->Diagnostic tools
-->Interpolation: posterior distribution, Gosh and Gelfand, Gneiting and Raftery ranked score, DIC, gain regarding the prior [Alex]
-->Simulate from model and look at the output [Peter]
-->Spend time at the end of the analysis - did you picked up features you believed were important [P. Craigmile]
->Little availability of the R packages to fit multivariate space/time models [P. Craigmile]
-->Habits carried on from first researchers (Tilmann) for e.g., diagnostics established by the authors of books, etc. [Tanya]
-->Leave-out data analysis - try to assess predictive power of the model [Tanya], improvement from new parameters
->Teaching multivariate spatial statistic -- master and PhD level, are they introduced, what book, how and what is introduced
-->Working groups for other than the basic concepts (discussion of papers) [P. Craigmile]
-->Lack of software [Peter], spBayes, randomfields, composite likelihood package (rigid construction)
-->Cross-correlation with fitted marginals, only bivariate setting [Tanya]
-->spacetime and spatiotemporal packages, but early versions [P. Craigmile]
->For both nonstationary and multivariate spatial models, users are only us, people in the room (chicken and egg) so that practitioners and students get involved with these models[P. Craigmile]
-->Create a Use R! book or a metapackage to help with the topic [Reinhart]
->Different software used?
->Due to model complexity - which model to pick? - people would not necessarily use them even if they are available [Johan]
-->Max-stable processes date back from 1970, but now in 2010 they are available to fit, and then there is explosion in the use - Disagree with the previous comment - practitioners start with usage before learning the theory, while theoreticians work backward [Mathieu]
-->Practicum in workshop is great to get opportunity to use these different techniques and software [Wendy]
-->Tutorial (like Paulo's online page) is wonderful for teaching experience with students, requires less support from the instructor [Nancy]
-->Provide the code of paper online, even if not in a package [Reinhart]
-->so that you can reproduce the results exactly [Peter], even when privacy issue to provide the data (simulation from the data)
-->even in journal publication (JRSS-B?), code and data is required nowadays, and it has been for a long time [David]