LSE — Fundamentals of Behavioural Science | 2023
This essay develops a dual-process account of self-defeating behaviours — actions that agents undertake despite knowing they are contrary to their own interests. Drawing on Evans and Stanovich's (2013) Type 1/Type 2 framework and Thaler and Shefrin's (1981) doer/planner model, the essay identifies the cognitive mechanisms that generate self-defeating conduct. It proposes a typology of cost-effective behavioural interventions and evaluates the autonomy–welfare trade-off inherent in paternalistic policy design.
The dominant framework in behavioural science distinguishes two types of cognitive processing. Evans and Stanovich (2013) synthesise decades of dual-process research into a robust taxonomy: Type 1 processes are fast, automatic, and associative; Type 2 processes are slow, deliberate, and rule-governed. The distinction is not binary but describes a continuum of autonomy — the degree to which processing is independent of working memory capacity.
Self-defeating behaviour arises from the tension between these two systems. Type 1 processing generates impulsive responses aligned with immediate reward; Type 2 processing recognises the long-run cost of those responses but is frequently unable to override them. Understanding why override fails — and under what conditions — is the empirical key to designing effective interventions.
The American Psychological Association defines self-defeating behaviours as actions or attitudes that prevent people from achieving their own goals, that they are aware of, and that they continue despite this awareness. This deliberate-yet-persistent structure is precisely what makes them analytically tractable: they are not failures of information, but failures of impulse control.
Thaler and Shefrin (1981) model the individual as a principal-agent dyad within a single mind. The planner (System 2) has long-run preferences and sets rules; the doer (System 1) executes actions in each period, maximising myopic utility. Self-defeating behaviour emerges when the doer overrides the planner's prior commitments — a form of intra-personal governance failure.
This model explains why commitment devices are effective: they are technologies that strengthen planner authority before temptation arises. Automatic pension enrolment, for instance, removes the doer's moment of discretion entirely — the planner's preferred outcome is achieved without requiring System 2 to win a within-period contest against System 1.
A particularly damaging class of self-defeating behaviour is the negative self-reflection spiral. Callan, Kay and Dawtry (2014) show that failure experiences trigger self-critical evaluation, which depletes cognitive resources, which increases reliance on System 1 heuristics, which produces further failures — a self-reinforcing loop. Beevers (2005) links this mechanism to depression: ruminative self-focus activates mood-congruent memories, suppresses corrective System 2 appraisal, and deepens negative affect. Interventions must break the loop, not just address the proximate failure.
The essay develops a two-dimensional typology of behavioural interventions based on (i) which cognitive system is targeted, and (ii) who deploys the intervention. This yields four cells, each with distinct welfare and autonomy implications.
Targets System 1 without agent involvement. Highest effectiveness, greatest autonomy cost.
Engages System 2 via external incentives or information. Effective when System 2 capacity is intact.
Agent trains automatic responses in advance. Preserves autonomy; requires initial System 2 investment.
Strengthens System 2 control capacity directly. Highest autonomy compatibility; requires sustained engagement.
Other-deployed interventions — particularly automatic defaults and bans — raise a fundamental normative question: when is it legitimate for a policy designer to override an agent's revealed preferences in the name of that agent's welfare? This is the classical liberal objection to paternalism, sharpened by the dual-process insight that "revealed preferences" may reflect System 1 impulses rather than the agent's own considered values.
Self-defeating behaviours, by definition, are contrary to the agent's own interests. The planner — the agent's deliberative self — would endorse the intervention. Tobacco taxes are welcomed by many smokers precisely because they strengthen the planner against the doer (Gruber & Mullainathan 2006). Where System 2 is systematically outmatched, welfare-enhancing intervention is justified.
Individuals may have legitimate higher-order preferences to experience and manage temptation without external interference. Paternalism assumes the policymaker knows the agent's true preferences better than the agent does — epistemically and normatively suspect. Coercive interventions like bans foreclose minority preferences that are not self-defeating for all individuals.
The essay concludes with Thaler and Sunstein's (2003) libertarian paternalism as the most defensible resolution. A policy is libertarian paternalistic if it steers choices towards the planner's preferred outcome while preserving the agent's freedom to opt out. Defaults achieve this: automatic pension enrolment is the canonical example — it serves the planner's long-run retirement preferences while allowing the doer to opt out at any time.
This approach is superior to both pure libertarianism (which ignores the systematic disadvantage of System 2 in within-period conflicts) and hard paternalism (which removes choice entirely). It is also cost-effective: by targeting architecture rather than prices or bans, it can achieve welfare gains without the distributional costs of taxation or the freedom costs of prohibition.