I want to understand the concept of a "value raster" and if I can use it in neighbourhood statistics.
It is just a synonym for the raster data you want to analyse. Unlike vector features, which are built from a geometry attribute (i.e. coordinates) and an attribute table, rasters are both geometry and attributes.
A raster is a grid of cells (pixels), each containing a VALUE (hence 'raster value'). The value can stand for whatever attribute one wishes, yet each raster represents one attribute only. E.g. one raster can be a slopes raster with its values standing for the slope in degrees or percentage; another raster will represent the aspect of a terrain in which each cell's value holds the azimuth to which a cell faces. Than another raster can be populated with categorical values, e.g. land use type, etc.
Sometime rasters can contain more than one attribute, using the concepts of raster datasets (or raster bands). In this case, there are several simple rasters that are combined and aligned together, thus forming a data structure which enables rasters to have more than one value per cell. Further reading on this issue and on rasters is reccomended: see this thread and arcgis help on raster datasets. In addition some basic reading on raster data from arcgis help.