This study was undertaken in Anambra state; Southeast Nigeria. The state has a high malaria transmission rate all year round and the annual incidence rate is between 10 to 35%. On the basis of discussions with Anambra State Ministry of Health (MOH) officials, 6 sites were chosen for the study. These were the three largest urban centres (Awka (state capital), Nnewi and Onitsha) from each of the three senatorial zones and one rural LGA randomly selected from each senatorial zone (Njikoka, Aguata and Ogbaru).
Then, one community from each of the three rural LGAs: Enugwu-Ukwu (Njikoka LGA), Ekwulobia (Aguata LGA) and Okpoko (Ogbaru LGA) was selected using two-stage sampling, by first stratifying the communities according to whether they have a general hospital and then randomly selecting the sites from those that have general hospitals
Sampling and sample size
The software for population survey in EPI Info 6 was used for sample size calculation. The parameters that were used for sample size calculation were a power of 80%, 95% confidence level and considering 2% as the proportion of people with malaria that used services from the least commonly visited providers (community health workers  for first treatment of malaria. The calculations assumed that all the socio-economic groups used the services equally. The last parameter was the study population, which was the number of people with malaria in the study sites. Anambra MOH estimates an average of 6% monthly malaria incidence rate in the state. Hence, using the minimum projected population of each rural site at 30,000 people and each urban area at 60,000 people, it is estimated that a minimum of 1800 and 3600 people will have malaria monthly in each rural and urban area. The calculated minimum sample sizes for the pooled data of urban sites was 720 (240 per urban site) and pooled data for rural site was 663 (221 per rural site) However making allowance for a 5% refusal rate and ensuring an adequate sample size for data analysis, a total sample size of 1200 in the urban areas (400 per urban site and a total of 1050 in rural areas (350 per rural site) was used (an overall sample size of 2,250).
Two-stage sampling was used to select households in each community where the questionnaire was administered. In the first stage, 20 representative streets (urban areas) and 5 hamlets (rural areas) were randomly selected and the number of households in the selected areas enumerated to produce the household list. In the second stage, households were systematically included at regular intervals down the list, the starting point being chosen at random . Information was obtained from respondents from the selected households. The primary respondent was a female caregiver or in her absence a male head of household in whose absence an adult representative was used.
A pre-tested interviewer-administered questionnaire was used to obtain information from respondents in the randomly selected households. Local educated residents of each community were recruited and trained as field workers to administer the questionnaire. The respondents were asked to rank and also rate their preferences for improved malaria treatment services. Respondents were given a list of different sources of treatment (home, public and private hospitals, public primary healthcare (PHC) centers, pharmacy shops, patent medicine dealers, trained mothers, herbalists and community health workers (CHWs). Colorful option cards that depicted the different providers were also shown to the respondents to help them in visualizing them and aid their ranking and rating.
The contingent ranking and rating of preferences of consumers for different providers was elicited after the different sources of treatment provision were explained to them. The ranking was done before the rating scale. They were first asked to rank the 3 they most preferred then rate each treatment source from 1 to 10. The ranking allowed respondents to state relative preferences amongst the top 3 and in each rating multiple options could be scored at the same level of preference. The questionnaire was also used to collect data on the general socio-economic and demographic characteristics of the respondents and their households, expenditures on food as well as value of home produced and consumed food, and their asset holdings (additional file 1).
Tabulations, testing of means and non-parametric tests were the major data analytical procedures used. In equity analysis, a SES index was used to categorize the households into SES quartiles: least poor, poor, very poor and poorest. Principal components analysis (PCA) was used to generate the SES index [22, 23] that was used to investigate the SES differences in preferences. The input to the PCA was information on ownership of key assets such as a motorcar, a motorcycle, a radio, a refrigerator, a television set, a grinding machine and a bicycle together with the cost of food. In the bivariate analysis the index was analysed as a categorical variable (divided into quartiles), with the ratio of the lowest SES to the highest SES computed as the measure of inequity. Comparison of data between urban and rural area was used to test for geographic differences in preferences. Equity ratios were computed to show the level of difference between the urban and rural areas and between the highest and lowest SES quartile. Chi-square test through cross-tabulations was used to test for relationship of the preferences with SES and geographic location respectively.
Ethical Clearance: The authors received the approval of the ethics committee of the College of Medicine, University of Nigeria, Enugu Campus before carrying out this study.