variable. Specific group differences were then evaluated using the lsmeans package (e.g., EFT-Cued vs. EFT-Uncued; Lenth, 2016).

Next, the effects of Temporal Perspective and Cue Presence (during the discounting task) on FTP scores were evaluated using the non-parametric equivalent of a one-way ANOVA (due to highly skewed distributions), using a recoded grouping variable in which each group (i.e., EFT-Cued, ERT-Cued, etc.) represented one level. Recoding with the use of this non-parametric technique was conducted because there are no non-parametric tests equivalent to a two-way ANOVA.

Mediator and Moderator Analyses

Mediator and moderator analyses were conducted with both the 6-month and 1-year indifference points. Mediation analyses were primarily conducted with beta regression, approximating the steps outlined by Baron and Kenny (1986). Because the mediation analysis evaluated if changes in temporal horizon mediated the effect of EFT on delay discounting, the analysis was confined to the EFT-Cued and ERT-Cued groups—the only groups for which significant effects of EFT on discounting and temporal horizon were observed (see Results below). The first step was to establish the significance of the difference in discounting between the EFT-Cued and ERT-Cued groups. In the second step, the significance of the difference in temporal horizon between these cued groups was established. Because FTP data were not on a 0 to 1 scale, the group differences could not be evaluated using beta regression. Therefore, the second step was approximated by conducting a Mann-Whitney U test. The final step of the mediation analysis was to test the difference in coefficients (a z-test; Cohen et al., 2003) for the EFT effect across regression models with and without temporal horizon as a predictor. Traditional mediation analysis (e.g., a Sobel test; Sobel, 1982) requires that the regression technique used for steps 1 and 2 be the same in order to estimate the mediated effect. Because of the limitations due to the scale and distributional properties of the (hypothesized) mediator and dependent variable, the coefficient test was used in lieu of the Sobel test.

Last, an exploratory analysis was conducted to determine if scores on the social desirability scale moderated EFT efficacy. This beta regression analysis was conducted using those in EFT groups only because EFT only reduced discounting in EFT-Cued. Thus, examining the interaction between SDS scores and Cue Presence among those in EFT groups tests the hypothesis that when the purpose of EFT is more readily apparent, individuals who tend to behave in more socially desirable ways show larger EFT effects (i.e., evidence a good-subject effect).

Results

A total of 468 individuals participated in the study, with two participants requesting their data be discarded. Of the remaining 466, 22% (n = 101) reported prior experience with EFT/ERT procedures and were therefore excluded from all analyses. Among the 365 EFT/ERT-naïve participants, 13 were excluded for failing to correctly complete the EFT/ERT tasks (or EFT/ERT and FTP tasks), or pass an attention check in the discounting assessment. The number of participants excluded for prior EFT/ERT history and failure to pass exclusionary criteria related to study tasks is provided in Table 1. An additional two participants encountered a survey error and did not complete all dependent measures. Thus, the final analytic N was 350.

Table 1.

Demographic Variables and Survey Durations for Participants in Experiment 1 (N = 350)

	EFT-Cued		EFT-Uncued		ERT-Cued		ERT-Uncued
Variable	% (n)	Median (Q1-Q3)	% (n)	Median (Q1-Q3)	% (n)	Median (Q1-Q3)	% (n)	Median (Q1-Q3)
n	24 (85)		26 (91)		26 (91)		24 (83)
Age (years)		33 (28–40)		35 (30–42)		36 (30–41)		34 (29–43)
Sex
Male	34 (29)		41 (37)		33 (30)		33 (27)
Female	66 (56)		59 (54)		67 (61)		67 (56)
Race
American Indian/Alaska Native	2 (2)		1 (1)		0 (0)		1 (1)
Asian	4 (3)		8 (7)		4 (4)		6 (5)
Black/African American	9 (8)		12 (11)		10 (9)		11 (9)
Hawaiian/Pacific Islander	1 (1)		0 (0)		0 (0)		0 (0)
Multiple	1 (1)		5 (5)		3 (3)		6 (5)
White	81 (69)		74 (67)		82 (75)		76 (63)
N/A	1 (1)		0 (0)		0 (0)		0 (0)
Ethnicity
Hispanic/Latino	6 (5)		6 (5)		11 (10)		8 (7)
Non-Hispanic/Latino	93 (79)		93 (85)		89 (81)		90 (75)
N/A	1 (1)		1 (1)		0 (0)		1 (1)
Education		16 (14–17)		16 (13–16)		16 (14–16)		16 (14–16)
Income		41 (21–66)		45 (25–70)		45 (28–65)		45 (30–60)
Survey duration		32 (25–47)		37 (27–52)		32 (24–41)		30 (23–41)
Prior EFT/ERT*	21 (23)		19 (22)		22 (25)		25 (31)
Data quality exclusions	3 (3)		3 (3)		0 (0)		8 (7)

Note. Percentages do not always sum to 100% due to rounding; age and years of education are rounded to the nearest year, income to the nearest $1,000 increment, and survey duration to the nearest minute.

^* Percentage calculated based on group n prior to any exclusions.

Table 1 provides descriptive statistics for demographic variables. Participants were predominantly white (> 73%), identified as non-Hispanic/Latino (89%) and had median ages in the mid-30s. There were no significant differences in demographic characteristics across the four groups. Participants in EFT-Uncued took longer to complete the survey (Mdn = 7.3 minutes longer) than those in ERT-Uncued (W = 4656, p = .01; no other comparisons significant). Only three significant correlations between discounting and potential covariates were detected (including variables such as vividness of episodic thought, etc.). Due to their small magnitude (ps ≤ .27) and inconsistent significance across groups, they were deemed unable to serve as meaningful covariates. These correlations are not further discussed but are shown in the Supplementary Materials, Figures S-1 through S-4.

Indifference points for individual participants, separated by group and large-reward delay are shown in Figure 1, panels A (6-month delay) and B (1-year delay). EFT reduced delay discounting at both delays but only when cues were presented during the discounting task (see Table 2 for parameter estimates). That is, the Cue Presence by Temporal Perspective interaction was significant for the models with both the 6-month (z = 2.11, p = .03) ² and 1-year indifference points, (z = 3.28, p = .001). The interaction effect was revealed by significantly higher indifference points in the EFT-Cued group (6-month delay Mdn = 70; 1-year delay, Mdn = 63) relative to all other groups (all ps ≤ .001); and a lack of significant differences between the EFT-Uncued group (6-month delay Mdn = 49; 1-year delay, Mdn = 43) and the ERT-Cued (6-month delay Mdn = 49; 1-year delay, Mdn = 24) and ERT-Uncued (6-month delay Mdn = 49; 1-year delay, Mdn = 37) groups. For the latter comparisons, all ps ≥ .48.

Figure 1.

Indifference points for the 6-month (panel A) and 1-year delays (panel B) from the discounting task for each group of participants. The height of the gray bars corresponds to the median indifference point, and the black bars correspond to the first and third quartiles. Open circles show individual participants’ indifference points. Asterisks indicate significant differences from all other groups (p < .001).

Table 2.

Parameter Estimates for the Beta Regressions Predicting Indifference Points in Experiment 1

Model/parameter	Beta	Std. error	Z	p
Model 1 (6-montd indifference point)
Intercept	0.70	0.12	5.68	<.001
Temporal perspective (present)	−0.56	0.17	−3.29	.001
Cue (uncued)	−0.62	0.17	−3.66	<.001
Temporal perspective X Cue	0.50	0.24	2.11	.035
Model 2 (1-year indifference point)
Intercept	0.47	0.12	3.81	<.001
Temporal perspective (present)	−0.84	0.17	−4.94	<.001
Cue (uncued)	−0.77	0.17	−4.48	<.001
Temporal perspective X Cue	0.79	0.24	3.28	.001

Note. For both models, the estimates for the intercept reflect the EFT-Cued group. Parameter estimates from this beta regression reflect the log-odds change in indifference points. The inverse logit can be used to obtain parameter estimates on the original scale of measurement: Exp(β)/[1+Exp(β)].

Temporal Perspective and Cue Presence also impacted temporal horizon. Figure 2 shows median (Panel A) and maximum (Panel B) extensions in the FTP task, separated by group; note the logarithmic y-axis. Across groups, there were significant differences in median future extension, χ²(3) = 13.88, p = .003. Pairwise comparisons revealed significantly shorter median future extensions in the ERT-Cued group (Mdn = 4.00 years) relative to the EFT-Cued (Mdn = 8.00 years, W = 4894.5, p = .0004), EFT-Uncued (Mdn = 6.50 years; W = 4842, p = .01), and ERT-Uncued groups (Mdn = 6.00 years; W = 2910, p = .03). No other between-groups comparisons were significant (ps ≥ .10). The same ordinal pattern of results (direction and statistical significance) was replicated when using maximum future extension from the FTP task as the dependent variable (see Figure 2, panel B).

Figure 2.

Median (panel A) and maximum (panel B) future extension on the future time perspective task. The height of the gray bars corresponds to the group median, and the black error bars indicate the first and third quartiles; note the logarithmic y-axis. Individual participant data are indicated by the open squares overlaid on the bars. * p < .05; ** p < .01

Mediation and Moderation Analyses

Changes in temporal horizon were evaluated for their potential mediating role in the effects of EFT/ERT on delay discounting. Due to the weaker effect of the independent variables on maximum extension scores, mediation analysis was conducted using median extension as the hypothesized mediator. Only those in the EFT groups providing valid FTP data were included in this analysis (n = 173). In the first step, Group (EFT-Cued vs. ERT-Cued) was a significant predictor of 6-month (z= 3.06, p = .002) and 1-year indifference points (z=4.53, p < .001). In the second step (the same non-parametric analysis as that above), median extension was found to differ across the EFT- and ERT-Cued groups, W = 4623, p = .0002. When including median extension as an additional predictor with Group in the model, its inclusion did not improve either of the models (ps ≥ .11), nor was it a significant predictor of indifference points at either the 6-month (p = .10) or 1-year delays (p = .25, see Table S-1 in supplementary materials for parameter estimates); including temporal extension did not account for a significant proportion of the EFT effect at either delay, zs ≤ 0.27.

Finally, cueing effects were not significantly moderated by social desirability (SDS scores). In regression models subset to EFT groups only (n = 176), Cue was a significant predictor of indifference points at the 6-month (z= 3.12, p = .02) and 1-year (z= 3.83, p < .001) delays. Subsequent inclusion of SDS scores and the interaction between SDS scores and Cue improved the models (ps <= .02; but only after the removal of overly influential observations), but the interaction was not a significant predictor of indifference points (see Table S-2 in the online supplementary materials for parameter estimates).

Discussion

Experiment 1 revealed that the effects of EFT on delay discounting were confined to conditions in which participants were explicitly cued to imagine future events during the discounting task. Two accounts of this cue-dependent effect of EFT will be discussed. First, consistent with the demand-characteristics hypothesis, the cues may inadvertently reveal to participants the experimenter’s hypothesis and participants oblige the experimenter by making more larger-later reward choices than they normally would. Difficult to reconcile with this hypothesis is one study reporting real-world benefits of EFT (weight loss) relative to a control procedure that convincingly equated demand characteristics (nutritional-information thinking; Sze, Daniel, Kilanowski, Collins, & Epstein, 2015).

A second account is that participants did not engage in EFT during the discounting task unless cues prompted them to do so. That is, delay discounting was not attenuated in the EFT-Uncued group because they were not prompted to (and subsequently did not) engage in EFT. This account could also explain why expanded future horizons (median extensions in the FTP) did not mediate the EFT effect on discounting in the EFT-Cued group – because cues were not provided during the FTP task. Arguing against the latter, time horizons were shortest in the ERT-Cued group, despite no ERT cues being presented in the FTP task itself. Why cueing ERT—but not EFT—during the discounting task would impact future time horizons is unclear. Thus, Experiment 2 was designed to further evaluate the demand characteristics hypothesis by isolating separate components of the EFT cues, components thought to either increase demand characteristics or cue EFT activities.

Experiment 2

In Experiment 2 we sought to separate EFT cues into their theoretically relevant and their demand-characteristic components. From a theoretical perspective, episodically thinking about one’s own future is the critical activity that impacts delay discounting (Benoit et al., 2011; Peters & Büchel, 2010). Therefore, in Experiment 2, one cue type prompted participants to think about the content of their prior EFT (e.g., enjoying their beach vacation) but without including an interval of time in the cue. A second cue type isolated the component of EFT cues thought to engender demand characteristics – the correspondence between the interval of time in the cue (1 year) and the delay to the larger-later reward in the discounting task (1 year). If cues high in theoretical relevance (alone) significantly reduce delay discounting, this would support the hypothesis that EFT during the discounting task is responsible for the reduction in delay discounting. It would also support the hypothesis that the cue-dependent EFT effect in Experiment 1 represents a failure of generalization; i.e., EFT did not occur in the novel, uncued discounting task. If, instead, cues high in demand characteristics (alone) reduce delay discounting, this would support the demand-characteristics hypothesis. To clarify the relation between awareness of the experimental hypothesis and delay discounting, participants in Experiment 2 completed a post-experiment questionnaire about the purpose of the cues presented during the discounting task. Multiple-choice questions prompted participants to indicate in what direction, if any, they thought the cues were designed to influence their choices.

Method

Participants

Participants were Amazon MTurk users between the ages of 18 and 60 years old. The eligibility criteria and contingencies for compensation were the same as those in Experiment 1, with the exception that participants were compensated $1.00, scaled based on the shorter survey duration (up to approximately 15 minutes). All other procedures were as in Experiment 1, and were approved by the Institutional Review Board at Utah State University (Experiment 3 in the same protocol noted previously).

Design

This experiment used a between-subjects design, with participants randomly assigned to one of five groups. Participants first engaged in episodic future (4 EFT groups) or recent thinking (1 ERT group); then, they completed the discounting task.

Cue content was designed to juxtapose the putative level of demand characteristics and active cue components as specified in the theory underlying EFT’s effects on delay discounting (Benoit et al., 2011; Peters & Büchel, 2010). Participants assigned to the EFT-Event group were given cues designed for high-theoretical relevance (cues only prompted participants to imagine the future event; e.g., “Now, imagine that you will be on the beach, enjoying vacation while answering…”) and low demand characteristics (the cue specified no delay to the episodic event). By contrast, participants in the EFT-Time group were given cues designed for high-demand characteristics (i.e., “Now, imagine 1 year from now while answering…”) and low theoretical relevance (the cue omitted the future event). Participants in the EFT-Typical groups were given the same cues used in the EFT-Cued group in Experiment 1; these cues were conceptualized as high in theoretical relevance and demand characteristics.

The control (EFT-Uncued) and sham (ERT-Time) groups were designed based on the following considerations. Typically, an ERT group serves as a neutral control, but the indifference points across the EFT-Uncued and ERT-Uncued groups in Experiment 1 were statistically equivalent³. Therefore, the EFT-Uncued group was chosen as the control, as it allowed us to hold constant temporal orientation of thought with all other EFT groups. However, if EFT-Time produced a reduction in discounting relative to EFT-Uncued, this could be because the time cue (i.e., “1 year”) caused spontaneous engagement in EFT. Therefore, the ERT-Time group was included as a sham-control group. This group engaged in the ERT task (described below) prior to exposure to cues high in demand characteristics.

Measures

Episodic future/recent thinking

Participants were prompted to identify one event, with the type of event (future or recent past) dependent on group assignment. The EFT groups were asked to imagine a positive, realistic event that could occur 1 year from now. The ERT group was asked to imagine a positive event that occurred yesterday between 12 PM and 4 PM. Participants were then prompted to vividly imagine (via answering elaboration questions) and rate (characteristics such as vividness, details, etc.) the identified event using the same questions/response formats as in Experiment 1. Exclusion criteria for EFT/ERT entries were as in Experiment 1. Three EFT-Event participants were excluded because they included a time-referent in their event description (e.g., vacation in one year).

Delay discounting

The delay discounting task from Experiment 1 was used (Du et al., 2002). Participants only completed the block of trials with the 1-year delay to the larger-later reward. The 6-month delay was omitted because (cued) EFT produced a nominally larger effect on discounting at the 1-year delay in Experiment 1 (d = 0.33 vs. d = 0.67 for the 6-month and 1-year delays, respectively).

Suspicion probe and in-task thoughts

After completing the discounting task, all participants were prompted to answer questions about thoughts they may have had while completing the survey. Participants were told that their answers were important and would help researchers understand the utility of a new psychological therapy, and their answers would not affect their compensation. Answers were presented in multiple-choice format and were mutually exclusive.

Participants in Cued groups were shown an example screen from the discounting task, complete with the assigned cue. They were then asked, “Why do you think we asked you to vividly imagine [cue text] while making your choices?” Participants in the EFT-Uncued group were asked, “Why do you think we asked you to vividly imagine [event text] before starting the money choice task?” Regardless of group, the three options were 1) “You thought it would increase my preference for the money now,” 2) “You thought it would increase my preference for the money in 1 year,” and 3) “Neither/I’m not sure.”

If participants selected options 1 or 2 they were asked several follow-up questions. First, when did they realize the purpose of the cue or, for participants in the EFT-Uncued group, the purpose of previously engaging in EFT. The two options were: 1) “While I was completing the money questions on the survey” and 2) “Just now, when you asked me about it on the previous page.” Participants were asked additional questions if they selected option 1, but due to a lack of statistical power (low proportion of participants selecting option 1), these data were not further analyzed. Details of these questions can be found in supplementary materials, along with descriptive statistics of the results (Table S-3).

Experimental history and demographics

Participants were asked the same questions regarding prior experience with discounting assessments and EFT/ERT procedures as in Experiment 1; this information was used for a-priori exclusion of data provided by those with prior EFT/ERT experience. For demographics, participants provided their age, sex, race, ethnicity, income, and highest level of education achieved.

Procedures

After accepting the HIT and providing informed consent, participants completed EFT or ERT as previously described. Next, participants completed the discounting task with (EFT-Typical, EFT-Event, EFT-Time, ERT-Time) or without cues (EFT-Uncued) presented at the top of the screen during the discounting task. Finally, participants completed the suspicion probe, in-task thought questions, prior experimental history, and demographics.

Data Analysis and Sample Size Determination

Sample size was based on the obtained effects in Experiment 1. Anticipating that the manipulation of cue contents would reduce the effects of EFT on discounting, we powered based on a medium-to-small main effect of Group (f = .19). Using a one-way ANOVA, and assuming .80 power and an alpha of .05, this effect size necessitated a sample size of N = 340 to achieve a significant main effect of group.

The same analytic approach used in Experiment 1 was used for Experiment 2 data. For the suspicion probe, group differences in the proportions of correct, incorrect, and no hypothesis deduced were compared using a chi-squared test across the EFT-Typical, EFT-Event, and EFT-Time groups. The test was restricted to these three groups to reveal if the EFT-Event cue (designed for theoretical relevance) was less likely to cause participants to deduce that the experimenter expected them to choose the larger-later reward.

Next, the proportion of correct (vs. incorrect and no) identifications of the hypothesis were compared to chance in each group using binomial tests. For this analysis, for all groups, a directional test was conducted (> .33). Similarly, the proportion of instances in which participants reported having identified the hypothesis during the discounting assessment (vs. only when prompted) was also compared to chance; this test was directional (> .50) for all groups except EFT-Uncued, which was performed as a non-directional test because the lack of cues in-task would presumably be less likely to lead to identification during the discounting task itself.

Results

A total of 465 MTurk users participated. Of these, 10 requested to have their data withdrawn and 84 participants data were excluded due to prior EFT/ERT experience. An additional 12 participants failed to follow instructions on the episodic thinking tasks and/or failed the attention check in the discounting assessment; and two participants encountered survey errors. Thus, the analytic N was 357, with 66 to 79 participants in each group.

Group ns and demographic characteristics are shown in Table 3. Participants tended to be in their early to mid-30s, and most had at least three years of education beyond high school. The majority of participants in each group identified as White (≥ 74%) and non-Hispanic/Latino (≥ 85%). Tests of demographic characteristics revealed no significant differences across groups. Similar to Experiment 1, within-group correlations between continuous demographic measures, variables related to episodic elaboration and in-task thoughts, and discounting revealed no significant, meaningful covariates for inclusion in later analyses (see Figures S-5 to S-9 in the Supplementary Materials).

Table 3.

Demographic Variables and Survey Durations for Participants in Experiment 2 (N = 357)

	EFT-Typical		EFT-Event		EFT-Time		ERT-Time		EFT-Uncued
		Median		Median		Median		Median		Median
Variable	% (n)	(Q1-Q3)	% (n)	(Q1-Q3)	% (n)	(Q1-Q3)	% (n)	(Q1-Q3)	% (n)	(Q1-Q3)
n	22 (79)		20 (71)		20 (71)		18 (66)		20 (70)
Age (years)		31 (27–36)		33 (28–43)		31 (26–39)		33 (28–38)		31 (26–38)
Sex
Male	49 (39)		45 (32)		52 (37)		36 (24)		37 (26)
Female	51 (40)		55 (39)		48 (34)		64 (42)		61 (43)
Race
American Indian/Alaska Native	1 (1)		1 (1)		0 (0)		2 (1)		1 (1)
Asian	9 (7)		6 (4)		6 (4)		5 (3)		7 (5)
Black/African American	5 (4)		7 (5)		4 (3)		8 (5)		11 (8)
Hawaiian/Pacific Islander	0 (0)		1 (1)		0 (0)		0 (0)		0 (0)
Multiple	3 (2)		1 (1)		6 (4)		0 (0)		1 (1)
White	81(64)		79 (56)		85 (60)		86 (57)		74 (52)
N/A	1 (1)		4 (3)		0 (0)		0 (0)		4 (3)
Ethnicity
Hispanic/Latino	9 (7)		14 (10)		6 (4)		8 (5)		7 (5)
Non-Hispanic/Latino	89 (71)		85 (60)		93 (66)		88 (58)		89 (62)
N/A	1 (1)		1 (1)		1 (1)		4 (3)		4 (3)
Education		16 (13–6)		15 (13–16)		16 (13–16)		16 (13–16)		16 (14–16)
Income		50 (30–65)		50 (30–66)		48 (33–56)		45 (20–61)		45 (30–60)
Survey duration		10 (8–13)		10(7–15)		10 (8–13)		10 (8–13)		11 (7–15)
Prior EFT/ERT*	11 (10)		21 (21)		17 (15)		26 (23)		17 (15)
Data quality exclusions*	2 (2)		8 (6)		4 (3)		0 (0)		1 (1)

Note. Percentages do not always sum to 100% due to rounding; age and years of education are rounded to the nearest year, income to the nearest $1,000 increment, and survey duration to the nearest minute.

^*Percentage calculated based on group n prior to any exclusions.

Cue Effects on Discounting

Group was a significant predictor of indifference points (χ²[4] =23.67, p < .001; see Figure 3). Participants given cues containing both event and time components (EFT-Typical) had significantly greater indifference points (i.e., less delay discounting) than those in the EFT-Uncued (z = 2.75, p = .006) and ERT-Time (z = 3.42, p = .0006) groups. Indifference points were only nominally higher in the EFT-Typical group than in the EFT-Time group (z = 1.56, p = .12). There was also no significant difference in indifference points between the EFT-Typical and EFT-Event groups (z = 0.81, p = .42; see Table 4 for parameter estimates).

Figure 3.

Indifference points for each group of participants in Experiment 2. The height of the gray bars corresponds to the median indifference point, and the black error bars correspond to the first and third quartiles. Open circles show individual participants’ data. * p < .05; ** p < .001

Table 4.

Parameter Estimates and Test Statistics for the Beta Regression Analysis Predicting Indifference Points by Group in Experiment 2

Term/parameter	Beta	Std. error	Z	p	χ2	df	p
Group					23.67	4	<.001
Intercept	0.56	0.13	4.43	<.001
EFT-Event	0.15	0.18	0.81	.42
EFT-Time	−0.28	0.18	−1.56	.12
ERT-Time	−0.62	0.18	−3.37	<.001
EFT-Uncued	−0.49	0.18	−2.73	.006

Note. The intercept reflects the estimate for the EFT-Cued group. All estimates are the log-odds change in indifference points. The inverse logit can be used to obtain parameter estimates on the original scale of measurement: Exp(β)/[1+Exp(β)].

The cues designed to dissociate demand characteristics and episodic thinking had differential effects on discounting. Indifference points in EFT-Event (designed to isolate the theoretically relevant cue component), were higher than those in the EFT-Time (z = 2.32, p = .02; but only after removing 5 overly influential observations⁴), ERT-Time (z = 4.13, p < .0001), and EFT-Uncued (z = 3.48, p = .001) groups. Cues designed to isolate the component thought to engender demand characteristics (EFT-Time) produced no significant reductions in discounting relative to the control group (EFT-Uncued; z = 1.16, p = .25), nor did sham cues (ERT-Time vs. EFT-Uncued, z = 0.68, p = .50).

Cue Effects on Awareness and Relations to Discounting

Figure 4 shows the percentages of participants within each group who identified correctly, incorrectly, or failed to deduce (N/A) the hypothesis that EFT would reduce delay discounting. Across the cued groups (left of the vertical line), there were no significant differences in the percentages of correct and incorrect deductions (or lack thereof), χ²(4) = 5.97, p = .20. However, participants in the EFT-Typical and EFT-Event groups correctly deduced the hypothesis at greater than chance levels (dotted line; p =.009, 95% CI [.37, 1.00]; and p < .001, 95% CI [0.43, 1.00], respectively); while participants in EFT-Time (p = .17, 95% CI [0.30, 1.00]) and ERT-Time did not (p = .25, 95% CI [.28, 1.00]). In the absence of cues (EFT-Uncued), participants were more likely than chance to deduce the purpose of engaging in the EFT task prior to the discounting assessment (p = .01, 95% CI [0.37, 1.00]), but ~70% of these participants came to this realization only after being prompted to do so (see Figure 5; p = .004; 95% CI [.17, .43]).

Figure 4.

Percentage of participants identifying the hypothesis correctly, incorrectly, or not deducing any hypothesis, by group. The vertical line separates episodic future thinking (EFT)-uncued due to the difference in question (purpose of prior EFT, not cues). The dotted line indicates chance-level distribution of participants across categories. * p < .05. ** p < .001.

Figure 5.

Percentage of participants who deduced a hypothesis and reported doing so while completing the discounting task (in-task) or only when prompted (just now), by group. The vertical line separates episodic future thinking (EFT)-uncued due to the difference in question (purpose of prior EFT, not cues). The dashed line indicates chance-level distribution of participants across categories. * p < .05. ** p < .01.

There were trends in when participants in cued EFT groups deduced a hypothesis (whether correct or incorrect; see data left of the vertical line in Figure 5), although no binomial tests reached statistical significance. Participants in the EFT-Typical (p = .06), EFT-Event (p = .06), and EFT-Time (p = .07) groups showed a nominally greater than chance tendency to deduce the hypothesis during the discounting assessments. When combined across these three groups to increase statistical power, individuals were more likely to identify the hypothesis during the discounting task when EFT cues were present, but the effect was small (p = .003; 95% CI [.54, 1.00]). Similarly, those receiving sham cues (ERT-Time) were more likely than chance to deduce the hypothesis during the discounting assessment, but the effect was also small (p = .049, 95% CI [.501, 1.00])

Figure 6 shows indifference points collapsed across the two groups whose discounting was reduced by cued-EFT (EFT-Typical and EFT-Event). Data are subset by whether the hypothesis they deduced was correct, incorrect, or they did not deduce a hypothesis (N/A). In a beta regression, participants’ deductions were a significant predictor of indifference points, χ²(2) = 8.40, p = .02 (see Table 5 for parameter estimates). Specifically, those correctly identifying the hypothesis had greater indifference points than those who incorrectly identified it, z = 2.74, p = .006. However, correctly deducing the hypothesis did not significantly impact indifference points relative to those who did not identify one. Instead, those who incorrectly deduced the hypothesis had significantly lower indifference points relative to those who reported not deducing a hypothesis, z = 2.26, p = .02 (although this comparison was n.s. before the removal of 5 overly influential observations, z = 1.40, p = .16). This pattern of results was nominally the same (direction and significance) when conducting this analysis at the level of individual groups, with the exception that the comparison of indifference points across those incorrectly deducing the hypothesis and those not deducing a hypothesis was n.s. for the EFT-Event group alone.

Figure 6.

Indifference points for participants in the EFT-Typical and EFT-Event groups, grouped by the correctness of their identified hypothesis or lack of identification. The height of the gray bars corresponds to the median indifference point, and the black bars correspond to the first and third quartiles. Open circles show individual participants’ data. * p < .05. ** 0 < .01.

Table 5.

Parameter Estimates and Test Statistics for the Beta Regression Predicting Indifference Points by Hypothesis Identification in Experiment 2

Term/Parameter	Beta	Std. error	Z	p	χ2	df	p
Identification					8.40	2	.015
Intercept (Correct)	0.78	0.13	5.97	<.001
Incorrect	−0.60	0.22	−2.77	.006
Don’t Know/Unsure	−0.02	0.23	−0.06	.950

Note. Model includes data from EFT-Typical and EFT-Event groups only. All estimates are the log-odds change in indifference points. The inverse logit can be used to obtain parameter estimates on the original scale of measurement: Exp(β)/[1+Exp(β)].

Discussion

In support of the validity of EFT, cues designed for high demand characteristics (EFT-Time and ERT-Time) did not reduce delay discounting relative to the control group (EFT-Uncued). Only the EFT groups provided either typical cues (EFT-Typical) or cues designed for theoretical relevance (EFT-Event) had significantly lower rates of delay discounting. Although both of these groups correctly identified the hypothesis at greater than chance levels, decreases in discounting were not mediated by demand characteristics, as discounting was not significantly different between those who correctly identified the hypothesis and those who did not deduce one. In sum, Experiment 2 yielded no evidence supporting the hypothesis that demand characteristics are responsible for the effects of EFT on delay discounting.

Notably, the present results pertaining to awareness of the hypothesis in EFT research are in contrast to the findings of Stein et al. (2018) who found no relation between EFT-exposed participants’ awareness of experimental hypotheses and discounting. These discrepant results may be due to the different methods used to probe suspicion. As previously discussed, Stein et al. arranged open-ended questions and may have been liberal in scoring the correctness of participants’ hypotheses. The survey methods used in the present Experiment 2 drew participant attention to the cues provided during the discounting task and provided closed-ended multiple-choice answers that pertained to discounting. While we may be accused of “leading the witness,” three procedural components were designed to reduce this possibility. First, participants were instructed that, “by providing honest answers, [they] can help us ensure that psychologists use effective therapies when treating their patients, rather than ineffective ones.” Second, the “Neither/I’m not sure” response alternatives did not force participants into reporting deductions they did not make. Third, the survey was sensitive to the possibility that participants deduced a hypothesis only when prompted to do so.

General Discussion

The results of Experiments 1 and 2 revealed two important findings for understanding the effects of EFT and its validity for reducing delay discounting. First, Experiment 1 is the first to reveal that EFT reduces delay discounting only when cues are arranged in the discounting task; this finding was replicated in Experiment 2. Thus, a brief course of EFT should not be expected to generalize to novel behaviors or contexts. Although time spent engaged in the EFT task was either inconsistently or not predictive of discounting in either experiment (see Figures S-1 to S-9 in Supplementary Materials), longer exposure to EFT training that is systematically designed to improve generalization (see Stokes & Baer, 1977) might prove useful in increasing EFT in challenging decision-making settings. Future research should explore such training regimens.

Second, the results of Experiment 2 support the position that the effects of EFT on delay discounting are a direct result of its theoretically relevant ingredients. Namely, the future-thinking must be episodic in nature to reduce delay discounting. Only those participants cued to think episodically about the future event (EFT-Typical & EFT-Event) had lower rates of delay discounting than those in the control group. Although participants in these groups were more likely than chance to correctly deduce the experimental hypothesis, this awareness was not related to reduced delay discounting relative to those who deduced no hypothesis.

While these two findings are clear, unresolved is the question concerning an expanded temporal horizon as the mechanism by which EFT reduces delay discounting (Lin & Epstein, 2014; Snider et al., 2016). Cued EFT reduced discounting in Experiment 1 but had no expansive effect on temporal horizons. While this may be because cues were not presented during the temporal-horizon task, such a rationale cannot explain why temporal horizons were significantly shorter in the ERT-Cued group. Clearly additional research is needed to resolve this issue.

In the meantime, alternative accounts of how EFT reduces delay discounting should be considered. Perhaps EFT (1) enhances attention to episodic events rather than time (e.g., Radu et al., 2011), (2) reduces the perceived duration to the larger-later outcome (i.e., it alters time perception, as opposed to temporal horizon), or (3) makes future events as concrete as the smaller-sooner reward (see Kim, Schnall, & White, 2013). Given that the effects of EFT are often dependent on the vividness of prospection (i.e., the extent to which one is able to imagine the details of said event; Peters & Büchel, 2010; and the degree of emotional intensity provoked by the imagination of those events; Benoit et al., 2011) change in construal is a candidate account to be explored. Future research in EFT exploring these accounts may prove useful in the further development of efficient and effective EFT procedures.

Limitations and Future Directions

Two limitations of the present research are noteworthy. First, the population recruited from was exclusively users of Amazon Mechanical Turk; therefore, these results may not apply to other populations. MTurk workers differ from the U.S. population in that a greater proportion are female (Ipeirotis, 2010; Paolacci, Chandler, & Ipeirotis, 2010), they are younger, and have higher levels of obtained education (Yank et al., 2017). Despite these differences, delay discounting phenomena originally established in the lab such as the magnitude effect (Johnson, Herrmann, & Johnson, 2015), and steeper discounting with lower levels of education and income (e.g., Jarmolowicz et al., 2012) have been replicated among participants recruited via Amazon Mechanical Turk, and the effects of EFT on discounting have been replicated across various populations (e.g., in children, Daniel et al., 2015; adults from the general community, Sasse, Peters, Büchel, & Brassen, 2015; Amazon MTurk workers, Sze et al., 2017; and those with drug abuse/dependence, Snider et al., 2016; and obesity, Daniel, Stanton, & Epstein, 2013a).

Second, and relatedly, participants attending experimental sessions in-person are often more motivated than those recruited and completing research online (Paolacci et al., 2010). Given that motivational variables are important to consider in issues of demand characteristics (Goldstein et al., 1972; Orne, 1962) there may be merit in replicating some components of the present experiments during in-person sessions. Such research should include a post-experiment questionnaire to probe awareness of the experimental hypothesis. Because in-person participants will falsely deny their awareness of the hypothesis when questioned by the primary experimenter (Nichols & Maner, 2008; Taylor & Shepperd, 1996), a novel experimenter should administer the questionnaire (Goldstein et al., 1972). Similarly, the effect of demand characteristics appears affected by social variables not necessarily tied to motivations for participation (e.g., how much the participant likes the experiment and experimenter, whether the participant is single vs. in a relationship; Nichols & Maner, 2008). The influences of such variables may also be more prominent when the participant has direct contact with the experimenter or research team more generally (whether in-person or online).

Conclusions

Two key findings failed to support the hypothesis that the benefits of EFT are due to demand characteristics. First, only those cues high in theoretical relevance reduced delay discounting; cues designed to accentuate demand characteristics alone did not. Second, while cues high in theoretical relevance produced a greater awareness of the hypothesis, the reductions in discounting produced by said cues were not related to awareness of the hypothesis. Thus, while it is concerning that the positive effects of EFT do not generalize to discounting in the absence of cues, the results of Experiment 2 support the internal validity of EFT in reducing delay discounting. Prior published EFT studies, with a variety of populations, settings, and dependent measures help to establish its external validity, and support future research designed to enhance participants’ independent use of EFT in health-impacting decision-making contexts.

Public Significance Statement.

Prior research indicated that the effects of Episodic Future Thinking (EFT), a method for reducing impulsive choice and addiction-related behaviors, may be due to participants’ ability to discern EFT’s hypothesized benefits. The present findings suggest awareness of the hypothesis does not account for EFT benefits. Instead, they suggest EFT effects have limited generalizability—something that will need to be addressed in future research. Uncovering mechanisms of EFT may help in developing interventions with more robust clinical utility.