6.5: Selecting Subsets of the Data
- Page ID
As we work on understanding the structure of a social network, there are occasions when we may wish to focus our attention on only a portion of the actors. Sometimes it's just a matter of clearing away "underbrush" of nodes that aren't "important". Sometimes it's a matter of selecting sets of actors for separate consideration.
UCINET has a number of built-in tools that can be useful for creating new datasets from existing datasets, that include only portions of the actors.
Data>Extract is a general-purpose tool that allows you to either "keep" or to "delete" rows, columns, or matrices for output to a new dataset. You may select the rows, columns, or relations (matrices) to keep by listing them in external data files, or by choosing the names of the rows, columns, or matrices from drop-down lists.
Data>Extract main component retains all the nodes and relations among nodes that are part of the largest component of a graph. In English: the information about the actor and connections among the actors who are part of the largest set of actors who are all connected is retained. If a graph contains several components (e.g. if there are some "isolates" or there are subgroups who have no connection to the largest group) only the largest will be retained. Many analyses require that all the nodes be connected. But, not all real networks actually are. So, you may wish to extract the largest component and analyze it.
Data>Subgraphs from partitions is a (somewhat more complicated) tool that lets you divide the cases into groups (partitions), and output separate data files for each group. The first step (after you've decided which cases fall in which partition), is to create an external data file that lists partition membership. Suppose I wanted to put nodes 1, 3, and 5 in one value of a partition (i.e. in one group) and cases 2, 4, and 6 in another. I'd create a data file that looked like: 1, 2, 1, 2, 1, 2. This says, put the first node in partition one, put the second node in partition two, put the third node in partition one, etc. This filename is supplied to the >Subgraphs from partitions dialog. You may also limit the process by electing to output only certain partitions (list them in the dialog window), and/or to create new datasets for a partition value only if there are more than some number (which you specify) of cases.
Many network analysis algorithms generate information on partition membership (and save partition membership information as files you can plug in to this utility). You might also want to impose your own partitions to identify people in one community, people of a particular gender, etc.
Data>Remove isolates creates a new dataset that contains all cases that are not isolated. An "isolate" is a case that has no connections at all to any other actors. Sometimes, when we collect information by doing a census of all the actors of a given type, or in a given location, some are "isolated". While this is usually an interesting social fact, we may wish to focus our attention on the community of actors who are connected (though not necessarily forming a single "component").
Data>Remove pendants creates a new dataset that contains all cases that are not "pendants". A "pendant" is a case that is connected to the graph by only one tie; cases like these will "dangle" off of more central cases that are more heavily connected. In looking at large graphs with many actors, we may wish to limit out attention to nodes that are connected to at least two other actors -- so as to focus attention on the "core" of the network. Removing isolates and pendants can help to clear some of the "clutter".
Data>Egonet is a tool that lets us extract particular actors and those in their immediate "neighborhood" as separate datasets. As we will see later on, the "ego-network" of a single actor, or of some selection of actors (all men, all cases with high between-ness, etc.) is often the focus of investigation more than the structure of the whole network.
An "ego-network" is the set of actors who are connected to a focal actor, along with the relations between ego and the alters, and any relations among the alters. The structure of ego networks (whether they are dense or thin, and whether they contain "structural holes") are often critical variables in understanding and predicting the behavior of "ego".
The Data>Egonet tool lets you list the "egos" or "focal nodes" you'd like to extract by using an external file list or by selecting their labels from a drop-down list. The dialog asks whether you want to include ego, or only to retain information on ego's neighbors; the most common, and default, choice is to include ego as well as ego's neighbors.
Data>Unpack is a tool for creating a new dataset that contains a subset of matrices from a larger dataset. For example, if we had stored information of both "liking" and "spouse" relation in a single dataset, we can use this utility to create separate data files for one or both relations. The relations to be "unpacked" are selected from a drop-down box.
Data>Join is a tool that can be used to combine separate sets of data into a new dataset. Often we collect attribute information about actors in several different settings (e.g. several classrooms in a school) and store these as separate files. Or, we may have multiple files that contain information about different attributes of actors (for example, one file might be things we know from outside sources like age, sex, etc.; another file might contain information on which partition of a graph each actor falls into). We might want to combine all the attribute information into a single file. Or, we might have information about different relations among the same set of actors, that have been stored as separate data files (as in the "liking" and "spouse" relations example).
Using Data>Join>Rows will combine two or more matrices (stored as separate files) into a single matrix that has rows for all nodes in each of the files. If I had separate files that listed the age of students in each of two classrooms, and I wanted to create a single file with all the students, the "rows" approach would be used.
Using Data>Join>Columns will combine two or more matrices (stored as separate files) into a single matrix that has the same number of rows as each of the input files, but appends the columns. If I had information on age and sex for actors A, B, and C in one file and information on centrality and degree for actors A, B, and C in another, I could do a column join to produce a file that listed age, sex, centrality, and degree for actors A, B, and C.
Using Data>Join>Matrices will combine information on multiple relations among the same sets of actors into a single file. Each input file has the same actors by actors array, but for different relations. The output file combines the multiple files into a three-dimensional array of actor by actor by relation.