Volume 11 Supplement 2
Population weighted raster maps can communicate findings of social audits: examples from three continents
© Mitchell et al; licensee BioMed Central Ltd. 2011
Published: 21 December 2011
Maps can portray trends, patterns, and spatial differences that might be overlooked in tabular data and are now widely used in health research. Little has been reported about the process of using maps to communicate epidemiological findings.
Population weighted raster maps show colour changes over the study area. Similar to the rasters of barometric pressure in a weather map, data are the health occurrence – a peak on the map represents a higher value of the indicator in question. The population relevance of each sentinel site, as determined in the stratified last stage random sample, combines with geography (inverse-distance weighting) to provide a population-weighted extension of each colour. This transforms the map to show population space rather than simply geographic space.
Maps allowed discussion of strategies to reduce violence against women in a context of political sensitivity about quoting summary indicator figures. Time-series maps showed planners how experiences of health services had deteriorated despite a reform programme; where in a country HIV risk behaviours were improving; and how knowledge of an economic development programme quickly fell off across a region. Change maps highlighted where indicators were improving and where they were deteriorating. Maps of potential impact of interventions, based on multivariate modelling, displayed how partial and full implementation of programmes could improve outcomes across a country. Scale depends on context. To support local planning, district maps or local government authority maps of health indicators were more useful than national maps; but multinational maps of outcomes were more useful for regional institutions. Mapping was useful to illustrate in which districts enrolment in religious schools – a rare occurrence - was more prevalent.
Population weighted raster maps can present social audit findings in an accessible and compelling way, increasing the use of evidence by planners with limited numeracy skills or little time to look at evidence. Maps complement epidemiological analysis, but they are not a substitute. Much less do they substitute for rigorous epidemiological designs, like randomised controlled trials.
A perennial challenge in social audit, as in most epidemiological research, is to collate and to present evidence in a way that is easily understood by those who most need the information. Visual presentation is a widely accepted strategy for knowledge translation, particularly in settings where people are less numerate or, more often, lack the time needed to comprehend statistics . Maps serve as a visual communication tool in many disciplines, including epidemiology . They can portray trends, patterns, and emphasise spatial differences easily overlooked in tabular data. For example, an overall average indicator of 50% can have regional variation, in an extreme case, from 0% in some communities to 100% in others. Maps allow us to identify hot-spots or outbreaks , and to identify the contributing environmental, geographic and social factors . Mapping has been used in evaluation of health services, particularly number, sizes, types, and locations of health services, resource allocation, and how these impact on health and well-being [5, 6].
Mapping in health research is not new. John Snow’s London Cholera maps of the mid-1800’s are part of epidemiological folklore, despite debate around the actual role the maps had in the identification of the source of the outbreak . Years before Snow's maps, spot maps illuminated yellow fever epidemics . Modern mapping technologies include remote sensing and geographic information systems (GIS). Hand-held global positioning systems, online mapping applications [9, 10], and the emergence of open source GIS  have increased the popularity of maps. These tools have illustrated topics such as dengue , malaria , cancer , polio , sexual risk behaviours , HIV stigma  and transactional sex . The term “health geomatics” includes the range of technologies (such as GIS) used to capture, analyse and map health phenomena [18, 19].
Much of the literature on mapping in health and health-related research focuses on the analytical approaches available through GIS, such as detection of clustering. Fewer published accounts cover use of the resulting maps to communicate risk and other evidence to planners and community members and how they can inform decision-taking [20, 21]. However, some recent examples do exist. For example, Geanuracos described the use of mapping as part of the Connect to Protect project in the United States and Puerto Rico, which helped planners identify areas in need of HIV and STI prevention planning . Jankowski described processes in urban and rural United States where GIS and its products played an integral role in participatory community-based decision taking around water resource planning . Joyce conducted qualitative research among public health decision makers in the United Kingdom and found that they appreciated GIS and mapping, but only as a complement to other sources of information . Driedger and colleagues explored interventions to increase uptake of GIS and mapping for decision making in early childhood learning services in Canada and found only a marginal increase in map use [25, 26].
A key concern of social audits  is to communicate survey findings to stakeholders at different levels – individual households, local organisations, local governments, national bodies, national governments, and regional groups of governments. In sample communities and local organisations and governments, this generates local evidence-based solutions that aggregate through the representative sample to a coherent policy contribution. At the aggregated level (state or country), the communication seeks to support understanding of the findings and to stimulate action to improve the situation. The evidence-stimulated dialogue on local solutions and actions – socialising evidence for participatory action (SEPA) -- is described elsewhere [28–30]. We present here the approach to producing maps of social audit findings and several ways in which maps have been used as part of SEPA, with examples from different countries and contexts.
Population weighted raster maps
The use of population weighted maps pre-dates modern GIS technologies. For example, in 1929, the Washington Post printed a map of the United States depicting each state with an area relative to its population rather than geographic size . More recently, in the early 1980’s, Taylor produced population weighted heating degree-day maps to help determine potential fuel requirements for Canada . In 1991, Upton demonstrated the use of population cartograms to display election results in the United Kingdom . Andersson and Mitchell described the use of population weighted raster maps, produced using CIETmap software, to display the findings of epidemiological studies , illustrating this in a paper exploring the social costs of landmines and the impact of mine risk reduction programmes in Afghanistan . Akin to weather and elevation maps, these population weighted raster maps show colour changes over the study area. Instead of elevations or barometric pressure, data are the epidemiological occurrence – a peak on the map represents a higher value of the indicator in question. The extent of spread of each colour is a function not only of space, as on a weather or elevation map, but also of population. The population relevance of each sentinel site, as determined in the stratified last stage random sample, combines with space (using inverse-distance weighted interpolation) to provide a population-weighted extension of each colour. This transforms the map to show population space instead of simply geographic space. Additional information about the construction of the maps can be found in the accompanying technical annex [see additional file 1 for the technical annex].
The resulting population weighted raster maps can be interpreted much like weather maps, where trends across regions are more informative than values at specific locations. Data are classified into four or five continuous class ranges represented with a legend. This helps to ensure that individual communities are not easily identified, which can be important when communicating sensitive data. There is evidence that sensitive information can sometimes be traced back to specific locations, including communities and households, even when these locations are not explicitly included on the original map .
A standard colour scheme means that darker areas on the maps represent areas in need of attention or investment. A green colour palette is used for maps that represent “coverage” or “programmes”; a brown colour palette for maps that represent “outcomes”; and a red colour palette for “change” maps. Vector overlays (such as administrative boundaries, roads, or other landmarks) are shown as black or grey lines and are often placed on top of the raster surface to provide visual reference points familiar to the map users. The interpolation itself is independent of the overlay, avoiding a situation where a health outcome changes abruptly at an arbitrary administrative boundary unrelated to the indicator being mapped.
Relying on the CIETmap freeware, we used population weighted raster maps, and extensions of this technique, for communication of evidence from social audits in different contexts and to different target audiences.
Presenting sensitive data
Publishing figures about a sensitive topic, such as abuse against women, can detract from efforts to tackle the underlying causes of the problem, as debate centres rather around the accuracy of the summary figures for a country or region. Our large survey in Pakistan in 2003 produced the first nationally representative figures about the levels of different kinds of abuse against women in that country . After the findings were shared with the government ministry that commissioned the research, some officials became concerned that the figures for the levels of different kinds of abuse revealed by the survey were too politically sensitive to be published. Yet they agreed the findings should be shared with government and other stakeholders in each province, in order to help them develop the most effective strategies for tackling the problem. They permitted the use of population weighted maps, including in large scale meetings with invited journalists, provided it did not lead to any “inflammatory headlines” or “negative publicity”.
We presented these maps of the levels of types of abuse in many meetings across Pakistan, attended by government and civil society stakeholders, as well as the media. In the presentation, for each type of abuse, the map showing the frequency across the country was followed by the results of a multivariate analysis of the variables related to the risk of that type of abuse. The presentation generated considerable interest and discussion, but at no point did anyone ask for the summary national or provincial figures for the levels of different kinds of abuse. Rather the discussion focused on what was responsible for the variation in levels of abuse, and then on how the underlying risk factors could be tackled. The media coverage from the meetings focused on constructive messages about what could help prevent abuse; the journalists did not have summary 'headline' figures to quote. Thus the maps had two advantages: they avoided simplistic, “shock” headlines which may have derailed the whole process of sharing the findings; and they helped to begin the discussion about underlying risk factors and the search for local solutions.
Presenting changes in time and space
Population weighted raster maps can illustrate temporal changes of programme coverage or impact, giving a visual impression that a table of numbers cannot match. Their advantage over charts (for example bar charts or line plots showing indicators at different time points) is that they show the variation in change across the entity. Maps summarise large amounts of data in an immediately accessible form.
A population weighted raster map for a given indicator at the first time point begins a series of maps. Subsequent maps in the series show the same indicator measured at later time points in the same panel of sample communities. Classification values must be fixed across all the maps in the series to ensure that legends are standardized and users can interpret changes on the map accurately. These maps can either be shown side by side, or presented as an animation where one map in the series ‘morphs’ into the next map. Time-series maps can show spatial differences over time that may not be readily apparent from a table. For example, it would be difficult to discern trends in a table showing a panel of 250 sites, with three columns of data showing the proportion of a given indicator across three surveys. Such information displayed on a times-series map can provide an almost instantaneous and visually compelling summary of regional variations in the indicator over time.
Three social audit cycles in Bangladesh in 1999, 2000 and 2003 documented public perceptions, use, and experience of government and other health services during a major health services reform programme . The three surveys visited the same nationally representative sample of 247 communities. Across the three surveys, household opinions of government health services deteriorated, and their use of government services fell. The satisfaction of users of government health services also fell, as did the proportion that reported they received all the prescribed medicines from the health facility.
Change maps are an alternative to time series maps. In this case, the amount of change in an indicator is mapped in a single population weighted raster map, rather than the actual value of the indicator across different time points. The legend is classified into three classes: increase, no change, and decrease. For most outcomes, “no change” can be defined as being within +/- 5%. For rare outcomes +/- 1% would be more appropriate. It is important to decide how “no change” is defined in each case, and make this clear to users of the maps. Change maps give an at-a-glance view of the way an indicator of programme coverage or outcome has changed, not only whether or not it has improved overall but also in which areas the change has been more or less marked. A disadvantage of change maps is that they do not give an indication of the actual level of the indicator; change from a low starting point looks the same as change from a much higher starting point. Therefore, it is often important to use change maps in conjunction with a map or other display of the actual level of the indicator at one of the time points.
Presenting impact scenarios
Presentation of the current level of programme coverage or health outcomes, or even of the way these have changed over time, is insufficient for planners to decide which interventions they should pursue in order to produce the most benefit for the most people – and of course, most cost-effectively. The best way to answer such questions is to conduct a pragmatic randomised cluster controlled trial (RCCT) or a stepped wedge roll-out, randomly allocating areas to receive the new programme first and comparing outcomes in areas with and without the programme after an appropriate time interval . Failing this, it is possible to model effects of different interventions, or combinations of interventions, based on data from cross sectional studies. This modelling of potential population gains is an important aspect of social audit . The challenge is to present the results of this modelling in a way that is accessible to planners, many of whom have limited background in the quantitative sciences. Mapping the potential outcomes of interventions shows how much difference could be made, and in which places the impact might be greatest.
Impact scenario mapping should rely on RCT evidence or on panel data after multivariate analysis. With the final models in place, potential ‘gains’ from an intervention (based on the risk difference and the proportion requiring intervention) are applied to the map, either increasing the value at each site (in the case of a positive phenomenon) or decreasing it (in the case of a negative phenomenon). Gains can either be applied across the map as a whole (for example, a gain of 5% applied to each site) or regional gains can be applied accordingly (for example, if there were different multivariate models for different areas of a country). Each set of intervention impact maps begins with the “current” map of a given indicator (the baseline map with no investment), followed by subsequent maps of different investment scenarios, such as what the map would look like if half of those requiring the intervention had it, and what the map would look like if all of those requiring the intervention had it.
Going to scale
For a map to be useful for the intended audience, the scale has to be right. For example, maps showing the variation of an indicator across the whole country are useful for national planners, but less useful for local planners, who need to know how the indicator varies across their area of jurisdiction. This is true whether the map is a population weighted raster map, showing variation of the indicator not demarcated by local authority boundaries, or a vector map displaying the average for each local authority. These maps and the figures underlying them can inform local authorities about how they compare with other local authorities, but they do not help them to plan how to intervene in different parts of their own area of authority.
At the other end of the spectrum, it can sometimes be useful to prepare maps showing variation across a number of countries, if the intended audience has an overarching role across the whole region. One has to decide whether to prepare vector maps with country as the block unit, or population weighted raster maps, in which the level of an indicator does not necessarily follow national borders.
Illustrating rare occurrences
Rare events or low prevalence indicators can be difficult to show graphically, especially when their geographic distribution is important. Yet it can be useful to get a picture of where these rare events are located.
Limitations and cautions
Maps are a visual aid but in some cases do not help to summarise the findings. For example, in some cases there may be large parts of a country or district that are inaccessible or have little or no population. We typically show areas with “no data collected” shaded in grey. For example, the maps of Pakistan (Figures 1, 2, 10, 11, 20 and 21) have grey areas which are districts or areas where no data was collected; in Botswana (Figures 5, 6, and 9) the grey area represents the central Kalahari reserve with little or no population; in Haripur (Figure 14b), the grey area is a large dam. In these cases, the maps are still useful. But in the case of Khairpur district (Figure 14c), much of the overall district area is mountain or desert with little or no population, and this renders the population weighted map of the district uninformative.
Sometimes, by chance, a random sample of sites across a domain might be concentrated in one part of the domain, or leave out a part. We avoid this whenever possible by including a geographic component as a stratification in the sample selection. Other methods, such as weighted kernel density estimation with an adaptive kernel, can help control for irregular sample distribution .
As with any presentation of evidence, it is easy to mislead with maps, intentionally or unintentionally, as described by Monmonier . For example, changing legend categories makes a change appear more impressive than it really is. The same rules of displaying data fairly apply to maps as to any other kind of illustration.
Because one sees a high proportion of outcome Y in one area of a map of a country, and a high proportion of exposure X in the same area of a second map of the country, one cannot conclude that X causes Y. This is a GIS corollary of the well-known ecological fallacy [52, 53]. The maps can open a fruitful discussion in this respect. It is not hard to guide the discussion with questions about individual relationships with outcome Y, both in areas where the outcome is common and in areas where the outcome is less common, and that the exposure X might not be the real cause of outcome Y at all. This can help to guide the formal epidemiological analysis of potential confounders.
Population weighted raster maps can present social audit findings in an accessible and compelling way. People with limited numeracy skills (or very little time to look at evidence) can readily appreciate spatial variation and changes over time in maps. Maps are powerful and persuasive and are a useful complement to epidemiological analysis, but they are not a substitute for in depth analysis. Much less do they substitute for rigorous epidemiological designs, like RCTs. It is important to match the type and scope of the map to the intended audience. Our examples illustrate the range of maps that can be produced and their usefulness and limitations.
List of abbreviations used
Geographic Information System
Local Government Authority
Non Government Organisation
Randomised Cluster Controlled Trial
Randomised Controlled Trial
Southern African Development Community
Spatial Development Initiative
Socialising Evidence for Participatory Action
We thank the CIET teams in Bangladesh, Botswana, Nigeria, Pakistan, and South Africa for their contributions to the studies that are drawn on in the paper. The Canadian International Development Centre (IDRC) funded the work in Afghanistan. The Organisation for Economic Co-operation and Development (OECD) funded the work in the Baltic States. The Canadian International Development Agency (CIDA) and the World Bank funded studies in Bangladesh. The European Union and IDRC funded studies in Southern Africa, and IDRC also funded the work in Nigeria. The UK Department for International Development (DFID) and CIDA funded studies in Pakistan.
This article has been published as part of BMC Health Services Research Volume 11 Supplement 2, 2011: Social audit: building the community voice into health service delivery and planning. The full contents of the supplement are available online at http://www.biomedcentral.com/1472-6963/11?issue=S2.
- Lipkus I, Hollands JG: The visual communication of risk. Journal of the National Cancer Institute Monographs. 1999, 25: 149-163.View ArticlePubMedGoogle Scholar
- Kandwal R, Garg PK, Garg RD: Health GIS and HIV/AIDS studies: Perspective and retrospective. Journal of Biomedical Informatics. 2009, 42: 748-755. 10.1016/j.jbi.2009.04.008.View ArticlePubMedGoogle Scholar
- Law D, Serre M, Christakos G, Leone P, Miller W: Spatial analysis and mapping of sexually transmitted diseases to optimise intervention and prevention strategies. Sexually Transmitted Infections. 2004, 80: 294-299. 10.1136/sti.2003.006700.PubMed CentralView ArticlePubMedGoogle Scholar
- Rytkönen MJ: Not all maps are equal: GIS and spatial analysis in epidemiology. Int J Circumpolar Health. 2004, 63: 9-24.PubMedGoogle Scholar
- McLafferty S: GIS and health care. Annual Review of Public Health. 2003, 24: 25-42. 10.1146/annurev.publhealth.24.012902.141012.View ArticlePubMedGoogle Scholar
- World Health Organization: Geographical information systems (GIS): mapping for epidemiological surveillance. Weekly Epidemiological Record. 1999, 74 (34): 281-285.Google Scholar
- McLeod KS: Our sense of Snow: the myth of John Snow in medical geography. Soc Sci Med. 2000, 50: 923-935. 10.1016/S0277-9536(99)00345-7.View ArticlePubMedGoogle Scholar
- Stevenson LG: Putting disease on the map: the early use of spot maps in the study of Yellow Fever. Journal of the History of Medicine and Allied Sciences. 1965, 20 (3): 226-261.View ArticlePubMedGoogle Scholar
- Chang AY, Parrales ME, Jimenez J, Sobieszczyk ME, Hammer SM, Copenhaver DJ, Kulkarni RP: Combining Google Earth and GIS mapping technologies in a dengue surveillance system for developing countries. Int J Health Geogr. 2009, 8: 49-10.1186/1476-072X-8-49.PubMed CentralView ArticlePubMedGoogle Scholar
- Gao S, Mioc D, Anton F, Yi X, Coleman DJ: Online GIS services for mapping and sharing disease information. Int J Health Geogr. 2008, 7: 8-10.1186/1476-072X-7-8.PubMed CentralView ArticlePubMedGoogle Scholar
- Steniger S, Bocher E: An overview of current free and open source desktop GIS developments. International Journal of Geographical Information Science. 2009, 23 (10): 1345-1370. 10.1080/13658810802634956.View ArticleGoogle Scholar
- Van Der Hoek W, Konradsen F, Amerasinghe P, Perera D, Piyaratne MK, Amerasinghe F: Towards a risk map of malaria for Sri Lanka: the importance of house location relative to vector breeding sites. International Journal of Epidemiology. 2003, 32: 280-285. 10.1093/ije/dyg055.View ArticlePubMedGoogle Scholar
- Bell BS, Hoskins RE, Pickle LW, Wartenberg D: Current practises in spatial analysis of cancer data: mapping health statistics to inform policymakers and the public. Int J Health Geogr. 2006, 5: 49-10.1186/1476-072X-5-49.PubMed CentralView ArticlePubMedGoogle Scholar
- Kamadjeu R: Tracking the polio virus down the Congo River: a case study on the use of Google Earth in public health planning and mapping. Int J Health Geogr. 2009, 8: 4-10.1186/1476-072X-8-4.PubMed CentralView ArticlePubMedGoogle Scholar
- Westercamp N, Moses S, Agot K, Ndinya-Achola JO, Parker C, Amolloh KO, Bailey RC: Spatial distribution and cluster analysis of sexual risk behaviours reported by young men in Kisumu, Kenya. Int J Health Geogr. 2010, 9: 24-10.1186/1476-072X-9-24.PubMed CentralView ArticlePubMedGoogle Scholar
- Kandwal R, Augustijn EW, Stein A, Miscione G, Garg PK, Garg RD: Geospatial analysis of HIV-related social stigma: a study of tested females across mandels of Andhra Pradesh in India. Int J Health Geogr. 2010, 9: 18-PubMed CentralView ArticlePubMedGoogle Scholar
- Ferguson AG, Morris CN: Mapping transactional sex on the Northern Corridor highway in Kenya. Health Place. 2007, 13: 504-519. 10.1016/j.healthplace.2006.05.009.View ArticlePubMedGoogle Scholar
- Alexander G, Kinman E, Miller L, Patrick T: Marginalization and health geomatics. Journal of Biomedical Informatics. 2003, 36: 400-407. 10.1016/j.jbi.2003.09.021.View ArticlePubMedGoogle Scholar
- Boulos M, Roudsari A, Carson E: Health Geomatics: An enabling suite of technologies in health and healthcare. Journal of Biomedical Informatics. 2001, 34: 195-219. 10.1006/jbin.2001.1015.View ArticlePubMedGoogle Scholar
- Mills JW, Curtis A: Geospatial approaches for disease risk communication in marginalized communities. Prog Community Health Partnersh. 2008, 2: 61-72. 10.1353/cpr.2008.0007.View ArticlePubMedGoogle Scholar
- Scotch M, Parmanto B, Gadd CS, Sharma RK: Exploring the role of GIS during community health assessment problem solving: experiences of public health professionals. Int J Health Geogr. 2006, 5: 39-10.1186/1476-072X-5-39.PubMed CentralView ArticlePubMedGoogle Scholar
- Geanuracos CG, Cunningham SD, Weiss G, Forte D, Henry Reid L, Ellen JM: Use of Geographic Information Systems for planning HIV prevention interventions for high-risk youths. American Journal of Public Health. 2007, 97 (11): 1974-1981. 10.2105/AJPH.2005.076851.PubMed CentralView ArticlePubMedGoogle Scholar
- Jankowski P: Towards participatory geographic information systems for community-based environmental decision making. Journal of Environmental Management. 2009, 90: 1966-1971. 10.1016/j.jenvman.2007.08.028.View ArticlePubMedGoogle Scholar
- Joyce K: "To me it's just another tool to help understand the evidence": public health decision-makers' perceptions of the value of geographical information systems (GIS). Health Place. 2009, 15: 801-810.View ArticlePubMedGoogle Scholar
- Driedger SM, Kothari A, Graham ID, Cooper E, Crighton EJ, Zahab M, Morrison J, Sawada M: If you build it, they still may not come: outcomes and process of implementing a community-based integrated knowledge translation mapping innovation. Implement Sci. 2010, 5: 47-10.1186/1748-5908-5-47.PubMed CentralView ArticlePubMedGoogle Scholar
- Driedger SM, Kothari A, Morrison J, Sawada M, Crighton EJ, Graham ID: Using participatory design to develop (public) health decision support systems through GIS. Int J Health Geogr. 2007, 6: 53-10.1186/1476-072X-6-53.View ArticleGoogle Scholar
- Andersson N: Building the community voice into planning: 25 years of methods development in social audit. BMC Health Serv Res. 2011, 11 (Suppl 2): S1-PubMed CentralView ArticlePubMedGoogle Scholar
- CIET: The SEPA model: CIET's approach to communication. CIET. 2006, http://www.ciet.org/_documents/SEPA%20Exec%200306.pdf,Google Scholar
- Omer K, Mhatre S, Ansari N, Laucirica J, Andersson N: Evidence-based training of frontline health workers for door-to-door health promotion: a pilot randomized controlled cluster trial with Lady Health Workers in Sindh Province, Pakistan. Patient Educ Couns. 2008, 72: 178-185. 10.1016/j.pec.2008.02.018.View ArticlePubMedGoogle Scholar
- Andersson N, Omer K, Caldwell D, Dambam MM, Maikudi AY, Effiong B, Ikpi E, Udofia E, Khan A, Ansari U, Ansari N, Hamel C: Male responsibility and maternal morbidity: a cross-sectional study in two Nigerian states. BMC Health Serv Res. 2011, 11 (Suppl 2): S7-PubMed CentralView ArticlePubMedGoogle Scholar
- Tobler W: Thirty Five Years of Computer Cartograms. Annals of the Association of American Geographers. 2004, 94 (1): 58-73. 10.1111/j.1467-8306.2004.09401004.x.View ArticleGoogle Scholar
- Taylor BL: Population-weighted heating degree days for Canada. Atmostphere-Ocean. 1981, 19 (3): 261-268. 10.1080/07055900.1981.9649113.View ArticleGoogle Scholar
- Upton GJ: Displaying election results. Political Geography Quarterly. 1991, 10 (3): 200-220. 10.1016/0260-9827(91)90034-R.View ArticleGoogle Scholar
- Andersson N, Mitchell S: CIETmap: Free GIS and epidemiology software from the CIET group, helping to build the community voice into planning. 2002, Presentation to World Congress of Epidemiology; Montreal, CanadaGoogle Scholar
- Andersson N, Mitchell S: Epidemiological geomatics in evaluation of mine risk education in Afghanistan: introducing population weighted raster maps. Int J Health Geogr. 2006, 5: 1-10.1186/1476-072X-5-1.PubMed CentralView ArticlePubMedGoogle Scholar
- Curtis AJ, Mills JW, Leitner M: Spatial confidentiality and GIS: re-engineering mortality locations from published maps about Hurricane Katrina. Int J Health Geogr. 2006, 5: 44-10.1186/1476-072X-5-44.PubMed CentralView ArticlePubMedGoogle Scholar
- Andersson N, Cockcroft A, Ansari N, Omer K, Chaudhry UU, Khan A, Pearson LW: Collecting reliable information about violence against women safely in household interviews: experience from a large-scale national survey in South Asia. Violence Against Women. 2009, 15 (4): 482-496. 10.1177/1077801208331063.View ArticlePubMedGoogle Scholar
- Cockcroft A, Andersson N, Milne D, Hossain MZ, Karim E: What did the public think of health services reform in Bangladesh? Three national community-based surveys 1999-2003. Health Res Policy Syst. 2007, 5: 1-10.1186/1478-4505-5-1.PubMed CentralView ArticlePubMedGoogle Scholar
- Andersson N, Cockcroft A, Ho-Foster A, Masisi M, Laetsang D: Survey of HIV and AIDS related knowledge, attitudes, and practice, Botswana 2007: Summary of key findings. 2008, Johannesburg: CIET TrustGoogle Scholar
- Mitchell S, Andersson N, Ngxowa N, Merhi S: A community-based impact assessment of the Wild Coast spatial development initiative, 1997-2004. Development Southern Africa. 2008, 25 (1): 119-132. 10.1080/03768350701837895.View ArticleGoogle Scholar
- CIETafrica: A decade of local economic development in the Wild Coast 1997-2007: the community view. Report to Conflict and Governance Facility (CAGE). 2007, Pretoria: CIET TrustGoogle Scholar
- Ansari U, Cockcroft A, Omer K, Ansari NM, Khan A, Choudhry UU, Andersson N: Devolution and public perceptions and experience of health services in Pakistan: linked cross sectional surveys in 2002 and 2004. BMC Health Serv Res. 2011, 11 (Suppl 2): S4-PubMed CentralView ArticlePubMedGoogle Scholar
- Cockcroft A, Andersson N, Omer K, Ansari N, Khan A, Chaudhry UU: Social audit of governance and delivery of public services. Baseline survey 2002, National report. 2003, Islamabad: National Reconstruction BureauGoogle Scholar
- Cockcroft A, Andersson N, Omer K, Ansari N, Khan A, Chaudhry UU, Saeed S: Social audit of governance and devlivery of public services. Pakistan 2004/05. National report. 2005, Islamabad: CIET and DTCEGoogle Scholar
- Andersson N: Proof of impact and pipeline planning: directions and challenges for social audit in the health sector. BMC Health Serv Res. 2011, 11 (Suppl 2): S16-PubMed CentralView ArticlePubMedGoogle Scholar
- Andersson N, Paredes-Solís S, Legorreta-Soberanis J, Cockcroft A, Sherr L: Breastfeeding in a complex emergency: four linked cross-sectional studies during the Bosnian conflict. Public Health Nutr. 2010, 13: 2097-2104. 10.1017/S1368980010001667.View ArticlePubMedGoogle Scholar
- Cockcroft A, Monasta L, Onishi J, Omer K: Service delivery survey, Bangladesh health and population sector programme, second cycle 2000. 2001, Dhaka: CIETcanada and Ministry of Health and Family Welfare, Government of BangladeshGoogle Scholar
- Cockcroft A, Andersson N, Paredes-Solís S, Caldwell D, Mitchell S, Milne D, Merhi S, Roche M, Konceviciute E, Ledogar RJ: An inter-country comparison of unofficial payments: results of a health sector social audit in the Baltic States. BMC Health Serv Res. 2008, 8: 15-10.1186/1472-6963-8-15.PubMed CentralView ArticlePubMedGoogle Scholar
- Cockcroft A, Andersson N, Milne D, Omer K, Ansari N, Khan A, Chaudhry UU: Challenging the myths about Madaris in Pakistan: a national household survey of enrolment and reasons for choosing religious schools. International Journal of Educational Development. 2009, 29: 342-349. 10.1016/j.ijedudev.2008.09.017.View ArticleGoogle Scholar
- Brunsdon C: Estimating probability surfaces for geographical point data: an adaptive kernel algorithm. Computers and Geosciences. 1995, 21 (7): 877-894. 10.1016/0098-3004(95)00020-9.View ArticleGoogle Scholar
- Monmonier M: Lying with Maps. Statistical Science. 2005, 20 (3): 215-222. 10.1214/088342305000000241.View ArticleGoogle Scholar
- Piantadosi S, Byar DP, Green SB: The ecological fallacy. American Journal of Epidemiology. 1988, 127 (5): 893-904.PubMedGoogle Scholar
- Robinson WS: Ecological correlations and the behaviour of individuals. American Sociological Review. 1950, 15 (3): 351-357. 10.2307/2087176.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.