Effect of Years in Retirement on Cognition, Considering Non-ignorable Missingness in Outcome

Abstract: Introduction: Research has suggested a positive association between cognitive decline and years in retirement. However, most prior studies are based on data with non-ignorable missingness in the response variable and performed with standard regression methods, requiring a missing at random assumption of the response. Hence, there are reasons to investigate if the association retains in case of nonignorable outcome missingness. An alternative to standard regression methods is based on identification sets, where the missingness mechanism is quantified by a parameter and hence considered non-ignorable. Objectives: The primary objective of this study was to investigate the connection between years in retirement and cognition based on SHARE data, controlling for age, physical health, early life conditions and socio-economic status, while considering non-ignorable missingness in the outcome. Methods: This study was based on data from the SHARE project and included a total of 9199 individuals, aged 50-99 years. Episodic memory was measured by a ten-word-list learning test, of which results were forming the cognition outcomes. Standard OLS regression and a regression method based on identification sets were used for the analysis, the latter method allowing for a missing not at random mechanism. Results: In line with previous research, a negative effect of years in retirement on cognition was found. A new result of this study was an obtained effect even when considering non-ignorable missingness in the outcome. Conclusion: The previously found connection and negative effect of years in retirement on cognition is further strengthened, since it also applies in case of non-ignorable missingness in the outcome.