Concluding commentsAwareness of the problems caused by spatial autocorrelation when using aggregated data in regression analysis is slowly spreading within the social sciences from the disciplines of geography, spatial econometrics, and regional science. Within sociology, for example, recent publications have emphasized the importance of space and place (e.g., Gieryn, 2000; Lobao, 2004; Lobao & Saenz, 2002; Tickamyer, 2000). In addition, a small number of sociologists have begun publishing research analyses where spatial processes have been brought into model specifications to correct for bias or inefficiency in parameter estimates that occur when spatial effects are ignored (e.g., Baller & Richardson, 2002; Baller et al., 2001; Deane, Beck, & Tolnay, 1998; Messner & Anselin, 2004; Sampson & Morenoff, 2004; Sampson, Morenoff, & Earls, 1999; Tolnay, 1995; Tolnay, Deane, & Beck, 1996). Unfortunately, however, this is still an emerging area where software developments have not kept pace with conceptual and theoretical advances—at least to the extent of making available relatively easy-to-use software. (GeoDa is emerging in ways that will soon contradict this statement, if it hasn’t already.) For example, in our reanalysis of the FL data, despite the fact that we have been able to deploy several useful software packageswith which to estimate spatial regression models, we were not able to fully implement the models we wished to estimate (e.g., properly weighted versions of models 3 and 4). As is evident from some ‘‘holes’’ (labeled ‘‘M’’) in Table 3, the kinds of regression diagnostics provided by the different packages differ (e.g., S-Plus did not provide a R2 statistic or AIC score for the weighted spatial error model). Finally, even when everything else matched up, the R2 statistic provided by SpaceStat differed from that reported by GeoDa and R, and the AIC score from SpaceStat also differed, but inconsequentially.