Political Science Computing: A Review of Trends in Computer Evolution and Political Science Research

ABSTRACT

The increase in computing power in the last four decades has had an enormous impact upon the way political scientists conduct research. In this article, we review the development of computing resources and the lifting of computational limitations of the use of political methodologies. Using descriptive analysis, we demonstrate that Moore's Law, or the doubling of power in computing hardware every 18 months, is closely associated with the adoption of advanced statistical methods and database availability in political science research. Our findings are surely not startling to researchers, but our analyses do highlight and provide important details as to the evolution of the profession over the last few decades. We aim to provide hints as to the future direction of political science research as computational capabilities continue to grow.

KEYWORDS:

• Electronic library database
• Moore's law
• multivariate models
• political methodology

Notes

1. Replication data are available at the JITP Dataverse referenced by Euel W. Elliott, Karl Ho, Jennifer S. Holmes, February 12, 2009, “Replication data for: Political Science Computing: A Review of Trends in Computer Evolution and Political Science Research,” hdl:1902.1/12360. Available at: http://hdl.handle.net/1902.1/12360.

2. For instance, number of articles published in APSR in 1970s, 1980s, and 1990s are 465, 438, and 427, respectively.

3. We include formal model articles in the theory category since these studies do not usually involve statistical methods. Given the recent boom of formal studies, this category actually sees an increase in proportion compared to the other categories.

4. When considering the use of such a scale, we weight between the ordinality (real order of scale) and generality (practical use), acknowledging that classification of methods can be fuzzy. In fact, certain studies may use a third category method (e.g., probit) with combined use of other more computationally intensive method which should fall into the highest category. The general rule is if more than one method is involved, it is coded using the more computationally intensive method. Otherwise, in a case not clearly stated, we adopt the more conservative approach in giving a lower score.

5. We have considered using other measures such as number of transistors and clock frequency, but MIPS proves to be a more valid measure that provides the speed and capacity to accommodate computational procedures actually involved in statistical methods.

6. This could be attributed to a surge of formal theory articles in the 1990s that attenuated the growth of the multivariate studies.