I analysed the 2001 and 2011 migration data by first converting it into a bipartite network - as the data available is state to district migration making it a directed bipartite graph. As I also have the migration flow values, which I used as weights, I now had a weighted directed bipartite graph.
The data cleaning and getting nodes and their coordinates in one Excel file and edge list with weights with the same IDs in the node file with new IDs given to states was a task in itself. While that was fun, network analysis and especially looking up kinds of analysis that will help me find what I want answered was even more fun.
The question I wanted to be answered was: Which districts, according to migration patterns of 2011 and 2001 combined, have growth potential? The results gave me two different kinds of districts which have potential - ones which can develop on their own without any help - as it is happening already with large in-migration as a proxy and another set which are so behind (given by proxy of migration data again) that migration has decreased to the districts between 2001 and 2011. Here migration is assumed to be positive as incoming migration means there are employment opportunities, as that is the major reason for migration in India.
I chose PageRank (PR) analysis to find the potential districts. PR was developed by Google for recognizing prominent pages in a website. Generally, if there are lots of edges connected to a node, that is identified as a prominent node (degree centrality) but in PR, a node is prominent even if there is say just one node connecting to it but that connection has some centrality or its inherent prominence, in turn increasing the prominence of this node with only one in-degree. In migration patterns, PR, especially in a directional bipartite graph, would give me the relative importance of a node as the PR values add up to 1. In other words, I have 640 districts, and if a random walked chose to migrate to any district, the probability of him going to any of the districts is 0.0015625 (1/640). In my PR analysis, the values based on migration patterns of 2011 and 2001 ranged from 0.000664 to 0.006547 showing how different districts (nodes) have different ‘prominence’.
The highest PR value was for Bangalore. I got high PR values for districts which have not seen much in-migration but may have a lot of out-migration (there is no data for out-migration in the Census). To see which are the two different kinds of districts of ‘prominence’: one which is already prominent, and another which needs to be focused on and developed are got by weighted in-degree values. PR values and weighted in-degree values have a correlation coefficient of 0.33.
The figure below shows the districts that are ‘prominent’. In the map on the right, you see the two different kinds of districts - Bihar and West Bengal clusters clearly need more development. The north east cluster and other clusters are silent in this analysis as I have taken the top 10 percent of the PR values and seen the in-degree values for those to show me the difference between the two kinds of districts.
The table below gives the list of top 20 and bottom 20 of the ones shown in the map (top 10 percent of PR values) and bottom 20 of the in-degree values which are not shown in map as well.
| In-degrees – Top 20 | In-degrees – Bottom 20 (in > 90th Percentile) | In-degrees – Bottom 20 (Not in Map) | Pagerank - Top 20 |
|---|---|---|---|
| Bangalore | Gorakhpur | Pashchimi Singhbhum | Bangalore |
| Surat | Gulbarga | Saharsa | Ghaziabad |
| Thane | Mumbai | Nawada | Siwan |
| Ghaziabad | Jaunpur | Pashchim Medinipur | North West |
| Ahmadabad | Dhanbad | Shahdol | Kolhapur |
| Gurgaon | Kolar | Sidhi | Gurgaon |
| North West | Hugli | Bilaspur | Purnia |
| South West | Haora | Bilaspur | North & Middle Andaman |
| Mumbai Suburban | Katihar | Khandwa (East Nimar) | Thane |
| Gautam Buddha Nagar | North 24 Parganas | North & Middle Andaman | Mumbai Suburban |
| North East | Nawada | Hazaribagh | Anantapur |
| Dehradun | Pashchim Medinipur | Kolkata | Sangli |
| Udham Singh Nagar | Sidhi | Pashchim Champaran | Ahmadabad |
| Alwar | North & Middle Andaman | Kamrup | Solapur |
| Jaipur | Kolkata | Darbhanga | Faridabad |
| Valsad | Pashchim Champaran | Purnia | Surat |
| Rangareddy | Darbhanga | Rupnagar | South West |
| Kolhapur | Purnia | Muzaffarpur | Pashchim Medinipur |
| Bharatpur | Muzaffarpur | Nalanda | Kolar |
| Indore | Siwan | Siwan | Neemuch |