The study was conducted in Muheza District near the coast of NE Tanzania, a site of malaria research for a number of years and where P. falciparum transmission is intense and perennial with seasonal peaks. 15 The local economy based on subsistence farming and commercial fruit growing. Child mortality is typical of that in Tanzania generally at 176/1000 children in the first 5 years of life 16.

The survey lasted between April and May 2008 to coincide with the rainy season and thus peak seasonal malaria transmission. In order to ensure that villages were typical of the high malaria transmission known to occur in the area we identified 72 villages that had each contributed more than 10 admissions for severe malaria to the local district hospital where such admissions were documented for 1 year in 2006–7. The district town was excluded as were villages that were known to have been the site of ITN trials where nets had been freely distributed. From the 68 eligible villages, four were randomly sampled and from each of these villages 10 'balozis' (a Tanzanian 10-house unit for which a register is kept in the village office) were randomly sampled and all houses within the selected balozis were visited for the survey. With adjustment for clustering the sample size was calculated to estimate a 20% prevalence of ITN ownership +/-5% with 95% confidence.

The responsible adult present in the house was interviewed regarding the number of nets in the house, their source, price (converted to U.S. dollars using the exchange rate at the time of the survey) and whether or not they were treated with insecticide in the previous 6 months or were long-acting insecticidal nets. Nets were inspected wherever possible but evidence of whether they were insecticidal was taken from the respondent only. Data were also collected on house construction and household possessions to compile an aggregated score of socio-economic status. Data on individual use of a bed net the previous night was collected and a finger-prick blood sample was obtained from all available household members for a malaria rapid diagnostic test (RDT) using Paracheck™, an HRP-2 based test that has compared well with expert slide reading 17.

Data were double entered in MS Access (MS Corp, Redmond, Va) and analysed using Stata 10 (Statacorp, College Rd, Tx). A socio-economic score was derived using principal component analysis of household assets as described by Filmer and Pritchet (2001) 18. The first principal component accounted for 40.3% of the observed variation in household assets and, using this component, quintiles of a weighted score were derived from the 6 most correlated assets (non-mud floor, non-grass roof, radio, bicycle, mobile phone and other than locally made oil burner as a light source).

The study was approved by the ethical review boards of the Tanzanian National Institute for Medical Research and LSHTM. Written consent was obtained from all study subjects or their parent/guardian if under 15 years of age.

Household asset scores are shown in Table 1. In addition to individual variation the mean household asset score varied by village (range 2.6 to 3.7). Overall, 401(67%) households had any net and the mean number of any nets and insecticidal nets per household member varied by household asset score (Figure 1). The ratio of household ownership of any net between the poorest and least poor was 0.58. Using regression analysis with adjustment for clustering between balozis, there was a 19.8% increase in the number of bed nets per person for each unit increase in household asset quintile (p < 0.001) and a 13.4% increase in the number of insecticidal nets per person (p < 0.001).

The mean price of nets acquired within the last year was $1.9 while for nets that were 3 or more years old it was $2.6 and for the poorest households these prices were $1.2 and $2.4 respectively (Figure 2). The source of bed nets varied by household asset quintile, with nets that had been distributed through a non-commercial 'malaria campaign' making up a third of nets in the poorest households but only 5% of nets in the least poor households (Figure 3).

Just over half of the nets (335/570, 59%) that had been purchased from a shop were insecticidal and for nets that had been bought using a TNVS voucher this fell to 125/250(50%). However, 165/214 (77%) of nets distributed through a 'malaria campaign', whether subsidized or free, were insecticidal. Nets used by adults (over 15 years of age) were less likely to be insecticidal than those used by children under the age of 15 years (OR 0.66, p = 0.012) although this did not vary between younger (less than five years) and older (5–15 years) children (OR 0.98, p = 0.95).

Factors associated with whether nets had been treated with insecticide were assessed using a logistic regression model; nets that were insecticidal were more likely than non-insecticidal nets to have been donated or obtained at a subsidized price, to have been acquired more recently and a higher proportion of nets in the poorest households had been treated with insecticide (Table 2).

The reported travelling time to the nearest shop selling bed nets was relatively short; 60% of households were within a 20 minute journey and 82% were within a one-hour journey of a shop selling nets and net ownership of nets did not vary systematically by travel time.

The findings of this study, like those of Schellenberg et al 19, reveal that while rural African villages often have an appearance of uniform poverty there are definite socioeconomic differences that are reflected in marked differences in the use of health interventions. As so often with health risks and provision of services, the equity in coverage of ITNs was double-edged; the poorest households had the lowest level of ITN possession and, independent of bed net use, were at greatest risk of current or recent P. falciparum infection. True equity (equality of access according to need) would actually require greater coverage of ITNs among the poorest households compared to the least poor, the opposite to the difference that we observed.

The socio-economic differential in coverage was less marked for ITNs compared to untreated nets and this was associated with recently acquired ITNs and non-commercial distribution (free or highly subsidized). The TNVS is likely to have contributed to improved bed net coverage although over half of the nets acquired under the scheme were not impregnated. This is disappointing as, although the voucher can be used to buy any net, basic nets in Tanzania are 'bundled' with insecticide from the manufacturer and raise the possibility of interference in the packaging. There were also anecdotal reports that shopkeepers, when presented with vouchers instead of cash, tended to increase the base price of a net and it seems likely that the poorest households are particularly vulnerable to these sort of sharp practices as they are less mobile and may be less aware of their rights.

There was also evidence of inequity based on gender; while net coverage of young boys was only slightly higher than girls, the difference was statistically significant and reflects similar differences that have been observed in children with severe malaria 20.

Overall almost half of the nets in the study were not impregnated. While this may represent an improvement over past rates it still suggests a failure to realise much of the individual benefit of sleeping under a bed net 21. There is well-documented evidence of the mass effect of ITNs 22 and high levels of ITN coverage have recently been associated with marked reductions in the burden of malaria 2324. In this study, non-impregnated nets were associated with paradoxically lower risk of RDT-positivity compared to impregnated nets although the effect was not significant. It seems likely that this was the result of residual confounding not controlled for by our asset score and suggests that the excess risk of malaria in the poorest households may be even higher than documented by the study.

Thus the case for free insecticide treatment as a social good seems to be particularly strong. While future net sales are likely to be restricted to long-acting insecticidal nets, re-impregnation of the huge numbers of existing untreated nets is unlikely to be achieved through social marketing; Guyatt et al 25 found that in Kenya, while high rates of insecticide treatment could be achieved through providing free insecticide, the rates declined dramatically when charges were introduced.

Tanzania has pioneered initiatives in socially marketed nets and our study provides, albeit incomplete, evidence on its effect where most circumstances favour its success. Our results demonstrate significant progress since 2005 when a survey in the study area found that bed net use in young children was only 22%, only 10% of which had been treated with insecticide 26. Many of these gains must be credited to social marketing.

In an rural area of Tanzania where social marketing had been rigorously promoted, Nathan et al 27 documented that the proportion of the poorest households with a bed net increased from less than a quarter in 1997 to more than a half in 2000 and an increase the ratio of 'any net in the household' between the poorest and least poor from 0.3 to 0.6. Coverage in our study area five years later showed a strikingly similar pattern and suggests a lag of several years between areas with intense promotion and those subject to the consequent national roll-out that in our study area has been supplemented by non-commercial distribution that accounted for a third of ITNs in the poorest households.

These results suggest that, in spite of significant gains, social marketing as a single strategy is insufficient to meet current targets or to further reduce the inequity in ITN coverage amongst the poorest communities 28. In Tanzania mass free ITN distribution is planned starting in 2009 but the effect of sustained free distribution on the existing commercial outlets and reliance on the continued donor support both remain uncertain 29.