Hello,
I am trying to estimate the effect of certain personality characteristics of CEO's on, for instance, firm value. I have a (unbalanced) panel data of firm-year observations, and for each firm-year information on the personality characteristics of the CEO.
Personality characteristics are, however, relatively stable over time (which you would expect from the literature). On a scale of 1-7, for instance, a CEO can have a value for extraversion in 2005 of 4,5, in 2006 around 4,7, and in 2007 around 4.4. The main difference therefore occur between CEO's,, rather than within CEOs. Many papers in this literature just average the personality value per CEO. So if a CEO is in power in 2005, 2006, and 2007, they take the average of 4,5/4,7 and 4.4. The disadvantage of this is that they almost always then use random effects models, because the values would drop out in a FE approach.
I would prefer not to dismiss/aggregate values like this per CEO. As a possible solution, I investigated the xthybrid option in Stata, and combined it with information from Schunck & Perales (2017, Stata Journal, on Within-and between clsuter effects in GLM models).
Is it correct to do then the following:
(a) Run an Xthybrid model. For instance:
xthybrid firm_value ceo_char1 ceo_char2 ceo_char3 firm_size, test
(b) Using the test statistics, investigate for which variables the difference between the within and between estimates are significantly different, and thus were the standard RE assumption would be violated.
Assume that the test indicates it is violated for ceo_char1
(c) Estimate a random effect models where for ceo_char 1, we include BOTH the average ceo_char1 for the unit , and the actual value for ceo_char1. In that sense, we include both the within and between effect for that specific variable.
Just wondering if that seems to make sense. Thanks!
I am trying to estimate the effect of certain personality characteristics of CEO's on, for instance, firm value. I have a (unbalanced) panel data of firm-year observations, and for each firm-year information on the personality characteristics of the CEO.
Personality characteristics are, however, relatively stable over time (which you would expect from the literature). On a scale of 1-7, for instance, a CEO can have a value for extraversion in 2005 of 4,5, in 2006 around 4,7, and in 2007 around 4.4. The main difference therefore occur between CEO's,, rather than within CEOs. Many papers in this literature just average the personality value per CEO. So if a CEO is in power in 2005, 2006, and 2007, they take the average of 4,5/4,7 and 4.4. The disadvantage of this is that they almost always then use random effects models, because the values would drop out in a FE approach.
I would prefer not to dismiss/aggregate values like this per CEO. As a possible solution, I investigated the xthybrid option in Stata, and combined it with information from Schunck & Perales (2017, Stata Journal, on Within-and between clsuter effects in GLM models).
Is it correct to do then the following:
(a) Run an Xthybrid model. For instance:
xthybrid firm_value ceo_char1 ceo_char2 ceo_char3 firm_size, test
(b) Using the test statistics, investigate for which variables the difference between the within and between estimates are significantly different, and thus were the standard RE assumption would be violated.
Assume that the test indicates it is violated for ceo_char1
(c) Estimate a random effect models where for ceo_char 1, we include BOTH the average ceo_char1 for the unit , and the actual value for ceo_char1. In that sense, we include both the within and between effect for that specific variable.
Just wondering if that seems to make sense. Thanks!
