In multivariate analysis involve analysing different data sets of the same individuals, sometimes it is called integration or multi-block analysis.
However I can't find how are these dataset defined:
- in psychology each block seems to be a psychological test, and the multi-block analysis usually aims to find the common underlying variables explained by different tests (aiming to measure the same: p. ex. intelligence)
- in economics each block might represent economics, political and demographic variables
As you might see, in these blocks sometimes there is a categorical variable
- in biology, with the new technologies enabling to measure thousand of gene expression and proteins. Which leads to recommendations like these:
The variables that compose each block have to be defined carefully: not only according to their nature (e.g. one block that contains all the voxels of an image, one block for all the metabolites, etc.) but also according to external information. Nowadays, a huge amount of external information is available and can be used to define each block. For example, a block that contains all the metabolites can be divided into several data blocks; hence, metabolites belonging to one pathway are gathered within the same block. A block that contains all the voxels of an image can be grouped by regions: voxels belonging to one specific region are then gathered within the same block. This grouping strategy makes more interpretable blocks and facilitates the interpretation
I have problems following this advice. I have a block of gene expression, that I could divide by protein-coding and other type of genes, and within protein-coding I could divide them further by pathway. But a single gene might be related in several pathways, so I can't (without duplicating the information) create a new dataset for each pathway. In addition, it is not clear the definition of those pathways.
Following a tidy design I thought that each block should represent a group of variables that closely interact and might be intra-correlated. But I don't have any other reasoning than the multivariate variable must summarize the whole block in a single number
How are the blocks defined? Or how should I divide my blocks (If I should at all) ?
Note that this is different from this question as it is about grouping different variables and not samples.
Also is there a multi-block tag or related tag?
