5 Reverse Causality?
The results above show that behavioural traits are an important determinant of who experi-
ences financial distress. However, if the behavioural traits are correlated with unobservables
that cannot be controlled for in the model, the measured effects might not be capturing the
true causal relationship between behavioural traits and financial distress. I consider this
issue for impulsiveness, which is the variable that has the largest effect in the probit re-
gressions. In this case, it might be argued that while impulsiveness increases the likelihood
of getting into financial difficulties, being in financial distress might cause respondents to
report that they’re impulsive. In order to assess this issue further, I employ instrumental
variable analysis.
The instrumental variable approach used here is motivated by Lusardi and Mitchell
(2007) who assess the effect of planning on net worth. In order to examine this issue, the
authors run a regression where net worth is the dependent variable and planning is one
of the independent variables in their model. In testing if reverse causality is a problem,
they run a ‘reverse’ regression where planning is the dependent variable and net worth is
one of the independent variables, and they instrument for net worth. Their results show
an insignificant coefficient on the instrumented net worth measure in their IV regression,
suggesting that reverse causality is not a problem and that their original results hold.
Following Lusardi and Mitchell’s methodology, I run a ‘reverse’ probit regression, where
impulsiveness is the dependent variable and independent variables include the same demo-
graphic and economic variables included in the previous models, plus the instrument for
financial distress - “Struggle to Keep Up”. I instrument for financial distress using infor-
mation on whether or not respondents have a long-standing illness. Twenty per cent of the
sample claim to have a long-standing illness.
The results are shown in Table 10. The IV regression shows no significant coefficient
on the financial distress variable, which would suggest that reverse causality in relation to
impulsiveness is not a problem in my regressions.11
11The first stage fit in the regression is good with the illness variable being statistically significant at the
1 per cent level.
27