Contribution
It is the first systematic and comprehensive review of attempts to reduce delay discounting or impulsive choice by manipulating the environment. Earlier reviews were either not systematic or not comprehensive (Introduction). Its result is a map: which kinds of manipulation move discounting, roughly how far, and how little is known about why.
Key insight
Discounting is not fixed. It can be shifted by how a choice is framed, by what is cued or imagined, by how the choice is arranged in time, and most of all by training with delays and bundles. But nearly every study measures discounting as the outcome and almost none tests what carries the change. So the literature says that discounting moves, not how.
Assumptions
- Discounting as a target. The review works “under the assumption that steeply discounting future outcomes plays a role in the development and/or continuation of maladaptive behaviors” (Discussion). The introduction states that the evidence “favors (but does not establish)” an etiological role.
- Discounting describes, it does not explain. Footnote 1: “delay discounting is a description of a pattern of choices …; it is not an explanation”. Opportunity costs, collection risk, magnitude sensitivity and failures of response–outcome coupling can all produce it.
- Inclusion (Method). PsycINFO and PLoS One searched on 1 March 2016, plus references, giving 1,233 articles. Excluded were pharmacological and neural manipulations, studies without a control or baseline, ambiguous measures, manipulations impractical outside the laboratory (reward magnitude, sign, income, commodity), unresolvable statistical violations, and categories hypothesized to increase discounting. 92 papers remained.
- Effect size (Method). Hedges’ g from between- or within-subject designs, with within-subject variances computed assuming r = .5 between measurements. Values were sometimes read off graphs. These are “relative, as opposed to absolute” (Discussion). Mixed-effects models were run by category and subcategory. Cross-category comparisons, moderators other than category and publication bias were deliberately not analysed.
Key results
- Overall (Results). “Overall, the interventions were successful in reducing discounting, Q(9) = 224.68, p < .0001”, with heterogeneity I² = 64%, across nine categories. In the authors’ order these are clinical interventions, episodic future thinking, framing, perspective taking, priming, cueing, context, learning-based approaches and environmental enrichment.
- Clinical interventions (Results, Table 1). Twelve studies. Overall B = 0.23, SE 0.08, z = 2.78, p = .005, I² = 38%. “For most of these studies, delay discounting was not a primary target.”
- Mindfulness-based approaches (Results, Table 1). Two studies,
subcategory estimate 0.32 (SE 0.17), z = 1.89, p = .059: “the reduction
in discounting produced by mindfulness interventions only approached
significance”.
- Hendrickson and Rasmussen (2013), college students, mindful eating against a nutrition-information control: g = 0.27 for food (significant) and 0.16 for money (n.s.).
- Morrison et al. (2014; NOTE-1425), college students, ACT against waitlist: g = 0.94, with the between-groups comparison at trend level (p < .06) and the within-subjects reduction significant.
- The review asks for “better controls (e.g., sham-therapy in lieu of wait-list controls)”.
- Contingency management (Results). Three smoking studies, inconsistent alone, significant together (0.36, z = 2.39). Other substance-use treatments gave a model-estimated d = .16. “These findings offer no simple support for the position that drug abstinence would reverse” discounting, and “much needed mediation analyses” are asked for.
- Reward bundling (Results). “If a single choice determines not just the next outcome, but a bundle of the same outcomes … this should reduce impulsive choice”. Studies “typically” found reductions (z = 4.57): Kirby and Guastello 2001 (five-reward bundles reversed many participants’ preference); Ainslie and Monterosso 2003 in rats; Hofmeyr et al. 2011 in smokers, not nonsmokers. After experience with 9-reward bundles, rats chose impulsively less even when the outcomes were unbundled (Stein, Smits, et al. 2013). Learning-based approaches overall: B = 0.62, SE 0.08, z = 7.43, I² = 60%.
- Mechanism (Theoretical Issues). “Within most of the categories of intervention strategies reviewed above, there is no clear theoretical understanding of how the manipulation influences delay discounting. Inclusion of formal mediation analyses is rare.” Demand characteristics are named as the chief threat, especially “when control procedures do not equate for expectancy of change (e.g., wait-list control groups)”.
- Conclusions. Brief manipulations (nature cues, episodic future thinking, adding delays, date and explicit-zero framing) reliably move one-off choices but are not designed for lasting change. The “largest and longest-lasting effects were produced by learning-based manipulations” (Abstract). EFT could be built into ACT (Summary).
Concepts
- Delay discounting — “the devaluation of an outcome because it is delayed”, approximately hyperbolic across species (Introduction).
- Impulsive choice — choosing the smaller-sooner reward (SSR) over the larger-later (LLR); the self-control choice is the LLR.
- Reward bundling — making one choice commit the chooser to a series of the same outcomes, which by the hyperbolic form favours the LLR.
- AUC — area under the indifference-point curve, the theory-neutral measure the authors recommend (Methodological Recommendations).
Connections
It builds on the addiction literature that treats steep discounting as a “trans-disease process” (Bickel et al. 2012), and on Ainslie’s bundling, tested by Kirby and Guastello and by Madden’s own laboratory. It covers studies published before 1 March 2016. So it cannot include Dixon et al. (2018; LIT-906) or Ge et al. (2024; LIT-916), the two direct mindfulness studies CLAIM-156 cites.
Bearing on the record
- CLAIM-156: partly, and only on the claim’s first fork. The claim’s defeat condition is that mindfulness’s effect on self-control is “fully mediated by a change in measured discounting”. To test that, a study must measure discounting and choice under temptation and ask which carries the change. This review shows that no such study existed by 2016 for any manipulation: mediation analyses are “rare”, and the outcome is discounting itself. What it adds is the size of the discounting change. Mindfulness and ACT moved measured discounting only at trend level (0.32, p = .06, two studies, both in college students, one against a waitlist). That is weak evidence that mindfulness acts on the valuations. It is consistent with the claim’s route, which needs no change in discounting, but it does not show that route either, since no non-discounting outcome was measured. It does not reach LIT-906 or LIT-916, which postdate its search.
- THEORY-067 and CLAIM-169. Outside what the claim waits on, the bundling subsection is a read summary of evidence for the bundling mechanism these entries say is untested. Bundled choice reduced impulsive choice in humans and rats, and in rats bundling experience carried over to unbundled choice. This review reports it at second hand; the bundling studies themselves are unread.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Only part of the review was read (see the history): the categories other than clinical interventions and bundling are known only from the abstract, the overall result and the discussion.
- Category-wise pooling. Effects were pooled within categories, often from few studies (two for mindfulness), with effect sizes partly estimated from graphs and an assumed correlation of .5. No publication bias assessment.
- Mostly hypothetical monetary outcomes in human studies (Methodological Recommendations).
- Demand characteristics. The authors flag them for manipulations that overlap the discounting task and for waitlist controls, which covers one of the two mindfulness studies.