This document shows the results of within-between models for the various psychological outcomes based on the data up to the date shown above. It will update dynamically as we get more data.

We have two versions of the non-committed income variable: calculated (what we calculate by subtracting reported housing, energy, water and council tax costs from what the participants tell us they brought in last month); and reported (the estimate that participants give us of how much money is left over in their bank accounts each month after paying their essential costs). The two income measures are correlated at: 0.62. We present modelling results using both.

We have four psychological outcomes: GAD (anxiety); PHQ (depression); time discounting (a measure of preference for smaller sooner over larger later rewards); and risk preference (a measure of preference for larger gambles over smaller safe options.)

For each outcome, there is potentially a between-subjects effect (e.g. if people whose average incomes over the study are lower are more anxious on average over the study, then there is a between-subjects effect of income on anxiety) and a within-subjects effect (if, in a month where a person’s income drops relative to the previous month, their anxiety goes up relative to the previous month, then there is a within-subjects effect of income on anxiety). We test both effects in the same model. We pool the two countries for the sake of this analysis, though there is a random effect of country to account for different baselines in the two countries.

For anxiety and depression, both the within- and between-subjects effects are really clear. The within-subjects effects are weaker than the between-subjects ones. For time discounting, we have a between-subjects effect but not a within-subjects effect; and for risk preference, we have no association with income.

In the following tables, the coefficients reflect the difference in anxiety or depression score per thousand euros difference in income.

Results with the calculated income variable as predictor

Variable Comment Between_effect Within_effect
GAD Higher = more anxious -0.158 ( 0.043 ), p = 0 -0.017 ( 0.007 ), p = 0.018
PHQ Higher = more depressed -0.154 ( 0.041 ), p = 0 -0.012 ( 0.007 ), p = 0.083
Time Higher = more impatient -0.362 ( 0.07 ), p = 0 -0.034 ( 0.014 ), p = 0.013
Risk Higher = more risk seeking 0.102 ( 0.042 ), p = 0.016 0.007 ( 0.013 ), p = 0.556

Results with the reported income variable as predictor

Variable Comment Between_effect Within_effect
GAD Higher = more anxious -0.244 ( 0.049 ), p = 0 -0.054 ( 0.013 ), p = 0
PHQ Higher = more depressed -0.242 ( 0.046 ), p = 0 -0.046 ( 0.012 ), p = 0
Time Higher = more impatient -0.49 ( 0.079 ), p = 0 -0.041 ( 0.024 ), p = 0.094
Risk Higher = more risk seeking 0.133 ( 0.048 ), p = 0.006 0.034 ( 0.022 ), p = 0.126

Here are the graphs to accompany the regression models shown above. The blue lines show the overall relationship in the sample; the small grey lines, the regression relationship for each individual participant. First, for anxiety score:

And depression score:

Subjective economic risk

We can perform a similar exercise with subjective economic risk. Here, the coefficients are in the opposite direction (positive, because higher subjective risk is associated with more anxiety and depression). Subjective economic risk is negatively correlated with calculated non-committed income at r = -0.24 and with reported non-committed income at r = -0.31.


The data already show very strong evidence that subjective financial risk is strongly associated with depression and anxiety at both the between and within-subjects levels. For time preference, subjective financial risk is a predictor at the between subjects but not within subjects levels. Subjective financial risk does not predict risk preference.


Variable Between_effect Within_effect
GAD anxiety 0.646 ( 0.045 ), p = 0 0.086 ( 0.009 ), p = 0
PHQ depression 0.617 ( 0.043 ), p = 0 0.079 ( 0.008 ), p = 0
Time discounting 0.53 ( 0.085 ), p = 0 0.019 ( 0.017 ), p = 0.252
Risk preference -0.062 ( 0.051 ), p = 0.225 -0.015 ( 0.015 ), p = 0.319

Here are the corresponding graphs, first for anxiety:

And depression: