Mind Over Money Trouble

Abstract

Households exposed to identical negative life events (a job loss, a divorce, a hospitalised illness) diverge sharply in their financial-hardship trajectories within twelve months. Some recover within a wave; others remain in hardship for the next three or four. Income decile, education, and pre-event wealth explain only part of the spread. The behavioural-economics literature on non-cognitive skills predicts that part of the residual reflects measurable differences in **locus of control**, the durable psychological orientation toward attributing outcomes to one's own actions versus to external forces.

This paper investigates locus of control as **psychological insurance** against the financial consequences of an exogenously timed adverse event, reducing both the probability of entering hardship and the persistence of hardship spells once they begin. The construct is measured via the Pearlin Mastery 7-item scale, fielded in panel waves 5, 9, 13, and 17. The behavioural mechanism: internal-locus households are more likely to engage in proactive search, planning, and recovery behaviours after a shock, shortening both the depth and duration of the hardship spell.

The existing analysis applies the classical identification stack: difference-in-differences around event timing, instrumental-variable estimation, panel fixed-effects, propensity-score matching, Oaxaca-Blinder decomposition, principal-component analysis on the Pearlin items (confirming the established single-factor structure), probit estimation, and Cox proportional-hazards modelling for hardship-spell duration. The existing classical evidence already points to internal locus of control as both reducing entry into hardship and shortening recovery time after a shock.

We propose extending the existing analysis with two ML layers: a Causal Forest that would recover the conditional locus-of-control buffer by life-event type, distinguishing the buffer's strength for employment shocks from health shocks, and a Random Survival Forest that would estimate the hardship-spell duration without the proportional-hazards constraint. The proposed extensions are the natural complement to the existing classical work; the writing-up phase is the bottleneck, not the underlying analysis.

Data & Methods