We thank Ghosh and Varadharajan for the attention they have given to our paper ‘Neighbourhood characteristics and prevalence and severity of depression: pooled analysis of eight Dutch cohort studies’.^{Reference Generaal, Hoogendijk, Stam, Henke, Rutters and Oosterman1} They raised some concerns about the multilevel regression analyses conducted for each of the eight contributing studies that were pooled in a meta-analysis.^{Reference Ghosh and Varadharajan2}

First, Ghosh and Varadharajan point out the potential issue of multicollinearity when highly correlated variables are entered within one regression model. However, this appears to be based on a misunderstanding. As indicated in the Method, all models run in our paper are univariable analyses performed for each neighbourhood variable in separate regression models. It is indeed true that different neighbourhood characteristics are moderately to strongly correlated. That is exactly the reason why we decided not to run multivariate analyses in which all environmental characteristics are entered within one model. By analysing them separately, we prevented the risk of multicollinearity and we provided better insights into which environmental characteristics are and are not associated with depression. We believe epidemiological studies that consider and compare multiple environmental characteristics, instead of focusing on only one characteristic, are very much needed as these give us a fuller understanding of the exposome relevance for mental health.

Second, Ghosh and Varadharajan indicated that non-normal distributions of depressive symptoms are not ideal for regression analyses. Indeed, in some of our cohorts the depressive symptom score was a bit skewed. However, we do not believe this has had impact on our overall results and conclusion. It is important to point out that the continuous depressive symptom score was only an outcome used in secondary sensitivity analyses. Our primary outcome measure was a dichotomous indicator of yes/no reporting significant depressive symptoms. Findings of secondary sensitivity analyses with a continuous outcome were very similar to that of primary analyses with a dichotomous outcome. In addition, as the skewness of the depressive symptom score was an issue in some but not in other cohorts, if this would have an impact, one would expect to see differences in associations across studies. However, our heterogeneity analyses showed in fact rather low heterogeneity in most results across the eight cohorts. So, we feel that also this issue did not impact on our results, which indicate – in a large-scale pooled analysis – that urbanisation and various socioeconomic, physical and social neighbourhood characteristics are associated with depression.