The Data set
The paper is based on the 2000–2001 Uganda Demographic and Health Survey (UDHS). The Demographic and Health Surveys are usually designed to collect information on demographic, health and family planning. The 2000–2001 survey was expanded in scope to include a detailed module on malaria. This module elicited information on the ownership and use of mosquito nets, and detailed treatment seeking behavior for malaria . The data is also publicly available at http://www.measuredhs.com.
Sample design and implementation
The sample consisted of a two-stage design. The first-stage sample frame was a list of Enumeration Areas (EAs) compiled from the 1991 Population Census. In this frame, the EAs were grouped by parish within a sub-county, by sub-county within a county and by county within a district. A total of 298 EAs (102 in urban areas and 196 in rural areas) were selected. Urban areas and districts included in the Delivery of Improved Services for Health (DISH) project and the Community Reproductive Health Project (CREHP) were over-sampled in order to produce estimates for these segments of the population. Within each selected EA, a complete household listing was done based on the 1991 census and updated in the field to provide the basis for the second-stage sampling. The number of households to be selected in each sampled EA was allocated proportionately to the number of households in the EA. However, due to insecurity, four out of 45 districts with approximately five per cent of the population were not included in the survey. A total of 8,792 households were selected in the sample, of which 8,234 were occupied. The short fall was largely due to structures that were found to be vacant. Of the existing households, 7,885 were successfully interviewed, yielding a household response rate of 96 per cent. About 42 per cent of the households had no children under-five years, and in those that did, the average was close to two children (1.8) per household, and the gross (for all households) average was one child per household.
Three questionnaires were used; the household's, the women's and the men's questionnaires. The household questionnaire was used to list all usual members and visitors in the selected households. Some basic information on sex, age, education and household characteristics, such as source of water, toilets, housing characteristics and ownership of various durable commodities was collected. Of particular interest to this paper, the household questionnaire collected information on whether the household owned a mosquito net and if children in that household slept under a mosquito net. The women's questionnaire was administered to all women aged 15–49 years of age. The women were asked questions related to their background, reproduction and sexual activity, and maternal and child care. For each surviving child born in the preceding five years, detailed information on mosquito net use, their illness and treatment seeking behavior was collected. The men's questionnaire was administered to all men aged 15–54 years of age living in every third household in the UDHS women sample. The men's questionnaire collected information similar to the women's questionnaire but omitted reproductive history and child care.
Individual child characteristics, those of the parents and the household are likely to determine the use of mosquito nets in children under-five years old. These characteristics included the age and sex of the child; those of the parents included their education level, employment status, and exposure to information on the usefulness of mosquito nets, mainly through the media. The household attributes likely to affect mosquito net use in under-five year olds are its income or socioeconomic status, who in the household takes final decisions on issues of expenditure, particularly in relation to health care, whether the household is located in a rural or urban setting, the number of children under five years of age, and whether the household is headed by a male or female. With regard to malaria transmission, the altitude and region where a household is located is also likely to be a determinant.
In this paper, the education level has three categories, 1) no formal education, 2) primary, and 3) secondary and post secondary levels, which correspond to zero, 1–7 years, 8–11 years and more than 11 years of formal schooling. The wealth index was used as a proxy for socioeconomic status of a household. Using factor analysis, the most important household owned durable goods were identified. These were used to categorize households into quintiles by socioeconomic status . For example, ownership of a radio or television is a measure of access to mass media; access to a telephone measures access to efficient means of communication; cupboard and refrigerator ownership indicates the capacity for hygienic storage of foods and utensils; lantern ownership indicates a source of lighting; and ownership of bicycle, motorcycle, boat/canoe, or private car shows the means of transport privately available to the household. Ownership of these items, in turn, has a bearing on the household's access to information and health care – including mosquito nets. Districts included in the survey were categorized as 1,000–1,500 metres, 1,500–2,000 metres, and 2,000–3,000 metres of altitude.
Household mosquito net ownership and use among the under-fives were first explored using descriptive statistics. Then multivariate analysis was used to examine differentials in mosquito net use. Where the outcome of interest is a binary choice variable, as in this case of whether a child usually sleeps under a mosquito net, estimation was by the logit model, primarily due to the desire to obtain estimates of odds ratios. [Estimation with the probit model, not unsurprisingly, makes no meaningful difference to the results, and accordingly, only results from the logit estimation are reported.].
In the model specification, whether children usually sleep under mosquito nets is considered, regardless of the net treatment status. This is because first, the number of children with treated mosquito nets is so small and second increasing mosquito net usage is a precondition for increasing mosquito net treatment. Once there is a mosquito net in the household, whether children sleep under it or not is no longer a question of the price rather personal and behavioural factors of both parents, in addition to characteristics of an individual child and the household.
Considering the characteristics of a child first, there is evidence in the literature, especially from South-East Asia that females are at a disadvantage when it comes to household resource allocation. [10, 11]. The age of a child is another factor that possibly could determine whether a child sleeps under a mosquito net. The hypothesis is that if there is more than one child in a household, then preference would be given to the young one. In a recently completed study in Uganda , a number of advantages were given in favour of having under-fives sleeping under a mosquito net, avoiding being bitten by mosquitoes (81%), avoiding getting malaria (62%), keeping warm (22%) and avoiding other pests (17%). Clearly, from the advantages mentioned, they all had to do with protecting the young child. It would, therefore, be reasonable to assume that the young child has preference.
Maternal characteristics which will influence whether children sleep under mosquito nets include her education, exposure to the media and whether she works outside the home. All these factors are related to exposure to information and the level of understanding regarding protective measures. Exposure to the media is defined as 1 if the mother reads a newspaper, listens to a radio or watches television at least once a week and 0 otherwise. The hypothesis is that if a mother is exposed to knowledge and has a higher understanding, then her children are more likely to sleep under the mosquito net once one is available within the household. Similar characteristics of the husband, such as education and employment are regarded as having the same effect.
The household characteristics considered include sex of the household head, wealth of the household, type of residence (urban or rural), number of children under-fives, and whether the mother has final say in seeking health care. There is evidence that women are more likely to spend their money on their children than men  and, therefore, in households headed by women, children are more likely to sleep under mosquito nets once available. This argument holds for households where the mother has the final say on seeking health care.
Household wealth or socioeconomic status is a tricky factor. It is expected that the rich are more likely to own mosquito nets, and, therefore, the proportion of children sleeping under mosquito nets is larger in rich households. However, is it the case when the net is already available in the household? It is proposed and argued that perceived vulnerability is higher among the poor because of the limited options available to them once the child gets sick. So it is likely that children from poorer households are likely to sleep under a mosquito net once it is available in the household.
Additionally, urban households generally have better incomes than rural households and therefore are thought to own mosquito nets, but what of their use among the under-fives? Whether it is also related to more perception of the level of vulnerability in rural areas or the inability for the wealth index to capture incomes levels, is unclear. It is proposed that under-fives in rural households are more likely to use mosquito nets, given that one is available in the household. Factors relating to the area of residence, including altitude and region, were also considered.
Based on the above, the equation for use of mosquito nets among under-fives is as follows:
∪ (C, H, F, M, A)
where C, H, F, M, A are vectors of the child, household, partner, husband or father, mother, and area characteristics.