Multilevel, individual and ecological linear regression analysis of systolic blood pressure (SBP) in 25 000 people living in the 39 neighbourhoods (hypothetical data)
| Multilevel regression analysis (empty model) | Single level individual analysis* | Single level ecological analysis† | |
|---|---|---|---|
| “Fixed effects” and “random effects” are expressions that are often used in MLRA. In very simple terms, fixed effects are used to model means whereas random effects are used to model variance. *In the individual level analysis we ignore the existence of the neighbourhood level. †In the single level ecological analysis we aggregate the individual level information on SBP to the neighbourhood level, neglecting the existence of the individual level. ‡The deviance expresses the goodness of fit of the model, and was calculated using the −2×log likelihood (IGLS). The Bayesian deviance information criterion (DIC) gave similar results. The multilevel models fit the data much better than the single level individual analysis. The ecological analysis is not comparable as it is based on 39 neighbourhoods rather than on 25 000 people. We refer elsewhere for an extended explanation of the deviance statistic.44,45 | |||
| Fixed effects | |||
| Mean SBP of the city | 130.2 | 130.2 | 130.0 |
| Random effects | |||
| Multilevel measures of health variation | |||
| Variance (SE) | |||
| Between neighbourhoods | 36.2 | – | 37.0 |
| Between individuals | 433.4 | 468.1 | |
| Intraclass correlation (ICC) | ICC = 0.08 | – | – |
| Deviance (goodness of fit of the model)‡ | 222764 | 224532 | – |