Data errors in small data sets can determine empirical findings

This paper provides an example of a model that yields widely divergent estimates when different stock market indexes are used to calculate two independent variables in Romer's 1990] model. Her model sought to explain consumer durable good production before the Great Crash (31 observations). She used the Cowles Commissions Series P Stock Price Index to calculate two independent variables. However, when this paper uses the S&P Index to calculate these variables, its estimates completely contradict Romer's findings. It discovered that one incorrect monthly observation in the S&P Index is responsible for this difference. It also found that robustness techniques serve to limit the impact of the errant observation, illustrating the importance of using robustness techniques in small data sets.