Table 3.

The simulated associations of serious mental illness with reduced earnings at the individual level among men and women separately in high-income and low- and middle-income countries^a

	Estimate (s.e.)
	High-income countries			Low- and middle-income countries
	Total	Male	Female	Total	Male	Female
Overall association
Association between serious mental illness and earnings in the total sampleb	0.32* (0.03)	0.53* (0.07)	0.19* (0.02)	0.33* (0.14)	0.29 (0.23)	0.35 (0.18)
Component effects
Effect of serious mental illness on probability of non-zero earningsc	0.14* (0.02)	0.16* (0.02)	0.14* (0.02)	0.05 (0.03)	0.07 (0.05)	0.05 (0.04)
Estimated effect of serious mental illness on earnings given non-zero earningsb	0.26* (0.04)	0.42* (0.08)	0.12* (0.03)	0.42 (0.28)	0.21 (0.30)	0.57 (0.44)
Decomposition of overall effectd
Due to difference in probability of non-zero earnings	0.39* (0.05)	0.31* (0.06)	0.55* (0.08)	0.27 (0.19)	0.50 (0.58)	0.18 (0.18)
Due to difference in earnings given non-zero earnings	0.49* (0.05)	0.56* (0.07)	0.36* (0.07)	0.66* (0.20)	0.45 (0.53)	0.75* (0.23)
Due to the interaction between the two components	0.12* (0.01)	0.13* (0.02)	0.09* (0.02)	0.07* (0.03)	0.05 (0.06)	0.08 (0.06)

a. High-income countries: Belgium, Germany, Israel, Italy, Japan, The Netherlands, Spain, USA, New Zealand; low- and middle-income countries: Brazil, Bulgaria, Colombia, India, Lebanon, Mexico, Nigeria, People’s Republic of China, South Africa.

b. The estimates reported in these rows summarise the results of individual-level simulations based on the coefficients in the best-fitting multiple regression model. (The coefficients from these models are not reported here, but are available from the authors.) That model was a generalised linear model that assumed a logarithmic link function between predictors and the outcome with prediction error variance proportional to the predicted values. A discussion of generalised linear model estimation is presented elsewhere.⁵ The simulation used the model coefficients to predict individual-level earnings twice for each respondent, once using the actual characteristics of the respondent and a second time based on the counterfactual assumption that none of the respondents had serious mental illness. Individual-level differences between these earnings estimates were averaged across all respondents with serious mental illness to estimate the expected mean individual-level decrease in earnings associated with serious mental illness in Part I of the current table. Standard errors were obtained by replicating the entire analysis in pseudo-samples using the method of jackknife repeated replication and using the distribution of estimates to generate an empirical estimate of the standard error.²⁸

c. The estimates reported in this row summarise the results of logistic regression analysis to predict any earnings v. no earnings.

d. Demographic rate standardisation²⁷ was then used to decompose the societal-level estimates into components due to the associations of serious mental illness with probability of having any earnings and with the amount earned by those with any earnings. A description of this method is presented elsewhere.⁵

*Significant at the 0.05 level, two-sided test.