Scottish Health Survey - topic report: The Glasgow Effect

Appendix 5: Overweight model development for men

The McFadden pseudo R ²s given below are the results when the same sample of 651 is used in each model.

In the original logistic regression model containing only age and residence (Model 1) residence was not significant with the reduced sample, however it was when the full available sample was used (odds ratio of 0.78). McFadden's pseudo R ² for this model was 0.054. Similarly residence was not significant when SIMD was added to the model for the restricted sample (Model 2), but was when the full sample was used (odds ratio of 0.80). This model had a McFadden's pseudo R ² of 0.066, showing it to be a better model than the Model 1.

When the socio-economic variables had been added (Model 3) the variables remaining in the model when all available data were used were: age, residence, equivalised income and marital status; however of these variables only equivalised income and marital status were significant with the reduced sample. The odds ratio for residence in Greater Glasgow and Clyde was significant when the full available sample was used, and the odds ratio was 0.80. McFadden's pseudo R ² for this model was 0.118, showing that this model provides a better fit than the previous models.

When the behavioural variables had been added (Model 4) the variables remaining in the model when all available data were used were: age, residence, marital status, smoking status, binge drinking and physical activity level; however of these variables only marital status was significant with the reduced sample. The odds ratio for residence in Greater Glasgow and Clyde was 0.81 when the full available sample was used. McFadden's pseudo R ² for this model was 0.117, showing that this model is a slightly worse fit than Model 3.

When the biological variables had been added (Model 5) the variables remaining in the model when all available data were used were: equivalised income, marital status, smoking status, cholesterol level and C-reactive protein. HDL cholesterol was not added to the model as only 34 men had low HDL, all of whom were overweight, causing a separation in the data. As it is not possible to re-run this model using a larger sample due to the inclusion of blood analytes, it is not possible to know whether, as in the earlier models described, residence in Greater Glasgow and Clyde would have been significant in a larger sample. McFadden's pseudo R ² for this model was 0.196, showing this model to fit the data best out of the models described here.