12THE ERROR COMPONENT INSPATIAL DATABACKGROUND: ERROR AND DATAQUALITY DEFINITIONSNo map can be picked apart into completelyindependent pellets of information. There issomething collective and comprehensive about thespatial representation. First, the map is spatiallycontinuous and comiected. There is also a deeperreason. Individual facts become more usefulinformation through an interaction with a structureof theory that provides a context to interpret theindividual facts. In common usage, the maindistinction between data and information arisesfrom meaning, but meaning is context dependent.To be more concrete, the process of converting aparticular fact into information must comprehendthe fitness of that fact for some particular purpose.This line of argument provides an importantintroduction to the role of data quality in aninformation system.Quality has various definitions in industrialengineering (Hayes and Romig 1977), but oneaccepted definition is `fitness for use' (Chrisman1983). Recently, the US National CommitteeDigital Cartographic Data Standards Task Force(DCDSTF 1988) has adopted this definitionformally for inclusion in a US national standard forexchange of spatial data. The standard requires aquality report that provides the basis for a user tomake the final judgement - the conversion toinformation by interpretation for a particular use.N R CHRISMANAlthough most data gathering disciplines treat error as an embarrassing issue to beexpunged, the error inherent in spatial data deserves closer attention and publicunderstanding. This chapter presents a review of the components of error in spatialdata, based on Sinton's scheme of roles in the measurement process and also on thecategories in the US Proposed Standard for Digital Cartographic Data Quality.This particular element of the US proposal has alsobeen adopted, at least in spirit, by a British proposal(Haywood 1986), and the French (Salge and Sclafer1989), among others.Quality is a neutral term, fitting for a nationalstandard, but the tough issues of quality are bestevoked by the loaded word `error'. In commonusage, error is a bad thing, and many professionsrelated to spatial data share this attitude. Forinstance, geodesists, surveyors andphotogrammetrists go to great lengths to reduce theerror in their measurement products. For thesedisciplines the full attention focuses on thereduction of deviation between positionalmeasurements and `ground truth'. Cartographers,perhaps because they often cannot remove all error,have two incompatible approaches, both designedto avoid the issue. One tendency is to generateauthoritative maps, through their official status orsome other form of paternalism. The traditionalapproach to standards places little emphasis on auser's determination of a particular need. The othertendency, more common in academic circles, adoptsthe communication paradigm and expects thecartographer to use all means to communicate themessage (Robinsonet al.1984). The communicationparadigm, like a paternalist agency, assumes thatthe map maker controls the process, particularly thejudgement of fitness for use. In summary, thedisciplines of mapping technology are bent onreducing error or minimizing its importance. Whilethis may be a reasonable approach to foster the165N R Chrismancurrent divisions of labour, it does not lead to thefull exploitation of spatial information.Error is not a problem unique to spatialevidence. Other disciplines have created othersolutions which are worth considering. Mostphysical, biological and social sciences integratedata collectors and data analysts into the samediscipline, while mapping places them in distinctguilds. Perhaps as a result, error bars or some otherestimates appear on the display of most physicalmeasurements. Also, even the popular presspresents a standard error of estimate for opinionpoll results. Reporting error is not a sign ofweakness in these other disciplines, because theerror estimate provides crucial information whichmust be preserved for correct interpretation. Themost developed scientific approach to error is thebody of statistical procedures which have developedover the past century. However, many of theadvanced techniques in statistics are not attuned tothe problems inherent in geographical information.For some attempts to understand the statisticalbasis of errors in spatial databases see White (1978),Goodchild and Dubuc (1987) and Goodchild andGopal (1989).Basic terms for errorBefore delving deeper, it is useful to present somefundamental terms to discuss error. Under thegeneral intent of describing data quality the goal isto describe fitness for some particular use. Manynumerically oriented disciplines have developed aconcept of error as a deviation (or distance)between a measurement and the `true' value.Different disciplines use different terms to refer tothis concept, and some of these terms conflict. Thischapter will follow the general practice of themapping sciences (see DCDSTF 1988: 28) and usethe term accuracy to refer to the closeness of anobservation to a true value (or one accepted astrue). This definition of accuracy implies anindependent, objective external reality and sets thehighest standard for the concept of accuracy. Insome contexts, a measurement system may produceinaccurate results that preserve local relationships.Statistically, such a condition arises from,systematic' error (as opposed to random error).Systematic error involves a uniform shift of values,hence the term `bias' which is applied in some166measurement sciences. Such systems are describedby cartographers as having `relative accuracy', butthis property is usually a sign that the process ofdata preparation (compilation) has not beencompleted. In the spirit of `fitness for use', relativeaccuracy may be perfectly viable for someapplications, while unfit for others which depend ongeodetic registration.The concept of accuracy is essentiallyindependent of the issue of resolution although bothcontribute to overall data quality. The resolution ofa measurement system is the minimum distancewhich can be recorded. Clearly, resolution sets alower bound on accuracy. It is considered goodpractice to make this minimum difference smallerthan the accuracy of the whole system, but a usershould not confuse the two. Both resolution andaccuracy can be applied to the various componentsof spatial information, both attributes (Dueker1979) and positions (Tobler 1988; see also Fisher1991 in this