Authors’ Response


Dear Sirs,

We are pleased to have the chance to respond to the questions raised in the letter by Muula referring to our article about the influence of neighbourhood characteristics on HIV risk in young women in Ndola. Given that the analysis used data that were not collected for this purpose, there are obviously limitations to what we could do. However, we believe that our results highlight the importance of considering context effects and we hope to trigger a discussion around these issues, which will lead to the availability of better data in the future.

For this analysis we grouped the original 32 census-tract based sampling clusters into 16 geographically defined neighbourhoods. Neighbourhood was therefore based on area of residence. While people may meet sexual partners outside this area, the area of residence is likely to influence many aspects of their lives.

Circumcision levels ranged from 0% to 25% between neighbourhoods (none had exactly 8%). We chose 8% as a cut-off as this resulted in about equal numbers of young women in each group.

The original questionnaire asked each household whether the following facilities were available in the cluster or not: all-weather road, health centre/ hospital, trading centre/market, electricity, running water, hotel/bar/boarding house. However, in practice, the response varied within clusters, suggesting that the answer reflected availability near the household rather than the cluster as a whole. Since this was not the focus of the original study, no information on distance was collected. For this analysis we defined uniform availability of water or electricity in a neighbourhood as everybody in the neighbourhood having it. Any misclassification is likely to have underestimated neighbourhood-level effects.

We did not attempt to identify the exact pathway of electricity or water availability on HIV risk. Rather we used those variables, together with three aggregate variables (education, occupation and employment levels among adults over 25 years of age), to build a neighbourhood socioeconomic status (SES) construct. We are well aware of the limitations of this construct, but believe it may capture some important contextual effects, which is supported by the strong effects found.

We did indeed attempt to explain how neighbourhood SES and proximity to a market or health centre exert their effect on HIV risk in young women by examining 1) their relationship to the proximate determinants (including behavioural factors) and 2) how their odds ratios are affected by controlling for variables thought to be on the causal pathway.

While some possible mediating factors were identified for SES, none of the data available on proximate determinants could explain the effects of market and health centre proximity. The latter could possibly be mediated by access to treatment for treatable STIs or be a proxy for other uncaptured aspects of the neighbourhood. We provided some suggestions in the discussion on how omitted factors and problems in measurement could have affected our ability to detect pathways.