Eingestellte DIW Publikationen 1/2 / 1972, S. 79-85
Herwig Birg
This study deals with the problem described as follows: For any given base point in time, a matrix describing certain conditions is known. For a second point in time, only parts of this same matrix are known, namely the sums of the rows and columns. From this information the elements of the complete matrix are to be estimated. The problem can be solved using the RAS procedure or the MODOP procedure. The results from the MODOP procedure depend on the values used to start the iteration process which is to solve the unknown elements of the matrix. Therefore, a means of arriving at the most suitable starting values is offered. This suggested procedure calls for finding the relative deviation between an element of the base matrix and the hypothetical value this same element would have if stochastic independence could be assumed. Then, on the basis of hypotheses on the development of these relative deviations over time, the unknown elements can be estimated from the sums of the rows and columns. The estimation procedure was demonstrated using the example of migration among the states of the Federal Republic of Germany.