"The paper proposes a memory saving decomposition of the design matrix to facilitate fixed effects estimation of the three-way error component model with high numbers of observations and groups. A common way to estimate such a model is to include two of the effects as dummy variables and to sweep out the other effect by the fixed effects trans-formation. If the number of groups is high, the design matrix that includes the dummy variables can be prohibitively large for computer packages that need to store the whole data set in memory. The decomposition of the design matrix proposed here shows a way of how to create the cross-product matrices for the least squares normal equations without explicitly creating the dummy variables for the group effects. As the cross-product matrices are of much lower dimension than the design matrix, this procedure reduces the computer memory required considerably. For example, a model computation shows that in a linked employer-employee data set with 20 million observations and 10 thousand firms, the memory requirement drops from 800 gigabytes to 1 gigabyte. The method is implemented in Stata by making use of the new Mata environment available in Stata 9.0. Besides implementing the memory-saving estimation method, the program also takes care of identification issues (grouping algorithm) and provides useful summary statistics. The paper presents the Stata program and comments its output." (author's abstract, IAB-Doku) ((en))
If you experience problems downloading a file, check if you have the
proper application to
view it first. Information about this may be contained
in the File-Format links below. In case of further problems read
the IDEAS help
page. Note that these files are not on the IDEAS
site. Please be patient as the files may be large.
Publisher Info
Paper provided by Institut für Arbeitsmarkt- und Berufsforschung (IAB), Nürnberg [Institute for Employment Research, Nuremberg, Germany] in its series FDZ Methodenreport with number
200603_en.
Find related papers by JEL classification: C23 - Mathematical and Quantitative Methods - - Single Equation Models; Single Variables - - - Models with Panel Data C63 - Mathematical and Quantitative Methods - - Mathematical Methods and Programming - - - Computational Techniques