Published on July 30th, 20130
Time-Series Cross-Section Analysis
By Tor G. Jakobsen
The method called time-series cross-section (TSCS) consists of time-series data observed on several units. It is commonly used in studies of conflict and peace, macroeconomics, and international political economy.
In the 1970s and early 80s the most used method within the above mention fields was cross-sectional regression. If a researcher wanted to make inferences about industrialized countries, he was stuck with around 20 units of analysis (as there were a limited number of 1st world countries). Such models were especially sensitive to outliers, like Norway and Luxembourg.
In statistics it is always an advantage to have as many cases as possible. The development of TSCS-modeling allowed for this. A researcher thus combines cross-section with time-series, and the number of cases increases. In TSCS data a unit could for example be a country, while the case would be a country in a given year.
TSCS data have similarities to panel data. In panel data one has a large number of units that are observed over a small number of waves (repeated interviews of the same persons). However, TSCS-data have the opposite structure. There is generally a relative small number of units (e.g., the countries of the world) that are observed over a considerable length of time (which could be months, quarters, or years).
As such, this type of data structure can be labeled as hierarchical, as the time points are a sub-group of countries. But there is also additional structure to the data, like time- and geographical dependency.
When running regressions on time-series data, like TSCS, one encounters several statistical challenges that need attention. This includes heteroskedasticity (that the error terms do not have the same variance), autocorrelation (that the last time-period’s values affect current values, and also spatial autocorrelation (e.g., an economic shock in one country could affect neighboring countries).
Yet, the advantages often outweigh the disadvantages. TSCS analysis increases the number of observation. It gives the researcher the ability to model time and space, which again increases his ability to show causation.
TSCS analysis can be used to model continuous variables like foreign direct investment or economic growth. It is also possible to model binary (0–1) dependent variables, like the onset of civil war. Such models represent a sub-group of TSCS, namely binary time-series cross-section (BTSCS). This method resembles survival analysis, and allows for multiple failures per unit.
Beck, Nathaniel (2008) “Time-Series Cross-Section Methods” in Janet M. Box-Steffensmeier, Henry E. Brady, & David Collier (eds.) The Oxford Handbook of Political Methodology. Oxford: Oxford University Press: 475–493.
Beck, Nathaniel, Jonathan N. Katz, & Richard Tucker (1998) “Taking Time Seriously: Time-Series-Cross-Section Analysis with a Binary Dependent Variable” American Journal of Political Science, 42(4): 1260–1288.