Addi van Bergen

Chapter 2 36 Risk of bias within studies In line with previous reviews we found almost no study using a valid measure for SE/ SI. Most studies used self-chosen indicators and in nearly half of the studies the data were not originally designed to measure SE/SI. The lack of valid measures for SE/SI prevents very firm conclusions being drawn from this review. We agree with previous reviews that the development and use of validated multidimensional measures [12, 13, 16, 56] is warranted in future research. As SE is a multi-interpretable concept that can be operationalised in various ways depending on one’s theoretical perspective, political positionandpurpose, it isnotobviousthatoneagreeduponmeasureforSE/SIwillsurface. This need not be problematic as long as choices are explicated and substantiated. Finally, seven of the 22 studies did not adjust for demographic and other potential confounding factors. As confounding may affect the results of our review through over-estimation, the evidence was also analysed without these seven studies and the inferences remained unchanged. It is important to note that in all observational studies, residual confounding may account for part of the associations observed. Strengths and limitations of this review The principal strengths of this review are its systematic approach, tactical search strategy, and clear conceptual framework. These made it possible, despite the great diversity of studies, to take a step further than previous reviews, which did not report on the relationship between SE/SI and health, or did not do this systematically, but merely as exemplary descriptions [2, 12-16]. Another strong points is the inclusion of papers in languages other than English. There are limitations too. The method we used to summarise the evidence is based on p-values. P-values give an indication of the compatibility of the data with the null-hypothesis of each paper, and not of the effect size or the importance of the results [57]. To enable interpretation of the results we reported for each paper effect sizes and/or other statistics in the tables and provided some qualitative context in the main text. Another limitation arises from the classification of health outcomes, which was not always straightforward. In a number of studies no clear distinction could be made between MH and PH components, for instance, when researchers considered other, non-congruent, classes of diseases. As these results were classified as GH, this category may have become somewhat ambiguous. Yet another limitation is that our review is not exhaustive. The downside of applying strict selection criteria is that e.g. studies not using a multidimensional construct of SE/SI were left out. As a previous review using a comprehensive search strategy yielded unmanageable amounts of 100,000 plus titles [14], we choose a narrow search strategy to identify papers that focus specifically on SE/SI and not on related subjects such as income, housing, social cohesion etc. This way we may have missed relevant papers not using the specific terms SE or SI in title or abstract, but we do not expect there to be many, for two reasons. First, studies on only one dimension of SE/SI or on a constituent element,