Addi van Bergen

Chapter 7 148 Limitations In this study, we faced several limitations. First, the current state of the relevant research did not allow us to quantify the strength of the association between SE and specific health outcomes. The method we used in chapter 2 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 manuscript but not of the effect size or the importance of the results. Due to the great diversity in health outcomes, we classified them into broader groups: mental health, physical health and general health. The classification was not always straightforward, particularly not for general health. Second, we have to mention potential bias due to selection in the studies in chapters 4, 5, 6 and 7. Persons without a fixed address or living in an institutional setting were a priori excluded from the sample. This group is estimated at approximately 0.2% and 1.6-1.8%, respectively, of the Dutch adult population (CBS Statline). As these tend to be vulnerable people with a high risk of SE, such as people experiencing homelessness, incarcerated people and frail, older people, this may lead to an underestimation of the prevalence of SE in the population. Selective non-response is another potential source of bias. In the Netherlands, the response rates in survey research are low and have decreased over time [19, 20]. Despite the use of strategies to reduce non-response rates concentrated on hard-to-reach groups and despite oversampling in deprived neighbourhoods and weighting to adjust for non-response bias, the possibility of some bias cannot be ruled out. The PHM is no exception: average response rates in the G4 declined from 50% in 2008 to 33% in 2016. Again, the tendency is towards an underestimation of the SE prevalence rates. Third, the classification of the SEI-HS index and dimension scores into categories involved a certain degree of arbitrariness. SE is a continuous phenomenon with no natural boundaries between being excluded or not or between some, moderate and strong exclusion. The main reason for classifying the SEI-HS was to enhance its applicability in public health policy. Policymakers require clear and simple data, and continuous scale scores will not do. We opted for the use of 85th and 95th percentile values in the Dutch adult population as cut-off scores. These fit the right-skewed distribution of the index and dimension scores, with the largest part of the population having low scores, a small part having very high scores, and a modest group in the middle. Our choice is also in line with the cut-off point of 1 SD above the mean used by Gijsbers [21] to define social exclusion. Fourth, widespread research across the Netherlands allowed us to extend the generalisability of the SEI-HS to the whole Dutch adult population, both urban and rural, but the generalisability to populations in other countries may be limited. The items of the SEI-HS measure aspects of SE in the Dutch context. Bottle banks, for example, are unknown in large parts of Turkey, and in southern countries such as India, the item “I have enough money to heat my home” is irrelevant. In low- and middle- income countries, items such as access to electricity, pipe water and sewerage as well