Example of characteristics data ready for phylogenetic software

Input-output tables cannot be directly used for the reconstruction of industrial trees. The first step is to transform them. Here we take the example the US intermediate flow data for 1972. Given these data we calculate both the input coefficients and the output coefficients, transform them into binary data (according to the 1% criterion), and further transform 0s to As and 1s to Cs. Then we get 70 characteristics for each of the 23 industries in our large set of industries. Example data are available from the main page in three different formats. You may want to inspect these formats. Here are data for US 1972 in a simple table format (that may be opened by e.g. MS Excel), in easily understandable MEGA1 format (usable for MEGA and MEGA2), and in the more complex MEGA2 format (that helps work in MEGA2).

This dataset is depicted in the table. Here the first 35 columns represent input characteristics and columns 36-70 represent output characteristics. It is possible to recognise several important patterns in the data depicted by the figure. For instance, nearly all inputs to an industry from itself is marked by a C (=1). Thus there is a significant intrasectoral trade. We also see that highly aggregated industries tend to have a larger number of significant deliveries that their less aggregated daughter industries. For instance, 351/2-:IndChem has a large number of significant inputs, while 3522:Drugs has a quite different pattern. The same type of difference is seen by comparing 382-:Machinery and 3825:Computers} as well as by comparing 383-:Electric and 3832:Communic.

The table also demonstrates why it is important to have lots of characteristics data. In principle, it is sufficient to have the the same number of industries and characteristics for the determination of a unique tree. But some characteristics gives no information since they have the same state values for all industries (e.g. characteristics 34 and 35). Other characteristics are only randomly different because of roundoff errors (0.001 gives a C while 0.0099999 gives an A) or statistical errors. Still others are different because of the different relative sizes of the intraindustrial flows. For these and many other reasons we would like to have lots of data (e.g. by using more disaggregate input-output data).