Dear all,
I am trying to implement a GLS regression based on the information given in an article, which would be this one: Triche, J. H., & Walden, E. (2018). The use of impression management strategies to manage stock market reactions to IT failures. Journal of the Association for Information Systems, 19(4), 1.
They perform (pooled) OLS and GLS regression. They specify their model on page 344, show results of the OLS regression on page 345 and for the GLS regression on page 346.
Their OLS model does not seem to rely on cluster-robust robust standard errors. Still the main issue with their data is non-indepenence across observations as some companies appear multiple times in the sample. I have trouble to figure out how to recreate the GLS regression they use in stata.
They share these thoughts on their procedure:
"In order to correct for potential correlations between the samples, we use generalized least square (GLS) regression analysis. GLS is preferred over ordinary least square (OLS) regression when the variances of the observation are heteroscedastic (i.e., differing variances), or when there is correlation between the samples (Greene, 2003). Although the data do not lend themselves to heteroscedasticity, the data might not be independently distributed. Since the data sample consists of companies that experienced multiple failures over the period of 2005-2012, and there may have been multiple announcements about the same IT failure, the samples may be nonindependent. GLS regression corrects for this potential issue."
I have taken a look at the xtgls-command, yet it does not seem to fit the context as its for panel-data.
Any thoughts on it would be welcome.
Kind regards
I am trying to implement a GLS regression based on the information given in an article, which would be this one: Triche, J. H., & Walden, E. (2018). The use of impression management strategies to manage stock market reactions to IT failures. Journal of the Association for Information Systems, 19(4), 1.
They perform (pooled) OLS and GLS regression. They specify their model on page 344, show results of the OLS regression on page 345 and for the GLS regression on page 346.
Their OLS model does not seem to rely on cluster-robust robust standard errors. Still the main issue with their data is non-indepenence across observations as some companies appear multiple times in the sample. I have trouble to figure out how to recreate the GLS regression they use in stata.
They share these thoughts on their procedure:
"In order to correct for potential correlations between the samples, we use generalized least square (GLS) regression analysis. GLS is preferred over ordinary least square (OLS) regression when the variances of the observation are heteroscedastic (i.e., differing variances), or when there is correlation between the samples (Greene, 2003). Although the data do not lend themselves to heteroscedasticity, the data might not be independently distributed. Since the data sample consists of companies that experienced multiple failures over the period of 2005-2012, and there may have been multiple announcements about the same IT failure, the samples may be nonindependent. GLS regression corrects for this potential issue."
I have taken a look at the xtgls-command, yet it does not seem to fit the context as its for panel-data.
Any thoughts on it would be welcome.
Kind regards
