LAUSR

Two articles recently published in the International Journal of Epidemiology came to different conclusions about the association between parental income and off-spring’s mental health diagnoses. Using panel data analysed with logistic regression and adjustment for observed confounders, Kinge et al. 1 reported a negative association in Norwegian data. Sariaslan et al. 2 implemented a discordant siblings design study and stratified Cox regression to reduce bias from unmeasured confounders (but also see Frisell et al . 3 ) and concluded that there is no causal association in Finnish data. The two articles describe their results in associative vs. causal language, but both studies ultim-ately report associations whose causal interpretation depends on untestable assumptions, and associations from both studies can be seen as potentially biased estimators of the causal effect in target population. To investigate if the diverging results are due to different study populations, study designs, statistical approaches or definitions of exposure and outcome, we analysed data from Norwegian children with Scandinavian parents born between 1997 and 2012. We implemented the analytical approaches used in the two articles and a conditional logistic regression model 4 that investigates effects of year-to-year income variation while controlling for unobserved unit-level heterogeneity. We used income percentiles as exposure, and diagnoses registered in the National Patient Register from child age 5 years on until child age 21 or the year 2017 as outcomes. The sample included 557 056 children with a prevalence of 4.8%, 2.1% and 4.6% for at-tention-deficit hyperactivity disorder (ADHD