In this section we focus the discussion on the changes that took place in sources of health care between 2002/3 and 2005/6 and on the implications of these changes for the poor and those who reside in the rural areas.
The odds of not seeking care increased in 2005/6 compared to 2002/3. In both surveys, one of the most frequent reasons given for not seeking care among those who did not consider their sickness mild was the high cost of seeking care. The influence of cost could not be tested using multi-level modelling because data on costs of seeking care were not readily identifiable or consistent between the two surveys.
However, studies done elsewhere have indicated that cost is often a barrier to seeking services especially for the poor [1, 4, 26, 27]. Investment in health services by the government remains low and falls below the estimated minimum to provide the basic health care package . This has resulted in gaps in service delivery such as lack of fully functional laboratories, stock-outs of medicines and supplies, and inadequately skilled, under-supervised and poorly motivated health workers. These gaps have resulted in use of private drug shops, pharmacies and laboratories even when consultation could be provided in the public facilities. This could also explain use of private clinics amidst "free" care in public facilities.
Another reason that was given for not seeking care was poor geographical access to health facilities. Although citing distance as a reason for not seeking care decreased by 43% among the rural residents there was no significant decline among the most poor. Not all rural residents are poor and distance as a barrier may not be perceived to the same degree by the poor and less poor. It is possible that further analysis may reveal that a majority of respondents who did not report distance as a barrier in rural areas may belong to less poor households, able to pay for transportation to far off facilities while the poor in rural areas cannot afford transport. Furthermore, those who chose not to seek care may have done so out of concern over costs of seeking care rather than severity of illness. Because of the lack of data on costs of health services, we were not able to assess this. On average, physical access, measured as the population living within 5 km of a health facility, increased from 49% in 1999 to 72% in 2004 [29, 30]. It is important to note that there is substantial variation in physical access [17, 30]. Although distance was not significant in multi-variable analysis as a predictor for actual reported utilisation, the common mention of distance as a barrier to seeking care may suggest that health facilities are still perceived, especially by the most poor, to be too far for them to reach easily. Studies done elsewhere have also indicated that distance from a public health facility reduces poor people's likelihood of accessing care [31–33]. We know that in several of the rural areas where majority of the poor live, facilities have been put up but are not very functional, due to the absences of health workers or medicines, and inadequate budgets to operate the new facilities. Given this scenario, a respondent may consider a health facility as not being there, which could influence responses.
The majority of the respondents who fell sick 30 days preceding the survey sought care from private clinics. It is possible that the increased use of clinics and health centres may be related to the reported increase in illness incidence from 28.3% to 39.5% between 2002/3 and 2005/6. However, a similar picture of increased use of clinics has been found in other developing countries [3, 5, 34, 35]. These surveys did not include information on the quality of services provided in the clinics. A study done in Tanzania indicated that even when they access services, the poor, the less educated, and the rural women were less likely to receive key ANC interventions . Limited research has been done in the private for profit sector in Uganda. However, available evidence indicates that the sector is still largely unregulated and concerns have been raised about the training of the health workers, and the quality of care provided in these health facilities, as well as in public facilities [17, 36, 37]. Similar concerns about the quality of the services provided by the PFP sector have been raised in other low income countries [5, 33].
Respondents were more likely to use PNFP and public facilities relative to PFP in 2005/6 than in 2002/3; more likely to use public if female or rural; and less likely to use public if less poor. It is plausible that this resulted from improved proximity to the health facilities, stemming from the decentralization policy, coupled with increased funding from debt relief, which resulted in the construction of more health centres. Indeed there was an increase in the utilisation of health centres especially among the most poor and the rural residents over the period. Many PNFP providers responded to subsidies to increase accessibility to services by the poor. These actions included reducing charges, flattening of fees, or even completely removing fees [4, 38]. Although this increased accessibility of services for the poor, in some cases it reduced the revenue base of the PNFP facilities. The costs of production (especially staff salaries) continued to rise and the subsidies that they received from government and contributions from their donors did not increase proportionately [19, 20].
The less poor (quintile 4) and least poor (quintile 5) were less likely to use government clinics relative to private clinics. This is expected because although government services are nominally free, there remain numerous problems related to the shortage of health workers, drugs, supplies and equipment, and so many of those who can afford to pay for better quality services go to the private sector [1, 26]. This socioeconomic gradient was not observed in the utilization of PNFP services as was the case with public facilities relative to PFP services. This could have resulted from the PNFP sub-sector, unlike the private sub-sector, making a deliberate effort to keep their fees affordable even to the lower socioeconomic groups. Secondly, it is possible that the technical quality of services offered by PNFP was better than what the PFP and public facilities offered. If this was the case they may have tended to attract the better off users as well. These two effects would tend to cancel each other reducing the possibility of having a socioeconomic gradient. A previous study in Uganda that compared health care outputs between public and PNFP showed that some other factor seems to be at work in PNFP facilities .
Overall, the use of traditional healers was negligible. These findings are consistent with other studies [26, 34]. However, this information could be under reported because of the stigma associated with their use.
Finally we would like to highlight some methodological and other considerations and how they might affect interpretation of the findings. In general secondary data analyses are limited by the fact that the objectives of the secondary analysis and the original surveys may not be well aligned. For instance, the cost of seeking care could not be modelled because of inconsistencies in the way cost information was captured. Severity of illness was also not modelled because this information was not available.
Given the design of the study, i.e., using two cross-sectional sample surveys, it is not possible to definitively relate the changes to the reforms. The allocation of primary sampling units between urban and rural used in the two surveys differed. However, proportionate allocation of households and the adjustments using weights  appear to have been adequate as this resulted in roughly equal percentages of rural households between the two surveys of 85.3% and 84.2% in the 2002/3 and 2005/6 surveys respectively. The two surveys differed in the composition of households belonging to different wealth quintiles. The quintiles are not equal, especially in 2002/3, due to lumping. In addition, although using the same items, the asset weights and cut-offs were done separately for each survey. The 2005/6 survey reported that a significant decline in poverty was observed in rural areas between 2002/3 and 2005/6 from 42.7% to 34.2% . Possible differences in illness incidence between the most poor and least poor might affect the results particularly where comparisons are made between 2002/3 and 2005/6 for rural residents.
We would also like to highlight the importance of taking cluster effects at different levels into consideration during analysis for surveys such as the UNHS. For instance, when we did multivariable analysis accounting for the effect of districts (GLLAMM), rural residence emerged as having a significant effect on use of public facilities, whereas it was not significant when we did not account for the district effects (Tables 2 and 3).