1)  Morgenthaler’s paper
a. Data in EDA mode doesn't represent an underlying population like statistical anaysis but corresponds more to a list or batch of numbers. Therefore there is no need to consider a model as we do in statistical approach. The aim of EDA mode is finding relevant iformation about the data without deeply analyzing the strength of the evidence. For those reasons, EDA is much more flexible than statistics which is a very rigorous method, solving all problems by following the same precise approach.
b. When part, even a small one, of the batch values are changes, median stay constant or is not so much affected so it's preferable to use median in EDA mode.
c. EDA is usefull to spot easily patterns of the data. When data are re-expressed to EDA, they are better visualized and pattern are determined more rapidly. For instance, it's quite usefull to re-scale the data with the logarithmic function in order to decrease the gap between extremes values and facilitate the conclusion.
d. For further understanding of the data, residuals might be usefull by comparing their box plot with box plot of the data or ploting each row residuals against the column effect.
2)
a. According to Anselin's paper, spatial outliers could be detected by linking a collection of specialized choropleth maps, also called box maps, but we could also use the 3D scatter plot, Moran scatter plot or Lisa map.
b. In order to assess sensitivity of Moran's I statistic results, GeoDa includes several options such as changing the number of permutations, rerunning them a certain amount of time and changing the significance cutoff value. This process is usefull to determine how precise and stable is the outliers indication when the the significance barrier is lowered.
c. Tobler's observation highlights principles of spatial autocorrelation. Geographic data are affected by their location , they are spatially autocorrelated. Their distribution cannot be random as they are influenced by the neighborhood.
d. GeoDa is much easier to use, it doesn't required any programming skills, operator only has to use a point and click interface and it contains more mapping capability compare to R. However, GeoDa is not customizable for now which constitutes a major disadvantage compare to R environment. The best solution would be to switch to R after being introduced to GeoDa's techniques.
e. GeoDa might be better to use with vector file analysis instead of raster analysis. Indeed, it is usefull for analysis of discrete geospatial data such as point coordinates or plygon boundary coordinates.