Two significant challenges face researchers tracking HIV-related socio-economic and demographic change over time in large cohort studies. Firstly, data collected in cohort studies established to describe the dynamics of HIV infection may contain no systematic data on household consumption expenditures which is an established measure of current and long-run household welfare. The second challenge is the choice of the unit of analysis in order to recognise and record impact; this is because most cohorts use the household as that unit. This means that the influence of factors outside that unit cannot easily be tracked. In this paper we show how a detailed understanding of the impact of HIV and AIDS on wider families and social networks, obtained through in-depth longitudinal research with a small number of households, can shed light on the findings from quantitative analysis from a larger cohort in the same population in rural Uganda. The findings of large-scale survey data from more than 2000 households over a 12-year period showed a lack of a strong association between poverty, HIV status and/or death of the household head. In-depth ethnographic research with 26 households in 1991/2 and a restudy of the same households in 2006/7 provide insights into the reasons for this finding: the choice of socio-economic indicators and support from other family and community members play a part in affecting survey findings on the impact of HIV at household level. One other factor is important in explaining the findings. HIV-infected family members from outside the household may drain resources from the household, so looking at the impact of HIV and AIDS on people's wider families provides pointers to why those who have not had an AIDS-related death in their own household may have failed to prosper. Our qualitative findings show that AIDS may well throw households into disarray and poverty, but more often reduces development and hinders families from getting out of poverty. Used strategically, small longitudinal studies can provide important information with which to explain patterns observed in large-scale quantitative datasets.