This text is a personal account of the research process that led to a dissertation about the long-term relevance of social and emotional factors for cognitive health. It describes how research ideas and methodological concerns developed out of the first results where people who had lived alone, especially after being widowed already in midlife, had a high risk increase for Alzheimer’s disease over two decades later. A second study investigated the associations between feelings of hopelessness in midlife and cognitive health in later life. Both of these studies used an epidemiological association approach with logistic regression as the main statistical method, including adjustments for several relevant variables. To address the methodological concerns in the epidemiological association approach, primarily the risk of reverse causation and confounding variables, a subsequent study applied a randomized control experimental design. In this study, the effects on brain-derived neurotrophic factor levels from different types of activities were measured in healthy elderly persons through a within-subject cross-over design. The case study illustrates the compromises that had to be made through this methodological shift in relation to the original research question—compromises in terms of real-life outcomes and exposures in exchange for experimental control and establishment of causality. The relative merits and limitations of three main approaches are finally summarized in light of these experiences: experimental studies on animals and on humans and epidemiological associations in humans. The examples are from research on dementia, but should be relevant for many other research questions.